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
)
110 struct rcu_data
*rdp
;
112 rdp
= &per_cpu(rcu_sched_data
, cpu
);
113 rdp
->passed_quiesc_completed
= rdp
->completed
;
115 rdp
->passed_quiesc
= 1;
116 rcu_preempt_note_context_switch(cpu
);
119 void rcu_bh_qs(int cpu
)
121 struct rcu_data
*rdp
;
123 rdp
= &per_cpu(rcu_bh_data
, cpu
);
124 rdp
->passed_quiesc_completed
= rdp
->completed
;
126 rdp
->passed_quiesc
= 1;
130 DEFINE_PER_CPU(struct rcu_dynticks
, rcu_dynticks
) = {
131 .dynticks_nesting
= 1,
134 #endif /* #ifdef CONFIG_NO_HZ */
136 static int blimit
= 10; /* Maximum callbacks per softirq. */
137 static int qhimark
= 10000; /* If this many pending, ignore blimit. */
138 static int qlowmark
= 100; /* Once only this many pending, use blimit. */
140 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
);
141 static int rcu_pending(int cpu
);
144 * Return the number of RCU-sched batches processed thus far for debug & stats.
146 long rcu_batches_completed_sched(void)
148 return rcu_sched_state
.completed
;
150 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched
);
153 * Return the number of RCU BH batches processed thus far for debug & stats.
155 long rcu_batches_completed_bh(void)
157 return rcu_bh_state
.completed
;
159 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
162 * Does the CPU have callbacks ready to be invoked?
165 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
167 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
171 * Does the current CPU require a yet-as-unscheduled grace period?
174 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
176 /* ACCESS_ONCE() because we are accessing outside of lock. */
177 return *rdp
->nxttail
[RCU_DONE_TAIL
] &&
178 ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
);
182 * Return the root node of the specified rcu_state structure.
184 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
186 return &rsp
->node
[0];
192 * If the specified CPU is offline, tell the caller that it is in
193 * a quiescent state. Otherwise, whack it with a reschedule IPI.
194 * Grace periods can end up waiting on an offline CPU when that
195 * CPU is in the process of coming online -- it will be added to the
196 * rcu_node bitmasks before it actually makes it online. The same thing
197 * can happen while a CPU is in the process of coming online. Because this
198 * race is quite rare, we check for it after detecting that the grace
199 * period has been delayed rather than checking each and every CPU
200 * each and every time we start a new grace period.
202 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
205 * If the CPU is offline, it is in a quiescent state. We can
206 * trust its state not to change because interrupts are disabled.
208 if (cpu_is_offline(rdp
->cpu
)) {
213 /* If preemptable RCU, no point in sending reschedule IPI. */
214 if (rdp
->preemptable
)
217 /* The CPU is online, so send it a reschedule IPI. */
218 if (rdp
->cpu
!= smp_processor_id())
219 smp_send_reschedule(rdp
->cpu
);
226 #endif /* #ifdef CONFIG_SMP */
231 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
233 * Enter nohz mode, in other words, -leave- the mode in which RCU
234 * read-side critical sections can occur. (Though RCU read-side
235 * critical sections can occur in irq handlers in nohz mode, a possibility
236 * handled by rcu_irq_enter() and rcu_irq_exit()).
238 void rcu_enter_nohz(void)
241 struct rcu_dynticks
*rdtp
;
243 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
244 local_irq_save(flags
);
245 rdtp
= &__get_cpu_var(rcu_dynticks
);
247 rdtp
->dynticks_nesting
--;
248 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
249 local_irq_restore(flags
);
253 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
255 * Exit nohz mode, in other words, -enter- the mode in which RCU
256 * read-side critical sections normally occur.
258 void rcu_exit_nohz(void)
261 struct rcu_dynticks
*rdtp
;
263 local_irq_save(flags
);
264 rdtp
= &__get_cpu_var(rcu_dynticks
);
266 rdtp
->dynticks_nesting
++;
267 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
268 local_irq_restore(flags
);
269 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
273 * rcu_nmi_enter - inform RCU of entry to NMI context
275 * If the CPU was idle with dynamic ticks active, and there is no
276 * irq handler running, this updates rdtp->dynticks_nmi to let the
277 * RCU grace-period handling know that the CPU is active.
279 void rcu_nmi_enter(void)
281 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
283 if (rdtp
->dynticks
& 0x1)
285 rdtp
->dynticks_nmi
++;
286 WARN_ON_ONCE(!(rdtp
->dynticks_nmi
& 0x1));
287 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
291 * rcu_nmi_exit - inform RCU of exit from NMI context
293 * If the CPU was idle with dynamic ticks active, and there is no
294 * irq handler running, this updates rdtp->dynticks_nmi to let the
295 * RCU grace-period handling know that the CPU is no longer active.
