Root/kernel/rcutree.c

1/*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
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.
8 *
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.
13 *
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.
17 *
18 * Copyright IBM Corporation, 2008
19 *
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
23 *
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
26 *
27 * For detailed explanation of Read-Copy Update mechanism see -
28 * Documentation/RCU
29 */
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>
49
50#include "rcutree.h"
51
52/* Data structures. */
53
54#define RCU_STATE_INITIALIZER(name) { \
55    .level = { &name.node[0] }, \
56    .levelcnt = { \
57        NUM_RCU_LVL_0, /* root of hierarchy. */ \
58        NUM_RCU_LVL_1, \
59        NUM_RCU_LVL_2, \
60        NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
61    }, \
62    .signaled = RCU_GP_IDLE, \
63    .gpnum = -300, \
64    .completed = -300, \
65    .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
66    .orphan_cbs_list = NULL, \
67    .orphan_cbs_tail = &name.orphan_cbs_list, \
68    .orphan_qlen = 0, \
69    .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
70    .n_force_qs = 0, \
71    .n_force_qs_ngp = 0, \
72}
73
74struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
75DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
76
77struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
78DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
79
80
81/*
82 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
83 * permit this function to be invoked without holding the root rcu_node
84 * structure's ->lock, but of course results can be subject to change.
85 */
86static int rcu_gp_in_progress(struct rcu_state *rsp)
87{
88    return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
89}
90
91/*
92 * Note a quiescent state. Because we do not need to know
93 * how many quiescent states passed, just if there was at least
94 * one since the start of the grace period, this just sets a flag.
95 */
96void rcu_sched_qs(int cpu)
97{
98    struct rcu_data *rdp;
99
100    rdp = &per_cpu(rcu_sched_data, cpu);
101    rdp->passed_quiesc_completed = rdp->completed;
102    barrier();
103    rdp->passed_quiesc = 1;
104    rcu_preempt_note_context_switch(cpu);
105}
106
107void rcu_bh_qs(int cpu)
108{
109    struct rcu_data *rdp;
110
111    rdp = &per_cpu(rcu_bh_data, cpu);
112    rdp->passed_quiesc_completed = rdp->completed;
113    barrier();
114    rdp->passed_quiesc = 1;
115}
116
117#ifdef CONFIG_NO_HZ
118DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
119    .dynticks_nesting = 1,
120    .dynticks = 1,
121};
122#endif /* #ifdef CONFIG_NO_HZ */
123
124static int blimit = 10; /* Maximum callbacks per softirq. */
125static int qhimark = 10000; /* If this many pending, ignore blimit. */
126static int qlowmark = 100; /* Once only this many pending, use blimit. */
127
128module_param(blimit, int, 0);
129module_param(qhimark, int, 0);
130module_param(qlowmark, int, 0);
131
132static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
133static int rcu_pending(int cpu);
134
135/*
136 * Return the number of RCU-sched batches processed thus far for debug & stats.
137 */
138long rcu_batches_completed_sched(void)
139{
140    return rcu_sched_state.completed;
141}
142EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
143
144/*
145 * Return the number of RCU BH batches processed thus far for debug & stats.
146 */
147long rcu_batches_completed_bh(void)
148{
149    return rcu_bh_state.completed;
150}
151EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
152
153/*
154 * Does the CPU have callbacks ready to be invoked?
155 */
156static int
157cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
158{
159    return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
160}
161
162/*
163 * Does the current CPU require a yet-as-unscheduled grace period?
164 */
165static int
166cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
167{
168    return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
169}
170
171/*
172 * Return the root node of the specified rcu_state structure.
173 */
174static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
175{
176    return &rsp->node[0];
177}
178
179/*
180 * Record the specified "completed" value, which is later used to validate
181 * dynticks counter manipulations and CPU-offline checks. Specify
182 * "rsp->completed - 1" to unconditionally invalidate any future dynticks
183 * manipulations and CPU-offline checks. Such invalidation is useful at
184 * the beginning of a grace period.
185 */
186static void dyntick_record_completed(struct rcu_state *rsp, long comp)
187{
188    rsp->dynticks_completed = comp;
189}
190
191#ifdef CONFIG_SMP
192
193/*
194 * Recall the previously recorded value of the completion for dynticks.
195 */
196static long dyntick_recall_completed(struct rcu_state *rsp)
197{
198    return rsp->dynticks_completed;
199}
200
201/*
202 * If the specified CPU is offline, tell the caller that it is in
203 * a quiescent state. Otherwise, whack it with a reschedule IPI.
204 * Grace periods can end up waiting on an offline CPU when that
205 * CPU is in the process of coming online -- it will be added to the
206 * rcu_node bitmasks before it actually makes it online. The same thing
207 * can happen while a CPU is in the process of coming online. Because this
208 * race is quite rare, we check for it after detecting that the grace
209 * period has been delayed rather than checking each and every CPU
210 * each and every time we start a new grace period.
211 */
212static int rcu_implicit_offline_qs(struct rcu_data *rdp)
213{
214    /*
215     * If the CPU is offline, it is in a quiescent state. We can
216     * trust its state not to change because interrupts are disabled.
217     */
218    if (cpu_is_offline(rdp->cpu)) {
219        rdp->offline_fqs++;
220        return 1;
221    }
222
223    /* If preemptable RCU, no point in sending reschedule IPI. */
224    if (rdp->preemptable)
225        return 0;
226
227    /* The CPU is online, so send it a reschedule IPI. */
228    if (rdp->cpu != smp_processor_id())
229        smp_send_reschedule(rdp->cpu);
230    else
231        set_need_resched();
232    rdp->resched_ipi++;
233    return 0;
234}
235
236#endif /* #ifdef CONFIG_SMP */
237
238#ifdef CONFIG_NO_HZ
239
240/**
241 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
242 *
243 * Enter nohz mode, in other words, -leave- the mode in which RCU
244 * read-side critical sections can occur. (Though RCU read-side
245 * critical sections can occur in irq handlers in nohz mode, a possibility
246 * handled by rcu_irq_enter() and rcu_irq_exit()).
247 */
248void rcu_enter_nohz(void)
249{
250    unsigned long flags;
251    struct rcu_dynticks *rdtp;
252
253    smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
254    local_irq_save(flags);
255    rdtp = &__get_cpu_var(rcu_dynticks);
256    rdtp->dynticks++;
257    rdtp->dynticks_nesting--;
258    WARN_ON_ONCE(rdtp->dynticks & 0x1);
259    local_irq_restore(flags);
260}
261
262/*
263 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
264 *
265 * Exit nohz mode, in other words, -enter- the mode in which RCU
266 * read-side critical sections normally occur.
267 */
268void rcu_exit_nohz(void)
269{
270    unsigned long flags;
271    struct rcu_dynticks *rdtp;
272
273    local_irq_save(flags);
274    rdtp = &__get_cpu_var(rcu_dynticks);
275    rdtp->dynticks++;
276    rdtp->dynticks_nesting++;
277    WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
278    local_irq_restore(flags);
279    smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
280}
281
282/**
283 * rcu_nmi_enter - inform RCU of entry to NMI context
284 *
285 * If the CPU was idle with dynamic ticks active, and there is no
286 * irq handler running, this updates rdtp->dynticks_nmi to let the
287 * RCU grace-period handling know that the CPU is active.
