blob: fe85600ba8c2bf628264fe33a3040982d95f2354 [file] [log] [blame]
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001/*
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 -
Paul E. McKenneya71fca52009-09-18 10:28:19 -070028 * Documentation/RCU
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010029 */
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>
Ingo Molnarc1dc0b92009-08-02 11:28:21 +020038#include <linux/nmi.h>
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010039#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>
Paul E. McKenneybbad9372010-04-02 16:17:17 -070049#include <linux/kernel_stat.h>
Paul E. McKenneya26ac242011-01-12 14:10:23 -080050#include <linux/wait.h>
51#include <linux/kthread.h>
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010052
Paul E. McKenney9f77da92009-08-22 13:56:45 -070053#include "rcutree.h"
54
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010055/* Data structures. */
56
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -080057static struct lock_class_key rcu_node_class[NUM_RCU_LVLS];
Peter Zijlstra88b91c72009-10-26 10:24:31 -070058
Paul E. McKenney4300aa62010-04-13 16:18:22 -070059#define RCU_STATE_INITIALIZER(structname) { \
60 .level = { &structname.node[0] }, \
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010061 .levelcnt = { \
62 NUM_RCU_LVL_0, /* root of hierarchy. */ \
63 NUM_RCU_LVL_1, \
64 NUM_RCU_LVL_2, \
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080065 NUM_RCU_LVL_3, \
66 NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010067 }, \
Paul E. McKenney83f5b012009-10-28 08:14:49 -070068 .signaled = RCU_GP_IDLE, \
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010069 .gpnum = -300, \
70 .completed = -300, \
Paul E. McKenney4300aa62010-04-13 16:18:22 -070071 .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname.onofflock), \
Paul E. McKenney4300aa62010-04-13 16:18:22 -070072 .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname.fqslock), \
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010073 .n_force_qs = 0, \
74 .n_force_qs_ngp = 0, \
Paul E. McKenney4300aa62010-04-13 16:18:22 -070075 .name = #structname, \
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010076}
77
Paul E. McKenneyd6714c22009-08-22 13:56:46 -070078struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
79DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +010080
Ingo Molnar6258c4f2009-03-25 16:42:24 +010081struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
82DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
Ingo Molnarb1f77b02009-03-13 03:20:49 +010083
Paul E. McKenney27f4d282011-02-07 12:47:15 -080084static struct rcu_state *rcu_state;
85
Paul E. McKenneybbad9372010-04-02 16:17:17 -070086int rcu_scheduler_active __read_mostly;
87EXPORT_SYMBOL_GPL(rcu_scheduler_active);
88
Ingo Molnarb1f77b02009-03-13 03:20:49 +010089/*
Paul E. McKenneya26ac242011-01-12 14:10:23 -080090 * Control variables for per-CPU and per-rcu_node kthreads. These
91 * handle all flavors of RCU.
92 */
93static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
Paul E. McKenneyd71df902011-03-29 17:48:28 -070094DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -070095DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu);
Paul E. McKenney5ece5ba2011-04-22 18:08:51 -070096DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
Paul E. McKenneya26ac242011-01-12 14:10:23 -080097static DEFINE_PER_CPU(wait_queue_head_t, rcu_cpu_wq);
Paul E. McKenneyd71df902011-03-29 17:48:28 -070098DEFINE_PER_CPU(char, rcu_cpu_has_work);
Paul E. McKenneya26ac242011-01-12 14:10:23 -080099static char rcu_kthreads_spawnable;
100
Paul E. McKenney0f962a52011-04-14 12:13:53 -0700101static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800102static void invoke_rcu_cpu_kthread(void);
Paul E. McKenneya26ac242011-01-12 14:10:23 -0800103
104#define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */
105
106/*
Paul E. McKenney4a298652011-04-03 21:33:51 -0700107 * Track the rcutorture test sequence number and the update version
108 * number within a given test. The rcutorture_testseq is incremented
109 * on every rcutorture module load and unload, so has an odd value
110 * when a test is running. The rcutorture_vernum is set to zero
111 * when rcutorture starts and is incremented on each rcutorture update.
112 * These variables enable correlating rcutorture output with the
113 * RCU tracing information.
114 */
115unsigned long rcutorture_testseq;
116unsigned long rcutorture_vernum;
117
118/*
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700119 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
120 * permit this function to be invoked without holding the root rcu_node
121 * structure's ->lock, but of course results can be subject to change.
122 */
123static int rcu_gp_in_progress(struct rcu_state *rsp)
124{
125 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
126}
127
128/*
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700129 * Note a quiescent state. Because we do not need to know
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100130 * how many quiescent states passed, just if there was at least
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700131 * one since the start of the grace period, this just sets a flag.
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100132 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700133void rcu_sched_qs(int cpu)
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100134{
Paul E. McKenney25502a62010-04-01 17:37:01 -0700135 struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700136
Paul E. McKenneyc64ac3c2009-11-10 13:37:22 -0800137 rdp->passed_quiesc_completed = rdp->gpnum - 1;
Paul E. McKenneyc3422be2009-09-13 09:15:10 -0700138 barrier();
139 rdp->passed_quiesc = 1;
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100140}
141
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700142void rcu_bh_qs(int cpu)
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100143{
Paul E. McKenney25502a62010-04-01 17:37:01 -0700144 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700145
Paul E. McKenneyc64ac3c2009-11-10 13:37:22 -0800146 rdp->passed_quiesc_completed = rdp->gpnum - 1;
Paul E. McKenneyc3422be2009-09-13 09:15:10 -0700147 barrier();
148 rdp->passed_quiesc = 1;
Ingo Molnarb1f77b02009-03-13 03:20:49 +0100149}
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100150
Paul E. McKenney25502a62010-04-01 17:37:01 -0700151/*
152 * Note a context switch. This is a quiescent state for RCU-sched,
153 * and requires special handling for preemptible RCU.
154 */
155void rcu_note_context_switch(int cpu)
156{
157 rcu_sched_qs(cpu);
158 rcu_preempt_note_context_switch(cpu);
159}
160
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100161#ifdef CONFIG_NO_HZ
Paul E. McKenney90a4d2c2009-01-04 11:41:11 -0800162DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
163 .dynticks_nesting = 1,
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700164 .dynticks = ATOMIC_INIT(1),
Paul E. McKenney90a4d2c2009-01-04 11:41:11 -0800165};
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100166#endif /* #ifdef CONFIG_NO_HZ */
167
168static int blimit = 10; /* Maximum callbacks per softirq. */
169static int qhimark = 10000; /* If this many pending, ignore blimit. */
170static int qlowmark = 100; /* Once only this many pending, use blimit. */
171
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700172module_param(blimit, int, 0);
173module_param(qhimark, int, 0);
174module_param(qlowmark, int, 0);
175
Paul E. McKenneya00e0d712011-02-08 17:14:39 -0800176int rcu_cpu_stall_suppress __read_mostly;
Paul E. McKenneyf2e0dd72010-07-14 14:38:30 -0700177module_param(rcu_cpu_stall_suppress, int, 0644);
Paul E. McKenney742734e2010-06-30 11:43:52 -0700178
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100179static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
Paul E. McKenneya1572292009-08-22 13:56:51 -0700180static int rcu_pending(int cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100181
182/*
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700183 * Return the number of RCU-sched batches processed thus far for debug & stats.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100184 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700185long rcu_batches_completed_sched(void)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100186{
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700187 return rcu_sched_state.completed;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100188}
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700189EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100190
191/*
192 * Return the number of RCU BH batches processed thus far for debug & stats.
193 */
194long rcu_batches_completed_bh(void)
195{
196 return rcu_bh_state.completed;
197}
198EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
199
200/*
Paul E. McKenneybf66f182010-01-04 15:09:10 -0800201 * Force a quiescent state for RCU BH.
202 */
203void rcu_bh_force_quiescent_state(void)
204{
205 force_quiescent_state(&rcu_bh_state, 0);
206}
207EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
208
209/*
Paul E. McKenney4a298652011-04-03 21:33:51 -0700210 * Record the number of times rcutorture tests have been initiated and
211 * terminated. This information allows the debugfs tracing stats to be
212 * correlated to the rcutorture messages, even when the rcutorture module
213 * is being repeatedly loaded and unloaded. In other words, we cannot
214 * store this state in rcutorture itself.
215 */
216void rcutorture_record_test_transition(void)
217{
218 rcutorture_testseq++;
219 rcutorture_vernum = 0;
220}
221EXPORT_SYMBOL_GPL(rcutorture_record_test_transition);
222
223/*
224 * Record the number of writer passes through the current rcutorture test.
225 * This is also used to correlate debugfs tracing stats with the rcutorture
226 * messages.
227 */
228void rcutorture_record_progress(unsigned long vernum)
229{
230 rcutorture_vernum++;
231}
232EXPORT_SYMBOL_GPL(rcutorture_record_progress);
233
234/*
Paul E. McKenneybf66f182010-01-04 15:09:10 -0800235 * Force a quiescent state for RCU-sched.
236 */
237void rcu_sched_force_quiescent_state(void)
238{
239 force_quiescent_state(&rcu_sched_state, 0);
240}
241EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);
242
243/*
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100244 * Does the CPU have callbacks ready to be invoked?
245 */
246static int
247cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
248{
249 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
250}
251
252/*
253 * Does the current CPU require a yet-as-unscheduled grace period?
254 */
255static int
256cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
257{
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700258 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100259}
260
261/*
262 * Return the root node of the specified rcu_state structure.
263 */
264static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
265{
266 return &rsp->node[0];
267}
268
269#ifdef CONFIG_SMP
270
271/*
272 * If the specified CPU is offline, tell the caller that it is in
273 * a quiescent state. Otherwise, whack it with a reschedule IPI.
274 * Grace periods can end up waiting on an offline CPU when that
275 * CPU is in the process of coming online -- it will be added to the
276 * rcu_node bitmasks before it actually makes it online. The same thing
277 * can happen while a CPU is in the process of coming online. Because this
278 * race is quite rare, we check for it after detecting that the grace
279 * period has been delayed rather than checking each and every CPU
280 * each and every time we start a new grace period.
281 */
282static int rcu_implicit_offline_qs(struct rcu_data *rdp)
283{
284 /*
285 * If the CPU is offline, it is in a quiescent state. We can
286 * trust its state not to change because interrupts are disabled.
287 */
288 if (cpu_is_offline(rdp->cpu)) {
289 rdp->offline_fqs++;
290 return 1;
291 }
292
Paul E. McKenney6cc68792011-03-02 13:15:15 -0800293 /* If preemptible RCU, no point in sending reschedule IPI. */
294 if (rdp->preemptible)
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700295 return 0;
296
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100297 /* The CPU is online, so send it a reschedule IPI. */
298 if (rdp->cpu != smp_processor_id())
299 smp_send_reschedule(rdp->cpu);
300 else
301 set_need_resched();
302 rdp->resched_ipi++;
303 return 0;
304}
305
306#endif /* #ifdef CONFIG_SMP */
307
308#ifdef CONFIG_NO_HZ
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100309
310/**
311 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
312 *
313 * Enter nohz mode, in other words, -leave- the mode in which RCU
314 * read-side critical sections can occur. (Though RCU read-side
315 * critical sections can occur in irq handlers in nohz mode, a possibility
316 * handled by rcu_irq_enter() and rcu_irq_exit()).
