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 /* Data structures. */
54 static struct lock_class_key rcu_root_class
;
56 #define RCU_STATE_INITIALIZER(name) { \
57 .level = { &name.node[0] }, \
59 NUM_RCU_LVL_0, /* root of hierarchy. */ \
62 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
64 .signaled = RCU_GP_IDLE, \
67 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
68 .orphan_cbs_list = NULL, \
69 .orphan_cbs_tail = &name.orphan_cbs_list, \
71 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
73 .n_force_qs_ngp = 0, \
76 struct rcu_state rcu_sched_state
= RCU_STATE_INITIALIZER(rcu_sched_state
);
77 DEFINE_PER_CPU(struct rcu_data
, rcu_sched_data
);
79 struct rcu_state rcu_bh_state
= RCU_STATE_INITIALIZER(rcu_bh_state
);
80 DEFINE_PER_CPU(struct rcu_data
, rcu_bh_data
);
84 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
85 * permit this function to be invoked without holding the root rcu_node
86 * structure's ->lock, but of course results can be subject to change.
88 static int rcu_gp_in_progress(struct rcu_state
*rsp
)
90 return ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
);
94 * Note a quiescent state. Because we do not need to know
95 * how many quiescent states passed, just if there was at least
96 * one since the start of the grace period, this just sets a flag.
98 void rcu_sched_qs(int cpu
)
100 struct rcu_data
*rdp
;
102 rdp
= &per_cpu(rcu_sched_data
, cpu
);
103 rdp
->passed_quiesc_completed
= rdp
->completed
;
105 rdp
->passed_quiesc
= 1;
106 rcu_preempt_note_context_switch(cpu
);
109 void rcu_bh_qs(int cpu
)
111 struct rcu_data
*rdp
;
113 rdp
= &per_cpu(rcu_bh_data
, cpu
);
114 rdp
->passed_quiesc_completed
= rdp
->completed
;
116 rdp
->passed_quiesc
= 1;
120 DEFINE_PER_CPU(struct rcu_dynticks
, rcu_dynticks
) = {
121 .dynticks_nesting
= 1,
124 #endif /* #ifdef CONFIG_NO_HZ */
126 static int blimit
= 10; /* Maximum callbacks per softirq. */
127 static int qhimark
= 10000; /* If this many pending, ignore blimit. */
128 static int qlowmark
= 100; /* Once only this many pending, use blimit. */
130 module_param(blimit
, int, 0);
131 module_param(qhimark
, int, 0);
132 module_param(qlowmark
, int, 0);
134 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
);
135 static int rcu_pending(int cpu
);
138 * Return the number of RCU-sched batches processed thus far for debug & stats.
140 long rcu_batches_completed_sched(void)
142 return rcu_sched_state
.completed
;
144 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched
);
147 * Return the number of RCU BH batches processed thus far for debug & stats.
149 long rcu_batches_completed_bh(void)
151 return rcu_bh_state
.completed
;
153 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
156 * Does the CPU have callbacks ready to be invoked?
159 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
161 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
165 * Does the current CPU require a yet-as-unscheduled grace period?
168 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
170 return *rdp
->nxttail
[RCU_DONE_TAIL
] && !rcu_gp_in_progress(rsp
);
174 * Return the root node of the specified rcu_state structure.
176 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
178 return &rsp
->node
[0];
182 * Record the specified "completed" value, which is later used to validate
183 * dynticks counter manipulations and CPU-offline checks. Specify
184 * "rsp->completed - 1" to unconditionally invalidate any future dynticks
185 * manipulations and CPU-offline checks. Such invalidation is useful at
186 * the beginning of a grace period.
188 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
190 rsp
->dynticks_completed
= comp
;
196 * Recall the previously recorded value of the completion for dynticks.
198 static long dyntick_recall_completed(struct rcu_state
*rsp
)
200 return rsp
->dynticks_completed
;
204 * If the specified CPU is offline, tell the caller that it is in
205 * a quiescent state. Otherwise, whack it with a reschedule IPI.
206 * Grace periods can end up waiting on an offline CPU when that
207 * CPU is in the process of coming online -- it will be added to the
208 * rcu_node bitmasks before it actually makes it online. The same thing
209 * can happen while a CPU is in the process of coming online. Because this
210 * race is quite rare, we check for it after detecting that the grace
211 * period has been delayed rather than checking each and every CPU
212 * each and every time we start a new grace period.
214 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
217 * If the CPU is offline, it is in a quiescent state. We can
218 * trust its state not to change because interrupts are disabled.
220 if (cpu_is_offline(rdp
->cpu
)) {
225 /* If preemptable RCU, no point in sending reschedule IPI. */
226 if (rdp
->preemptable
)
229 /* The CPU is online, so send it a reschedule IPI. */
230 if (rdp
->cpu
!= smp_processor_id())
231 smp_send_reschedule(rdp
->cpu
);
238 #endif /* #ifdef CONFIG_SMP */
243 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
245 * Enter nohz mode, in other words, -leave- the mode in which RCU
246 * read-side critical sections can occur. (Though RCU read-side
247 * critical sections can occur in irq handlers in nohz mode, a possibility
248 * handled by rcu_irq_enter() and rcu_irq_exit()).
250 void rcu_enter_nohz(void)
253 struct rcu_dynticks
*rdtp
;
255 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
256 local_irq_save(flags
);
257 rdtp
= &__get_cpu_var(rcu_dynticks
);
259 rdtp
->dynticks_nesting
--;
260 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
261 local_irq_restore(flags
);
265 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
267 * Exit nohz mode, in other words, -enter- the mode in which RCU
268 * read-side critical sections normally occur.
270 void rcu_exit_nohz(void)
273 struct rcu_dynticks
*rdtp
;
275 local_irq_save(flags
);
276 rdtp
= &__get_cpu_var(rcu_dynticks
);
278 rdtp
->dynticks_nesting
++;
279 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
280 local_irq_restore(flags
);
281 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
285 * rcu_nmi_enter - inform RCU of entry to NMI context
287 * If the CPU was idle with dynamic ticks active, and there is no
288 * irq handler running, this updates rdtp->dynticks_nmi to let the
289 * RCU grace-period handling know that the CPU is active.