288 */
289void rcu_nmi_enter(void)
290{
291    struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
292
293    if (rdtp->dynticks & 0x1)
294        return;
295    rdtp->dynticks_nmi++;
296    WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
297    smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
298}
299
300/**
301 * rcu_nmi_exit - inform RCU of exit from NMI context
302 *
303 * If the CPU was idle with dynamic ticks active, and there is no
304 * irq handler running, this updates rdtp->dynticks_nmi to let the
305 * RCU grace-period handling know that the CPU is no longer active.
306 */
307void rcu_nmi_exit(void)
308{
309    struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
310
311    if (rdtp->dynticks & 0x1)
312        return;
313    smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
314    rdtp->dynticks_nmi++;
315    WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
316}
317
318/**
319 * rcu_irq_enter - inform RCU of entry to hard irq context
320 *
321 * If the CPU was idle with dynamic ticks active, this updates the
322 * rdtp->dynticks to let the RCU handling know that the CPU is active.
323 */
324void rcu_irq_enter(void)
325{
326    struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
327
328    if (rdtp->dynticks_nesting++)
329        return;
330    rdtp->dynticks++;
331    WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
332    smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
333}
334
335/**
336 * rcu_irq_exit - inform RCU of exit from hard irq context
337 *
338 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
339 * to put let the RCU handling be aware that the CPU is going back to idle
340 * with no ticks.
341 */
342void rcu_irq_exit(void)
343{
344    struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
345
346    if (--rdtp->dynticks_nesting)
347        return;
348    smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
349    rdtp->dynticks++;
350    WARN_ON_ONCE(rdtp->dynticks & 0x1);
351
352    /* If the interrupt queued a callback, get out of dyntick mode. */
353    if (__get_cpu_var(rcu_sched_data).nxtlist ||
354        __get_cpu_var(rcu_bh_data).nxtlist)
355        set_need_resched();
356}
357
358#ifdef CONFIG_SMP
359
360/*
361 * Snapshot the specified CPU's dynticks counter so that we can later
362 * credit them with an implicit quiescent state. Return 1 if this CPU
363 * is in dynticks idle mode, which is an extended quiescent state.
364 */
365static int dyntick_save_progress_counter(struct rcu_data *rdp)
366{
367    int ret;
368    int snap;
369    int snap_nmi;
370
371    snap = rdp->dynticks->dynticks;
372    snap_nmi = rdp->dynticks->dynticks_nmi;
373    smp_mb(); /* Order sampling of snap with end of grace period. */
374    rdp->dynticks_snap = snap;
375    rdp->dynticks_nmi_snap = snap_nmi;
376    ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
377    if (ret)
378        rdp->dynticks_fqs++;
379    return ret;
380}
381
382/*
383 * Return true if the specified CPU has passed through a quiescent
384 * state by virtue of being in or having passed through an dynticks
385 * idle state since the last call to dyntick_save_progress_counter()
386 * for this same CPU.
387 */
388static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
389{
390    long curr;
391    long curr_nmi;
392    long snap;
393    long snap_nmi;
394
395    curr = rdp->dynticks->dynticks;
396    snap = rdp->dynticks_snap;
397    curr_nmi = rdp->dynticks->dynticks_nmi;
398    snap_nmi = rdp->dynticks_nmi_snap;
399    smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
400
401    /*
402     * If the CPU passed through or entered a dynticks idle phase with
403     * no active irq/NMI handlers, then we can safely pretend that the CPU
404     * already acknowledged the request to pass through a quiescent
405     * state. Either way, that CPU cannot possibly be in an RCU
406     * read-side critical section that started before the beginning
407     * of the current RCU grace period.
408     */
409    if ((curr != snap || (curr & 0x1) == 0) &&
410        (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
411        rdp->dynticks_fqs++;
412        return 1;
413    }
414
415    /* Go check for the CPU being offline. */
416    return rcu_implicit_offline_qs(rdp);
417}
418
419#endif /* #ifdef CONFIG_SMP */
420
421#else /* #ifdef CONFIG_NO_HZ */
422
423#ifdef CONFIG_SMP
424
425static int dyntick_save_progress_counter(struct rcu_data *rdp)
426{
427    return 0;
428}
429
430static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
431{
432    return rcu_implicit_offline_qs(rdp);
433}
434
435#endif /* #ifdef CONFIG_SMP */
436
437#endif /* #else #ifdef CONFIG_NO_HZ */
438
439#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
440
441static void record_gp_stall_check_time(struct rcu_state *rsp)
442{
443    rsp->gp_start = jiffies;
444    rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
445}
446
447static void print_other_cpu_stall(struct rcu_state *rsp)
448{
449    int cpu;
450    long delta;
451    unsigned long flags;
452    struct rcu_node *rnp = rcu_get_root(rsp);
453
454    /* Only let one CPU complain about others per time interval. */
455
456    spin_lock_irqsave(&rnp->lock, flags);
457    delta = jiffies - rsp->jiffies_stall;
458    if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
459        spin_unlock_irqrestore(&rnp->lock, flags);
460        return;
461    }
462    rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
463
464    /*
465     * Now rat on any tasks that got kicked up to the root rcu_node
466     * due to CPU offlining.
467     */
468    rcu_print_task_stall(rnp);
469    spin_unlock_irqrestore(&rnp->lock, flags);
470
471    /* OK, time to rat on our buddy... */
472
473    printk(KERN_ERR "INFO: RCU detected CPU stalls:");
474    rcu_for_each_leaf_node(rsp, rnp) {
475        rcu_print_task_stall(rnp);
476        if (rnp->qsmask == 0)
477            continue;
478        for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
479            if (rnp->qsmask & (1UL << cpu))
480                printk(" %d", rnp->grplo + cpu);
481    }
482    printk(" (detected by %d, t=%ld jiffies)\n",
483           smp_processor_id(), (long)(jiffies - rsp->gp_start));
484    trigger_all_cpu_backtrace();
485
486    force_quiescent_state(rsp, 0); /* Kick them all. */
487}
488
489static void print_cpu_stall(struct rcu_state *rsp)
490{
491    unsigned long flags;
492    struct rcu_node *rnp = rcu_get_root(rsp);
493
494    printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
495            smp_processor_id(), jiffies - rsp->gp_start);
496    trigger_all_cpu_backtrace();
497
498    spin_lock_irqsave(&rnp->lock, flags);
499    if ((long)(jiffies - rsp->jiffies_stall) >= 0)
500        rsp->jiffies_stall =
501            jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
502    spin_unlock_irqrestore(&rnp->lock, flags);
503
504    set_need_resched(); /* kick ourselves to get things going. */
505}
506
507static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
508{
509    long delta;
510    struct rcu_node *rnp;
511
512    delta = jiffies - rsp->jiffies_stall;
513    rnp = rdp->mynode;
514    if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
515
516        /* We haven't checked in, so go dump stack. */
517        print_cpu_stall(rsp);
518
519    } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
520
521        /* They had two time units to dump stack, so complain. */
522        print_other_cpu_stall(rsp);
523    }
524}
525
526#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
527
528static void record_gp_stall_check_time(struct rcu_state *rsp)
529{
530}
531
532static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
533{
534}
535
536#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
537
538/*
539 * Update CPU-local rcu_data state to record the newly noticed grace period.