317 */
318void rcu_enter_nohz(void)
319{
320 unsigned long flags;
321 struct rcu_dynticks *rdtp;
322
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100323 local_irq_save(flags);
324 rdtp = &__get_cpu_var(rcu_dynticks);
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700325 if (--rdtp->dynticks_nesting) {
326 local_irq_restore(flags);
327 return;
328 }
329 /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
330 smp_mb__before_atomic_inc(); /* See above. */
331 atomic_inc(&rdtp->dynticks);
332 smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */
333 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100334 local_irq_restore(flags);
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700335
336 /* If the interrupt queued a callback, get out of dyntick mode. */
337 if (in_irq() &&
338 (__get_cpu_var(rcu_sched_data).nxtlist ||
339 __get_cpu_var(rcu_bh_data).nxtlist ||
340 rcu_preempt_needs_cpu(smp_processor_id())))
341 set_need_resched();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100342}
343
344/*
345 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
346 *
347 * Exit nohz mode, in other words, -enter- the mode in which RCU
348 * read-side critical sections normally occur.
349 */
350void rcu_exit_nohz(void)
351{
352 unsigned long flags;
353 struct rcu_dynticks *rdtp;
354
355 local_irq_save(flags);
356 rdtp = &__get_cpu_var(rcu_dynticks);
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700357 if (rdtp->dynticks_nesting++) {
358 local_irq_restore(flags);
359 return;
360 }
361 smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */
362 atomic_inc(&rdtp->dynticks);
363 /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
364 smp_mb__after_atomic_inc(); /* See above. */
365 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100366 local_irq_restore(flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100367}
368
369/**
370 * rcu_nmi_enter - inform RCU of entry to NMI context
371 *
372 * If the CPU was idle with dynamic ticks active, and there is no
373 * irq handler running, this updates rdtp->dynticks_nmi to let the
374 * RCU grace-period handling know that the CPU is active.
375 */
376void rcu_nmi_enter(void)
377{
378 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
379
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700380 if (rdtp->dynticks_nmi_nesting == 0 &&
381 (atomic_read(&rdtp->dynticks) & 0x1))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100382 return;
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700383 rdtp->dynticks_nmi_nesting++;
384 smp_mb__before_atomic_inc(); /* Force delay from prior write. */
385 atomic_inc(&rdtp->dynticks);
386 /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
387 smp_mb__after_atomic_inc(); /* See above. */
388 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100389}
390
391/**
392 * rcu_nmi_exit - inform RCU of exit from NMI context
393 *
394 * If the CPU was idle with dynamic ticks active, and there is no
395 * irq handler running, this updates rdtp->dynticks_nmi to let the
396 * RCU grace-period handling know that the CPU is no longer active.
397 */
398void rcu_nmi_exit(void)
399{
400 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
401
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700402 if (rdtp->dynticks_nmi_nesting == 0 ||
403 --rdtp->dynticks_nmi_nesting != 0)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100404 return;
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700405 /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
406 smp_mb__before_atomic_inc(); /* See above. */
407 atomic_inc(&rdtp->dynticks);
408 smp_mb__after_atomic_inc(); /* Force delay to next write. */
409 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100410}
411
412/**
413 * rcu_irq_enter - inform RCU of entry to hard irq context
414 *
415 * If the CPU was idle with dynamic ticks active, this updates the
416 * rdtp->dynticks to let the RCU handling know that the CPU is active.
417 */
418void rcu_irq_enter(void)
419{
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700420 rcu_exit_nohz();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100421}
422
423/**
424 * rcu_irq_exit - inform RCU of exit from hard irq context
425 *
426 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
427 * to put let the RCU handling be aware that the CPU is going back to idle
428 * with no ticks.
429 */
430void rcu_irq_exit(void)
431{
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700432 rcu_enter_nohz();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100433}
434
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100435#ifdef CONFIG_SMP
436
437/*
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100438 * Snapshot the specified CPU's dynticks counter so that we can later
439 * credit them with an implicit quiescent state. Return 1 if this CPU
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700440 * is in dynticks idle mode, which is an extended quiescent state.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100441 */
442static int dyntick_save_progress_counter(struct rcu_data *rdp)
443{
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700444 rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
445 return 0;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100446}
447
448/*
449 * Return true if the specified CPU has passed through a quiescent
450 * state by virtue of being in or having passed through an dynticks
451 * idle state since the last call to dyntick_save_progress_counter()
452 * for this same CPU.
453 */
454static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
455{
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700456 unsigned long curr;
457 unsigned long snap;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100458
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700459 curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks);
460 snap = (unsigned long)rdp->dynticks_snap;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100461
462 /*
463 * If the CPU passed through or entered a dynticks idle phase with
464 * no active irq/NMI handlers, then we can safely pretend that the CPU
465 * already acknowledged the request to pass through a quiescent
466 * state. Either way, that CPU cannot possibly be in an RCU
467 * read-side critical section that started before the beginning
468 * of the current RCU grace period.
469 */
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700470 if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100471 rdp->dynticks_fqs++;
472 return 1;
473 }
474
475 /* Go check for the CPU being offline. */
476 return rcu_implicit_offline_qs(rdp);
477}
478
479#endif /* #ifdef CONFIG_SMP */
480
481#else /* #ifdef CONFIG_NO_HZ */
482
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100483#ifdef CONFIG_SMP
484
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100485static int dyntick_save_progress_counter(struct rcu_data *rdp)
486{
487 return 0;
488}
489
490static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
491{
492 return rcu_implicit_offline_qs(rdp);
493}
494
495#endif /* #ifdef CONFIG_SMP */
496
497#endif /* #else #ifdef CONFIG_NO_HZ */
498
Paul E. McKenney742734e2010-06-30 11:43:52 -0700499int rcu_cpu_stall_suppress __read_mostly;
Paul E. McKenneyc68de202010-04-15 10:12:40 -0700500
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100501static void record_gp_stall_check_time(struct rcu_state *rsp)
502{
503 rsp->gp_start = jiffies;
504 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
505}
506
507static void print_other_cpu_stall(struct rcu_state *rsp)
508{
509 int cpu;
510 long delta;
511 unsigned long flags;
512 struct rcu_node *rnp = rcu_get_root(rsp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100513
514 /* Only let one CPU complain about others per time interval. */
515
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800516 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100517 delta = jiffies - rsp->jiffies_stall;
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700518 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800519 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100520 return;
521 }
522 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700523
524 /*
525 * Now rat on any tasks that got kicked up to the root rcu_node
526 * due to CPU offlining.
527 */
528 rcu_print_task_stall(rnp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800529 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100530
Paul E. McKenney8cdd32a2010-08-09 14:23:03 -0700531 /*
532 * OK, time to rat on our buddy...
533 * See Documentation/RCU/stallwarn.txt for info on how to debug
534 * RCU CPU stall warnings.
535 */
Paul E. McKenney4300aa62010-04-13 16:18:22 -0700536 printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {",
537 rsp->name);
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700538 rcu_for_each_leaf_node(rsp, rnp) {
Paul E. McKenney3acd9eb2010-02-22 17:05:03 -0800539 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700540 rcu_print_task_stall(rnp);
Paul E. McKenney3acd9eb2010-02-22 17:05:03 -0800541 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700542 if (rnp->qsmask == 0)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100543 continue;
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700544 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
545 if (rnp->qsmask & (1UL << cpu))
546 printk(" %d", rnp->grplo + cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100547 }
Paul E. McKenney4300aa62010-04-13 16:18:22 -0700548 printk("} (detected by %d, t=%ld jiffies)\n",
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100549 smp_processor_id(), (long)(jiffies - rsp->gp_start));
Ingo Molnarc1dc0b92009-08-02 11:28:21 +0200550 trigger_all_cpu_backtrace();
551
Paul E. McKenney1ed509a2010-02-22 17:05:05 -0800552 /* If so configured, complain about tasks blocking the grace period. */
553
554 rcu_print_detail_task_stall(rsp);
555
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100556 force_quiescent_state(rsp, 0); /* Kick them all. */
557}
558
559static void print_cpu_stall(struct rcu_state *rsp)
560{
561 unsigned long flags;
562 struct rcu_node *rnp = rcu_get_root(rsp);
563
Paul E. McKenney8cdd32a2010-08-09 14:23:03 -0700564 /*
565 * OK, time to rat on ourselves...
566 * See Documentation/RCU/stallwarn.txt for info on how to debug
567 * RCU CPU stall warnings.
568 */
Paul E. McKenney4300aa62010-04-13 16:18:22 -0700569 printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n",
570 rsp->name, smp_processor_id(), jiffies - rsp->gp_start);
Ingo Molnarc1dc0b92009-08-02 11:28:21 +0200571 trigger_all_cpu_backtrace();
572
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800573 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800574 if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100575 rsp->jiffies_stall =
576 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800577 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Ingo Molnarc1dc0b92009-08-02 11:28:21 +0200578
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100579 set_need_resched(); /* kick ourselves to get things going. */
580}
581
582static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
583{
584 long delta;
585 struct rcu_node *rnp;
586
Paul E. McKenney742734e2010-06-30 11:43:52 -0700587 if (rcu_cpu_stall_suppress)
Paul E. McKenneyc68de202010-04-15 10:12:40 -0700588 return;
Dongdong Deng4ee0a602010-09-28 16:32:43 +0800589 delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100590 rnp = rdp->mynode;
Dongdong Deng4ee0a602010-09-28 16:32:43 +0800591 if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && delta >= 0) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100592
593 /* We haven't checked in, so go dump stack. */
594 print_cpu_stall(rsp);
595
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700596 } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100597
598 /* They had two time units to dump stack, so complain. */
599 print_other_cpu_stall(rsp);
600 }
601}
602
Paul E. McKenneyc68de202010-04-15 10:12:40 -0700603static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
604{
Paul E. McKenney742734e2010-06-30 11:43:52 -0700605 rcu_cpu_stall_suppress = 1;
Paul E. McKenneyc68de202010-04-15 10:12:40 -0700606 return NOTIFY_DONE;
607}
608
Paul E. McKenney53d84e02010-08-10 14:28:53 -0700609/**
610 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
611 *
612 * Set the stall-warning timeout way off into the future, thus preventing
613 * any RCU CPU stall-warning messages from appearing in the current set of
614 * RCU grace periods.
615 *
616 * The caller must disable hard irqs.
617 */
618void rcu_cpu_stall_reset(void)
619{
620 rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2;
621 rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2;
622 rcu_preempt_stall_reset();
623}
624
Paul E. McKenneyc68de202010-04-15 10:12:40 -0700625static struct notifier_block rcu_panic_block = {
626 .notifier_call = rcu_panic,
627};
628
629static void __init check_cpu_stall_init(void)
630{
631 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
632}
633
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100634/*
635 * Update CPU-local rcu_data state to record the newly noticed grace period.
636 * This is used both when we started the grace period and when we notice
Paul E. McKenney91603062009-11-02 13:52:29 -0800637 * that someone else started the grace period. The caller must hold the
638 * ->lock of the leaf rcu_node structure corresponding to the current CPU,
639 * and must have irqs disabled.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100640 */
Paul E. McKenney91603062009-11-02 13:52:29 -0800641static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
642{
643 if (rdp->gpnum != rnp->gpnum) {
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800644 /*
645 * If the current grace period is waiting for this CPU,
646 * set up to detect a quiescent state, otherwise don't
647 * go looking for one.