291 void rcu_nmi_enter(void)
293 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
295 if (rdtp
->dynticks
& 0x1)
297 rdtp
->dynticks_nmi
++;
298 WARN_ON_ONCE(!(rdtp
->dynticks_nmi
& 0x1));
299 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
303 * rcu_nmi_exit - inform RCU of exit from NMI context
305 * If the CPU was idle with dynamic ticks active, and there is no
306 * irq handler running, this updates rdtp->dynticks_nmi to let the
307 * RCU grace-period handling know that the CPU is no longer active.
309 void rcu_nmi_exit(void)
311 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
313 if (rdtp
->dynticks
& 0x1)
315 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
316 rdtp
->dynticks_nmi
++;
317 WARN_ON_ONCE(rdtp
->dynticks_nmi
& 0x1);
321 * rcu_irq_enter - inform RCU of entry to hard irq context
323 * If the CPU was idle with dynamic ticks active, this updates the
324 * rdtp->dynticks to let the RCU handling know that the CPU is active.
326 void rcu_irq_enter(void)
328 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
330 if (rdtp
->dynticks_nesting
++)
333 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
334 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
338 * rcu_irq_exit - inform RCU of exit from hard irq context
340 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
341 * to put let the RCU handling be aware that the CPU is going back to idle
344 void rcu_irq_exit(void)
346 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
348 if (--rdtp
->dynticks_nesting
)
350 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
352 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
354 /* If the interrupt queued a callback, get out of dyntick mode. */
355 if (__get_cpu_var(rcu_sched_data
).nxtlist
||
356 __get_cpu_var(rcu_bh_data
).nxtlist
)
363 * Snapshot the specified CPU's dynticks counter so that we can later
364 * credit them with an implicit quiescent state. Return 1 if this CPU
365 * is in dynticks idle mode, which is an extended quiescent state.
367 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
373 snap
= rdp
->dynticks
->dynticks
;
374 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
375 smp_mb(); /* Order sampling of snap with end of grace period. */
376 rdp
->dynticks_snap
= snap
;
377 rdp
->dynticks_nmi_snap
= snap_nmi
;
378 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
385 * Return true if the specified CPU has passed through a quiescent
386 * state by virtue of being in or having passed through an dynticks
387 * idle state since the last call to dyntick_save_progress_counter()
390 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
397 curr
= rdp
->dynticks
->dynticks
;
398 snap
= rdp
->dynticks_snap
;
399 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
400 snap_nmi
= rdp
->dynticks_nmi_snap
;
401 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
404 * If the CPU passed through or entered a dynticks idle phase with
405 * no active irq/NMI handlers, then we can safely pretend that the CPU
406 * already acknowledged the request to pass through a quiescent
407 * state. Either way, that CPU cannot possibly be in an RCU
408 * read-side critical section that started before the beginning
409 * of the current RCU grace period.
411 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
412 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
417 /* Go check for the CPU being offline. */
418 return rcu_implicit_offline_qs(rdp
);
421 #endif /* #ifdef CONFIG_SMP */
423 #else /* #ifdef CONFIG_NO_HZ */
427 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
432 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
434 return rcu_implicit_offline_qs(rdp
);
437 #endif /* #ifdef CONFIG_SMP */
439 #endif /* #else #ifdef CONFIG_NO_HZ */
441 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
443 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
445 rsp
->gp_start
= jiffies
;
446 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
449 static void print_other_cpu_stall(struct rcu_state
*rsp
)
454 struct rcu_node
*rnp
= rcu_get_root(rsp
);
456 /* Only let one CPU complain about others per time interval. */
458 spin_lock_irqsave(&rnp
->lock
, flags
);
459 delta
= jiffies
- rsp
->jiffies_stall
;
460 if (delta
< RCU_STALL_RAT_DELAY
|| !rcu_gp_in_progress(rsp
)) {
461 spin_unlock_irqrestore(&rnp
->lock
, flags
);
464 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
467 * Now rat on any tasks that got kicked up to the root rcu_node
468 * due to CPU offlining.
470 rcu_print_task_stall(rnp
);
471 spin_unlock_irqrestore(&rnp
->lock
, flags
);
473 /* OK, time to rat on our buddy... */
475 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
476 rcu_for_each_leaf_node(rsp
, rnp
) {
477 rcu_print_task_stall(rnp
);
478 if (rnp
->qsmask
== 0)
480 for (cpu
= 0; cpu
<= rnp
->grphi
- rnp
->grplo
; cpu
++)
481 if (rnp
->qsmask
& (1UL << cpu
))
482 printk(" %d", rnp
->grplo
+ cpu
);
484 printk(" (detected by %d, t=%ld jiffies)\n",
485 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
486 trigger_all_cpu_backtrace();
488 force_quiescent_state(rsp
, 0); /* Kick them all. */
491 static void print_cpu_stall(struct rcu_state
*rsp
)
494 struct rcu_node
*rnp
= rcu_get_root(rsp
);
496 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
497 smp_processor_id(), jiffies
- rsp
->gp_start
);
498 trigger_all_cpu_backtrace();
500 spin_lock_irqsave(&rnp
->lock
, flags
);
501 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
503 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
504 spin_unlock_irqrestore(&rnp
->lock
, flags
);
506 set_need_resched(); /* kick ourselves to get things going. */
509 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
512 struct rcu_node
*rnp
;
514 delta
= jiffies
- rsp
->jiffies_stall
;
516 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
518 /* We haven't checked in, so go dump stack. */
519 print_cpu_stall(rsp
);
521 } else if (rcu_gp_in_progress(rsp
) && delta
>= RCU_STALL_RAT_DELAY
) {
523 /* They had two time units to dump stack, so complain. */
524 print_other_cpu_stall(rsp
);
528 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
530 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
534 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
538 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
541 * Update CPU-local rcu_data state to record the newly noticed grace period.