540 * This is used both when we started the grace period and when we notice
541 * that someone else started the grace period. The caller must hold the
542 * ->lock of the leaf rcu_node structure corresponding to the current CPU,
543 * and must have irqs disabled.
544 */
545static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
546{
547    if (rdp->gpnum != rnp->gpnum) {
548        rdp->qs_pending = 1;
549        rdp->passed_quiesc = 0;
550        rdp->gpnum = rnp->gpnum;
551    }
552}
553
554static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
555{
556    unsigned long flags;
557    struct rcu_node *rnp;
558
559    local_irq_save(flags);
560    rnp = rdp->mynode;
561    if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */
562        !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */
563        local_irq_restore(flags);
564        return;
565    }
566    __note_new_gpnum(rsp, rnp, rdp);
567    spin_unlock_irqrestore(&rnp->lock, flags);
568}
569
570/*
571 * Did someone else start a new RCU grace period start since we last
572 * checked? Update local state appropriately if so. Must be called
573 * on the CPU corresponding to rdp.
574 */
575static int
576check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
577{
578    unsigned long flags;
579    int ret = 0;
580
581    local_irq_save(flags);
582    if (rdp->gpnum != rsp->gpnum) {
583        note_new_gpnum(rsp, rdp);
584        ret = 1;
585    }
586    local_irq_restore(flags);
587    return ret;
588}
589
590/*
591 * Advance this CPU's callbacks, but only if the current grace period
592 * has ended. This may be called only from the CPU to whom the rdp
593 * belongs. In addition, the corresponding leaf rcu_node structure's
594 * ->lock must be held by the caller, with irqs disabled.
595 */
596static void
597__rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
598{
599    /* Did another grace period end? */
600    if (rdp->completed != rnp->completed) {
601
602        /* Advance callbacks. No harm if list empty. */
603        rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
604        rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
605        rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
606
607        /* Remember that we saw this grace-period completion. */
608        rdp->completed = rnp->completed;
609    }
610}
611
612/*
613 * Advance this CPU's callbacks, but only if the current grace period
614 * has ended. This may be called only from the CPU to whom the rdp
615 * belongs.
616 */
617static void
618rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
619{
620    unsigned long flags;
621    struct rcu_node *rnp;
622
623    local_irq_save(flags);
624    rnp = rdp->mynode;
625    if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */
626        !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */
627        local_irq_restore(flags);
628        return;
629    }
630    __rcu_process_gp_end(rsp, rnp, rdp);
631    spin_unlock_irqrestore(&rnp->lock, flags);
632}
633
634/*
635 * Do per-CPU grace-period initialization for running CPU. The caller
636 * must hold the lock of the leaf rcu_node structure corresponding to
637 * this CPU.
638 */
639static void
640rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
641{
642    /* Prior grace period ended, so advance callbacks for current CPU. */
643    __rcu_process_gp_end(rsp, rnp, rdp);
644
645    /*
646     * Because this CPU just now started the new grace period, we know
647     * that all of its callbacks will be covered by this upcoming grace
648     * period, even the ones that were registered arbitrarily recently.
649     * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
650     *
651     * Other CPUs cannot be sure exactly when the grace period started.
652     * Therefore, their recently registered callbacks must pass through
653     * an additional RCU_NEXT_READY stage, so that they will be handled
654     * by the next RCU grace period.
655     */
656    rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
657    rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
658
659    /* Set state so that this CPU will detect the next quiescent state. */
660    __note_new_gpnum(rsp, rnp, rdp);
661}
662
663/*
664 * Start a new RCU grace period if warranted, re-initializing the hierarchy
665 * in preparation for detecting the next grace period. The caller must hold
666 * the root node's ->lock, which is released before return. Hard irqs must
667 * be disabled.
668 */
669static void
670rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
671    __releases(rcu_get_root(rsp)->lock)
672{
673    struct rcu_data *rdp = rsp->rda[smp_processor_id()];
674    struct rcu_node *rnp = rcu_get_root(rsp);
675
676    if (!cpu_needs_another_gp(rsp, rdp)) {
677        spin_unlock_irqrestore(&rnp->lock, flags);
678        return;
679    }
680
681    /* Advance to a new grace period and initialize state. */
682    rsp->gpnum++;
683    WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
684    rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
685    rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
686    record_gp_stall_check_time(rsp);
687    dyntick_record_completed(rsp, rsp->completed - 1);
688
689    /* Special-case the common single-level case. */
690    if (NUM_RCU_NODES == 1) {
691        rcu_preempt_check_blocked_tasks(rnp);
692        rnp->qsmask = rnp->qsmaskinit;
693        rnp->gpnum = rsp->gpnum;
694        rnp->completed = rsp->completed;
695        rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
696        rcu_start_gp_per_cpu(rsp, rnp, rdp);
697        spin_unlock_irqrestore(&rnp->lock, flags);
698        return;
699    }
700
701    spin_unlock(&rnp->lock); /* leave irqs disabled. */
702
703
704    /* Exclude any concurrent CPU-hotplug operations. */
705    spin_lock(&rsp->onofflock); /* irqs already disabled. */
706
707    /*
708     * Set the quiescent-state-needed bits in all the rcu_node
709     * structures for all currently online CPUs in breadth-first
710     * order, starting from the root rcu_node structure. This
711     * operation relies on the layout of the hierarchy within the
712     * rsp->node[] array. Note that other CPUs will access only
713     * the leaves of the hierarchy, which still indicate that no
714     * grace period is in progress, at least until the corresponding
715     * leaf node has been initialized. In addition, we have excluded
716     * CPU-hotplug operations.
717     *
718     * Note that the grace period cannot complete until we finish
719     * the initialization process, as there will be at least one
720     * qsmask bit set in the root node until that time, namely the
721     * one corresponding to this CPU, due to the fact that we have
722     * irqs disabled.