648 */
Paul E. McKenney91603062009-11-02 13:52:29 -0800649 rdp->gpnum = rnp->gpnum;
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800650 if (rnp->qsmask & rdp->grpmask) {
651 rdp->qs_pending = 1;
652 rdp->passed_quiesc = 0;
653 } else
654 rdp->qs_pending = 0;
Paul E. McKenney91603062009-11-02 13:52:29 -0800655 }
656}
657
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100658static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
659{
Paul E. McKenney91603062009-11-02 13:52:29 -0800660 unsigned long flags;
661 struct rcu_node *rnp;
662
663 local_irq_save(flags);
664 rnp = rdp->mynode;
665 if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800666 !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
Paul E. McKenney91603062009-11-02 13:52:29 -0800667 local_irq_restore(flags);
668 return;
669 }
670 __note_new_gpnum(rsp, rnp, rdp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800671 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100672}
673
674/*
675 * Did someone else start a new RCU grace period start since we last
676 * checked? Update local state appropriately if so. Must be called
677 * on the CPU corresponding to rdp.
678 */
679static int
680check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
681{
682 unsigned long flags;
683 int ret = 0;
684
685 local_irq_save(flags);
686 if (rdp->gpnum != rsp->gpnum) {
687 note_new_gpnum(rsp, rdp);
688 ret = 1;
689 }
690 local_irq_restore(flags);
691 return ret;
692}
693
694/*
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800695 * Advance this CPU's callbacks, but only if the current grace period
696 * has ended. This may be called only from the CPU to whom the rdp
697 * belongs. In addition, the corresponding leaf rcu_node structure's
698 * ->lock must be held by the caller, with irqs disabled.
699 */
700static void
701__rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
702{
703 /* Did another grace period end? */
704 if (rdp->completed != rnp->completed) {
705
706 /* Advance callbacks. No harm if list empty. */
707 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
708 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
709 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
710
711 /* Remember that we saw this grace-period completion. */
712 rdp->completed = rnp->completed;
Frederic Weisbecker20377f32010-12-10 22:11:10 +0100713
714 /*
Frederic Weisbecker5ff8e6f2010-12-10 22:11:11 +0100715 * If we were in an extended quiescent state, we may have
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800716 * missed some grace periods that others CPUs handled on
Frederic Weisbecker5ff8e6f2010-12-10 22:11:11 +0100717 * our behalf. Catch up with this state to avoid noting
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800718 * spurious new grace periods. If another grace period
719 * has started, then rnp->gpnum will have advanced, so
720 * we will detect this later on.
Frederic Weisbecker5ff8e6f2010-12-10 22:11:11 +0100721 */
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800722 if (ULONG_CMP_LT(rdp->gpnum, rdp->completed))
Frederic Weisbecker5ff8e6f2010-12-10 22:11:11 +0100723 rdp->gpnum = rdp->completed;
724
725 /*
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800726 * If RCU does not need a quiescent state from this CPU,
727 * then make sure that this CPU doesn't go looking for one.
Frederic Weisbecker20377f32010-12-10 22:11:10 +0100728 */
Paul E. McKenney121dfc42010-12-10 15:02:47 -0800729 if ((rnp->qsmask & rdp->grpmask) == 0)
Frederic Weisbecker20377f32010-12-10 22:11:10 +0100730 rdp->qs_pending = 0;
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800731 }
732}
733
734/*
735 * Advance this CPU's callbacks, but only if the current grace period
736 * has ended. This may be called only from the CPU to whom the rdp
737 * belongs.
738 */
739static void
740rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
741{
742 unsigned long flags;
743 struct rcu_node *rnp;
744
745 local_irq_save(flags);
746 rnp = rdp->mynode;
747 if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800748 !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800749 local_irq_restore(flags);
750 return;
751 }
752 __rcu_process_gp_end(rsp, rnp, rdp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800753 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800754}
755
756/*
757 * Do per-CPU grace-period initialization for running CPU. The caller
758 * must hold the lock of the leaf rcu_node structure corresponding to
759 * this CPU.
760 */
761static void
762rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
763{
764 /* Prior grace period ended, so advance callbacks for current CPU. */
765 __rcu_process_gp_end(rsp, rnp, rdp);
766
767 /*
768 * Because this CPU just now started the new grace period, we know
769 * that all of its callbacks will be covered by this upcoming grace
770 * period, even the ones that were registered arbitrarily recently.
771 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
772 *
773 * Other CPUs cannot be sure exactly when the grace period started.
774 * Therefore, their recently registered callbacks must pass through
775 * an additional RCU_NEXT_READY stage, so that they will be handled
776 * by the next RCU grace period.
777 */
778 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
779 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
Paul E. McKenney91603062009-11-02 13:52:29 -0800780
781 /* Set state so that this CPU will detect the next quiescent state. */
782 __note_new_gpnum(rsp, rnp, rdp);
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800783}
784
785/*
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100786 * Start a new RCU grace period if warranted, re-initializing the hierarchy
787 * in preparation for detecting the next grace period. The caller must hold
788 * the root node's ->lock, which is released before return. Hard irqs must
789 * be disabled.
790 */
791static void
792rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
793 __releases(rcu_get_root(rsp)->lock)
794{
Lai Jiangshan394f99a2010-06-28 16:25:04 +0800795 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100796 struct rcu_node *rnp = rcu_get_root(rsp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100797
Paul E. McKenney07079d52010-01-04 15:09:02 -0800798 if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) {
Paul E. McKenney46a1e342010-01-04 15:09:09 -0800799 if (cpu_needs_another_gp(rsp, rdp))
800 rsp->fqs_need_gp = 1;
Paul E. McKenneyb32e9eb2009-11-12 22:35:03 -0800801 if (rnp->completed == rsp->completed) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800802 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenneyb32e9eb2009-11-12 22:35:03 -0800803 return;
804 }
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800805 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenneyb32e9eb2009-11-12 22:35:03 -0800806
807 /*
808 * Propagate new ->completed value to rcu_node structures
809 * so that other CPUs don't have to wait until the start
810 * of the next grace period to process their callbacks.
811 */
812 rcu_for_each_node_breadth_first(rsp, rnp) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800813 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenneyb32e9eb2009-11-12 22:35:03 -0800814 rnp->completed = rsp->completed;
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800815 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenneyb32e9eb2009-11-12 22:35:03 -0800816 }
817 local_irq_restore(flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100818 return;
819 }
820
821 /* Advance to a new grace period and initialize state. */
822 rsp->gpnum++;
Paul E. McKenneyc3422be2009-09-13 09:15:10 -0700823 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100824 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
825 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100826 record_gp_stall_check_time(rsp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100827
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100828 /* Special-case the common single-level case. */
829 if (NUM_RCU_NODES == 1) {
Paul E. McKenneyb0e165c02009-09-13 09:15:09 -0700830 rcu_preempt_check_blocked_tasks(rnp);
Paul E. McKenney28ecd582009-09-18 09:50:17 -0700831 rnp->qsmask = rnp->qsmaskinit;
Paul E. McKenneyde078d82009-09-08 15:54:36 -0700832 rnp->gpnum = rsp->gpnum;
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800833 rnp->completed = rsp->completed;
Paul E. McKenneyc12172c2009-01-04 20:30:06 -0800834 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800835 rcu_start_gp_per_cpu(rsp, rnp, rdp);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800836 rcu_preempt_boost_start_gp(rnp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800837 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100838 return;
839 }
840
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800841 raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100842
843
844 /* Exclude any concurrent CPU-hotplug operations. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800845 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100846
847 /*
Paul E. McKenneyb835db12009-09-08 15:54:37 -0700848 * Set the quiescent-state-needed bits in all the rcu_node
849 * structures for all currently online CPUs in breadth-first
850 * order, starting from the root rcu_node structure. This
851 * operation relies on the layout of the hierarchy within the
852 * rsp->node[] array. Note that other CPUs will access only
853 * the leaves of the hierarchy, which still indicate that no
854 * grace period is in progress, at least until the corresponding
855 * leaf node has been initialized. In addition, we have excluded
856 * CPU-hotplug operations.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100857 *
858 * Note that the grace period cannot complete until we finish
859 * the initialization process, as there will be at least one
860 * qsmask bit set in the root node until that time, namely the
Paul E. McKenneyb835db12009-09-08 15:54:37 -0700861 * one corresponding to this CPU, due to the fact that we have
862 * irqs disabled.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100863 */
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700864 rcu_for_each_node_breadth_first(rsp, rnp) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800865 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenneyb0e165c02009-09-13 09:15:09 -0700866 rcu_preempt_check_blocked_tasks(rnp);
Paul E. McKenney49e29122009-09-18 09:50:19 -0700867 rnp->qsmask = rnp->qsmaskinit;
Paul E. McKenneyde078d82009-09-08 15:54:36 -0700868 rnp->gpnum = rsp->gpnum;
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800869 rnp->completed = rsp->completed;
870 if (rnp == rdp->mynode)
871 rcu_start_gp_per_cpu(rsp, rnp, rdp);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800872 rcu_preempt_boost_start_gp(rnp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800873 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100874 }
875
Paul E. McKenney83f5b012009-10-28 08:14:49 -0700876 rnp = rcu_get_root(rsp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800877 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100878 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800879 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
880 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100881}
882
883/*
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800884 * Report a full set of quiescent states to the specified rcu_state
885 * data structure. This involves cleaning up after the prior grace
886 * period and letting rcu_start_gp() start up the next grace period
887 * if one is needed. Note that the caller must hold rnp->lock, as
888 * required by rcu_start_gp(), which will release it.
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700889 */
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800890static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700891 __releases(rcu_get_root(rsp)->lock)
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700892{
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700893 unsigned long gp_duration;
894
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -0700895 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
Paul E. McKenneye59fb312010-09-07 10:38:22 -0700896
897 /*
898 * Ensure that all grace-period and pre-grace-period activity
899 * is seen before the assignment to rsp->completed.
900 */
901 smp_mb(); /* See above block comment. */
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700902 gp_duration = jiffies - rsp->gp_start;
903 if (gp_duration > rsp->gp_max)
904 rsp->gp_max = gp_duration;
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700905 rsp->completed = rsp->gpnum;
Paul E. McKenney83f5b012009-10-28 08:14:49 -0700906 rsp->signaled = RCU_GP_IDLE;
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700907 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
908}
909
910/*
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800911 * Similar to rcu_report_qs_rdp(), for which it is a helper function.
912 * Allows quiescent states for a group of CPUs to be reported at one go
913 * to the specified rcu_node structure, though all the CPUs in the group
914 * must be represented by the same rcu_node structure (which need not be
915 * a leaf rcu_node structure, though it often will be). That structure's
916 * lock must be held upon entry, and it is released before return.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100917 */
918static void
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800919rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
920 struct rcu_node *rnp, unsigned long flags)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100921 __releases(rnp->lock)
922{
Paul E. McKenney28ecd582009-09-18 09:50:17 -0700923 struct rcu_node *rnp_c;
924
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100925 /* Walk up the rcu_node hierarchy. */
926 for (;;) {
927 if (!(rnp->qsmask & mask)) {
928
929 /* Our bit has already been cleared, so done. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800930 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100931 return;
932 }
933 rnp->qsmask &= ~mask;
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800934 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100935
936 /* Other bits still set at this level, so done. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800937 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100938 return;
939 }
940 mask = rnp->grpmask;
941 if (rnp->parent == NULL) {
942
943 /* No more levels. Exit loop holding root lock. */
944
945 break;
946 }
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800947 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney28ecd582009-09-18 09:50:17 -0700948 rnp_c = rnp;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100949 rnp = rnp->parent;
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800950 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney28ecd582009-09-18 09:50:17 -0700951 WARN_ON_ONCE(rnp_c->qsmask);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100952 }
953
954 /*
955 * Get here if we are the last CPU to pass through a quiescent
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800956 * state for this grace period. Invoke rcu_report_qs_rsp()
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700957 * to clean up and start the next grace period if one is needed.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100958 */
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800959 rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100960}
961
962/*
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800963 * Record a quiescent state for the specified CPU to that CPU's rcu_data
964 * structure. This must be either called from the specified CPU, or
965 * called when the specified CPU is known to be offline (and when it is
966 * also known that no other CPU is concurrently trying to help the offline
967 * CPU). The lastcomp argument is used to make sure we are still in the
968 * grace period of interest. We don't want to end the current grace period
969 * based on quiescent states detected in an earlier grace period!
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100970 */
971static void
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800972rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100973{
974 unsigned long flags;
975 unsigned long mask;
976 struct rcu_node *rnp;
977
978 rnp = rdp->mynode;
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800979 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney560d4bc2009-11-13 19:51:38 -0800980 if (lastcomp != rnp->completed) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100981
982 /*
983 * Someone beat us to it for this grace period, so leave.
984 * The race with GP start is resolved by the fact that we
985 * hold the leaf rcu_node lock, so that the per-CPU bits
986 * cannot yet be initialized -- so we would simply find our
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800987 * CPU's bit already cleared in rcu_report_qs_rnp() if this
988 * race occurred.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100989 */
990 rdp->passed_quiesc = 0; /* try again later! */
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800991 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100992 return;
993 }
994 mask = rdp->grpmask;
995 if ((rnp->qsmask & mask) == 0) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800996 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100997 } else {
998 rdp->qs_pending = 0;
999
1000 /*
1001 * This GP can't end until cpu checks in, so all of our
1002 * callbacks can be processed during the next GP.
1003 */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001004 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1005
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -08001006 rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001007 }
1008}
1009
1010/*
1011 * Check to see if there is a new grace period of which this CPU
1012 * is not yet aware, and if so, set up local rcu_data state for it.
1013 * Otherwise, see if this CPU has just passed through its first
1014 * quiescent state for this grace period, and record that fact if so.
1015 */
1016static void
1017rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
1018{
1019 /* If there is now a new grace period, record and return. */
1020 if (check_for_new_grace_period(rsp, rdp))
1021 return;
1022
1023 /*
1024 * Does this CPU still need to do its part for current grace period?
1025 * If no, return and let the other CPUs do their part as well.
1026 */
1027 if (!rdp->qs_pending)
1028 return;
1029
1030 /*
1031 * Was there a quiescent state since the beginning of the grace
1032 * period? If no, then exit and wait for the next call.
1033 */
1034 if (!rdp->passed_quiesc)
1035 return;
1036
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -08001037 /*
1038 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
1039 * judge of that).
1040 */
1041 rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001042}
1043
1044#ifdef CONFIG_HOTPLUG_CPU
1045
1046/*
Lai Jiangshan29494be2010-10-20 14:13:06 +08001047 * Move a dying CPU's RCU callbacks to online CPU's callback list.
1048 * Synchronization is not required because this function executes
1049 * in stop_machine() context.
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001050 */
Lai Jiangshan29494be2010-10-20 14:13:06 +08001051static void rcu_send_cbs_to_online(struct rcu_state *rsp)
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001052{
1053 int i;
Lai Jiangshan29494be2010-10-20 14:13:06 +08001054 /* current DYING CPU is cleared in the cpu_online_mask */
1055 int receive_cpu = cpumask_any(cpu_online_mask);
Lai Jiangshan394f99a2010-06-28 16:25:04 +08001056 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
Lai Jiangshan29494be2010-10-20 14:13:06 +08001057 struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu);
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001058
1059 if (rdp->nxtlist == NULL)
1060 return; /* irqs disabled, so comparison is stable. */
Lai Jiangshan29494be2010-10-20 14:13:06 +08001061
1062 *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
1063 receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1064 receive_rdp->qlen += rdp->qlen;
1065 receive_rdp->n_cbs_adopted += rdp->qlen;
1066 rdp->n_cbs_orphaned += rdp->qlen;
1067
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001068 rdp->nxtlist = NULL;
1069 for (i = 0; i < RCU_NEXT_SIZE; i++)
1070 rdp->nxttail[i] = &rdp->nxtlist;
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001071 rdp->qlen = 0;
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001072}
1073
1074/*
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001075 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
1076 * and move all callbacks from the outgoing CPU to the current one.
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001077 * There can only be one CPU hotplug operation at a time, so no other
1078 * CPU can be attempting to update rcu_cpu_kthread_task.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001079 */
1080static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
1081{
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001082 unsigned long flags;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001083 unsigned long mask;
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -08001084 int need_report = 0;
Lai Jiangshan394f99a2010-06-28 16:25:04 +08001085 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001086 struct rcu_node *rnp;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001087 struct task_struct *t;
1088
1089 /* Stop the CPU's kthread. */
1090 t = per_cpu(rcu_cpu_kthread_task, cpu);
1091 if (t != NULL) {
1092 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1093 kthread_stop(t);
1094 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001095
1096 /* Exclude any attempts to start a new grace period. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001097 raw_spin_lock_irqsave(&rsp->onofflock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001098
1099 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
Paul E. McKenney28ecd582009-09-18 09:50:17 -07001100 rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001101 mask = rdp->grpmask; /* rnp->grplo is constant. */
1102 do {
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001103 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001104 rnp->qsmaskinit &= ~mask;
1105 if (rnp->qsmaskinit != 0) {
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001106 if (rnp != rdp->mynode)
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001107 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001108 break;
1109 }
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001110 if (rnp == rdp->mynode)
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -08001111 need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001112 else
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001113 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001114 mask = rnp->grpmask;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001115 rnp = rnp->parent;
1116 } while (rnp != NULL);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001117
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001118 /*
1119 * We still hold the leaf rcu_node structure lock here, and
1120 * irqs are still disabled. The reason for this subterfuge is
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -08001121 * because invoking rcu_report_unblock_qs_rnp() with ->onofflock
1122 * held leads to deadlock.
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001123 */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001124 raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001125 rnp = rdp->mynode;
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -08001126 if (need_report & RCU_OFL_TASKS_NORM_GP)
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -08001127 rcu_report_unblock_qs_rnp(rnp, flags);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -08001128 else
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001129 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -08001130 if (need_report & RCU_OFL_TASKS_EXP_GP)
1131 rcu_report_exp_rnp(rsp, rnp);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001132
1133 /*
1134 * If there are no more online CPUs for this rcu_node structure,
1135 * kill the rcu_node structure's kthread. Otherwise, adjust its
1136 * affinity.
1137 */
1138 t = rnp->node_kthread_task;
1139 if (t != NULL &&
1140 rnp->qsmaskinit == 0) {
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001141 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001142 rnp->node_kthread_task = NULL;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001143 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1144 kthread_stop(t);
1145 rcu_stop_boost_kthread(rnp);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001146 } else
Paul E. McKenney0f962a52011-04-14 12:13:53 -07001147 rcu_node_kthread_setaffinity(rnp, -1);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001148}
1149
1150/*
1151 * Remove the specified CPU from the RCU hierarchy and move any pending
1152 * callbacks that it might have to the current CPU. This code assumes
1153 * that at least one CPU in the system will remain running at all times.
1154 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
1155 */
1156static void rcu_offline_cpu(int cpu)
1157{
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001158 __rcu_offline_cpu(cpu, &rcu_sched_state);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001159 __rcu_offline_cpu(cpu, &rcu_bh_state);
Paul E. McKenney33f76142009-08-24 09:42:01 -07001160 rcu_preempt_offline_cpu(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001161}
1162
1163#else /* #ifdef CONFIG_HOTPLUG_CPU */
1164
Lai Jiangshan29494be2010-10-20 14:13:06 +08001165static void rcu_send_cbs_to_online(struct rcu_state *rsp)
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07001166{
1167}
1168
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001169static void rcu_offline_cpu(int cpu)
1170{
1171}
1172
1173#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1174
1175/*
1176 * Invoke any RCU callbacks that have made it to the end of their grace
1177 * period. Thottle as specified by rdp->blimit.
1178 */
Paul E. McKenney37c72e52009-10-14 10:15:55 -07001179static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001180{
1181 unsigned long flags;
1182 struct rcu_head *next, *list, **tail;
1183 int count;
1184
1185 /* If no callbacks are ready, just return.*/
1186 if (!cpu_has_callbacks_ready_to_invoke(rdp))
1187 return;
1188
1189 /*
1190 * Extract the list of ready callbacks, disabling to prevent
1191 * races with call_rcu() from interrupt handlers.
1192 */
1193 local_irq_save(flags);
1194 list = rdp->nxtlist;
1195 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
1196 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
1197 tail = rdp->nxttail[RCU_DONE_TAIL];
1198 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
1199 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
1200 rdp->nxttail[count] = &rdp->nxtlist;
1201 local_irq_restore(flags);
1202
1203 /* Invoke callbacks. */
1204 count = 0;
1205 while (list) {
1206 next = list->next;
1207 prefetch(next);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -04001208 debug_rcu_head_unqueue(list);
Lai Jiangshan9ab15442011-03-18 11:15:47 +08001209 __rcu_reclaim(list);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001210 list = next;
1211 if (++count >= rdp->blimit)
1212 break;
1213 }
1214
1215 local_irq_save(flags);
1216
1217 /* Update count, and requeue any remaining callbacks. */
1218 rdp->qlen -= count;
Paul E. McKenney269dcc12010-09-07 14:23:09 -07001219 rdp->n_cbs_invoked += count;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001220 if (list != NULL) {
1221 *tail = rdp->nxtlist;
1222 rdp->nxtlist = list;
1223 for (count = 0; count < RCU_NEXT_SIZE; count++)
1224 if (&rdp->nxtlist == rdp->nxttail[count])
1225 rdp->nxttail[count] = tail;
1226 else
1227 break;
1228 }
1229
1230 /* Reinstate batch limit if we have worked down the excess. */
1231 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1232 rdp->blimit = blimit;
1233
Paul E. McKenney37c72e52009-10-14 10:15:55 -07001234 /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1235 if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
1236 rdp->qlen_last_fqs_check = 0;
1237 rdp->n_force_qs_snap = rsp->n_force_qs;
1238 } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
1239 rdp->qlen_last_fqs_check = rdp->qlen;
1240
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001241 local_irq_restore(flags);
1242
1243 /* Re-raise the RCU softirq if there are callbacks remaining. */
1244 if (cpu_has_callbacks_ready_to_invoke(rdp))
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001245 invoke_rcu_cpu_kthread();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001246}
1247
1248/*
1249 * Check to see if this CPU is in a non-context-switch quiescent state
1250 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1251 * Also schedule the RCU softirq handler.
1252 *
1253 * This function must be called with hardirqs disabled. It is normally
1254 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1255 * false, there is no point in invoking rcu_check_callbacks().
1256 */
1257void rcu_check_callbacks(int cpu, int user)
1258{
1259 if (user ||
Paul E. McKenneya6826042009-02-25 18:03:42 -08001260 (idle_cpu(cpu) && rcu_scheduler_active &&
1261 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001262
1263 /*
1264 * Get here if this CPU took its interrupt from user
1265 * mode or from the idle loop, and if this is not a
1266 * nested interrupt. In this case, the CPU is in
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001267 * a quiescent state, so note it.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001268 *
1269 * No memory barrier is required here because both
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001270 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1271 * variables that other CPUs neither access nor modify,
1272 * at least not while the corresponding CPU is online.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001273 */
1274
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001275 rcu_sched_qs(cpu);
1276 rcu_bh_qs(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001277
1278 } else if (!in_softirq()) {
1279
1280 /*
1281 * Get here if this CPU did not take its interrupt from
1282 * softirq, in other words, if it is not interrupting
1283 * a rcu_bh read-side critical section. This is an _bh
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001284 * critical section, so note it.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001285 */
1286
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001287 rcu_bh_qs(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001288 }
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07001289 rcu_preempt_check_callbacks(cpu);
Paul E. McKenneyd21670a2010-04-14 17:39:26 -07001290 if (rcu_pending(cpu))
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001291 invoke_rcu_cpu_kthread();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001292}
1293
1294#ifdef CONFIG_SMP
1295
1296/*
1297 * Scan the leaf rcu_node structures, processing dyntick state for any that
1298 * have not yet encountered a quiescent state, using the function specified.