542 * This is used both when we started the grace period and when we notice
543 * that someone else started the grace period. The caller must hold the
544 * ->lock of the leaf rcu_node structure corresponding to the current CPU,
545 * and must have irqs disabled.
547 static void __note_new_gpnum(struct rcu_state
*rsp
, struct rcu_node
*rnp
, struct rcu_data
*rdp
)
549 if (rdp
->gpnum
!= rnp
->gpnum
) {
551 rdp
->passed_quiesc
= 0;
552 rdp
->gpnum
= rnp
->gpnum
;
556 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
559 struct rcu_node
*rnp
;
561 local_irq_save(flags
);
563 if (rdp
->gpnum
== ACCESS_ONCE(rnp
->gpnum
) || /* outside lock. */
564 !spin_trylock(&rnp
->lock
)) { /* irqs already off, retry later. */
565 local_irq_restore(flags
);
568 __note_new_gpnum(rsp
, rnp
, rdp
);
569 spin_unlock_irqrestore(&rnp
->lock
, flags
);
573 * Did someone else start a new RCU grace period start since we last
574 * checked? Update local state appropriately if so. Must be called
575 * on the CPU corresponding to rdp.
578 check_for_new_grace_period(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
583 local_irq_save(flags
);
584 if (rdp
->gpnum
!= rsp
->gpnum
) {
585 note_new_gpnum(rsp
, rdp
);
588 local_irq_restore(flags
);
593 * Advance this CPU's callbacks, but only if the current grace period
594 * has ended. This may be called only from the CPU to whom the rdp
595 * belongs. In addition, the corresponding leaf rcu_node structure's
596 * ->lock must be held by the caller, with irqs disabled.
599 __rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_node
*rnp
, struct rcu_data
*rdp
)
601 /* Did another grace period end? */
602 if (rdp
->completed
!= rnp
->completed
) {
604 /* Advance callbacks. No harm if list empty. */
605 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
606 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
607 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
609 /* Remember that we saw this grace-period completion. */
610 rdp
->completed
= rnp
->completed
;
615 * Advance this CPU's callbacks, but only if the current grace period
616 * has ended. This may be called only from the CPU to whom the rdp
620 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
623 struct rcu_node
*rnp
;
625 local_irq_save(flags
);
627 if (rdp
->completed
== ACCESS_ONCE(rnp
->completed
) || /* outside lock. */
628 !spin_trylock(&rnp
->lock
)) { /* irqs already off, retry later. */
629 local_irq_restore(flags
);
632 __rcu_process_gp_end(rsp
, rnp
, rdp
);
633 spin_unlock_irqrestore(&rnp
->lock
, flags
);
637 * Do per-CPU grace-period initialization for running CPU. The caller
638 * must hold the lock of the leaf rcu_node structure corresponding to
642 rcu_start_gp_per_cpu(struct rcu_state
*rsp
, struct rcu_node
*rnp
, struct rcu_data
*rdp
)
644 /* Prior grace period ended, so advance callbacks for current CPU. */
645 __rcu_process_gp_end(rsp
, rnp
, rdp
);
648 * Because this CPU just now started the new grace period, we know
649 * that all of its callbacks will be covered by this upcoming grace
650 * period, even the ones that were registered arbitrarily recently.
651 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
653 * Other CPUs cannot be sure exactly when the grace period started.
654 * Therefore, their recently registered callbacks must pass through
655 * an additional RCU_NEXT_READY stage, so that they will be handled
656 * by the next RCU grace period.
658 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
659 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
661 /* Set state so that this CPU will detect the next quiescent state. */
662 __note_new_gpnum(rsp
, rnp
, rdp
);
666 * Start a new RCU grace period if warranted, re-initializing the hierarchy
667 * in preparation for detecting the next grace period. The caller must hold
668 * the root node's ->lock, which is released before return. Hard irqs must
672 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
673 __releases(rcu_get_root(rsp
)->lock
)
675 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
676 struct rcu_node
*rnp
= rcu_get_root(rsp
);
678 if (!cpu_needs_another_gp(rsp
, rdp
)) {
679 spin_unlock_irqrestore(&rnp
->lock
, flags
);
683 /* Advance to a new grace period and initialize state. */
685 WARN_ON_ONCE(rsp
->signaled
== RCU_GP_INIT
);
686 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
687 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
688 record_gp_stall_check_time(rsp
);
689 dyntick_record_completed(rsp
, rsp
->completed
- 1);
691 /* Special-case the common single-level case. */
692 if (NUM_RCU_NODES
== 1) {
693 rcu_preempt_check_blocked_tasks(rnp
);
694 rnp
->qsmask
= rnp
->qsmaskinit
;
695 rnp
->gpnum
= rsp
->gpnum
;
696 rnp
->completed
= rsp
->completed
;
697 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
698 rcu_start_gp_per_cpu(rsp
, rnp
, rdp
);
699 spin_unlock_irqrestore(&rnp
->lock
, flags
);
703 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
706 /* Exclude any concurrent CPU-hotplug operations. */
707 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
710 * Set the quiescent-state-needed bits in all the rcu_node
711 * structures for all currently online CPUs in breadth-first
712 * order, starting from the root rcu_node structure. This
713 * operation relies on the layout of the hierarchy within the
714 * rsp->node[] array. Note that other CPUs will access only
715 * the leaves of the hierarchy, which still indicate that no
716 * grace period is in progress, at least until the corresponding
717 * leaf node has been initialized. In addition, we have excluded
718 * CPU-hotplug operations.