723     */
724    rcu_for_each_node_breadth_first(rsp, rnp) {
725        spin_lock(&rnp->lock); /* irqs already disabled. */
726        rcu_preempt_check_blocked_tasks(rnp);
727        rnp->qsmask = rnp->qsmaskinit;
728        rnp->gpnum = rsp->gpnum;
729        rnp->completed = rsp->completed;
730        if (rnp == rdp->mynode)
731            rcu_start_gp_per_cpu(rsp, rnp, rdp);
732        spin_unlock(&rnp->lock); /* irqs remain disabled. */
733    }
734
735    rnp = rcu_get_root(rsp);
736    spin_lock(&rnp->lock); /* irqs already disabled. */
737    rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
738    spin_unlock(&rnp->lock); /* irqs remain disabled. */
739    spin_unlock_irqrestore(&rsp->onofflock, flags);
740}
741
742/*
743 * Clean up after the prior grace period and let rcu_start_gp() start up
744 * the next grace period if one is needed. Note that the caller must
745 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
746 */
747static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
748    __releases(rcu_get_root(rsp)->lock)
749{
750    WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
751    rsp->completed = rsp->gpnum;
752    rsp->signaled = RCU_GP_IDLE;
753    rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
754}
755
756/*
757 * Similar to cpu_quiet(), for which it is a helper function. Allows
758 * a group of CPUs to be quieted at one go, though all the CPUs in the
759 * group must be represented by the same leaf rcu_node structure.
760 * That structure's lock must be held upon entry, and it is released
761 * before return.
762 */
763static void
764cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
765          unsigned long flags)
766    __releases(rnp->lock)
767{
768    struct rcu_node *rnp_c;
769
770    /* Walk up the rcu_node hierarchy. */
771    for (;;) {
772        if (!(rnp->qsmask & mask)) {
773
774            /* Our bit has already been cleared, so done. */
775            spin_unlock_irqrestore(&rnp->lock, flags);
776            return;
777        }
778        rnp->qsmask &= ~mask;
779        if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
780
781            /* Other bits still set at this level, so done. */
782            spin_unlock_irqrestore(&rnp->lock, flags);
783            return;
784        }
785        mask = rnp->grpmask;
786        if (rnp->parent == NULL) {
787
788            /* No more levels. Exit loop holding root lock. */
789
790            break;
791        }
792        spin_unlock_irqrestore(&rnp->lock, flags);
793        rnp_c = rnp;
794        rnp = rnp->parent;
795        spin_lock_irqsave(&rnp->lock, flags);
796        WARN_ON_ONCE(rnp_c->qsmask);
797    }
798
799    /*
800     * Get here if we are the last CPU to pass through a quiescent
801     * state for this grace period. Invoke cpu_quiet_msk_finish()
802     * to clean up and start the next grace period if one is needed.
803     */
804    cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
805}
806
807/*
808 * Record a quiescent state for the specified CPU, which must either be
809 * the current CPU. The lastcomp argument is used to make sure we are
810 * still in the grace period of interest. We don't want to end the current
811 * grace period based on quiescent states detected in an earlier grace
812 * period!
813 */
814static void
815cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
816{
817    unsigned long flags;
818    unsigned long mask;
819    struct rcu_node *rnp;
820
821    rnp = rdp->mynode;
822    spin_lock_irqsave(&rnp->lock, flags);
823    if (lastcomp != ACCESS_ONCE(rsp->completed)) {
824
825        /*
826         * Someone beat us to it for this grace period, so leave.
827         * The race with GP start is resolved by the fact that we
828         * hold the leaf rcu_node lock, so that the per-CPU bits
829         * cannot yet be initialized -- so we would simply find our
830         * CPU's bit already cleared in cpu_quiet_msk() if this race
831         * occurred.
832         */
833        rdp->passed_quiesc = 0; /* try again later! */
834        spin_unlock_irqrestore(&rnp->lock, flags);
835        return;
836    }
837    mask = rdp->grpmask;
838    if ((rnp->qsmask & mask) == 0) {
839        spin_unlock_irqrestore(&rnp->lock, flags);
840    } else {
841        rdp->qs_pending = 0;
842
843        /*
844         * This GP can't end until cpu checks in, so all of our
845         * callbacks can be processed during the next GP.
846         */
847        rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
848
849        cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
850    }
851}
852
853/*
854 * Check to see if there is a new grace period of which this CPU
855 * is not yet aware, and if so, set up local rcu_data state for it.
856 * Otherwise, see if this CPU has just passed through its first
857 * quiescent state for this grace period, and record that fact if so.
858 */
859static void
860rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
861{
862    /* If there is now a new grace period, record and return. */
863    if (check_for_new_grace_period(rsp, rdp))
864        return;
865
866    /*
867     * Does this CPU still need to do its part for current grace period?
868     * If no, return and let the other CPUs do their part as well.
869     */
870    if (!rdp->qs_pending)
871        return;
872
873    /*
874     * Was there a quiescent state since the beginning of the grace
875     * period? If no, then exit and wait for the next call.
876     */
877    if (!rdp->passed_quiesc)
878        return;
879
880    /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
881    cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
882}
883
884#ifdef CONFIG_HOTPLUG_CPU
885
886/*
887 * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
888 * specified flavor of RCU. The callbacks will be adopted by the next
889 * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
890 * comes first. Because this is invoked from the CPU_DYING notifier,
891 * irqs are already disabled.
892 */
893static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
894{
895    int i;
896    struct rcu_data *rdp = rsp->rda[smp_processor_id()];
897
898    if (rdp->nxtlist == NULL)
899        return; /* irqs disabled, so comparison is stable. */
900    spin_lock(&rsp->onofflock); /* irqs already disabled. */
901    *rsp->orphan_cbs_tail = rdp->nxtlist;
902    rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
903    rdp->nxtlist = NULL;
904    for (i = 0; i < RCU_NEXT_SIZE; i++)
905        rdp->nxttail[i] = &rdp->nxtlist;
906    rsp->orphan_qlen += rdp->qlen;
907    rdp->qlen = 0;
908    spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
909}
910
911/*
912 * Adopt previously orphaned RCU callbacks.
913 */
914static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
915{
916    unsigned long flags;
917    struct rcu_data *rdp;
918
919    spin_lock_irqsave(&rsp->onofflock, flags);
920    rdp = rsp->rda[smp_processor_id()];
921    if (rsp->orphan_cbs_list == NULL) {
922        spin_unlock_irqrestore(&rsp->onofflock, flags);
923        return;
924    }
925    *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
926    rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
927    rdp->qlen += rsp->orphan_qlen;
928    rsp->orphan_cbs_list = NULL;
929    rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
930    rsp->orphan_qlen = 0;
931    spin_unlock_irqrestore(&rsp->onofflock, flags);
932}
933
934/*
935 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
936 * and move all callbacks from the outgoing CPU to the current one.
937 */
938static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
939{
940    unsigned long flags;
941    long lastcomp;
942    unsigned long mask;
943    struct rcu_data *rdp = rsp->rda[cpu];
944    struct rcu_node *rnp;
945
946    /* Exclude any attempts to start a new grace period. */
947    spin_lock_irqsave(&rsp->onofflock, flags);
948
949    /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
950    rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
951    mask = rdp->grpmask; /* rnp->grplo is constant. */
952    do {
953        spin_lock(&rnp->lock); /* irqs already disabled. */
954        rnp->qsmaskinit &= ~mask;
955        if (rnp->qsmaskinit != 0) {
956            spin_unlock(&rnp->lock); /* irqs remain disabled. */
957            break;
958        }
959
960        /*
961         * If there was a task blocking the current grace period,
962         * and if all CPUs have checked in, we need to propagate
963         * the quiescent state up the rcu_node hierarchy. But that
964         * is inconvenient at the moment due to deadlock issues if
965         * this should end the current grace period. So set the
966         * offlined CPU's bit in ->qsmask in order to force the
967         * next force_quiescent_state() invocation to clean up this
968         * mess in a deadlock-free manner.