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001299 * Also initiate boosting for any threads blocked on the root rcu_node.
1300 *
Paul E. McKenneyee47eb92010-01-04 15:09:07 -08001301 * The caller must have suppressed start of new grace periods.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001302 */
Paul E. McKenney45f014c52e2010-01-04 15:09:08 -08001303static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001304{
1305 unsigned long bit;
1306 int cpu;
1307 unsigned long flags;
1308 unsigned long mask;
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -07001309 struct rcu_node *rnp;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001310
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -07001311 rcu_for_each_leaf_node(rsp, rnp) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001312 mask = 0;
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001313 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenneyee47eb92010-01-04 15:09:07 -08001314 if (!rcu_gp_in_progress(rsp)) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001315 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney0f10dc822010-01-04 15:09:06 -08001316 return;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001317 }
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -07001318 if (rnp->qsmask == 0) {
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001319 rcu_initiate_boost(rnp);
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001320 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001321 continue;
1322 }
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -07001323 cpu = rnp->grplo;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001324 bit = 1;
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -07001325 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
Lai Jiangshan394f99a2010-06-28 16:25:04 +08001326 if ((rnp->qsmask & bit) != 0 &&
1327 f(per_cpu_ptr(rsp->rda, cpu)))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001328 mask |= bit;
1329 }
Paul E. McKenney45f014c52e2010-01-04 15:09:08 -08001330 if (mask != 0) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001331
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -08001332 /* rcu_report_qs_rnp() releases rnp->lock. */
1333 rcu_report_qs_rnp(mask, rsp, rnp, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001334 continue;
1335 }
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001336 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001337 }
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001338 rnp = rcu_get_root(rsp);
1339 raw_spin_lock_irqsave(&rnp->lock, flags);
1340 if (rnp->qsmask == 0)
1341 rcu_initiate_boost(rnp);
1342 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001343}
1344
1345/*
1346 * Force quiescent states on reluctant CPUs, and also detect which
1347 * CPUs are in dyntick-idle mode.
1348 */
1349static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1350{
1351 unsigned long flags;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001352 struct rcu_node *rnp = rcu_get_root(rsp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001353
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -07001354 if (!rcu_gp_in_progress(rsp))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001355 return; /* No grace period in progress, nothing to force. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001356 if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001357 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1358 return; /* Someone else is already on the job. */
1359 }
Paul E. McKenney20133cf2010-02-22 17:05:01 -08001360 if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies))
Paul E. McKenneyf96e9232010-01-04 15:09:00 -08001361 goto unlock_fqs_ret; /* no emergency and done recently. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001362 rsp->n_force_qs++;
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001363 raw_spin_lock(&rnp->lock); /* irqs already disabled */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001364 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
Paul E. McKenney560d4bc2009-11-13 19:51:38 -08001365 if(!rcu_gp_in_progress(rsp)) {
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001366 rsp->n_force_qs_ngp++;
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001367 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
Paul E. McKenneyf96e9232010-01-04 15:09:00 -08001368 goto unlock_fqs_ret; /* no GP in progress, time updated. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001369 }
Paul E. McKenney07079d52010-01-04 15:09:02 -08001370 rsp->fqs_active = 1;
Paul E. McKenneyf3a8b5c2010-01-04 15:09:03 -08001371 switch (rsp->signaled) {
Paul E. McKenney83f5b012009-10-28 08:14:49 -07001372 case RCU_GP_IDLE:
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001373 case RCU_GP_INIT:
1374
Paul E. McKenney83f5b012009-10-28 08:14:49 -07001375 break; /* grace period idle or initializing, ignore. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001376
1377 case RCU_SAVE_DYNTICK:
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001378 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1379 break; /* So gcc recognizes the dead code. */
1380
Lai Jiangshanf2614142010-03-28 11:15:20 +08001381 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1382
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001383 /* Record dyntick-idle state. */
Paul E. McKenney45f014c52e2010-01-04 15:09:08 -08001384 force_qs_rnp(rsp, dyntick_save_progress_counter);
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001385 raw_spin_lock(&rnp->lock); /* irqs already disabled */
Paul E. McKenneyee47eb92010-01-04 15:09:07 -08001386 if (rcu_gp_in_progress(rsp))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001387 rsp->signaled = RCU_FORCE_QS;
Paul E. McKenneyee47eb92010-01-04 15:09:07 -08001388 break;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001389
1390 case RCU_FORCE_QS:
1391
1392 /* Check dyntick-idle state, send IPI to laggarts. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001393 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
Paul E. McKenney45f014c52e2010-01-04 15:09:08 -08001394 force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001395
1396 /* Leave state in case more forcing is required. */
1397
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001398 raw_spin_lock(&rnp->lock); /* irqs already disabled */
Paul E. McKenneyf96e9232010-01-04 15:09:00 -08001399 break;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001400 }
Paul E. McKenney07079d52010-01-04 15:09:02 -08001401 rsp->fqs_active = 0;
Paul E. McKenney46a1e342010-01-04 15:09:09 -08001402 if (rsp->fqs_need_gp) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001403 raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */
Paul E. McKenney46a1e342010-01-04 15:09:09 -08001404 rsp->fqs_need_gp = 0;
1405 rcu_start_gp(rsp, flags); /* releases rnp->lock */
1406 return;
1407 }
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001408 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
Paul E. McKenneyf96e9232010-01-04 15:09:00 -08001409unlock_fqs_ret:
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001410 raw_spin_unlock_irqrestore(&rsp->fqslock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001411}
1412
1413#else /* #ifdef CONFIG_SMP */
1414
1415static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1416{
1417 set_need_resched();
1418}
1419
1420#endif /* #else #ifdef CONFIG_SMP */
1421
1422/*
1423 * This does the RCU processing work from softirq context for the
1424 * specified rcu_state and rcu_data structures. This may be called
1425 * only from the CPU to whom the rdp belongs.
1426 */
1427static void
1428__rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1429{
1430 unsigned long flags;
1431
Paul E. McKenney2e597552009-08-15 09:53:48 -07001432 WARN_ON_ONCE(rdp->beenonline == 0);
1433
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001434 /*
1435 * If an RCU GP has gone long enough, go check for dyntick
1436 * idle CPUs and, if needed, send resched IPIs.
1437 */
Paul E. McKenney20133cf2010-02-22 17:05:01 -08001438 if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001439 force_quiescent_state(rsp, 1);
1440
1441 /*
1442 * Advance callbacks in response to end of earlier grace
1443 * period that some other CPU ended.
1444 */
1445 rcu_process_gp_end(rsp, rdp);
1446
1447 /* Update RCU state based on any recent quiescent states. */
1448 rcu_check_quiescent_state(rsp, rdp);
1449
1450 /* Does this CPU require a not-yet-started grace period? */
1451 if (cpu_needs_another_gp(rsp, rdp)) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -08001452 raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001453 rcu_start_gp(rsp, flags); /* releases above lock */
1454 }
1455
1456 /* If there are callbacks ready, invoke them. */
Paul E. McKenney37c72e52009-10-14 10:15:55 -07001457 rcu_do_batch(rsp, rdp);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001458}
1459
1460/*
1461 * Do softirq processing for the current CPU.
1462 */
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001463static void rcu_process_callbacks(void)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001464{
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001465 __rcu_process_callbacks(&rcu_sched_state,
1466 &__get_cpu_var(rcu_sched_data));
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001467 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07001468 rcu_preempt_process_callbacks();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001469
Paul E. McKenneya47cd882010-02-26 16:38:56 -08001470 /* If we are last CPU on way to dyntick-idle mode, accelerate it. */
1471 rcu_needs_cpu_flush();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001472}
1473
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001474/*
1475 * Wake up the current CPU's kthread. This replaces raise_softirq()
1476 * in earlier versions of RCU. Note that because we are running on
1477 * the current CPU with interrupts disabled, the rcu_cpu_kthread_task
1478 * cannot disappear out from under us.
1479 */
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001480static void invoke_rcu_cpu_kthread(void)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001481{
1482 unsigned long flags;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001483
1484 local_irq_save(flags);
Paul E. McKenneyf0a07ae2011-02-23 11:10:52 -08001485 __this_cpu_write(rcu_cpu_has_work, 1);
1486 if (__this_cpu_read(rcu_cpu_kthread_task) == NULL) {
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001487 local_irq_restore(flags);
1488 return;
1489 }
Paul E. McKenneyf0a07ae2011-02-23 11:10:52 -08001490 wake_up(&__get_cpu_var(rcu_cpu_wq));
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001491 local_irq_restore(flags);
1492}
1493
1494/*
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001495 * Wake up the specified per-rcu_node-structure kthread.
1496 * The caller must hold ->lock.
1497 */
1498static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1499{
1500 struct task_struct *t;
1501
1502 t = rnp->node_kthread_task;
1503 if (t != NULL)
1504 wake_up_process(t);
1505}
1506
1507/*
Paul E. McKenneye3995a22011-04-18 15:31:26 -07001508 * Set the specified CPU's kthread to run RT or not, as specified by
1509 * the to_rt argument. The CPU-hotplug locks are held, so the task
1510 * is not going away.
1511 */
1512static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1513{
1514 int policy;
1515 struct sched_param sp;
1516 struct task_struct *t;
1517
1518 t = per_cpu(rcu_cpu_kthread_task, cpu);
1519 if (t == NULL)
1520 return;
1521 if (to_rt) {
1522 policy = SCHED_FIFO;
1523 sp.sched_priority = RCU_KTHREAD_PRIO;
1524 } else {
1525 policy = SCHED_NORMAL;
1526 sp.sched_priority = 0;
1527 }
1528 sched_setscheduler_nocheck(t, policy, &sp);
1529}
1530
1531/*
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001532 * Timer handler to initiate the waking up of per-CPU kthreads that
1533 * have yielded the CPU due to excess numbers of RCU callbacks.
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001534 * We wake up the per-rcu_node kthread, which in turn will wake up
1535 * the booster kthread.
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001536 */
1537static void rcu_cpu_kthread_timer(unsigned long arg)
1538{
1539 unsigned long flags;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001540 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001541 struct rcu_node *rnp = rdp->mynode;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001542
1543 raw_spin_lock_irqsave(&rnp->lock, flags);
1544 rnp->wakemask |= rdp->grpmask;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001545 invoke_rcu_node_kthread(rnp);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001546 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1547}
1548
1549/*
1550 * Drop to non-real-time priority and yield, but only after posting a
1551 * timer that will cause us to regain our real-time priority if we
1552 * remain preempted. Either way, we restore our real-time priority
1553 * before returning.
1554 */
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001555static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001556{
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001557 struct sched_param sp;
1558 struct timer_list yield_timer;
1559
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001560 setup_timer_on_stack(&yield_timer, f, arg);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001561 mod_timer(&yield_timer, jiffies + 2);
1562 sp.sched_priority = 0;
1563 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
Paul E. McKenneybaa1ae02011-03-26 22:01:35 -07001564 set_user_nice(current, 19);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001565 schedule();
1566 sp.sched_priority = RCU_KTHREAD_PRIO;
1567 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1568 del_timer(&yield_timer);
1569}
1570
1571/*
1572 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1573 * This can happen while the corresponding CPU is either coming online
1574 * or going offline. We cannot wait until the CPU is fully online
1575 * before starting the kthread, because the various notifier functions
1576 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1577 * the corresponding CPU is online.
1578 *
1579 * Return 1 if the kthread needs to stop, 0 otherwise.
1580 *
1581 * Caller must disable bh. This function can momentarily enable it.
1582 */
1583static int rcu_cpu_kthread_should_stop(int cpu)
1584{
1585 while (cpu_is_offline(cpu) ||
1586 !cpumask_equal(&current->cpus_allowed, cpumask_of(cpu)) ||
1587 smp_processor_id() != cpu) {
1588 if (kthread_should_stop())
1589 return 1;
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -07001590 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1591 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001592 local_bh_enable();
1593 schedule_timeout_uninterruptible(1);
1594 if (!cpumask_equal(&current->cpus_allowed, cpumask_of(cpu)))
1595 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1596 local_bh_disable();
1597 }
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -07001598 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001599 return 0;
1600}
1601
1602/*
1603 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1604 * earlier RCU softirq.
1605 */
1606static int rcu_cpu_kthread(void *arg)
1607{
1608 int cpu = (int)(long)arg;
1609 unsigned long flags;
1610 int spincnt = 0;
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001611 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001612 wait_queue_head_t *wqp = &per_cpu(rcu_cpu_wq, cpu);
1613 char work;
1614 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1615
1616 for (;;) {
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001617 *statusp = RCU_KTHREAD_WAITING;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001618 wait_event_interruptible(*wqp,
1619 *workp != 0 || kthread_should_stop());
1620 local_bh_disable();
1621 if (rcu_cpu_kthread_should_stop(cpu)) {
1622 local_bh_enable();
1623 break;
1624 }
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001625 *statusp = RCU_KTHREAD_RUNNING;
Paul E. McKenney5ece5ba2011-04-22 18:08:51 -07001626 per_cpu(rcu_cpu_kthread_loops, cpu)++;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001627 local_irq_save(flags);
1628 work = *workp;
1629 *workp = 0;
1630 local_irq_restore(flags);
1631 if (work)
1632 rcu_process_callbacks();
1633 local_bh_enable();
1634 if (*workp != 0)
1635 spincnt++;
1636 else
1637 spincnt = 0;
1638 if (spincnt > 10) {
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001639 *statusp = RCU_KTHREAD_YIELDING;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001640 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001641 spincnt = 0;
1642 }
1643 }
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001644 *statusp = RCU_KTHREAD_STOPPED;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001645 return 0;
1646}
1647
1648/*
1649 * Spawn a per-CPU kthread, setting up affinity and priority.
1650 * Because the CPU hotplug lock is held, no other CPU will be attempting
1651 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1652 * attempting to access it during boot, but the locking in kthread_bind()
1653 * will enforce sufficient ordering.
1654 */
1655static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1656{
1657 struct sched_param sp;
1658 struct task_struct *t;
1659
1660 if (!rcu_kthreads_spawnable ||
1661 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1662 return 0;
1663 t = kthread_create(rcu_cpu_kthread, (void *)(long)cpu, "rcuc%d", cpu);
1664 if (IS_ERR(t))
1665 return PTR_ERR(t);
1666 kthread_bind(t, cpu);
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -07001667 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001668 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1669 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1670 wake_up_process(t);
1671 sp.sched_priority = RCU_KTHREAD_PRIO;
1672 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1673 return 0;
1674}
1675
1676/*
1677 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1678 * kthreads when needed. We ignore requests to wake up kthreads
1679 * for offline CPUs, which is OK because force_quiescent_state()
1680 * takes care of this case.
1681 */
1682static int rcu_node_kthread(void *arg)
1683{
1684 int cpu;
1685 unsigned long flags;
1686 unsigned long mask;
1687 struct rcu_node *rnp = (struct rcu_node *)arg;
1688 struct sched_param sp;
1689 struct task_struct *t;
1690
1691 for (;;) {
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001692 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001693 wait_event_interruptible(rnp->node_wq, rnp->wakemask != 0 ||
1694 kthread_should_stop());
1695 if (kthread_should_stop())
1696 break;
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001697 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001698 raw_spin_lock_irqsave(&rnp->lock, flags);
1699 mask = rnp->wakemask;
1700 rnp->wakemask = 0;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001701 rcu_initiate_boost(rnp);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001702 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1703 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1704 if ((mask & 0x1) == 0)
1705 continue;
1706 preempt_disable();
1707 t = per_cpu(rcu_cpu_kthread_task, cpu);
1708 if (!cpu_online(cpu) || t == NULL) {
1709 preempt_enable();
1710 continue;
1711 }
1712 per_cpu(rcu_cpu_has_work, cpu) = 1;
1713 sp.sched_priority = RCU_KTHREAD_PRIO;
1714 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1715 preempt_enable();
1716 }
1717 }
Paul E. McKenneyd71df902011-03-29 17:48:28 -07001718 rnp->node_kthread_status = RCU_KTHREAD_STOPPED;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001719 return 0;
1720}
1721
1722/*
1723 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001724 * served by the rcu_node in question. The CPU hotplug lock is still
1725 * held, so the value of rnp->qsmaskinit will be stable.
Paul E. McKenney0f962a52011-04-14 12:13:53 -07001726 *
1727 * We don't include outgoingcpu in the affinity set, use -1 if there is
1728 * no outgoing CPU. If there are no CPUs left in the affinity set,
1729 * this function allows the kthread to execute on any CPU.
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001730 */
Paul E. McKenney0f962a52011-04-14 12:13:53 -07001731static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001732{
1733 cpumask_var_t cm;
1734 int cpu;
1735 unsigned long mask = rnp->qsmaskinit;
1736
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001737 if (rnp->node_kthread_task == NULL || mask == 0)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001738 return;
1739 if (!alloc_cpumask_var(&cm, GFP_KERNEL))
1740 return;
1741 cpumask_clear(cm);
1742 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
Paul E. McKenney0f962a52011-04-14 12:13:53 -07001743 if ((mask & 0x1) && cpu != outgoingcpu)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001744 cpumask_set_cpu(cpu, cm);
Paul E. McKenney0f962a52011-04-14 12:13:53 -07001745 if (cpumask_weight(cm) == 0) {
1746 cpumask_setall(cm);
1747 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++)
1748 cpumask_clear_cpu(cpu, cm);
1749 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1750 }
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001751 set_cpus_allowed_ptr(rnp->node_kthread_task, cm);
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001752 rcu_boost_kthread_setaffinity(rnp, cm);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001753 free_cpumask_var(cm);
1754}
1755
1756/*
1757 * Spawn a per-rcu_node kthread, setting priority and affinity.
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001758 * Called during boot before online/offline can happen, or, if
1759 * during runtime, with the main CPU-hotplug locks held. So only
1760 * one of these can be executing at a time.
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001761 */
1762static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp,
1763 struct rcu_node *rnp)
1764{
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001765 unsigned long flags;
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001766 int rnp_index = rnp - &rsp->node[0];
1767 struct sched_param sp;
1768 struct task_struct *t;
1769
1770 if (!rcu_kthreads_spawnable ||
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001771 rnp->qsmaskinit == 0)
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001772 return 0;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001773 if (rnp->node_kthread_task == NULL) {
1774 t = kthread_create(rcu_node_kthread, (void *)rnp,
1775 "rcun%d", rnp_index);
1776 if (IS_ERR(t))
1777 return PTR_ERR(t);
1778 raw_spin_lock_irqsave(&rnp->lock, flags);
1779 rnp->node_kthread_task = t;
1780 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1781 wake_up_process(t);
1782 sp.sched_priority = 99;
1783 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1784 }
1785 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001786}
1787
1788/*
1789 * Spawn all kthreads -- called as soon as the scheduler is running.
1790 */
1791static int __init rcu_spawn_kthreads(void)
1792{
1793 int cpu;
1794 struct rcu_node *rnp;
1795
1796 rcu_kthreads_spawnable = 1;
1797 for_each_possible_cpu(cpu) {
1798 init_waitqueue_head(&per_cpu(rcu_cpu_wq, cpu));
1799 per_cpu(rcu_cpu_has_work, cpu) = 0;
1800 if (cpu_online(cpu))
1801 (void)rcu_spawn_one_cpu_kthread(cpu);
1802 }
Paul E. McKenney27f4d282011-02-07 12:47:15 -08001803 rnp = rcu_get_root(rcu_state);
1804 init_waitqueue_head(&rnp->node_wq);
1805 rcu_init_boost_waitqueue(rnp);
1806 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1807 if (NUM_RCU_NODES > 1)
1808 rcu_for_each_leaf_node(rcu_state, rnp) {
1809 init_waitqueue_head(&rnp->node_wq);
1810 rcu_init_boost_waitqueue(rnp);
1811 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1812 }
Paul E. McKenneya26ac242011-01-12 14:10:23 -08001813 return 0;
1814}
1815early_initcall(rcu_spawn_kthreads);
1816
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001817static void
1818__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1819 struct rcu_state *rsp)
1820{
1821 unsigned long flags;
1822 struct rcu_data *rdp;
1823
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -04001824 debug_rcu_head_queue(head);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001825 head->func = func;
1826 head->next = NULL;
1827
1828 smp_mb(); /* Ensure RCU update seen before callback registry. */
1829
1830 /*
1831 * Opportunistically note grace-period endings and beginnings.
1832 * Note that we might see a beginning right after we see an
1833 * end, but never vice versa, since this CPU has to pass through
1834 * a quiescent state betweentimes.
1835 */
1836 local_irq_save(flags);
Lai Jiangshan394f99a2010-06-28 16:25:04 +08001837 rdp = this_cpu_ptr(rsp->rda);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001838
1839 /* Add the callback to our list. */
1840 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1841 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1842
Paul E. McKenney37c72e52009-10-14 10:15:55 -07001843 /*
1844 * Force the grace period if too many callbacks or too long waiting.
1845 * Enforce hysteresis, and don't invoke force_quiescent_state()
1846 * if some other CPU has recently done so. Also, don't bother
1847 * invoking force_quiescent_state() if the newly enqueued callback
1848 * is the only one waiting for a grace period to complete.
1849 */
1850 if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
Paul E. McKenneyb52573d2010-12-14 17:36:02 -08001851
1852 /* Are we ignoring a completed grace period? */
1853 rcu_process_gp_end(rsp, rdp);
1854 check_for_new_grace_period(rsp, rdp);
1855
1856 /* Start a new grace period if one not already started. */
1857 if (!rcu_gp_in_progress(rsp)) {
1858 unsigned long nestflag;
1859 struct rcu_node *rnp_root = rcu_get_root(rsp);
1860
1861 raw_spin_lock_irqsave(&rnp_root->lock, nestflag);
1862 rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */
1863 } else {
1864 /* Give the grace period a kick. */
1865 rdp->blimit = LONG_MAX;
1866 if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1867 *rdp->nxttail[RCU_DONE_TAIL] != head)
1868 force_quiescent_state(rsp, 0);
1869 rdp->n_force_qs_snap = rsp->n_force_qs;
1870 rdp->qlen_last_fqs_check = rdp->qlen;
1871 }
Paul E. McKenney20133cf2010-02-22 17:05:01 -08001872 } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001873 force_quiescent_state(rsp, 1);
1874 local_irq_restore(flags);
1875}
1876
1877/*
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001878 * Queue an RCU-sched callback for invocation after a grace period.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001879 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001880void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001881{
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001882 __call_rcu(head, func, &rcu_sched_state);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001883}
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07001884EXPORT_SYMBOL_GPL(call_rcu_sched);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001885
1886/*
1887 * Queue an RCU for invocation after a quicker grace period.
1888 */
1889void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1890{
1891 __call_rcu(head, func, &rcu_bh_state);
1892}
1893EXPORT_SYMBOL_GPL(call_rcu_bh);
1894
Paul E. McKenney6ebb2372009-11-22 08:53:50 -08001895/**
1896 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
1897 *
1898 * Control will return to the caller some time after a full rcu-sched
1899 * grace period has elapsed, in other words after all currently executing
1900 * rcu-sched read-side critical sections have completed. These read-side
1901 * critical sections are delimited by rcu_read_lock_sched() and
1902 * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(),
1903 * local_irq_disable(), and so on may be used in place of
1904 * rcu_read_lock_sched().
1905 *
1906 * This means that all preempt_disable code sequences, including NMI and
1907 * hardware-interrupt handlers, in progress on entry will have completed
1908 * before this primitive returns. However, this does not guarantee that
1909 * softirq handlers will have completed, since in some kernels, these
1910 * handlers can run in process context, and can block.
1911 *
1912 * This primitive provides the guarantees made by the (now removed)
1913 * synchronize_kernel() API. In contrast, synchronize_rcu() only
1914 * guarantees that rcu_read_lock() sections will have completed.
1915 * In "classic RCU", these two guarantees happen to be one and
1916 * the same, but can differ in realtime RCU implementations.
1917 */
1918void synchronize_sched(void)
1919{
1920 struct rcu_synchronize rcu;
1921
1922 if (rcu_blocking_is_gp())
1923 return;
1924
Paul E. McKenney72d5a9f2010-05-10 17:12:17 -07001925 init_rcu_head_on_stack(&rcu.head);
Paul E. McKenney6ebb2372009-11-22 08:53:50 -08001926 init_completion(&rcu.completion);
1927 /* Will wake me after RCU finished. */
1928 call_rcu_sched(&rcu.head, wakeme_after_rcu);
1929 /* Wait for it. */
1930 wait_for_completion(&rcu.completion);
Paul E. McKenney72d5a9f2010-05-10 17:12:17 -07001931 destroy_rcu_head_on_stack(&rcu.head);
Paul E. McKenney6ebb2372009-11-22 08:53:50 -08001932}
1933EXPORT_SYMBOL_GPL(synchronize_sched);
1934
1935/**
1936 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
1937 *
1938 * Control will return to the caller some time after a full rcu_bh grace
1939 * period has elapsed, in other words after all currently executing rcu_bh
1940 * read-side critical sections have completed. RCU read-side critical
1941 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
1942 * and may be nested.
1943 */
1944void synchronize_rcu_bh(void)
1945{
1946 struct rcu_synchronize rcu;
1947
1948 if (rcu_blocking_is_gp())
1949 return;
1950
Paul E. McKenney72d5a9f2010-05-10 17:12:17 -07001951 init_rcu_head_on_stack(&rcu.head);
Paul E. McKenney6ebb2372009-11-22 08:53:50 -08001952 init_completion(&rcu.completion);
1953 /* Will wake me after RCU finished. */
1954 call_rcu_bh(&rcu.head, wakeme_after_rcu);
1955 /* Wait for it. */
1956 wait_for_completion(&rcu.completion);
Paul E. McKenney72d5a9f2010-05-10 17:12:17 -07001957 destroy_rcu_head_on_stack(&rcu.head);
Paul E. McKenney6ebb2372009-11-22 08:53:50 -08001958}
1959EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
1960
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001961/*
1962 * Check to see if there is any immediate RCU-related work to be done
1963 * by the current CPU, for the specified type of RCU, returning 1 if so.
1964 * The checks are in order of increasing expense: checks that can be
1965 * carried out against CPU-local state are performed first. However,
1966 * we must check for CPU stalls first, else we might not get a chance.
1967 */
1968static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1969{
Paul E. McKenney2f51f982009-11-13 19:51:39 -08001970 struct rcu_node *rnp = rdp->mynode;
1971
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001972 rdp->n_rcu_pending++;
1973
1974 /* Check for CPU stalls, if enabled. */
1975 check_cpu_stall(rsp, rdp);
1976
1977 /* Is the RCU core waiting for a quiescent state from this CPU? */
Paul E. McKenneyd21670a2010-04-14 17:39:26 -07001978 if (rdp->qs_pending && !rdp->passed_quiesc) {
Paul E. McKenneyd25eb942010-03-18 21:36:51 -07001979
1980 /*
1981 * If force_quiescent_state() coming soon and this CPU
1982 * needs a quiescent state, and this is either RCU-sched
1983 * or RCU-bh, force a local reschedule.
1984 */
Paul E. McKenneyd21670a2010-04-14 17:39:26 -07001985 rdp->n_rp_qs_pending++;
Paul E. McKenney6cc68792011-03-02 13:15:15 -08001986 if (!rdp->preemptible &&
Paul E. McKenneyd25eb942010-03-18 21:36:51 -07001987 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1,
1988 jiffies))
1989 set_need_resched();
Paul E. McKenneyd21670a2010-04-14 17:39:26 -07001990 } else if (rdp->qs_pending && rdp->passed_quiesc) {
1991 rdp->n_rp_report_qs++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001992 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07001993 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001994
1995 /* Does this CPU have callbacks ready to invoke? */
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07001996 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1997 rdp->n_rp_cb_ready++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01001998 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07001999 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002000
2001 /* Has RCU gone idle with this CPU needing another grace period? */
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002002 if (cpu_needs_another_gp(rsp, rdp)) {
2003 rdp->n_rp_cpu_needs_gp++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002004 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002005 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002006
2007 /* Has another RCU grace period completed? */
Paul E. McKenney2f51f982009-11-13 19:51:39 -08002008 if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002009 rdp->n_rp_gp_completed++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002010 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002011 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002012
2013 /* Has a new RCU grace period started? */
Paul E. McKenney2f51f982009-11-13 19:51:39 -08002014 if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002015 rdp->n_rp_gp_started++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002016 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002017 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002018
2019 /* Has an RCU GP gone long enough to send resched IPIs &c? */
Paul E. McKenneyfc2219d42009-09-23 09:50:41 -07002020 if (rcu_gp_in_progress(rsp) &&
Paul E. McKenney20133cf2010-02-22 17:05:01 -08002021 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) {
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002022 rdp->n_rp_need_fqs++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002023 return 1;
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002024 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002025
2026 /* nothing to do */
Paul E. McKenney7ba5c842009-04-13 21:31:17 -07002027 rdp->n_rp_need_nothing++;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002028 return 0;
2029}
2030
2031/*
2032 * Check to see if there is any immediate RCU-related work to be done
2033 * by the current CPU, returning 1 if so. This function is part of the
2034 * RCU implementation; it is -not- an exported member of the RCU API.
2035 */
Paul E. McKenneya1572292009-08-22 13:56:51 -07002036static int rcu_pending(int cpu)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002037{
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07002038 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002039 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
2040 rcu_preempt_pending(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002041}
2042
2043/*
2044 * Check to see if any future RCU-related work will need to be done
2045 * by the current CPU, even if none need be done immediately, returning
Paul E. McKenney8bd93a22010-02-22 17:04:59 -08002046 * 1 if so.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002047 */
Paul E. McKenney8bd93a22010-02-22 17:04:59 -08002048static int rcu_needs_cpu_quick_check(int cpu)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002049{
2050 /* RCU callbacks either ready or pending? */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -07002051 return per_cpu(rcu_sched_data, cpu).nxtlist ||
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002052 per_cpu(rcu_bh_data, cpu).nxtlist ||
2053 rcu_preempt_needs_cpu(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002054}
2055
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002056static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
2057static atomic_t rcu_barrier_cpu_count;
2058static DEFINE_MUTEX(rcu_barrier_mutex);
2059static struct completion rcu_barrier_completion;
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002060
2061static void rcu_barrier_callback(struct rcu_head *notused)
2062{
2063 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
2064 complete(&rcu_barrier_completion);
2065}
2066
2067/*
2068 * Called with preemption disabled, and from cross-cpu IRQ context.
2069 */
2070static void rcu_barrier_func(void *type)
2071{
2072 int cpu = smp_processor_id();
2073 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
2074 void (*call_rcu_func)(struct rcu_head *head,
2075 void (*func)(struct rcu_head *head));
2076
2077 atomic_inc(&rcu_barrier_cpu_count);
2078 call_rcu_func = type;
2079 call_rcu_func(head, rcu_barrier_callback);
2080}
2081
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002082/*
2083 * Orchestrate the specified type of RCU barrier, waiting for all
2084 * RCU callbacks of the specified type to complete.
2085 */
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07002086static void _rcu_barrier(struct rcu_state *rsp,
2087 void (*call_rcu_func)(struct rcu_head *head,
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002088 void (*func)(struct rcu_head *head)))
2089{
2090 BUG_ON(in_interrupt());
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07002091 /* Take mutex to serialize concurrent rcu_barrier() requests. */
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002092 mutex_lock(&rcu_barrier_mutex);
2093 init_completion(&rcu_barrier_completion);
2094 /*
2095 * Initialize rcu_barrier_cpu_count to 1, then invoke
2096 * rcu_barrier_func() on each CPU, so that each CPU also has
2097 * incremented rcu_barrier_cpu_count. Only then is it safe to
2098 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
2099 * might complete its grace period before all of the other CPUs
2100 * did their increment, causing this function to return too
Paul E. McKenney2d999e02010-10-20 12:06:18 -07002101 * early. Note that on_each_cpu() disables irqs, which prevents
2102 * any CPUs from coming online or going offline until each online
2103 * CPU has queued its RCU-barrier callback.
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002104 */
2105 atomic_set(&rcu_barrier_cpu_count, 1);
2106 on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
2107 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
2108 complete(&rcu_barrier_completion);
2109 wait_for_completion(&rcu_barrier_completion);
2110 mutex_unlock(&rcu_barrier_mutex);
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002111}
2112
2113/**
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002114 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
2115 */
2116void rcu_barrier_bh(void)
2117{
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07002118 _rcu_barrier(&rcu_bh_state, call_rcu_bh);
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002119}
2120EXPORT_SYMBOL_GPL(rcu_barrier_bh);
2121
2122/**
2123 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
2124 */
2125void rcu_barrier_sched(void)
2126{
Paul E. McKenneye74f4c42009-10-06 21:48:17 -07002127 _rcu_barrier(&rcu_sched_state, call_rcu_sched);
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002128}
2129EXPORT_SYMBOL_GPL(rcu_barrier_sched);
2130
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002131/*
Paul E. McKenney27569622009-08-15 09:53:46 -07002132 * Do boot-time initialization of a CPU's per-CPU RCU data.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002133 */
Paul E. McKenney27569622009-08-15 09:53:46 -07002134static void __init
2135rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002136{
2137 unsigned long flags;
2138 int i;
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002139 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
Paul E. McKenney27569622009-08-15 09:53:46 -07002140 struct rcu_node *rnp = rcu_get_root(rsp);
2141
2142 /* Set up local state, ensuring consistent view of global state. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002143 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney27569622009-08-15 09:53:46 -07002144 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
2145 rdp->nxtlist = NULL;
2146 for (i = 0; i < RCU_NEXT_SIZE; i++)
2147 rdp->nxttail[i] = &rdp->nxtlist;
2148 rdp->qlen = 0;
2149#ifdef CONFIG_NO_HZ
2150 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
2151#endif /* #ifdef CONFIG_NO_HZ */
2152 rdp->cpu = cpu;
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002153 raw_spin_unlock_irqrestore(&rnp->lock, flags);
Paul E. McKenney27569622009-08-15 09:53:46 -07002154}
2155
2156/*
2157 * Initialize a CPU's per-CPU RCU data. Note that only one online or
2158 * offline event can be happening at a given time. Note also that we
2159 * can accept some slop in the rsp->completed access due to the fact
2160 * that this CPU cannot possibly have any RCU callbacks in flight yet.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002161 */
2162static void __cpuinit
Paul E. McKenney6cc68792011-03-02 13:15:15 -08002163rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002164{
2165 unsigned long flags;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002166 unsigned long mask;
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002167 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002168 struct rcu_node *rnp = rcu_get_root(rsp);
2169
2170 /* Set up local state, ensuring consistent view of global state. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002171 raw_spin_lock_irqsave(&rnp->lock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002172 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
2173 rdp->qs_pending = 1; /* so set up to respond to current GP. */
2174 rdp->beenonline = 1; /* We have now been online. */
Paul E. McKenney6cc68792011-03-02 13:15:15 -08002175 rdp->preemptible = preemptible;
Paul E. McKenney37c72e52009-10-14 10:15:55 -07002176 rdp->qlen_last_fqs_check = 0;
2177 rdp->n_force_qs_snap = rsp->n_force_qs;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002178 rdp->blimit = blimit;
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002179 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002180
2181 /*
2182 * A new grace period might start here. If so, we won't be part
2183 * of it, but that is OK, as we are currently in a quiescent state.
2184 */
2185
2186 /* Exclude any attempts to start a new GP on large systems. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002187 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002188
2189 /* Add CPU to rcu_node bitmasks. */
2190 rnp = rdp->mynode;
2191 mask = rdp->grpmask;
2192 do {
2193 /* Exclude any attempts to start a new GP on small systems. */
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002194 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002195 rnp->qsmaskinit |= mask;
2196 mask = rnp->grpmask;
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -08002197 if (rnp == rdp->mynode) {
2198 rdp->gpnum = rnp->completed; /* if GP in progress... */
2199 rdp->completed = rnp->completed;
2200 rdp->passed_quiesc_completed = rnp->completed - 1;
2201 }
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002202 raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002203 rnp = rnp->parent;
2204 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
2205
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002206 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002207}
2208
2209static void __cpuinit rcu_online_cpu(int cpu)
2210{
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002211 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
2212 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
2213 rcu_preempt_init_percpu_data(cpu);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002214}
2215
Paul E. McKenneya26ac242011-01-12 14:10:23 -08002216static void __cpuinit rcu_online_kthreads(int cpu)
2217{
Paul E. McKenney27f4d282011-02-07 12:47:15 -08002218 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08002219 struct rcu_node *rnp = rdp->mynode;
2220
2221 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
2222 if (rcu_kthreads_spawnable) {
2223 (void)rcu_spawn_one_cpu_kthread(cpu);
2224 if (rnp->node_kthread_task == NULL)
Paul E. McKenney27f4d282011-02-07 12:47:15 -08002225 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08002226 }
2227}
2228
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002229/*
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002230 * Handle CPU online/offline notification events.
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002231 */
Paul E. McKenney9f680ab2009-11-22 08:53:49 -08002232static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
2233 unsigned long action, void *hcpu)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002234{
2235 long cpu = (long)hcpu;
Paul E. McKenney27f4d282011-02-07 12:47:15 -08002236 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08002237 struct rcu_node *rnp = rdp->mynode;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002238
2239 switch (action) {
2240 case CPU_UP_PREPARE:
2241 case CPU_UP_PREPARE_FROZEN:
2242 rcu_online_cpu(cpu);
Paul E. McKenneya26ac242011-01-12 14:10:23 -08002243 rcu_online_kthreads(cpu);
2244 break;
2245 case CPU_ONLINE:
Paul E. McKenney0f962a52011-04-14 12:13:53 -07002246 case CPU_DOWN_FAILED:
2247 rcu_node_kthread_setaffinity(rnp, -1);
Paul E. McKenneye3995a22011-04-18 15:31:26 -07002248 rcu_cpu_kthread_setrt(cpu, 1);
Paul E. McKenney0f962a52011-04-14 12:13:53 -07002249 break;
2250 case CPU_DOWN_PREPARE:
2251 rcu_node_kthread_setaffinity(rnp, cpu);
Paul E. McKenneye3995a22011-04-18 15:31:26 -07002252 rcu_cpu_kthread_setrt(cpu, 0);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002253 break;
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002254 case CPU_DYING:
2255 case CPU_DYING_FROZEN:
2256 /*
Paul E. McKenney2d999e02010-10-20 12:06:18 -07002257 * The whole machine is "stopped" except this CPU, so we can
2258 * touch any data without introducing corruption. We send the
2259 * dying CPU's callbacks to an arbitrarily chosen online CPU.
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002260 */
Lai Jiangshan29494be2010-10-20 14:13:06 +08002261 rcu_send_cbs_to_online(&rcu_bh_state);
2262 rcu_send_cbs_to_online(&rcu_sched_state);
2263 rcu_preempt_send_cbs_to_online();
Paul E. McKenneyd0ec7742009-10-06 21:48:16 -07002264 break;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002265 case CPU_DEAD:
2266 case CPU_DEAD_FROZEN:
2267 case CPU_UP_CANCELED:
2268 case CPU_UP_CANCELED_FROZEN:
2269 rcu_offline_cpu(cpu);
2270 break;
2271 default:
2272 break;
2273 }
2274 return NOTIFY_OK;
2275}
2276
2277/*
Paul E. McKenneybbad9372010-04-02 16:17:17 -07002278 * This function is invoked towards the end of the scheduler's initialization
2279 * process. Before this is called, the idle task might contain
2280 * RCU read-side critical sections (during which time, this idle
2281 * task is booting the system). After this function is called, the
2282 * idle tasks are prohibited from containing RCU read-side critical
2283 * sections. This function also enables RCU lockdep checking.
2284 */
2285void rcu_scheduler_starting(void)
2286{
2287 WARN_ON(num_online_cpus() != 1);
2288 WARN_ON(nr_context_switches() > 0);
2289 rcu_scheduler_active = 1;
2290}
2291
2292/*
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002293 * Compute the per-level fanout, either using the exact fanout specified
2294 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
2295 */
2296#ifdef CONFIG_RCU_FANOUT_EXACT
2297static void __init rcu_init_levelspread(struct rcu_state *rsp)
2298{
2299 int i;
2300
Paul E. McKenney0209f642010-12-14 16:07:52 -08002301 for (i = NUM_RCU_LVLS - 1; i > 0; i--)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002302 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
Paul E. McKenney0209f642010-12-14 16:07:52 -08002303 rsp->levelspread[0] = RCU_FANOUT_LEAF;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002304}
2305#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
2306static void __init rcu_init_levelspread(struct rcu_state *rsp)
2307{
2308 int ccur;
2309 int cprv;
2310 int i;
2311
2312 cprv = NR_CPUS;
2313 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
2314 ccur = rsp->levelcnt[i];
2315 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
2316 cprv = ccur;
2317 }
2318}
2319#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
2320
2321/*
2322 * Helper function for rcu_init() that initializes one rcu_state structure.
2323 */
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002324static void __init rcu_init_one(struct rcu_state *rsp,
2325 struct rcu_data __percpu *rda)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002326{
Paul E. McKenneyb6407e82010-01-04 16:04:02 -08002327 static char *buf[] = { "rcu_node_level_0",
2328 "rcu_node_level_1",
2329 "rcu_node_level_2",
2330 "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002331 int cpustride = 1;
2332 int i;
2333 int j;
2334 struct rcu_node *rnp;
2335
Paul E. McKenneyb6407e82010-01-04 16:04:02 -08002336 BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */
2337
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002338 /* Initialize the level-tracking arrays. */
2339
2340 for (i = 1; i < NUM_RCU_LVLS; i++)
2341 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
2342 rcu_init_levelspread(rsp);
2343
2344 /* Initialize the elements themselves, starting from the leaves. */
2345
2346 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
2347 cpustride *= rsp->levelspread[i];
2348 rnp = rsp->level[i];
2349 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
Paul E. McKenney1304afb2010-02-22 17:05:02 -08002350 raw_spin_lock_init(&rnp->lock);
Paul E. McKenneyb6407e82010-01-04 16:04:02 -08002351 lockdep_set_class_and_name(&rnp->lock,
2352 &rcu_node_class[i], buf[i]);
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002353 rnp->gpnum = 0;
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002354 rnp->qsmask = 0;
2355 rnp->qsmaskinit = 0;
2356 rnp->grplo = j * cpustride;
2357 rnp->grphi = (j + 1) * cpustride - 1;
2358 if (rnp->grphi >= NR_CPUS)
2359 rnp->grphi = NR_CPUS - 1;
2360 if (i == 0) {
2361 rnp->grpnum = 0;
2362 rnp->grpmask = 0;
2363 rnp->parent = NULL;
2364 } else {
2365 rnp->grpnum = j % rsp->levelspread[i - 1];
2366 rnp->grpmask = 1UL << rnp->grpnum;
2367 rnp->parent = rsp->level[i - 1] +
2368 j / rsp->levelspread[i - 1];
2369 }
2370 rnp->level = i;
Paul E. McKenney12f5f522010-11-29 21:56:39 -08002371 INIT_LIST_HEAD(&rnp->blkd_tasks);
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002372 }
2373 }
Lai Jiangshan0c340292010-03-28 11:12:30 +08002374
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002375 rsp->rda = rda;
Lai Jiangshan0c340292010-03-28 11:12:30 +08002376 rnp = rsp->level[NUM_RCU_LVLS - 1];
2377 for_each_possible_cpu(i) {
Paul E. McKenney4a90a062010-04-14 16:48:11 -07002378 while (i > rnp->grphi)
Lai Jiangshan0c340292010-03-28 11:12:30 +08002379 rnp++;
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002380 per_cpu_ptr(rsp->rda, i)->mynode = rnp;
Lai Jiangshan0c340292010-03-28 11:12:30 +08002381 rcu_boot_init_percpu_data(i, rsp);
2382 }
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002383}
2384
Paul E. McKenney9f680ab2009-11-22 08:53:49 -08002385void __init rcu_init(void)
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002386{
Paul E. McKenney017c4262010-01-14 16:10:58 -08002387 int cpu;
Paul E. McKenney9f680ab2009-11-22 08:53:49 -08002388
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002389 rcu_bootup_announce();
Lai Jiangshan394f99a2010-06-28 16:25:04 +08002390 rcu_init_one(&rcu_sched_state, &rcu_sched_data);
2391 rcu_init_one(&rcu_bh_state, &rcu_bh_data);
Paul E. McKenneyf41d9112009-08-22 13:56:52 -07002392 __rcu_init_preempt();
Paul E. McKenney9f680ab2009-11-22 08:53:49 -08002393
2394 /*
2395 * We don't need protection against CPU-hotplug here because
2396 * this is called early in boot, before either interrupts
2397 * or the scheduler are operational.
2398 */
2399 cpu_notifier(rcu_cpu_notify, 0);
Paul E. McKenney017c4262010-01-14 16:10:58 -08002400 for_each_online_cpu(cpu)
2401 rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
Paul E. McKenneyc68de202010-04-15 10:12:40 -07002402 check_cpu_stall_init();
Paul E. McKenney64db4cf2008-12-18 21:55:32 +01002403}
2404
Paul E. McKenney1eba8f82009-09-23 09:50:42 -07002405#include "rcutree_plugin.h"