720 * Note that the grace period cannot complete until we finish
721 * the initialization process, as there will be at least one
722 * qsmask bit set in the root node until that time, namely the
723 * one corresponding to this CPU, due to the fact that we have
726 rcu_for_each_node_breadth_first(rsp
, rnp
) {
727 spin_lock(&rnp
->lock
); /* irqs already disabled. */
728 rcu_preempt_check_blocked_tasks(rnp
);
729 rnp
->qsmask
= rnp
->qsmaskinit
;
730 rnp
->gpnum
= rsp
->gpnum
;
731 rnp
->completed
= rsp
->completed
;
732 if (rnp
== rdp
->mynode
)
733 rcu_start_gp_per_cpu(rsp
, rnp
, rdp
);
734 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
737 rnp
= rcu_get_root(rsp
);
738 spin_lock(&rnp
->lock
); /* irqs already disabled. */
739 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
740 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
741 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
745 * Clean up after the prior grace period and let rcu_start_gp() start up
746 * the next grace period if one is needed. Note that the caller must
747 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
749 static void cpu_quiet_msk_finish(struct rcu_state
*rsp
, unsigned long flags
)
750 __releases(rcu_get_root(rsp
)->lock
)
752 WARN_ON_ONCE(!rcu_gp_in_progress(rsp
));
753 rsp
->completed
= rsp
->gpnum
;
754 rsp
->signaled
= RCU_GP_IDLE
;
755 rcu_start_gp(rsp
, flags
); /* releases root node's rnp->lock. */
759 * Similar to cpu_quiet(), for which it is a helper function. Allows
760 * a group of CPUs to be quieted at one go, though all the CPUs in the
761 * group must be represented by the same leaf rcu_node structure.
762 * That structure's lock must be held upon entry, and it is released
766 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
768 __releases(rnp
->lock
)
770 struct rcu_node
*rnp_c
;
772 /* Walk up the rcu_node hierarchy. */
774 if (!(rnp
->qsmask
& mask
)) {
776 /* Our bit has already been cleared, so done. */
777 spin_unlock_irqrestore(&rnp
->lock
, flags
);
780 rnp
->qsmask
&= ~mask
;
781 if (rnp
->qsmask
!= 0 || rcu_preempted_readers(rnp
)) {
783 /* Other bits still set at this level, so done. */
784 spin_unlock_irqrestore(&rnp
->lock
, flags
);
788 if (rnp
->parent
== NULL
) {
790 /* No more levels. Exit loop holding root lock. */
794 spin_unlock_irqrestore(&rnp
->lock
, flags
);
797 spin_lock_irqsave(&rnp
->lock
, flags
);
798 WARN_ON_ONCE(rnp_c
->qsmask
);
802 * Get here if we are the last CPU to pass through a quiescent
803 * state for this grace period. Invoke cpu_quiet_msk_finish()
804 * to clean up and start the next grace period if one is needed.
806 cpu_quiet_msk_finish(rsp
, flags
); /* releases rnp->lock. */
810 * Record a quiescent state for the specified CPU, which must either be
811 * the current CPU. The lastcomp argument is used to make sure we are
812 * still in the grace period of interest. We don't want to end the current
813 * grace period based on quiescent states detected in an earlier grace
817 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
821 struct rcu_node
*rnp
;
824 spin_lock_irqsave(&rnp
->lock
, flags
);
825 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
828 * Someone beat us to it for this grace period, so leave.
829 * The race with GP start is resolved by the fact that we
830 * hold the leaf rcu_node lock, so that the per-CPU bits
831 * cannot yet be initialized -- so we would simply find our
832 * CPU's bit already cleared in cpu_quiet_msk() if this race
835 rdp
->passed_quiesc
= 0; /* try again later! */
836 spin_unlock_irqrestore(&rnp
->lock
, flags
);
840 if ((rnp
->qsmask
& mask
) == 0) {
841 spin_unlock_irqrestore(&rnp
->lock
, flags
);
846 * This GP can't end until cpu checks in, so all of our
847 * callbacks can be processed during the next GP.
849 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
851 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
856 * Check to see if there is a new grace period of which this CPU
857 * is not yet aware, and if so, set up local rcu_data state for it.
858 * Otherwise, see if this CPU has just passed through its first
859 * quiescent state for this grace period, and record that fact if so.
862 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
864 /* If there is now a new grace period, record and return. */
865 if (check_for_new_grace_period(rsp
, rdp
))
869 * Does this CPU still need to do its part for current grace period?
870 * If no, return and let the other CPUs do their part as well.
872 if (!rdp
->qs_pending
)
876 * Was there a quiescent state since the beginning of the grace
877 * period? If no, then exit and wait for the next call.
879 if (!rdp
->passed_quiesc
)
882 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
883 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
886 #ifdef CONFIG_HOTPLUG_CPU
889 * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
890 * specified flavor of RCU. The callbacks will be adopted by the next
891 * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
892 * comes first. Because this is invoked from the CPU_DYING notifier,
893 * irqs are already disabled.
895 static void rcu_send_cbs_to_orphanage(struct rcu_state
*rsp
)
898 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
900 if (rdp
->nxtlist
== NULL
)
901 return; /* irqs disabled, so comparison is stable. */
902 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
903 *rsp
->orphan_cbs_tail
= rdp
->nxtlist
;
904 rsp
->orphan_cbs_tail
= rdp
->nxttail
[RCU_NEXT_TAIL
];
906 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
907 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
908 rsp
->orphan_qlen
+= rdp
->qlen
;
910 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
914 * Adopt previously orphaned RCU callbacks.
916 static void rcu_adopt_orphan_cbs(struct rcu_state
*rsp
)
919 struct rcu_data
*rdp
;
921 spin_lock_irqsave(&rsp
->onofflock
, flags
);
922 rdp
= rsp
->rda
[smp_processor_id()];
923 if (rsp
->orphan_cbs_list
== NULL
) {
924 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
927 *rdp
->nxttail
[RCU_NEXT_TAIL
] = rsp
->orphan_cbs_list
;
928 rdp
->nxttail
[RCU_NEXT_TAIL
] = rsp
->orphan_cbs_tail
;
929 rdp
->qlen
+= rsp
->orphan_qlen
;
930 rsp
->orphan_cbs_list
= NULL
;
931 rsp
->orphan_cbs_tail
= &rsp
->orphan_cbs_list
;
932 rsp
->orphan_qlen
= 0;
933 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
937 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
938 * and move all callbacks from the outgoing CPU to the current one.
940 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
945 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
946 struct rcu_node
*rnp
;
948 /* Exclude any attempts to start a new grace period. */
949 spin_lock_irqsave(&rsp
->onofflock
, flags
);
951 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
952 rnp
= rdp
->mynode
; /* this is the outgoing CPU's rnp. */
953 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
955 spin_lock(&rnp
->lock
); /* irqs already disabled. */
956 rnp
->qsmaskinit
&= ~mask
;
957 if (rnp
->qsmaskinit
!= 0) {
958 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
963 * If there was a task blocking the current grace period,
964 * and if all CPUs have checked in, we need to propagate
965 * the quiescent state up the rcu_node hierarchy. But that
966 * is inconvenient at the moment due to deadlock issues if
967 * this should end the current grace period. So set the
968 * offlined CPU's bit in ->qsmask in order to force the
969 * next force_quiescent_state() invocation to clean up this
970 * mess in a deadlock-free manner.
972 if (rcu_preempt_offline_tasks(rsp
, rnp
, rdp
) && !rnp
->qsmask
)
976 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
978 } while (rnp
!= NULL
);
979 lastcomp
= rsp
->completed
;
981 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
983 rcu_adopt_orphan_cbs(rsp
);
987 * Remove the specified CPU from the RCU hierarchy and move any pending
988 * callbacks that it might have to the current CPU. This code assumes
989 * that at least one CPU in the system will remain running at all times.
990 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
992 static void rcu_offline_cpu(int cpu
)
994 __rcu_offline_cpu(cpu
, &rcu_sched_state
);
995 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
996 rcu_preempt_offline_cpu(cpu
);
999 #else /* #ifdef CONFIG_HOTPLUG_CPU */
1001 static void rcu_send_cbs_to_orphanage(struct rcu_state
*rsp
)
1005 static void rcu_adopt_orphan_cbs(struct rcu_state
*rsp
)
1009 static void rcu_offline_cpu(int cpu
)
1013 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1016 * Invoke any RCU callbacks that have made it to the end of their grace
1017 * period. Thottle as specified by rdp->blimit.
1019 static void rcu_do_batch(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1021 unsigned long flags
;
1022 struct rcu_head
*next
, *list
, **tail
;
1025 /* If no callbacks are ready, just return.*/
1026 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
1030 * Extract the list of ready callbacks, disabling to prevent
1031 * races with call_rcu() from interrupt handlers.
1033 local_irq_save(flags
);
1034 list
= rdp
->nxtlist
;
1035 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
1036 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
1037 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
1038 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
1039 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
1040 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
1041 local_irq_restore(flags
);
1043 /* Invoke callbacks. */
1050 if (++count
>= rdp
->blimit
)
1054 local_irq_save(flags
);
1056 /* Update count, and requeue any remaining callbacks. */
1059 *tail
= rdp
->nxtlist
;
1060 rdp
->nxtlist
= list
;
1061 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
1062 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
1063 rdp
->nxttail
[count
] = tail
;
1068 /* Reinstate batch limit if we have worked down the excess. */
1069 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
1070 rdp
->blimit
= blimit
;
1072 /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1073 if (rdp
->qlen
== 0 && rdp
->qlen_last_fqs_check
!= 0) {
1074 rdp
->qlen_last_fqs_check
= 0;
1075 rdp
->n_force_qs_snap
= rsp
->n_force_qs
;
1076 } else if (rdp
->qlen
< rdp
->qlen_last_fqs_check
- qhimark
)
1077 rdp
->qlen_last_fqs_check
= rdp
->qlen
;
1079 local_irq_restore(flags
);
1081 /* Re-raise the RCU softirq if there are callbacks remaining. */
1082 if (cpu_has_callbacks_ready_to_invoke(rdp
))
1083 raise_softirq(RCU_SOFTIRQ
);
1087 * Check to see if this CPU is in a non-context-switch quiescent state
1088 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1089 * Also schedule the RCU softirq handler.
1091 * This function must be called with hardirqs disabled. It is normally
1092 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1093 * false, there is no point in invoking rcu_check_callbacks().
1095 void rcu_check_callbacks(int cpu
, int user
)
1097 if (!rcu_pending(cpu
))
1098 return; /* if nothing for RCU to do. */
1100 (idle_cpu(cpu
) && rcu_scheduler_active
&&
1101 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
1104 * Get here if this CPU took its interrupt from user
1105 * mode or from the idle loop, and if this is not a
1106 * nested interrupt. In this case, the CPU is in
1107 * a quiescent state, so note it.
1109 * No memory barrier is required here because both
1110 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1111 * variables that other CPUs neither access nor modify,
1112 * at least not while the corresponding CPU is online.
1118 } else if (!in_softirq()) {
1121 * Get here if this CPU did not take its interrupt from
1122 * softirq, in other words, if it is not interrupting
1123 * a rcu_bh read-side critical section. This is an _bh
1124 * critical section, so note it.
1129 rcu_preempt_check_callbacks(cpu
);
1130 raise_softirq(RCU_SOFTIRQ
);
1136 * Scan the leaf rcu_node structures, processing dyntick state for any that
1137 * have not yet encountered a quiescent state, using the function specified.
1138 * Returns 1 if the current grace period ends while scanning (possibly
1139 * because we made it end).