969         */
970        if (rcu_preempt_offline_tasks(rsp, rnp, rdp) && !rnp->qsmask)
971            rnp->qsmask |= mask;
972
973        mask = rnp->grpmask;
974        spin_unlock(&rnp->lock); /* irqs remain disabled. */
975        rnp = rnp->parent;
976    } while (rnp != NULL);
977    lastcomp = rsp->completed;
978
979    spin_unlock_irqrestore(&rsp->onofflock, flags);
980
981    rcu_adopt_orphan_cbs(rsp);
982}
983
984/*
985 * Remove the specified CPU from the RCU hierarchy and move any pending
986 * callbacks that it might have to the current CPU. This code assumes
987 * that at least one CPU in the system will remain running at all times.
988 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
989 */
990static void rcu_offline_cpu(int cpu)
991{
992    __rcu_offline_cpu(cpu, &rcu_sched_state);
993    __rcu_offline_cpu(cpu, &rcu_bh_state);
994    rcu_preempt_offline_cpu(cpu);
995}
996
997#else /* #ifdef CONFIG_HOTPLUG_CPU */
998
999static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
1000{
1001}
1002
1003static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
1004{
1005}
1006
1007static void rcu_offline_cpu(int cpu)
1008{
1009}
1010
1011#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1012
1013/*
1014 * Invoke any RCU callbacks that have made it to the end of their grace
1015 * period. Thottle as specified by rdp->blimit.
1016 */
1017static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
1018{
1019    unsigned long flags;
1020    struct rcu_head *next, *list, **tail;
1021    int count;
1022
1023    /* If no callbacks are ready, just return.*/
1024    if (!cpu_has_callbacks_ready_to_invoke(rdp))
1025        return;
1026
1027    /*
1028     * Extract the list of ready callbacks, disabling to prevent
1029     * races with call_rcu() from interrupt handlers.
1030     */
1031    local_irq_save(flags);
1032    list = rdp->nxtlist;
1033    rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
1034    *rdp->nxttail[RCU_DONE_TAIL] = NULL;
1035    tail = rdp->nxttail[RCU_DONE_TAIL];
1036    for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
1037        if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
1038            rdp->nxttail[count] = &rdp->nxtlist;
1039    local_irq_restore(flags);
1040
1041    /* Invoke callbacks. */
1042    count = 0;
1043    while (list) {
1044        next = list->next;
1045        prefetch(next);
1046        list->func(list);
1047        list = next;
1048        if (++count >= rdp->blimit)
1049            break;
1050    }
1051
1052    local_irq_save(flags);
1053
1054    /* Update count, and requeue any remaining callbacks. */
1055    rdp->qlen -= count;
1056    if (list != NULL) {
1057        *tail = rdp->nxtlist;
1058        rdp->nxtlist = list;
1059        for (count = 0; count < RCU_NEXT_SIZE; count++)
1060            if (&rdp->nxtlist == rdp->nxttail[count])
1061                rdp->nxttail[count] = tail;
1062            else
1063                break;
1064    }
1065
1066    /* Reinstate batch limit if we have worked down the excess. */
1067    if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1068        rdp->blimit = blimit;
1069
1070    /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1071    if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
1072        rdp->qlen_last_fqs_check = 0;
1073        rdp->n_force_qs_snap = rsp->n_force_qs;
1074    } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
1075        rdp->qlen_last_fqs_check = rdp->qlen;
1076
1077    local_irq_restore(flags);
1078
1079    /* Re-raise the RCU softirq if there are callbacks remaining. */
1080    if (cpu_has_callbacks_ready_to_invoke(rdp))
1081        raise_softirq(RCU_SOFTIRQ);
1082}
1083
1084/*
1085 * Check to see if this CPU is in a non-context-switch quiescent state
1086 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1087 * Also schedule the RCU softirq handler.
1088 *
1089 * This function must be called with hardirqs disabled. It is normally
1090 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1091 * false, there is no point in invoking rcu_check_callbacks().
1092 */
1093void rcu_check_callbacks(int cpu, int user)
1094{
1095    if (!rcu_pending(cpu))
1096        return; /* if nothing for RCU to do. */
1097    if (user ||
1098        (idle_cpu(cpu) && rcu_scheduler_active &&
1099         !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1100
1101        /*
1102         * Get here if this CPU took its interrupt from user
1103         * mode or from the idle loop, and if this is not a
1104         * nested interrupt. In this case, the CPU is in
1105         * a quiescent state, so note it.
1106         *
1107         * No memory barrier is required here because both
1108         * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1109         * variables that other CPUs neither access nor modify,
1110         * at least not while the corresponding CPU is online.
1111         */
1112
1113        rcu_sched_qs(cpu);
1114        rcu_bh_qs(cpu);
1115
1116    } else if (!in_softirq()) {
1117
1118        /*
1119         * Get here if this CPU did not take its interrupt from
1120         * softirq, in other words, if it is not interrupting
1121         * a rcu_bh read-side critical section. This is an _bh
1122         * critical section, so note it.
1123         */
1124
1125        rcu_bh_qs(cpu);
1126    }
1127    rcu_preempt_check_callbacks(cpu);
1128    raise_softirq(RCU_SOFTIRQ);
1129}
1130
1131#ifdef CONFIG_SMP
1132
1133/*
1134 * Scan the leaf rcu_node structures, processing dyntick state for any that
1135 * have not yet encountered a quiescent state, using the function specified.
1136 * Returns 1 if the current grace period ends while scanning (possibly
1137 * because we made it end).
1138 */
1139static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1140                   int (*f)(struct rcu_data *))
1141{
1142    unsigned long bit;
1143    int cpu;
1144    unsigned long flags;
1145    unsigned long mask;
1146    struct rcu_node *rnp;
1147
1148    rcu_for_each_leaf_node(rsp, rnp) {
1149        mask = 0;
1150        spin_lock_irqsave(&rnp->lock, flags);
1151        if (rsp->completed != lastcomp) {
1152            spin_unlock_irqrestore(&rnp->lock, flags);
1153            return 1;
1154        }
1155        if (rnp->qsmask == 0) {
1156            spin_unlock_irqrestore(&rnp->lock, flags);
1157            continue;
1158        }
1159        cpu = rnp->grplo;
1160        bit = 1;
1161        for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1162            if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1163                mask |= bit;
1164        }
1165        if (mask != 0 && rsp->completed == lastcomp) {
1166
1167            /* cpu_quiet_msk() releases rnp->lock. */
1168            cpu_quiet_msk(mask, rsp, rnp, flags);
1169            continue;
1170        }
1171        spin_unlock_irqrestore(&rnp->lock, flags);
1172    }
1173    return 0;
1174}
1175
1176/*
1177 * Force quiescent states on reluctant CPUs, and also detect which
1178 * CPUs are in dyntick-idle mode.