297 void rcu_nmi_exit(void)
299 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
301 if (rdtp
->dynticks
& 0x1)
303 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
304 rdtp
->dynticks_nmi
++;
305 WARN_ON_ONCE(rdtp
->dynticks_nmi
& 0x1);
309 * rcu_irq_enter - inform RCU of entry to hard irq context
311 * If the CPU was idle with dynamic ticks active, this updates the
312 * rdtp->dynticks to let the RCU handling know that the CPU is active.
314 void rcu_irq_enter(void)
316 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
318 if (rdtp
->dynticks_nesting
++)
321 WARN_ON_ONCE(!(rdtp
->dynticks
& 0x1));
322 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
326 * rcu_irq_exit - inform RCU of exit from hard irq context
328 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
329 * to put let the RCU handling be aware that the CPU is going back to idle
332 void rcu_irq_exit(void)
334 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
336 if (--rdtp
->dynticks_nesting
)
338 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
340 WARN_ON_ONCE(rdtp
->dynticks
& 0x1);
342 /* If the interrupt queued a callback, get out of dyntick mode. */
343 if (__get_cpu_var(rcu_sched_data
).nxtlist
||
344 __get_cpu_var(rcu_bh_data
).nxtlist
)
349 * Record the specified "completed" value, which is later used to validate
350 * dynticks counter manipulations. Specify "rsp->completed - 1" to
351 * unconditionally invalidate any future dynticks manipulations (which is
352 * useful at the beginning of a grace period).
354 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
356 rsp
->dynticks_completed
= comp
;
362 * Recall the previously recorded value of the completion for dynticks.
364 static long dyntick_recall_completed(struct rcu_state
*rsp
)
366 return rsp
->dynticks_completed
;
370 * Snapshot the specified CPU's dynticks counter so that we can later
371 * credit them with an implicit quiescent state. Return 1 if this CPU
372 * is already in a quiescent state courtesy of dynticks idle mode.
374 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
380 snap
= rdp
->dynticks
->dynticks
;
381 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
382 smp_mb(); /* Order sampling of snap with end of grace period. */
383 rdp
->dynticks_snap
= snap
;
384 rdp
->dynticks_nmi_snap
= snap_nmi
;
385 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
392 * Return true if the specified CPU has passed through a quiescent
393 * state by virtue of being in or having passed through an dynticks
394 * idle state since the last call to dyntick_save_progress_counter()
397 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
404 curr
= rdp
->dynticks
->dynticks
;
405 snap
= rdp
->dynticks_snap
;
406 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
407 snap_nmi
= rdp
->dynticks_nmi_snap
;
408 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
411 * If the CPU passed through or entered a dynticks idle phase with
412 * no active irq/NMI handlers, then we can safely pretend that the CPU
413 * already acknowledged the request to pass through a quiescent
414 * state. Either way, that CPU cannot possibly be in an RCU
415 * read-side critical section that started before the beginning
416 * of the current RCU grace period.
418 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
419 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
424 /* Go check for the CPU being offline. */
425 return rcu_implicit_offline_qs(rdp
);
428 #endif /* #ifdef CONFIG_SMP */
430 #else /* #ifdef CONFIG_NO_HZ */
432 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
439 * If there are no dynticks, then the only way that a CPU can passively
440 * be in a quiescent state is to be offline. Unlike dynticks idle, which
441 * is a point in time during the prior (already finished) grace period,
442 * an offline CPU is always in a quiescent state, and thus can be
443 * unconditionally applied. So just return the current value of completed.