1141 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1142 int (*f
)(struct rcu_data
*))
1146 unsigned long flags
;
1148 struct rcu_node
*rnp
;
1150 rcu_for_each_leaf_node(rsp
, rnp
) {
1152 spin_lock_irqsave(&rnp
->lock
, flags
);
1153 if (rsp
->completed
!= lastcomp
) {
1154 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1157 if (rnp
->qsmask
== 0) {
1158 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1163 for (; cpu
<= rnp
->grphi
; cpu
++, bit
<<= 1) {
1164 if ((rnp
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1167 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1169 /* cpu_quiet_msk() releases rnp->lock. */
1170 cpu_quiet_msk(mask
, rsp
, rnp
, flags
);
1173 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1179 * Force quiescent states on reluctant CPUs, and also detect which
1180 * CPUs are in dyntick-idle mode.
1182 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1184 unsigned long flags
;
1186 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1190 if (!rcu_gp_in_progress(rsp
))
1191 return; /* No grace period in progress, nothing to force. */
1192 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1193 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1194 return; /* Someone else is already on the job. */
1197 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1198 goto unlock_ret
; /* no emergency and done recently. */
1200 spin_lock(&rnp
->lock
);
1201 lastcomp
= rsp
->completed
;
1202 signaled
= rsp
->signaled
;
1203 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1204 if (lastcomp
== rsp
->gpnum
) {
1205 rsp
->n_force_qs_ngp
++;
1206 spin_unlock(&rnp
->lock
);
1207 goto unlock_ret
; /* no GP in progress, time updated. */
1209 spin_unlock(&rnp
->lock
);
1214 break; /* grace period idle or initializing, ignore. */
1216 case RCU_SAVE_DYNTICK
:
1218 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1219 break; /* So gcc recognizes the dead code. */
1221 /* Record dyntick-idle state. */
1222 if (rcu_process_dyntick(rsp
, lastcomp
,
1223 dyntick_save_progress_counter
))
1225 /* fall into next case. */
1227 case RCU_SAVE_COMPLETED
:
1229 /* Update state, record completion counter. */
1231 spin_lock(&rnp
->lock
);
1232 if (lastcomp
== rsp
->completed
&&
1233 rsp
->signaled
== signaled
) {
1234 rsp
->signaled
= RCU_FORCE_QS
;
1235 dyntick_record_completed(rsp
, lastcomp
);
1236 forcenow
= signaled
== RCU_SAVE_COMPLETED
;
1238 spin_unlock(&rnp
->lock
);
1241 /* fall into next case. */
1245 /* Check dyntick-idle state, send IPI to laggarts. */
1246 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1247 rcu_implicit_dynticks_qs
))
1250 /* Leave state in case more forcing is required. */
1255 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1258 #else /* #ifdef CONFIG_SMP */
1260 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1265 #endif /* #else #ifdef CONFIG_SMP */
1268 * This does the RCU processing work from softirq context for the
1269 * specified rcu_state and rcu_data structures. This may be called
1270 * only from the CPU to whom the rdp belongs.
1273 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1275 unsigned long flags
;
1277 WARN_ON_ONCE(rdp
->beenonline
== 0);
1280 * If an RCU GP has gone long enough, go check for dyntick
1281 * idle CPUs and, if needed, send resched IPIs.
1283 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1284 force_quiescent_state(rsp
, 1);
1287 * Advance callbacks in response to end of earlier grace
1288 * period that some other CPU ended.
1290 rcu_process_gp_end(rsp
, rdp
);
1292 /* Update RCU state based on any recent quiescent states. */
1293 rcu_check_quiescent_state(rsp
, rdp
);
1295 /* Does this CPU require a not-yet-started grace period? */
1296 if (cpu_needs_another_gp(rsp
, rdp
)) {
1297 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1298 rcu_start_gp(rsp
, flags
); /* releases above lock */
1301 /* If there are callbacks ready, invoke them. */
1302 rcu_do_batch(rsp
, rdp
);
1306 * Do softirq processing for the current CPU.
1308 static void rcu_process_callbacks(struct softirq_action
*unused
)
1311 * Memory references from any prior RCU read-side critical sections
1312 * executed by the interrupted code must be seen before any RCU
1313 * grace-period manipulations below.
1315 smp_mb(); /* See above block comment. */
1317 __rcu_process_callbacks(&rcu_sched_state
,
1318 &__get_cpu_var(rcu_sched_data
));
1319 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1320 rcu_preempt_process_callbacks();
1323 * Memory references from any later RCU read-side critical sections
1324 * executed by the interrupted code must be seen after any RCU
1325 * grace-period manipulations above.
1327 smp_mb(); /* See above block comment. */
1331 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1332 struct rcu_state
*rsp
)
1334 unsigned long flags
;
1335 struct rcu_data
*rdp
;
1340 smp_mb(); /* Ensure RCU update seen before callback registry. */
1343 * Opportunistically note grace-period endings and beginnings.
1344 * Note that we might see a beginning right after we see an
1345 * end, but never vice versa, since this CPU has to pass through
1346 * a quiescent state betweentimes.
1348 local_irq_save(flags
);
1349 rdp
= rsp
->rda
[smp_processor_id()];
1350 rcu_process_gp_end(rsp
, rdp
);
1351 check_for_new_grace_period(rsp
, rdp
);
1353 /* Add the callback to our list. */
1354 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1355 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1357 /* Start a new grace period if one not already started. */
1358 if (!rcu_gp_in_progress(rsp
)) {
1359 unsigned long nestflag
;
1360 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1362 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1363 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1367 * Force the grace period if too many callbacks or too long waiting.
1368 * Enforce hysteresis, and don't invoke force_quiescent_state()
1369 * if some other CPU has recently done so. Also, don't bother
1370 * invoking force_quiescent_state() if the newly enqueued callback
1371 * is the only one waiting for a grace period to complete.