1179 */
1180static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1181{
1182    unsigned long flags;
1183    long lastcomp;
1184    struct rcu_node *rnp = rcu_get_root(rsp);
1185    u8 signaled;
1186    u8 forcenow;
1187
1188    if (!rcu_gp_in_progress(rsp))
1189        return; /* No grace period in progress, nothing to force. */
1190    if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1191        rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1192        return; /* Someone else is already on the job. */
1193    }
1194    if (relaxed &&
1195        (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1196        goto unlock_ret; /* no emergency and done recently. */
1197    rsp->n_force_qs++;
1198    spin_lock(&rnp->lock);
1199    lastcomp = rsp->completed;
1200    signaled = rsp->signaled;
1201    rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1202    if (lastcomp == rsp->gpnum) {
1203        rsp->n_force_qs_ngp++;
1204        spin_unlock(&rnp->lock);
1205        goto unlock_ret; /* no GP in progress, time updated. */
1206    }
1207    spin_unlock(&rnp->lock);
1208    switch (signaled) {
1209    case RCU_GP_IDLE:
1210    case RCU_GP_INIT:
1211
1212        break; /* grace period idle or initializing, ignore. */
1213
1214    case RCU_SAVE_DYNTICK:
1215
1216        if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1217            break; /* So gcc recognizes the dead code. */
1218
1219        /* Record dyntick-idle state. */
1220        if (rcu_process_dyntick(rsp, lastcomp,
1221                    dyntick_save_progress_counter))
1222            goto unlock_ret;
1223        /* fall into next case. */
1224
1225    case RCU_SAVE_COMPLETED:
1226
1227        /* Update state, record completion counter. */
1228        forcenow = 0;
1229        spin_lock(&rnp->lock);
1230        if (lastcomp == rsp->completed &&
1231            rsp->signaled == signaled) {
1232            rsp->signaled = RCU_FORCE_QS;
1233            dyntick_record_completed(rsp, lastcomp);
1234            forcenow = signaled == RCU_SAVE_COMPLETED;
1235        }
1236        spin_unlock(&rnp->lock);
1237        if (!forcenow)
1238            break;
1239        /* fall into next case. */
1240
1241    case RCU_FORCE_QS:
1242
1243        /* Check dyntick-idle state, send IPI to laggarts. */
1244        if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1245                    rcu_implicit_dynticks_qs))
1246            goto unlock_ret;
1247
1248        /* Leave state in case more forcing is required. */
1249
1250        break;
1251    }
1252unlock_ret:
1253    spin_unlock_irqrestore(&rsp->fqslock, flags);
1254}
1255
1256#else /* #ifdef CONFIG_SMP */
1257
1258static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1259{
1260    set_need_resched();
1261}
1262
1263#endif /* #else #ifdef CONFIG_SMP */
1264
1265/*
1266 * This does the RCU processing work from softirq context for the
1267 * specified rcu_state and rcu_data structures. This may be called
1268 * only from the CPU to whom the rdp belongs.
1269 */
1270static void
1271__rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1272{
1273    unsigned long flags;
1274
1275    WARN_ON_ONCE(rdp->beenonline == 0);
1276
1277    /*
1278     * If an RCU GP has gone long enough, go check for dyntick
1279     * idle CPUs and, if needed, send resched IPIs.
1280     */
1281    if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1282        force_quiescent_state(rsp, 1);
1283
1284    /*
1285     * Advance callbacks in response to end of earlier grace
1286     * period that some other CPU ended.
1287     */
1288    rcu_process_gp_end(rsp, rdp);
1289
1290    /* Update RCU state based on any recent quiescent states. */
1291    rcu_check_quiescent_state(rsp, rdp);
1292
1293    /* Does this CPU require a not-yet-started grace period? */
1294    if (cpu_needs_another_gp(rsp, rdp)) {
1295        spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1296        rcu_start_gp(rsp, flags); /* releases above lock */
1297    }
1298
1299    /* If there are callbacks ready, invoke them. */
1300    rcu_do_batch(rsp, rdp);
1301}
1302
1303/*
1304 * Do softirq processing for the current CPU.
1305 */
1306static void rcu_process_callbacks(struct softirq_action *unused)
1307{
1308    /*
1309     * Memory references from any prior RCU read-side critical sections
1310     * executed by the interrupted code must be seen before any RCU
1311     * grace-period manipulations below.
1312     */
1313    smp_mb(); /* See above block comment. */
1314
1315    __rcu_process_callbacks(&rcu_sched_state,
1316                &__get_cpu_var(rcu_sched_data));
1317    __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1318    rcu_preempt_process_callbacks();
1319
1320    /*
1321     * Memory references from any later RCU read-side critical sections
1322     * executed by the interrupted code must be seen after any RCU
1323     * grace-period manipulations above.
1324     */
1325    smp_mb(); /* See above block comment. */
1326}
1327
1328static void
1329__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1330       struct rcu_state *rsp)
1331{
1332    unsigned long flags;
1333    struct rcu_data *rdp;
1334
1335    head->func = func;
1336    head->next = NULL;
1337
1338    smp_mb(); /* Ensure RCU update seen before callback registry. */
1339
1340    /*
1341     * Opportunistically note grace-period endings and beginnings.
1342     * Note that we might see a beginning right after we see an
1343     * end, but never vice versa, since this CPU has to pass through
1344     * a quiescent state betweentimes.
1345     */
1346    local_irq_save(flags);
1347    rdp = rsp->rda[smp_processor_id()];
1348    rcu_process_gp_end(rsp, rdp);
1349    check_for_new_grace_period(rsp, rdp);
1350
1351    /* Add the callback to our list. */
1352    *rdp->nxttail[RCU_NEXT_TAIL] = head;
1353    rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1354
1355    /* Start a new grace period if one not already started. */
1356    if (!rcu_gp_in_progress(rsp)) {
1357        unsigned long nestflag;
1358        struct rcu_node *rnp_root = rcu_get_root(rsp);
1359
1360        spin_lock_irqsave(&rnp_root->lock, nestflag);
1361        rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1362    }
1363
1364    /*
1365     * Force the grace period if too many callbacks or too long waiting.
1366     * Enforce hysteresis, and don't invoke force_quiescent_state()
1367     * if some other CPU has recently done so. Also, don't bother
1368     * invoking force_quiescent_state() if the newly enqueued callback
1369     * is the only one waiting for a grace period to complete.
1370     */
1371    if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
1372        rdp->blimit = LONG_MAX;
1373        if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1374            *rdp->nxttail[RCU_DONE_TAIL] != head)
1375            force_quiescent_state(rsp, 0);
1376        rdp->n_force_qs_snap = rsp->n_force_qs;
1377        rdp->qlen_last_fqs_check = rdp->qlen;
1378    } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1379        force_quiescent_state(rsp, 1);
1380    local_irq_restore(flags);
1381}
1382
1383/*
1384 * Queue an RCU-sched callback for invocation after a grace period.