445 static long dyntick_recall_completed(struct rcu_state
*rsp
)
447 return rsp
->completed
;
450 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
455 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
457 return rcu_implicit_offline_qs(rdp
);
460 #endif /* #ifdef CONFIG_SMP */
462 #endif /* #else #ifdef CONFIG_NO_HZ */
464 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
466 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
468 rsp
->gp_start
= jiffies
;
469 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
472 static void print_other_cpu_stall(struct rcu_state
*rsp
)
477 struct rcu_node
*rnp
= rcu_get_root(rsp
);
478 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
479 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
481 /* Only let one CPU complain about others per time interval. */
483 spin_lock_irqsave(&rnp
->lock
, flags
);
484 delta
= jiffies
- rsp
->jiffies_stall
;
485 if (delta
< RCU_STALL_RAT_DELAY
|| rsp
->gpnum
== rsp
->completed
) {
486 spin_unlock_irqrestore(&rnp
->lock
, flags
);
489 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
490 spin_unlock_irqrestore(&rnp
->lock
, flags
);
492 /* OK, time to rat on our buddy... */
494 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
495 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
496 rcu_print_task_stall(rnp
);
497 if (rnp_cur
->qsmask
== 0)
499 for (cpu
= 0; cpu
<= rnp_cur
->grphi
- rnp_cur
->grplo
; cpu
++)
500 if (rnp_cur
->qsmask
& (1UL << cpu
))
501 printk(" %d", rnp_cur
->grplo
+ cpu
);
503 printk(" (detected by %d, t=%ld jiffies)\n",
504 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
505 trigger_all_cpu_backtrace();
507 force_quiescent_state(rsp
, 0); /* Kick them all. */
510 static void print_cpu_stall(struct rcu_state
*rsp
)
513 struct rcu_node
*rnp
= rcu_get_root(rsp
);
515 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
516 smp_processor_id(), jiffies
- rsp
->gp_start
);
517 trigger_all_cpu_backtrace();
519 spin_lock_irqsave(&rnp
->lock
, flags
);
520 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
522 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
523 spin_unlock_irqrestore(&rnp
->lock
, flags
);
525 set_need_resched(); /* kick ourselves to get things going. */
528 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
531 struct rcu_node
*rnp
;
533 delta
= jiffies
- rsp
->jiffies_stall
;
535 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
537 /* We haven't checked in, so go dump stack. */
538 print_cpu_stall(rsp
);
540 } else if (rsp
->gpnum
!= rsp
->completed
&&
541 delta
>= RCU_STALL_RAT_DELAY
) {
543 /* They had two time units to dump stack, so complain. */
544 print_other_cpu_stall(rsp
);
548 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
550 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
554 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
558 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
561 * Update CPU-local rcu_data state to record the newly noticed grace period.
562 * This is used both when we started the grace period and when we notice
563 * that someone else started the grace period.
565 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
568 rdp
->passed_quiesc
= 0;
569 rdp
->gpnum
= rsp
->gpnum
;
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 * Start a new RCU grace period if warranted, re-initializing the hierarchy
594 * in preparation for detecting the next grace period. The caller must hold
595 * the root node's ->lock, which is released before return. Hard irqs must
599 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
600 __releases(rcu_get_root(rsp
)->lock
)
602 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
603 struct rcu_node
*rnp
= rcu_get_root(rsp
);
604 struct rcu_node
*rnp_cur
;
605 struct rcu_node
*rnp_end
;
607 if (!cpu_needs_another_gp(rsp
, rdp
)) {
608 spin_unlock_irqrestore(&rnp
->lock
, flags
);
612 /* Advance to a new grace period and initialize state. */
614 WARN_ON_ONCE(rsp
->signaled
== RCU_GP_INIT
);
615 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
616 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
617 record_gp_stall_check_time(rsp
);
618 dyntick_record_completed(rsp
, rsp
->completed
- 1);
619 note_new_gpnum(rsp
, rdp
);
622 * Because we are first, we know that all our callbacks will
623 * be covered by this upcoming grace period, even the ones
624 * that were registered arbitrarily recently.
626 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
627 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
629 /* Special-case the common single-level case. */
630 if (NUM_RCU_NODES
== 1) {
631 rcu_preempt_check_blocked_tasks(rnp
);
632 rnp
->qsmask
= rnp
->qsmaskinit
;
633 rnp
->gpnum
= rsp
->gpnum
;
634 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
635 spin_unlock_irqrestore(&rnp
->lock
, flags
);
639 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
642 /* Exclude any concurrent CPU-hotplug operations. */
643 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
646 * Set the quiescent-state-needed bits in all the rcu_node
647 * structures for all currently online CPUs in breadth-first
648 * order, starting from the root rcu_node structure. This
649 * operation relies on the layout of the hierarchy within the
650 * rsp->node[] array. Note that other CPUs will access only
651 * the leaves of the hierarchy, which still indicate that no
652 * grace period is in progress, at least until the corresponding
653 * leaf node has been initialized. In addition, we have excluded
654 * CPU-hotplug operations.
656 * Note that the grace period cannot complete until we finish
657 * the initialization process, as there will be at least one
658 * qsmask bit set in the root node until that time, namely the
659 * one corresponding to this CPU, due to the fact that we have
662 rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
663 for (rnp_cur
= &rsp
->node
[0]; rnp_cur
< rnp_end
; rnp_cur
++) {
664 spin_lock(&rnp_cur
->lock
); /* irqs already disabled. */
665 rcu_preempt_check_blocked_tasks(rnp
);
666 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
667 rnp
->gpnum
= rsp
->gpnum
;
668 spin_unlock(&rnp_cur
->lock
); /* irqs already disabled. */
671 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
672 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
676 * Advance this CPU's callbacks, but only if the current grace period
677 * has ended. This may be called only from the CPU to whom the rdp
681 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
686 local_irq_save(flags
);
687 completed_snap
= ACCESS_ONCE(rsp
->completed
); /* outside of lock. */
689 /* Did another grace period end? */
690 if (rdp
->completed
!= completed_snap
) {
692 /* Advance callbacks. No harm if list empty. */
693 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
694 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
695 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
697 /* Remember that we saw this grace-period completion. */
698 rdp
->completed
= completed_snap
;
700 local_irq_restore(flags
);
704 * Clean up after the prior grace period and let rcu_start_gp() start up
705 * the next grace period if one is needed. Note that the caller must
706 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
708 static void cpu_quiet_msk_finish(struct rcu_state
*rsp
, unsigned long flags
)
709 __releases(rnp
->lock
)
711 WARN_ON_ONCE(rsp
->completed
== rsp
->gpnum
);
712 rsp
->completed
= rsp
->gpnum
;
713 rcu_process_gp_end(rsp
, rsp
->rda
[smp_processor_id()]);
714 rcu_start_gp(rsp
, flags
); /* releases root node's rnp->lock. */
718 * Similar to cpu_quiet(), for which it is a helper function. Allows
719 * a group of CPUs to be quieted at one go, though all the CPUs in the
720 * group must be represented by the same leaf rcu_node structure.
721 * That structure's lock must be held upon entry, and it is released
725 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
727 __releases(rnp
->lock
)
729 struct rcu_node
*rnp_c
;
731 /* Walk up the rcu_node hierarchy. */
733 if (!(rnp
->qsmask
& mask
)) {
735 /* Our bit has already been cleared, so done. */
736 spin_unlock_irqrestore(&rnp
->lock
, flags
);
739 rnp
->qsmask
&= ~mask
;
740 if (rnp
->qsmask
!= 0 || rcu_preempted_readers(rnp
)) {
742 /* Other bits still set at this level, so done. */
743 spin_unlock_irqrestore(&rnp
->lock
, flags
);
747 if (rnp
->parent
== NULL
) {
749 /* No more levels. Exit loop holding root lock. */
753 spin_unlock_irqrestore(&rnp
->lock
, flags
);
756 spin_lock_irqsave(&rnp
->lock
, flags
);
757 WARN_ON_ONCE(rnp_c
->qsmask
);
761 * Get here if we are the last CPU to pass through a quiescent
762 * state for this grace period. Invoke cpu_quiet_msk_finish()
763 * to clean up and start the next grace period if one is needed.
765 cpu_quiet_msk_finish(rsp
, flags
); /* releases rnp->lock. */
769 * Record a quiescent state for the specified CPU, which must either be
770 * the current CPU or an offline CPU. The lastcomp argument is used to
771 * make sure we are still in the grace period of interest. We don't want
772 * to end the current grace period based on quiescent states detected in
773 * an earlier grace period!
776 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
780 struct rcu_node
*rnp
;
783 spin_lock_irqsave(&rnp
->lock
, flags
);
784 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
787 * Someone beat us to it for this grace period, so leave.
788 * The race with GP start is resolved by the fact that we
789 * hold the leaf rcu_node lock, so that the per-CPU bits
790 * cannot yet be initialized -- so we would simply find our
791 * CPU's bit already cleared in cpu_quiet_msk() if this race
794 rdp
->passed_quiesc
= 0; /* try again later! */
795 spin_unlock_irqrestore(&rnp
->lock
, flags
);
799 if ((rnp
->qsmask
& mask
) == 0) {
800 spin_unlock_irqrestore(&rnp
->lock
, flags
);
805 * This GP can't end until cpu checks in, so all of our
806 * callbacks can be processed during the next GP.
808 rdp
= rsp
->rda
[smp_processor_id()];
809 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
811 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
816 * Check to see if there is a new grace period of which this CPU
817 * is not yet aware, and if so, set up local rcu_data state for it.
818 * Otherwise, see if this CPU has just passed through its first
819 * quiescent state for this grace period, and record that fact if so.
822 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
824 /* If there is now a new grace period, record and return. */
825 if (check_for_new_grace_period(rsp
, rdp
))
829 * Does this CPU still need to do its part for current grace period?
830 * If no, return and let the other CPUs do their part as well.
832 if (!rdp
->qs_pending
)
836 * Was there a quiescent state since the beginning of the grace
837 * period? If no, then exit and wait for the next call.
839 if (!rdp
->passed_quiesc
)
842 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
843 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
846 #ifdef CONFIG_HOTPLUG_CPU
849 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
850 * and move all callbacks from the outgoing CPU to the current one.