1373 if (unlikely(++rdp
->qlen
> rdp
->qlen_last_fqs_check
+ qhimark
)) {
1374 rdp
->blimit
= LONG_MAX
;
1375 if (rsp
->n_force_qs
== rdp
->n_force_qs_snap
&&
1376 *rdp
->nxttail
[RCU_DONE_TAIL
] != head
)
1377 force_quiescent_state(rsp
, 0);
1378 rdp
->n_force_qs_snap
= rsp
->n_force_qs
;
1379 rdp
->qlen_last_fqs_check
= rdp
->qlen
;
1380 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1381 force_quiescent_state(rsp
, 1);
1382 local_irq_restore(flags
);
1386 * Queue an RCU-sched callback for invocation after a grace period.
1388 void call_rcu_sched(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1390 __call_rcu(head
, func
, &rcu_sched_state
);
1392 EXPORT_SYMBOL_GPL(call_rcu_sched
);
1395 * Queue an RCU for invocation after a quicker grace period.
1397 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1399 __call_rcu(head
, func
, &rcu_bh_state
);
1401 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1404 * Check to see if there is any immediate RCU-related work to be done
1405 * by the current CPU, for the specified type of RCU, returning 1 if so.
1406 * The checks are in order of increasing expense: checks that can be
1407 * carried out against CPU-local state are performed first. However,
1408 * we must check for CPU stalls first, else we might not get a chance.
1410 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1412 rdp
->n_rcu_pending
++;
1414 /* Check for CPU stalls, if enabled. */
1415 check_cpu_stall(rsp
, rdp
);
1417 /* Is the RCU core waiting for a quiescent state from this CPU? */
1418 if (rdp
->qs_pending
) {
1419 rdp
->n_rp_qs_pending
++;
1423 /* Does this CPU have callbacks ready to invoke? */
1424 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1425 rdp
->n_rp_cb_ready
++;
1429 /* Has RCU gone idle with this CPU needing another grace period? */
1430 if (cpu_needs_another_gp(rsp
, rdp
)) {
1431 rdp
->n_rp_cpu_needs_gp
++;
1435 /* Has another RCU grace period completed? */
1436 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1437 rdp
->n_rp_gp_completed
++;
1441 /* Has a new RCU grace period started? */
1442 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1443 rdp
->n_rp_gp_started
++;
1447 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1448 if (rcu_gp_in_progress(rsp
) &&
1449 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1450 rdp
->n_rp_need_fqs
++;
1455 rdp
->n_rp_need_nothing
++;
1460 * Check to see if there is any immediate RCU-related work to be done
1461 * by the current CPU, returning 1 if so. This function is part of the
1462 * RCU implementation; it is -not- an exported member of the RCU API.
1464 static int rcu_pending(int cpu
)
1466 return __rcu_pending(&rcu_sched_state
, &per_cpu(rcu_sched_data
, cpu
)) ||
1467 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
)) ||
1468 rcu_preempt_pending(cpu
);
1472 * Check to see if any future RCU-related work will need to be done
1473 * by the current CPU, even if none need be done immediately, returning
1474 * 1 if so. This function is part of the RCU implementation; it is -not-
1475 * an exported member of the RCU API.
1477 int rcu_needs_cpu(int cpu
)
1479 /* RCU callbacks either ready or pending? */
1480 return per_cpu(rcu_sched_data
, cpu
).nxtlist
||
1481 per_cpu(rcu_bh_data
, cpu
).nxtlist
||
1482 rcu_preempt_needs_cpu(cpu
);
1485 static DEFINE_PER_CPU(struct rcu_head
, rcu_barrier_head
) = {NULL
};
1486 static atomic_t rcu_barrier_cpu_count
;
1487 static DEFINE_MUTEX(rcu_barrier_mutex
);
1488 static struct completion rcu_barrier_completion
;
1490 static void rcu_barrier_callback(struct rcu_head
*notused
)
1492 if (atomic_dec_and_test(&rcu_barrier_cpu_count
))
1493 complete(&rcu_barrier_completion
);
1497 * Called with preemption disabled, and from cross-cpu IRQ context.
1499 static void rcu_barrier_func(void *type
)
1501 int cpu
= smp_processor_id();
1502 struct rcu_head
*head
= &per_cpu(rcu_barrier_head
, cpu
);
1503 void (*call_rcu_func
)(struct rcu_head
*head
,
1504 void (*func
)(struct rcu_head
*head
));
1506 atomic_inc(&rcu_barrier_cpu_count
);
1507 call_rcu_func
= type
;
1508 call_rcu_func(head
, rcu_barrier_callback
);
1512 * Orchestrate the specified type of RCU barrier, waiting for all
1513 * RCU callbacks of the specified type to complete.
1515 static void _rcu_barrier(struct rcu_state
*rsp
,
1516 void (*call_rcu_func
)(struct rcu_head
*head
,
1517 void (*func
)(struct rcu_head
*head
)))
1519 BUG_ON(in_interrupt());
1520 /* Take mutex to serialize concurrent rcu_barrier() requests. */
1521 mutex_lock(&rcu_barrier_mutex
);
1522 init_completion(&rcu_barrier_completion
);
1524 * Initialize rcu_barrier_cpu_count to 1, then invoke
1525 * rcu_barrier_func() on each CPU, so that each CPU also has
1526 * incremented rcu_barrier_cpu_count. Only then is it safe to
1527 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1528 * might complete its grace period before all of the other CPUs
1529 * did their increment, causing this function to return too
1532 atomic_set(&rcu_barrier_cpu_count
, 1);
1533 preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
1534 rcu_adopt_orphan_cbs(rsp
);
1535 on_each_cpu(rcu_barrier_func
, (void *)call_rcu_func
, 1);
1536 preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
1537 if (atomic_dec_and_test(&rcu_barrier_cpu_count
))
1538 complete(&rcu_barrier_completion
);
1539 wait_for_completion(&rcu_barrier_completion
);
1540 mutex_unlock(&rcu_barrier_mutex
);
1544 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1546 void rcu_barrier_bh(void)
1548 _rcu_barrier(&rcu_bh_state
, call_rcu_bh
);
1550 EXPORT_SYMBOL_GPL(rcu_barrier_bh
);
1553 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1555 void rcu_barrier_sched(void)
1557 _rcu_barrier(&rcu_sched_state
, call_rcu_sched
);
1559 EXPORT_SYMBOL_GPL(rcu_barrier_sched
);
1562 * Do boot-time initialization of a CPU's per-CPU RCU data.
1565 rcu_boot_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1567 unsigned long flags
;
1569 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1570 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1572 /* Set up local state, ensuring consistent view of global state. */
1573 spin_lock_irqsave(&rnp
->lock
, flags
);
1574 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1575 rdp
->nxtlist
= NULL
;
1576 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1577 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1580 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1581 #endif /* #ifdef CONFIG_NO_HZ */
1583 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1587 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1588 * offline event can be happening at a given time. Note also that we
1589 * can accept some slop in the rsp->completed access due to the fact
1590 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1592 static void __cpuinit
1593 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
, int preemptable
)
1595 unsigned long flags
;
1597 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1598 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1600 /* Set up local state, ensuring consistent view of global state. */
1601 spin_lock_irqsave(&rnp
->lock
, flags
);
1602 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1603 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1604 rdp
->beenonline
= 1; /* We have now been online. */
1605 rdp
->preemptable
= preemptable
;
1606 rdp
->qlen_last_fqs_check
= 0;
1607 rdp
->n_force_qs_snap
= rsp
->n_force_qs
;
1608 rdp
->blimit
= blimit
;
1609 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1612 * A new grace period might start here. If so, we won't be part
1613 * of it, but that is OK, as we are currently in a quiescent state.
1616 /* Exclude any attempts to start a new GP on large systems. */
1617 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1619 /* Add CPU to rcu_node bitmasks. */
1621 mask
= rdp
->grpmask
;
1623 /* Exclude any attempts to start a new GP on small systems. */
1624 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1625 rnp
->qsmaskinit
|= mask
;
1626 mask
= rnp
->grpmask
;
1627 if (rnp
== rdp
->mynode
) {
1628 rdp
->gpnum
= rnp
->completed
; /* if GP in progress... */
1629 rdp
->completed
= rnp
->completed
;
1630 rdp
->passed_quiesc_completed
= rnp
->completed
- 1;
1632 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1634 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1636 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
1639 static void __cpuinit
rcu_online_cpu(int cpu
)
1641 rcu_init_percpu_data(cpu
, &rcu_sched_state
, 0);
1642 rcu_init_percpu_data(cpu
, &rcu_bh_state
, 0);
1643 rcu_preempt_init_percpu_data(cpu
);
1647 * Handle CPU online/offline notification events.
1649 int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1650 unsigned long action
, void *hcpu
)
1652 long cpu
= (long)hcpu
;
1655 case CPU_UP_PREPARE
:
1656 case CPU_UP_PREPARE_FROZEN
:
1657 rcu_online_cpu(cpu
);
1660 case CPU_DYING_FROZEN
:
1662 * preempt_disable() in _rcu_barrier() prevents stop_machine(),
1663 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
1664 * returns, all online cpus have queued rcu_barrier_func().
1665 * The dying CPU clears its cpu_online_mask bit and
1666 * moves all of its RCU callbacks to ->orphan_cbs_list
1667 * in the context of stop_machine(), so subsequent calls
1668 * to _rcu_barrier() will adopt these callbacks and only
1669 * then queue rcu_barrier_func() on all remaining CPUs.
1671 rcu_send_cbs_to_orphanage(&rcu_bh_state
);
1672 rcu_send_cbs_to_orphanage(&rcu_sched_state
);
1673 rcu_preempt_send_cbs_to_orphanage();
1676 case CPU_DEAD_FROZEN
:
1677 case CPU_UP_CANCELED
:
1678 case CPU_UP_CANCELED_FROZEN
:
1679 rcu_offline_cpu(cpu
);
1688 * Compute the per-level fanout, either using the exact fanout specified
1689 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1691 #ifdef CONFIG_RCU_FANOUT_EXACT
1692 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1696 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1697 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1699 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1700 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1707 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1708 ccur
= rsp
->levelcnt
[i
];
1709 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1713 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1716 * Helper function for rcu_init() that initializes one rcu_state structure.
1718 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1723 struct rcu_node
*rnp
;
1725 /* Initialize the level-tracking arrays. */
1727 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1728 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1729 rcu_init_levelspread(rsp
);
1731 /* Initialize the elements themselves, starting from the leaves. */
1733 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1734 cpustride
*= rsp
->levelspread
[i
];
1735 rnp
= rsp
->level
[i
];
1736 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1737 spin_lock_init(&rnp
->lock
);
1740 rnp
->qsmaskinit
= 0;
1741 rnp
->grplo
= j
* cpustride
;
1742 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1743 if (rnp
->grphi
>= NR_CPUS
)
1744 rnp
->grphi
= NR_CPUS
- 1;
1750 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1751 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1752 rnp
->parent
= rsp
->level
[i
- 1] +
1753 j
/ rsp
->levelspread
[i
- 1];
1756 INIT_LIST_HEAD(&rnp
->blocked_tasks
[0]);
1757 INIT_LIST_HEAD(&rnp
->blocked_tasks
[1]);
1760 lockdep_set_class(&rcu_get_root(rsp
)->lock
, &rcu_root_class
);
1764 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1765 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1768 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1772 struct rcu_node *rnp; \
1774 rcu_init_one(rsp); \
1775 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1777 for_each_possible_cpu(i) { \
1778 if (i > rnp[j].grphi) \
1780 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1781 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1782 rcu_boot_init_percpu_data(i, rsp); \
1786 void __init
__rcu_init(void)
1788 rcu_bootup_announce();
1789 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1790 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1791 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1792 RCU_INIT_FLAVOR(&rcu_sched_state
, rcu_sched_data
);
1793 RCU_INIT_FLAVOR(&rcu_bh_state
, rcu_bh_data
);
1794 __rcu_init_preempt();
1795 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1798 #include "rcutree_plugin.h"