1385 */
1386void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1387{
1388    __call_rcu(head, func, &rcu_sched_state);
1389}
1390EXPORT_SYMBOL_GPL(call_rcu_sched);
1391
1392/*
1393 * Queue an RCU for invocation after a quicker grace period.
1394 */
1395void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1396{
1397    __call_rcu(head, func, &rcu_bh_state);
1398}
1399EXPORT_SYMBOL_GPL(call_rcu_bh);
1400
1401/*
1402 * Check to see if there is any immediate RCU-related work to be done
1403 * by the current CPU, for the specified type of RCU, returning 1 if so.
1404 * The checks are in order of increasing expense: checks that can be
1405 * carried out against CPU-local state are performed first. However,
1406 * we must check for CPU stalls first, else we might not get a chance.
1407 */
1408static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1409{
1410    rdp->n_rcu_pending++;
1411
1412    /* Check for CPU stalls, if enabled. */
1413    check_cpu_stall(rsp, rdp);
1414
1415    /* Is the RCU core waiting for a quiescent state from this CPU? */
1416    if (rdp->qs_pending) {
1417        rdp->n_rp_qs_pending++;
1418        return 1;
1419    }
1420
1421    /* Does this CPU have callbacks ready to invoke? */
1422    if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1423        rdp->n_rp_cb_ready++;
1424        return 1;
1425    }
1426
1427    /* Has RCU gone idle with this CPU needing another grace period? */
1428    if (cpu_needs_another_gp(rsp, rdp)) {
1429        rdp->n_rp_cpu_needs_gp++;
1430        return 1;
1431    }
1432
1433    /* Has another RCU grace period completed? */
1434    if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1435        rdp->n_rp_gp_completed++;
1436        return 1;
1437    }
1438
1439    /* Has a new RCU grace period started? */
1440    if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1441        rdp->n_rp_gp_started++;
1442        return 1;
1443    }
1444
1445    /* Has an RCU GP gone long enough to send resched IPIs &c? */
1446    if (rcu_gp_in_progress(rsp) &&
1447        ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1448        rdp->n_rp_need_fqs++;
1449        return 1;
1450    }
1451
1452    /* nothing to do */
1453    rdp->n_rp_need_nothing++;
1454    return 0;
1455}
1456
1457/*
1458 * Check to see if there is any immediate RCU-related work to be done
1459 * by the current CPU, returning 1 if so. This function is part of the
1460 * RCU implementation; it is -not- an exported member of the RCU API.
1461 */
1462static int rcu_pending(int cpu)
1463{
1464    return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1465           __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1466           rcu_preempt_pending(cpu);
1467}
1468
1469/*
1470 * Check to see if any future RCU-related work will need to be done
1471 * by the current CPU, even if none need be done immediately, returning
1472 * 1 if so. This function is part of the RCU implementation; it is -not-
1473 * an exported member of the RCU API.
1474 */
1475int rcu_needs_cpu(int cpu)
1476{
1477    /* RCU callbacks either ready or pending? */
1478    return per_cpu(rcu_sched_data, cpu).nxtlist ||
1479           per_cpu(rcu_bh_data, cpu).nxtlist ||
1480           rcu_preempt_needs_cpu(cpu);
1481}
1482
1483static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
1484static atomic_t rcu_barrier_cpu_count;
1485static DEFINE_MUTEX(rcu_barrier_mutex);
1486static struct completion rcu_barrier_completion;
1487
1488static void rcu_barrier_callback(struct rcu_head *notused)
1489{
1490    if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1491        complete(&rcu_barrier_completion);
1492}
1493
1494/*
1495 * Called with preemption disabled, and from cross-cpu IRQ context.
1496 */
1497static void rcu_barrier_func(void *type)
1498{
1499    int cpu = smp_processor_id();
1500    struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
1501    void (*call_rcu_func)(struct rcu_head *head,
1502                  void (*func)(struct rcu_head *head));
1503
1504    atomic_inc(&rcu_barrier_cpu_count);
1505    call_rcu_func = type;
1506    call_rcu_func(head, rcu_barrier_callback);
1507}
1508
1509/*
1510 * Orchestrate the specified type of RCU barrier, waiting for all
1511 * RCU callbacks of the specified type to complete.
1512 */
1513static void _rcu_barrier(struct rcu_state *rsp,
1514             void (*call_rcu_func)(struct rcu_head *head,
1515                           void (*func)(struct rcu_head *head)))
1516{
1517    BUG_ON(in_interrupt());
1518    /* Take mutex to serialize concurrent rcu_barrier() requests. */
1519    mutex_lock(&rcu_barrier_mutex);
1520    init_completion(&rcu_barrier_completion);
1521    /*
1522     * Initialize rcu_barrier_cpu_count to 1, then invoke
1523     * rcu_barrier_func() on each CPU, so that each CPU also has
1524     * incremented rcu_barrier_cpu_count. Only then is it safe to
1525     * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1526     * might complete its grace period before all of the other CPUs
1527     * did their increment, causing this function to return too
1528     * early.
1529     */
1530    atomic_set(&rcu_barrier_cpu_count, 1);
1531    preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
1532    rcu_adopt_orphan_cbs(rsp);
1533    on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
1534    preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
1535    if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1536        complete(&rcu_barrier_completion);
1537    wait_for_completion(&rcu_barrier_completion);
1538    mutex_unlock(&rcu_barrier_mutex);
1539}
1540
1541/**
1542 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1543 */
1544void rcu_barrier_bh(void)
1545{
1546    _rcu_barrier(&rcu_bh_state, call_rcu_bh);
1547}
1548EXPORT_SYMBOL_GPL(rcu_barrier_bh);
1549
1550/**
1551 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1552 */
1553void rcu_barrier_sched(void)
1554{
1555    _rcu_barrier(&rcu_sched_state, call_rcu_sched);
1556}
1557EXPORT_SYMBOL_GPL(rcu_barrier_sched);
1558
1559/*
1560 * Do boot-time initialization of a CPU's per-CPU RCU data.
1561 */
1562static void __init
1563rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1564{
1565    unsigned long flags;
1566    int i;
1567    struct rcu_data *rdp = rsp->rda[cpu];
1568    struct rcu_node *rnp = rcu_get_root(rsp);
1569
1570    /* Set up local state, ensuring consistent view of global state. */
1571    spin_lock_irqsave(&rnp->lock, flags);
1572    rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1573    rdp->nxtlist = NULL;
1574    for (i = 0; i < RCU_NEXT_SIZE; i++)
1575        rdp->nxttail[i] = &rdp->nxtlist;
1576    rdp->qlen = 0;
1577#ifdef CONFIG_NO_HZ
1578    rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1579#endif /* #ifdef CONFIG_NO_HZ */
1580    rdp->cpu = cpu;
1581    spin_unlock_irqrestore(&rnp->lock, flags);
1582}
1583
1584/*
1585 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1586 * offline event can be happening at a given time. Note also that we
1587 * can accept some slop in the rsp->completed access due to the fact
1588 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1589 */
1590static void __cpuinit
1591rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1592{
1593    unsigned long flags;
1594    unsigned long mask;
1595    struct rcu_data *rdp = rsp->rda[cpu];
1596    struct rcu_node *rnp = rcu_get_root(rsp);
1597
1598    /* Set up local state, ensuring consistent view of global state. */
1599    spin_lock_irqsave(&rnp->lock, flags);
1600    rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1601    rdp->qs_pending = 1; /* so set up to respond to current GP. */
1602    rdp->beenonline = 1; /* We have now been online. */
1603    rdp->preemptable = preemptable;
1604    rdp->qlen_last_fqs_check = 0;
1605    rdp->n_force_qs_snap = rsp->n_force_qs;
1606    rdp->blimit = blimit;
1607    spin_unlock(&rnp->lock); /* irqs remain disabled. */
1608
1609    /*
1610     * A new grace period might start here. If so, we won't be part
1611     * of it, but that is OK, as we are currently in a quiescent state.