852 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
858 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
859 struct rcu_data
*rdp_me
;
860 struct rcu_node
*rnp
;
862 /* Exclude any attempts to start a new grace period. */
863 spin_lock_irqsave(&rsp
->onofflock
, flags
);
865 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
866 rnp
= rdp
->mynode
; /* this is the outgoing CPU's rnp. */
867 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
869 spin_lock(&rnp
->lock
); /* irqs already disabled. */
870 rnp
->qsmaskinit
&= ~mask
;
871 if (rnp
->qsmaskinit
!= 0) {
872 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
875 rcu_preempt_offline_tasks(rsp
, rnp
, rdp
);
877 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
879 } while (rnp
!= NULL
);
880 lastcomp
= rsp
->completed
;
882 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
884 /* Being offline is a quiescent state, so go record it. */
885 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
888 * Move callbacks from the outgoing CPU to the running CPU.
889 * Note that the outgoing CPU is now quiscent, so it is now
890 * (uncharacteristically) safe to access its rcu_data structure.
891 * Note also that we must carefully retain the order of the
892 * outgoing CPU's callbacks in order for rcu_barrier() to work
893 * correctly. Finally, note that we start all the callbacks
894 * afresh, even those that have passed through a grace period
895 * and are therefore ready to invoke. The theory is that hotplug
896 * events are rare, and that if they are frequent enough to
897 * indefinitely delay callbacks, you have far worse things to
900 rdp_me
= rsp
->rda
[smp_processor_id()];
901 if (rdp
->nxtlist
!= NULL
) {
902 *rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxtlist
;
903 rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
905 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
906 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
907 rdp_me
->qlen
+= rdp
->qlen
;
910 local_irq_restore(flags
);
914 * Remove the specified CPU from the RCU hierarchy and move any pending
915 * callbacks that it might have to the current CPU. This code assumes
916 * that at least one CPU in the system will remain running at all times.
917 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
919 static void rcu_offline_cpu(int cpu
)
921 __rcu_offline_cpu(cpu
, &rcu_sched_state
);
922 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
923 rcu_preempt_offline_cpu(cpu
);
926 #else /* #ifdef CONFIG_HOTPLUG_CPU */
928 static void rcu_offline_cpu(int cpu
)
932 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
935 * Invoke any RCU callbacks that have made it to the end of their grace
936 * period. Thottle as specified by rdp->blimit.
938 static void rcu_do_batch(struct rcu_data
*rdp
)
941 struct rcu_head
*next
, *list
, **tail
;
944 /* If no callbacks are ready, just return.*/
945 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
949 * Extract the list of ready callbacks, disabling to prevent
950 * races with call_rcu() from interrupt handlers.
952 local_irq_save(flags
);
954 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
955 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
956 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
957 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
958 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
959 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
960 local_irq_restore(flags
);
962 /* Invoke callbacks. */
969 if (++count
>= rdp
->blimit
)
973 local_irq_save(flags
);
975 /* Update count, and requeue any remaining callbacks. */
978 *tail
= rdp
->nxtlist
;
980 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
981 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
982 rdp
->nxttail
[count
] = tail
;
987 /* Reinstate batch limit if we have worked down the excess. */
988 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
989 rdp
->blimit
= blimit
;
991 local_irq_restore(flags
);
993 /* Re-raise the RCU softirq if there are callbacks remaining. */
994 if (cpu_has_callbacks_ready_to_invoke(rdp
))
995 raise_softirq(RCU_SOFTIRQ
);
999 * Check to see if this CPU is in a non-context-switch quiescent state
1000 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1001 * Also schedule the RCU softirq handler.
1003 * This function must be called with hardirqs disabled. It is normally
1004 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1005 * false, there is no point in invoking rcu_check_callbacks().
1007 void rcu_check_callbacks(int cpu
, int user
)
1009 if (!rcu_pending(cpu
))
1010 return; /* if nothing for RCU to do. */
1012 (idle_cpu(cpu
) && rcu_scheduler_active
&&
1013 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
1016 * Get here if this CPU took its interrupt from user
1017 * mode or from the idle loop, and if this is not a
1018 * nested interrupt. In this case, the CPU is in
1019 * a quiescent state, so note it.
1021 * No memory barrier is required here because both
1022 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1023 * variables that other CPUs neither access nor modify,
1024 * at least not while the corresponding CPU is online.
1030 } else if (!in_softirq()) {
1033 * Get here if this CPU did not take its interrupt from
1034 * softirq, in other words, if it is not interrupting
1035 * a rcu_bh read-side critical section. This is an _bh
1036 * critical section, so note it.
1041 rcu_preempt_check_callbacks(cpu
);
1042 raise_softirq(RCU_SOFTIRQ
);
1048 * Scan the leaf rcu_node structures, processing dyntick state for any that
1049 * have not yet encountered a quiescent state, using the function specified.
1050 * Returns 1 if the current grace period ends while scanning (possibly
1051 * because we made it end).
1053 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1054 int (*f
)(struct rcu_data
*))
1058 unsigned long flags
;
1060 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
1061 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
1063 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
1065 spin_lock_irqsave(&rnp_cur
->lock
, flags
);
1066 if (rsp
->completed
!= lastcomp
) {
1067 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1070 if (rnp_cur
->qsmask
== 0) {
1071 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1074 cpu
= rnp_cur
->grplo
;
1076 for (; cpu
<= rnp_cur
->grphi
; cpu
++, bit
<<= 1) {
1077 if ((rnp_cur
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1080 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1082 /* cpu_quiet_msk() releases rnp_cur->lock. */
1083 cpu_quiet_msk(mask
, rsp
, rnp_cur
, flags
);
1086 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1092 * Force quiescent states on reluctant CPUs, and also detect which
1093 * CPUs are in dyntick-idle mode.
1095 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1097 unsigned long flags
;
1099 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1102 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
))
1103 return; /* No grace period in progress, nothing to force. */
1104 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1105 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1106 return; /* Someone else is already on the job. */
1109 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1110 goto unlock_ret
; /* no emergency and done recently. */
1112 spin_lock(&rnp
->lock
);
1113 lastcomp
= rsp
->completed
;
1114 signaled
= rsp
->signaled
;
1115 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1116 if (lastcomp
== rsp
->gpnum
) {
1117 rsp
->n_force_qs_ngp
++;
1118 spin_unlock(&rnp
->lock
);
1119 goto unlock_ret
; /* no GP in progress, time updated. */
1121 spin_unlock(&rnp
->lock
);
1125 break; /* grace period still initializing, ignore. */
1127 case RCU_SAVE_DYNTICK
:
1129 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1130 break; /* So gcc recognizes the dead code. */
1132 /* Record dyntick-idle state. */
1133 if (rcu_process_dyntick(rsp
, lastcomp
,
1134 dyntick_save_progress_counter
))
1137 /* Update state, record completion counter. */
1138 spin_lock(&rnp
->lock
);
1139 if (lastcomp
== rsp
->completed
) {
1140 rsp
->signaled
= RCU_FORCE_QS
;
1141 dyntick_record_completed(rsp
, lastcomp
);
1143 spin_unlock(&rnp
->lock
);
1148 /* Check dyntick-idle state, send IPI to laggarts. */
1149 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1150 rcu_implicit_dynticks_qs
))
1153 /* Leave state in case more forcing is required. */
1158 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1161 #else /* #ifdef CONFIG_SMP */
1163 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1168 #endif /* #else #ifdef CONFIG_SMP */
1171 * This does the RCU processing work from softirq context for the
1172 * specified rcu_state and rcu_data structures. This may be called
1173 * only from the CPU to whom the rdp belongs.
1176 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1178 unsigned long flags
;
1180 WARN_ON_ONCE(rdp
->beenonline
== 0);
1183 * If an RCU GP has gone long enough, go check for dyntick
1184 * idle CPUs and, if needed, send resched IPIs.
1186 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1187 force_quiescent_state(rsp
, 1);
1190 * Advance callbacks in response to end of earlier grace
1191 * period that some other CPU ended.
1193 rcu_process_gp_end(rsp
, rdp
);
1195 /* Update RCU state based on any recent quiescent states. */
1196 rcu_check_quiescent_state(rsp
, rdp
);
1198 /* Does this CPU require a not-yet-started grace period? */
1199 if (cpu_needs_another_gp(rsp
, rdp
)) {
1200 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1201 rcu_start_gp(rsp
, flags
); /* releases above lock */
1204 /* If there are callbacks ready, invoke them. */
1209 * Do softirq processing for the current CPU.
1211 static void rcu_process_callbacks(struct softirq_action
*unused
)
1214 * Memory references from any prior RCU read-side critical sections
1215 * executed by the interrupted code must be seen before any RCU
1216 * grace-period manipulations below.
1218 smp_mb(); /* See above block comment. */
1220 __rcu_process_callbacks(&rcu_sched_state
,
1221 &__get_cpu_var(rcu_sched_data
));
1222 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1223 rcu_preempt_process_callbacks();
1226 * Memory references from any later RCU read-side critical sections
1227 * executed by the interrupted code must be seen after any RCU
1228 * grace-period manipulations above.
1230 smp_mb(); /* See above block comment. */
1234 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1235 struct rcu_state
*rsp
)
1237 unsigned long flags
;
1238 struct rcu_data
*rdp
;
1243 smp_mb(); /* Ensure RCU update seen before callback registry. */
1246 * Opportunistically note grace-period endings and beginnings.
1247 * Note that we might see a beginning right after we see an
1248 * end, but never vice versa, since this CPU has to pass through
1249 * a quiescent state betweentimes.
1251 local_irq_save(flags
);
1252 rdp
= rsp
->rda
[smp_processor_id()];
1253 rcu_process_gp_end(rsp
, rdp
);
1254 check_for_new_grace_period(rsp
, rdp
);
1256 /* Add the callback to our list. */
1257 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1258 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1260 /* Start a new grace period if one not already started. */
1261 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
)) {
1262 unsigned long nestflag
;
1263 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1265 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1266 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1269 /* Force the grace period if too many callbacks or too long waiting. */
1270 if (unlikely(++rdp
->qlen
> qhimark
)) {
1271 rdp
->blimit
= LONG_MAX
;
1272 force_quiescent_state(rsp
, 0);
1273 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1274 force_quiescent_state(rsp
, 1);
1275 local_irq_restore(flags
);
1279 * Queue an RCU-sched callback for invocation after a grace period.
1281 void call_rcu_sched(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1283 __call_rcu(head
, func
, &rcu_sched_state
);
1285 EXPORT_SYMBOL_GPL(call_rcu_sched
);
1288 * Queue an RCU for invocation after a quicker grace period.
1290 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1292 __call_rcu(head
, func
, &rcu_bh_state
);
1294 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1297 * Check to see if there is any immediate RCU-related work to be done
1298 * by the current CPU, for the specified type of RCU, returning 1 if so.
1299 * The checks are in order of increasing expense: checks that can be
1300 * carried out against CPU-local state are performed first. However,
1301 * we must check for CPU stalls first, else we might not get a chance.
1303 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1305 rdp
->n_rcu_pending
++;
1307 /* Check for CPU stalls, if enabled. */
1308 check_cpu_stall(rsp
, rdp
);
1310 /* Is the RCU core waiting for a quiescent state from this CPU? */
1311 if (rdp
->qs_pending
) {
1312 rdp
->n_rp_qs_pending
++;
1316 /* Does this CPU have callbacks ready to invoke? */
1317 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1318 rdp
->n_rp_cb_ready
++;
1322 /* Has RCU gone idle with this CPU needing another grace period? */
1323 if (cpu_needs_another_gp(rsp
, rdp
)) {
1324 rdp
->n_rp_cpu_needs_gp
++;
1328 /* Has another RCU grace period completed? */
1329 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1330 rdp
->n_rp_gp_completed
++;
1334 /* Has a new RCU grace period started? */
1335 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1336 rdp
->n_rp_gp_started
++;
1340 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1341 if (ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
) &&
1342 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1343 rdp
->n_rp_need_fqs
++;
1348 rdp
->n_rp_need_nothing
++;
1353 * Check to see if there is any immediate RCU-related work to be done
1354 * by the current CPU, returning 1 if so. This function is part of the
1355 * RCU implementation; it is -not- an exported member of the RCU API.
1357 static int rcu_pending(int cpu
)
1359 return __rcu_pending(&rcu_sched_state
, &per_cpu(rcu_sched_data
, cpu
)) ||
1360 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
)) ||
1361 rcu_preempt_pending(cpu
);
1365 * Check to see if any future RCU-related work will need to be done
1366 * by the current CPU, even if none need be done immediately, returning
1367 * 1 if so. This function is part of the RCU implementation; it is -not-
1368 * an exported member of the RCU API.
1370 int rcu_needs_cpu(int cpu
)
1372 /* RCU callbacks either ready or pending? */
1373 return per_cpu(rcu_sched_data
, cpu
).nxtlist
||
1374 per_cpu(rcu_bh_data
, cpu
).nxtlist
||
1375 rcu_preempt_needs_cpu(cpu
);
1379 * Do boot-time initialization of a CPU's per-CPU RCU data.
1382 rcu_boot_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1384 unsigned long flags
;
1386 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1387 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1389 /* Set up local state, ensuring consistent view of global state. */
1390 spin_lock_irqsave(&rnp
->lock
, flags
);
1391 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1392 rdp
->nxtlist
= NULL
;
1393 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1394 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1397 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1398 #endif /* #ifdef CONFIG_NO_HZ */
1400 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1404 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1405 * offline event can be happening at a given time. Note also that we
1406 * can accept some slop in the rsp->completed access due to the fact
1407 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1409 static void __cpuinit
1410 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
, int preemptable
)
1412 unsigned long flags
;
1415 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1416 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1418 /* Set up local state, ensuring consistent view of global state. */
1419 spin_lock_irqsave(&rnp
->lock
, flags
);
1420 lastcomp
= rsp
->completed
;
1421 rdp
->completed
= lastcomp
;
1422 rdp
->gpnum
= lastcomp
;
1423 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1424 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1425 rdp
->beenonline
= 1; /* We have now been online. */
1426 rdp
->preemptable
= preemptable
;
1427 rdp
->passed_quiesc_completed
= lastcomp
- 1;
1428 rdp
->blimit
= blimit
;
1429 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1432 * A new grace period might start here. If so, we won't be part
1433 * of it, but that is OK, as we are currently in a quiescent state.
1436 /* Exclude any attempts to start a new GP on large systems. */
1437 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1439 /* Add CPU to rcu_node bitmasks. */
1441 mask
= rdp
->grpmask
;
1443 /* Exclude any attempts to start a new GP on small systems. */
1444 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1445 rnp
->qsmaskinit
|= mask
;
1446 mask
= rnp
->grpmask
;
1447 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1449 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1451 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
1454 * A new grace period might start here. If so, we will be part of
1455 * it, and its gpnum will be greater than ours, so we will
1456 * participate. It is also possible for the gpnum to have been
1457 * incremented before this function was called, and the bitmasks
1458 * to not be filled out until now, in which case we will also
1459 * participate due to our gpnum being behind.
1462 /* Since it is coming online, the CPU is in a quiescent state. */
1463 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
1464 local_irq_restore(flags
);
1467 static void __cpuinit
rcu_online_cpu(int cpu
)
1469 rcu_init_percpu_data(cpu
, &rcu_sched_state
, 0);
1470 rcu_init_percpu_data(cpu
, &rcu_bh_state
, 0);
1471 rcu_preempt_init_percpu_data(cpu
);
1475 * Handle CPU online/offline notification events.
1477 int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1478 unsigned long action
, void *hcpu
)
1480 long cpu
= (long)hcpu
;
1483 case CPU_UP_PREPARE
:
1484 case CPU_UP_PREPARE_FROZEN
:
1485 rcu_online_cpu(cpu
);
1488 case CPU_DEAD_FROZEN
:
1489 case CPU_UP_CANCELED
:
1490 case CPU_UP_CANCELED_FROZEN
:
1491 rcu_offline_cpu(cpu
);
1500 * Compute the per-level fanout, either using the exact fanout specified
1501 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1503 #ifdef CONFIG_RCU_FANOUT_EXACT
1504 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1508 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1509 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1511 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1512 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1519 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1520 ccur
= rsp
->levelcnt
[i
];
1521 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1525 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1528 * Helper function for rcu_init() that initializes one rcu_state structure.
1530 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1535 struct rcu_node
*rnp
;
1537 /* Initialize the level-tracking arrays. */
1539 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1540 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1541 rcu_init_levelspread(rsp
);
1543 /* Initialize the elements themselves, starting from the leaves. */
1545 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1546 cpustride
*= rsp
->levelspread
[i
];
1547 rnp
= rsp
->level
[i
];
1548 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1549 spin_lock_init(&rnp
->lock
);
1552 rnp
->qsmaskinit
= 0;
1553 rnp
->grplo
= j
* cpustride
;
1554 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1555 if (rnp
->grphi
>= NR_CPUS
)
1556 rnp
->grphi
= NR_CPUS
- 1;
1562 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1563 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1564 rnp
->parent
= rsp
->level
[i
- 1] +
1565 j
/ rsp
->levelspread
[i
- 1];
1568 INIT_LIST_HEAD(&rnp
->blocked_tasks
[0]);
1569 INIT_LIST_HEAD(&rnp
->blocked_tasks
[1]);
1575 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1576 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1579 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1581 rcu_init_one(rsp); \
1582 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1584 for_each_possible_cpu(i) { \
1585 if (i > rnp[j].grphi) \
1587 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1588 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1589 rcu_boot_init_percpu_data(i, rsp); \
1593 #ifdef CONFIG_TREE_PREEMPT_RCU
1595 void __init
__rcu_init_preempt(void)
1597 int i
; /* All used by RCU_INIT_FLAVOR(). */
1599 struct rcu_node
*rnp
;
1601 RCU_INIT_FLAVOR(&rcu_preempt_state
, rcu_preempt_data
);
1604 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1606 void __init
__rcu_init_preempt(void)
1610 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1612 void __init
__rcu_init(void)
1614 int i
; /* All used by RCU_INIT_FLAVOR(). */
1616 struct rcu_node
*rnp
;
1618 rcu_bootup_announce();
1619 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1620 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1621 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1622 RCU_INIT_FLAVOR(&rcu_sched_state
, rcu_sched_data
);
1623 RCU_INIT_FLAVOR(&rcu_bh_state
, rcu_bh_data
);
1624 __rcu_init_preempt();
1625 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1628 module_param(blimit
, int, 0);
1629 module_param(qhimark
, int, 0);
1630 module_param(qlowmark
, int, 0);