1612     */
1613
1614    /* Exclude any attempts to start a new GP on large systems. */
1615    spin_lock(&rsp->onofflock); /* irqs already disabled. */
1616
1617    /* Add CPU to rcu_node bitmasks. */
1618    rnp = rdp->mynode;
1619    mask = rdp->grpmask;
1620    do {
1621        /* Exclude any attempts to start a new GP on small systems. */
1622        spin_lock(&rnp->lock); /* irqs already disabled. */
1623        rnp->qsmaskinit |= mask;
1624        mask = rnp->grpmask;
1625        if (rnp == rdp->mynode) {
1626            rdp->gpnum = rnp->completed; /* if GP in progress... */
1627            rdp->completed = rnp->completed;
1628            rdp->passed_quiesc_completed = rnp->completed - 1;
1629        }
1630        spin_unlock(&rnp->lock); /* irqs already disabled. */
1631        rnp = rnp->parent;
1632    } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1633
1634    spin_unlock_irqrestore(&rsp->onofflock, flags);
1635}
1636
1637static void __cpuinit rcu_online_cpu(int cpu)
1638{
1639    rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1640    rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1641    rcu_preempt_init_percpu_data(cpu);
1642}
1643
1644/*
1645 * Handle CPU online/offline notification events.
1646 */
1647int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1648                 unsigned long action, void *hcpu)
1649{
1650    long cpu = (long)hcpu;
1651
1652    switch (action) {
1653    case CPU_UP_PREPARE:
1654    case CPU_UP_PREPARE_FROZEN:
1655        rcu_online_cpu(cpu);
1656        break;
1657    case CPU_DYING:
1658    case CPU_DYING_FROZEN:
1659        /*
1660         * preempt_disable() in _rcu_barrier() prevents stop_machine(),
1661         * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
1662         * returns, all online cpus have queued rcu_barrier_func().
1663         * The dying CPU clears its cpu_online_mask bit and
1664         * moves all of its RCU callbacks to ->orphan_cbs_list
1665         * in the context of stop_machine(), so subsequent calls
1666         * to _rcu_barrier() will adopt these callbacks and only
1667         * then queue rcu_barrier_func() on all remaining CPUs.
1668         */
1669        rcu_send_cbs_to_orphanage(&rcu_bh_state);
1670        rcu_send_cbs_to_orphanage(&rcu_sched_state);
1671        rcu_preempt_send_cbs_to_orphanage();
1672        break;
1673    case CPU_DEAD:
1674    case CPU_DEAD_FROZEN:
1675    case CPU_UP_CANCELED:
1676    case CPU_UP_CANCELED_FROZEN:
1677        rcu_offline_cpu(cpu);
1678        break;
1679    default:
1680        break;
1681    }
1682    return NOTIFY_OK;
1683}
1684
1685/*
1686 * Compute the per-level fanout, either using the exact fanout specified
1687 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1688 */
1689#ifdef CONFIG_RCU_FANOUT_EXACT
1690static void __init rcu_init_levelspread(struct rcu_state *rsp)
1691{
1692    int i;
1693
1694    for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1695        rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1696}
1697#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1698static void __init rcu_init_levelspread(struct rcu_state *rsp)
1699{
1700    int ccur;
1701    int cprv;
1702    int i;
1703
1704    cprv = NR_CPUS;
1705    for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1706        ccur = rsp->levelcnt[i];
1707        rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1708        cprv = ccur;
1709    }
1710}
1711#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1712
1713/*
1714 * Helper function for rcu_init() that initializes one rcu_state structure.
1715 */
1716static void __init rcu_init_one(struct rcu_state *rsp)
1717{
1718    int cpustride = 1;
1719    int i;
1720    int j;
1721    struct rcu_node *rnp;
1722
1723    /* Initialize the level-tracking arrays. */
1724
1725    for (i = 1; i < NUM_RCU_LVLS; i++)
1726        rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1727    rcu_init_levelspread(rsp);
1728
1729    /* Initialize the elements themselves, starting from the leaves. */
1730
1731    for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1732        cpustride *= rsp->levelspread[i];
1733        rnp = rsp->level[i];
1734        for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1735            if (rnp != rcu_get_root(rsp))
1736                spin_lock_init(&rnp->lock);
1737            rnp->gpnum = 0;
1738            rnp->qsmask = 0;
1739            rnp->qsmaskinit = 0;
1740            rnp->grplo = j * cpustride;
1741            rnp->grphi = (j + 1) * cpustride - 1;
1742            if (rnp->grphi >= NR_CPUS)
1743                rnp->grphi = NR_CPUS - 1;
1744            if (i == 0) {
1745                rnp->grpnum = 0;
1746                rnp->grpmask = 0;
1747                rnp->parent = NULL;
1748            } else {
1749                rnp->grpnum = j % rsp->levelspread[i - 1];
1750                rnp->grpmask = 1UL << rnp->grpnum;
1751                rnp->parent = rsp->level[i - 1] +
1752                          j / rsp->levelspread[i - 1];
1753            }
1754            rnp->level = i;
1755            INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1756            INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1757        }
1758    }
1759    spin_lock_init(&rcu_get_root(rsp)->lock);
1760}
1761
1762/*
1763 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1764 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1765 * structure.
1766 */
1767#define RCU_INIT_FLAVOR(rsp, rcu_data) \
1768do { \
1769    int i; \
1770    int j; \
1771    struct rcu_node *rnp; \
1772    \
1773    rcu_init_one(rsp); \
1774    rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1775    j = 0; \
1776    for_each_possible_cpu(i) { \
1777        if (i > rnp[j].grphi) \
1778            j++; \
1779        per_cpu(rcu_data, i).mynode = &rnp[j]; \
1780        (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1781        rcu_boot_init_percpu_data(i, rsp); \
1782    } \
1783} while (0)
1784
1785void __init __rcu_init(void)
1786{
1787    rcu_bootup_announce();
1788#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1789    printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1790#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1791    RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1792    RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1793    __rcu_init_preempt();
1794    open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1795}
1796
1797#include "rcutree_plugin.h"
1798

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive