2 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition
3 * Internal non-public definitions that provide either classic
4 * or preemptible semantics.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 * Copyright (c) 2010 Linaro
22 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
25 #include <linux/kthread.h>
26 #include <linux/module.h>
27 #include <linux/debugfs.h>
28 #include <linux/seq_file.h>
30 /* Global control variables for rcupdate callback mechanism. */
32 struct rcu_head *rcucblist; /* List of pending callbacks (CBs). */
33 struct rcu_head **donetail; /* ->next pointer of last "done" CB. */
34 struct rcu_head **curtail; /* ->next pointer of last CB. */
35 RCU_TRACE(long qlen); /* Number of pending CBs. */
36 RCU_TRACE(char *name); /* Name of RCU type. */
39 /* Definition for rcupdate control block. */
40 static struct rcu_ctrlblk rcu_sched_ctrlblk = {
41 .donetail = &rcu_sched_ctrlblk.rcucblist,
42 .curtail = &rcu_sched_ctrlblk.rcucblist,
43 RCU_TRACE(.name = "rcu_sched")
46 static struct rcu_ctrlblk rcu_bh_ctrlblk = {
47 .donetail = &rcu_bh_ctrlblk.rcucblist,
48 .curtail = &rcu_bh_ctrlblk.rcucblist,
49 RCU_TRACE(.name = "rcu_bh")
52 #ifdef CONFIG_DEBUG_LOCK_ALLOC
53 int rcu_scheduler_active __read_mostly;
54 EXPORT_SYMBOL_GPL(rcu_scheduler_active);
55 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
57 #ifdef CONFIG_TINY_PREEMPT_RCU
59 #include <linux/delay.h>
61 /* Global control variables for preemptible RCU. */
62 struct rcu_preempt_ctrlblk {
63 struct rcu_ctrlblk rcb; /* curtail: ->next ptr of last CB for GP. */
64 struct rcu_head **nexttail;
65 /* Tasks blocked in a preemptible RCU */
66 /* read-side critical section while an */
67 /* preemptible-RCU grace period is in */
68 /* progress must wait for a later grace */
69 /* period. This pointer points to the */
70 /* ->next pointer of the last task that */
71 /* must wait for a later grace period, or */
72 /* to &->rcb.rcucblist if there is no */
74 struct list_head blkd_tasks;
75 /* Tasks blocked in RCU read-side critical */
76 /* section. Tasks are placed at the head */
77 /* of this list and age towards the tail. */
78 struct list_head *gp_tasks;
79 /* Pointer to the first task blocking the */
80 /* current grace period, or NULL if there */
81 /* is no such task. */
82 struct list_head *exp_tasks;
83 /* Pointer to first task blocking the */
84 /* current expedited grace period, or NULL */
85 /* if there is no such task. If there */
86 /* is no current expedited grace period, */
87 /* then there cannot be any such task. */
88 #ifdef CONFIG_RCU_BOOST
89 struct list_head *boost_tasks;
90 /* Pointer to first task that needs to be */
91 /* priority-boosted, or NULL if no priority */
92 /* boosting is needed. If there is no */
93 /* current or expedited grace period, there */
94 /* can be no such task. */
95 #endif /* #ifdef CONFIG_RCU_BOOST */
96 u8 gpnum; /* Current grace period. */
97 u8 gpcpu; /* Last grace period blocked by the CPU. */
98 u8 completed; /* Last grace period completed. */
99 /* If all three are equal, RCU is idle. */
100 #ifdef CONFIG_RCU_BOOST
101 unsigned long boost_time; /* When to start boosting (jiffies) */
102 #endif /* #ifdef CONFIG_RCU_BOOST */
103 #ifdef CONFIG_RCU_TRACE
104 unsigned long n_grace_periods;
105 #ifdef CONFIG_RCU_BOOST
106 unsigned long n_tasks_boosted;
107 /* Total number of tasks boosted. */
108 unsigned long n_exp_boosts;
109 /* Number of tasks boosted for expedited GP. */
110 unsigned long n_normal_boosts;
111 /* Number of tasks boosted for normal GP. */
112 unsigned long n_balk_blkd_tasks;
113 /* Refused to boost: no blocked tasks. */
114 unsigned long n_balk_exp_gp_tasks;
115 /* Refused to boost: nothing blocking GP. */
116 unsigned long n_balk_boost_tasks;
117 /* Refused to boost: already boosting. */
118 unsigned long n_balk_notyet;
119 /* Refused to boost: not yet time. */
120 unsigned long n_balk_nos;
121 /* Refused to boost: not sure why, though. */
122 /* This can happen due to race conditions. */
123 #endif /* #ifdef CONFIG_RCU_BOOST */
124 #endif /* #ifdef CONFIG_RCU_TRACE */
127 static struct rcu_preempt_ctrlblk rcu_preempt_ctrlblk = {
128 .rcb.donetail = &rcu_preempt_ctrlblk.rcb.rcucblist,
129 .rcb.curtail = &rcu_preempt_ctrlblk.rcb.rcucblist,
130 .nexttail = &rcu_preempt_ctrlblk.rcb.rcucblist,
131 .blkd_tasks = LIST_HEAD_INIT(rcu_preempt_ctrlblk.blkd_tasks),
132 RCU_TRACE(.rcb.name = "rcu_preempt")
135 static void rcu_read_unlock_special(struct task_struct *t);
136 static int rcu_preempted_readers_exp(void);
137 static void rcu_report_exp_done(void);
140 * Return true if the CPU has not yet responded to the current grace period.
142 static int rcu_cpu_blocking_cur_gp(void)
144 return rcu_preempt_ctrlblk.gpcpu != rcu_preempt_ctrlblk.gpnum;
148 * Check for a running RCU reader. Because there is only one CPU,
149 * there can be but one running RCU reader at a time. ;-)
151 * Returns zero if there are no running readers. Returns a positive
152 * number if there is at least one reader within its RCU read-side
153 * critical section. Returns a negative number if an outermost reader
154 * is in the midst of exiting from its RCU read-side critical section
156 * Returns zero if there are no running readers. Returns a positive
157 * number if there is at least one reader within its RCU read-side
158 * critical section. Returns a negative number if an outermost reader
159 * is in the midst of exiting from its RCU read-side critical section.
161 static int rcu_preempt_running_reader(void)
163 return current->rcu_read_lock_nesting;
167 * Check for preempted RCU readers blocking any grace period.
168 * If the caller needs a reliable answer, it must disable hard irqs.
170 static int rcu_preempt_blocked_readers_any(void)
172 return !list_empty(&rcu_preempt_ctrlblk.blkd_tasks);
176 * Check for preempted RCU readers blocking the current grace period.
177 * If the caller needs a reliable answer, it must disable hard irqs.
179 static int rcu_preempt_blocked_readers_cgp(void)
181 return rcu_preempt_ctrlblk.gp_tasks != NULL;
185 * Return true if another preemptible-RCU grace period is needed.
187 static int rcu_preempt_needs_another_gp(void)
189 return *rcu_preempt_ctrlblk.rcb.curtail != NULL;
193 * Return true if a preemptible-RCU grace period is in progress.
194 * The caller must disable hardirqs.
196 static int rcu_preempt_gp_in_progress(void)
198 return rcu_preempt_ctrlblk.completed != rcu_preempt_ctrlblk.gpnum;
202 * Advance a ->blkd_tasks-list pointer to the next entry, instead
203 * returning NULL if at the end of the list.
205 static struct list_head *rcu_next_node_entry(struct task_struct *t)
207 struct list_head *np;
209 np = t->rcu_node_entry.next;
210 if (np == &rcu_preempt_ctrlblk.blkd_tasks)
215 #ifdef CONFIG_RCU_TRACE
217 #ifdef CONFIG_RCU_BOOST
218 static void rcu_initiate_boost_trace(void);
219 #endif /* #ifdef CONFIG_RCU_BOOST */
222 * Dump additional statistice for TINY_PREEMPT_RCU.
224 static void show_tiny_preempt_stats(struct seq_file *m)
226 seq_printf(m, "rcu_preempt: qlen=%ld gp=%lu g%u/p%u/c%u tasks=%c%c%c\n",
227 rcu_preempt_ctrlblk.rcb.qlen,
228 rcu_preempt_ctrlblk.n_grace_periods,
229 rcu_preempt_ctrlblk.gpnum,
230 rcu_preempt_ctrlblk.gpcpu,
231 rcu_preempt_ctrlblk.completed,
232 "T."[list_empty(&rcu_preempt_ctrlblk.blkd_tasks)],
233 "N."[!rcu_preempt_ctrlblk.gp_tasks],
234 "E."[!rcu_preempt_ctrlblk.exp_tasks]);
235 #ifdef CONFIG_RCU_BOOST
236 seq_printf(m, "%sttb=%c ntb=%lu neb=%lu nnb=%lu j=%04x bt=%04x\n",
238 "B."[!rcu_preempt_ctrlblk.boost_tasks],
239 rcu_preempt_ctrlblk.n_tasks_boosted,
240 rcu_preempt_ctrlblk.n_exp_boosts,
241 rcu_preempt_ctrlblk.n_normal_boosts,
242 (int)(jiffies & 0xffff),
243 (int)(rcu_preempt_ctrlblk.boost_time & 0xffff));
244 seq_printf(m, "%s: nt=%lu egt=%lu bt=%lu ny=%lu nos=%lu\n",
246 rcu_preempt_ctrlblk.n_balk_blkd_tasks,
247 rcu_preempt_ctrlblk.n_balk_exp_gp_tasks,
248 rcu_preempt_ctrlblk.n_balk_boost_tasks,
249 rcu_preempt_ctrlblk.n_balk_notyet,
250 rcu_preempt_ctrlblk.n_balk_nos);
251 #endif /* #ifdef CONFIG_RCU_BOOST */
254 #endif /* #ifdef CONFIG_RCU_TRACE */
256 #ifdef CONFIG_RCU_BOOST
258 #include "rtmutex_common.h"
260 #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO
262 /* Controls for rcu_kthread() kthread. */
263 static struct task_struct *rcu_kthread_task;
264 static DECLARE_WAIT_QUEUE_HEAD(rcu_kthread_wq);
265 static unsigned long have_rcu_kthread_work;
268 * Carry out RCU priority boosting on the task indicated by ->boost_tasks,
269 * and advance ->boost_tasks to the next task in the ->blkd_tasks list.
271 static int rcu_boost(void)
275 struct task_struct *t;
276 struct list_head *tb;
278 if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
279 rcu_preempt_ctrlblk.exp_tasks == NULL)
280 return 0; /* Nothing to boost. */
282 raw_local_irq_save(flags);
285 * Recheck with irqs disabled: all tasks in need of boosting
286 * might exit their RCU read-side critical sections on their own
287 * if we are preempted just before disabling irqs.
289 if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
290 rcu_preempt_ctrlblk.exp_tasks == NULL) {
291 raw_local_irq_restore(flags);
296 * Preferentially boost tasks blocking expedited grace periods.
297 * This cannot starve the normal grace periods because a second
298 * expedited grace period must boost all blocked tasks, including
299 * those blocking the pre-existing normal grace period.
301 if (rcu_preempt_ctrlblk.exp_tasks != NULL) {
302 tb = rcu_preempt_ctrlblk.exp_tasks;
303 RCU_TRACE(rcu_preempt_ctrlblk.n_exp_boosts++);
305 tb = rcu_preempt_ctrlblk.boost_tasks;
306 RCU_TRACE(rcu_preempt_ctrlblk.n_normal_boosts++);
308 RCU_TRACE(rcu_preempt_ctrlblk.n_tasks_boosted++);
311 * We boost task t by manufacturing an rt_mutex that appears to
312 * be held by task t. We leave a pointer to that rt_mutex where
313 * task t can find it, and task t will release the mutex when it
314 * exits its outermost RCU read-side critical section. Then
315 * simply acquiring this artificial rt_mutex will boost task
316 * t's priority. (Thanks to tglx for suggesting this approach!)
318 t = container_of(tb, struct task_struct, rcu_node_entry);
319 rt_mutex_init_proxy_locked(&mtx, t);
320 t->rcu_boost_mutex = &mtx;
321 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BOOSTED;
322 raw_local_irq_restore(flags);
324 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
326 return ACCESS_ONCE(rcu_preempt_ctrlblk.boost_tasks) != NULL ||
327 ACCESS_ONCE(rcu_preempt_ctrlblk.exp_tasks) != NULL;
331 * Check to see if it is now time to start boosting RCU readers blocking
332 * the current grace period, and, if so, tell the rcu_kthread_task to
333 * start boosting them. If there is an expedited boost in progress,
334 * we wait for it to complete.
336 * If there are no blocked readers blocking the current grace period,
337 * return 0 to let the caller know, otherwise return 1. Note that this
338 * return value is independent of whether or not boosting was done.
340 static int rcu_initiate_boost(void)
342 if (!rcu_preempt_blocked_readers_cgp() &&
343 rcu_preempt_ctrlblk.exp_tasks == NULL) {
344 RCU_TRACE(rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++);
347 if (rcu_preempt_ctrlblk.exp_tasks != NULL ||
348 (rcu_preempt_ctrlblk.gp_tasks != NULL &&
349 rcu_preempt_ctrlblk.boost_tasks == NULL &&
350 ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))) {
351 if (rcu_preempt_ctrlblk.exp_tasks == NULL)
352 rcu_preempt_ctrlblk.boost_tasks =
353 rcu_preempt_ctrlblk.gp_tasks;
354 invoke_rcu_callbacks();
356 RCU_TRACE(rcu_initiate_boost_trace());
360 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
363 * Do priority-boost accounting for the start of a new grace period.
365 static void rcu_preempt_boost_start_gp(void)
367 rcu_preempt_ctrlblk.boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
370 #else /* #ifdef CONFIG_RCU_BOOST */
373 * If there is no RCU priority boosting, we don't initiate boosting,
374 * but we do indicate whether there are blocked readers blocking the
375 * current grace period.
377 static int rcu_initiate_boost(void)
379 return rcu_preempt_blocked_readers_cgp();
383 * If there is no RCU priority boosting, nothing to do at grace-period start.
385 static void rcu_preempt_boost_start_gp(void)
389 #endif /* else #ifdef CONFIG_RCU_BOOST */
392 * Record a preemptible-RCU quiescent state for the specified CPU. Note
393 * that this just means that the task currently running on the CPU is
394 * in a quiescent state. There might be any number of tasks blocked
395 * while in an RCU read-side critical section.
397 * Unlike the other rcu_*_qs() functions, callers to this function
398 * must disable irqs in order to protect the assignment to
399 * ->rcu_read_unlock_special.
401 * Because this is a single-CPU implementation, the only way a grace
402 * period can end is if the CPU is in a quiescent state. The reason is
403 * that a blocked preemptible-RCU reader can exit its critical section
404 * only if the CPU is running it at the time. Therefore, when the
405 * last task blocking the current grace period exits its RCU read-side
406 * critical section, neither the CPU nor blocked tasks will be stopping
407 * the current grace period. (In contrast, SMP implementations
408 * might have CPUs running in RCU read-side critical sections that
409 * block later grace periods -- but this is not possible given only
412 static void rcu_preempt_cpu_qs(void)
414 /* Record both CPU and task as having responded to current GP. */
415 rcu_preempt_ctrlblk.gpcpu = rcu_preempt_ctrlblk.gpnum;
416 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
418 /* If there is no GP then there is nothing more to do. */
419 if (!rcu_preempt_gp_in_progress())
422 * Check up on boosting. If there are readers blocking the
423 * current grace period, leave.
425 if (rcu_initiate_boost())
428 /* Advance callbacks. */
429 rcu_preempt_ctrlblk.completed = rcu_preempt_ctrlblk.gpnum;
430 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.rcb.curtail;
431 rcu_preempt_ctrlblk.rcb.curtail = rcu_preempt_ctrlblk.nexttail;
433 /* If there are no blocked readers, next GP is done instantly. */
434 if (!rcu_preempt_blocked_readers_any())
435 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.nexttail;
437 /* If there are done callbacks, cause them to be invoked. */
438 if (*rcu_preempt_ctrlblk.rcb.donetail != NULL)
439 invoke_rcu_callbacks();
443 * Start a new RCU grace period if warranted. Hard irqs must be disabled.
445 static void rcu_preempt_start_gp(void)
447 if (!rcu_preempt_gp_in_progress() && rcu_preempt_needs_another_gp()) {
449 /* Official start of GP. */
450 rcu_preempt_ctrlblk.gpnum++;
451 RCU_TRACE(rcu_preempt_ctrlblk.n_grace_periods++);
453 /* Any blocked RCU readers block new GP. */
454 if (rcu_preempt_blocked_readers_any())
455 rcu_preempt_ctrlblk.gp_tasks =
456 rcu_preempt_ctrlblk.blkd_tasks.next;
458 /* Set up for RCU priority boosting. */
459 rcu_preempt_boost_start_gp();
461 /* If there is no running reader, CPU is done with GP. */
462 if (!rcu_preempt_running_reader())
463 rcu_preempt_cpu_qs();
468 * We have entered the scheduler, and the current task might soon be
469 * context-switched away from. If this task is in an RCU read-side
470 * critical section, we will no longer be able to rely on the CPU to
471 * record that fact, so we enqueue the task on the blkd_tasks list.
472 * If the task started after the current grace period began, as recorded
473 * by ->gpcpu, we enqueue at the beginning of the list. Otherwise
474 * before the element referenced by ->gp_tasks (or at the tail if
475 * ->gp_tasks is NULL) and point ->gp_tasks at the newly added element.
476 * The task will dequeue itself when it exits the outermost enclosing
477 * RCU read-side critical section. Therefore, the current grace period
478 * cannot be permitted to complete until the ->gp_tasks pointer becomes
481 * Caller must disable preemption.
483 void rcu_preempt_note_context_switch(void)
485 struct task_struct *t = current;
488 local_irq_save(flags); /* must exclude scheduler_tick(). */
489 if (rcu_preempt_running_reader() > 0 &&
490 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
492 /* Possibly blocking in an RCU read-side critical section. */
493 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
496 * If this CPU has already checked in, then this task
497 * will hold up the next grace period rather than the
498 * current grace period. Queue the task accordingly.
499 * If the task is queued for the current grace period
500 * (i.e., this CPU has not yet passed through a quiescent
501 * state for the current grace period), then as long
502 * as that task remains queued, the current grace period
505 list_add(&t->rcu_node_entry, &rcu_preempt_ctrlblk.blkd_tasks);
506 if (rcu_cpu_blocking_cur_gp())
507 rcu_preempt_ctrlblk.gp_tasks = &t->rcu_node_entry;
508 } else if (rcu_preempt_running_reader() < 0 &&
509 t->rcu_read_unlock_special) {
511 * Complete exit from RCU read-side critical section on
512 * behalf of preempted instance of __rcu_read_unlock().
514 rcu_read_unlock_special(t);
518 * Either we were not in an RCU read-side critical section to
519 * begin with, or we have now recorded that critical section
520 * globally. Either way, we can now note a quiescent state
521 * for this CPU. Again, if we were in an RCU read-side critical
522 * section, and if that critical section was blocking the current
523 * grace period, then the fact that the task has been enqueued
524 * means that current grace period continues to be blocked.
526 rcu_preempt_cpu_qs();
527 local_irq_restore(flags);
531 * Tiny-preemptible RCU implementation for rcu_read_lock().
532 * Just increment ->rcu_read_lock_nesting, shared state will be updated
535 void __rcu_read_lock(void)
537 current->rcu_read_lock_nesting++;
538 barrier(); /* needed if we ever invoke rcu_read_lock in rcutiny.c */
540 EXPORT_SYMBOL_GPL(__rcu_read_lock);
543 * Handle special cases during rcu_read_unlock(), such as needing to
544 * notify RCU core processing or task having blocked during the RCU
545 * read-side critical section.
547 static noinline void rcu_read_unlock_special(struct task_struct *t)
552 struct list_head *np;
556 * NMI handlers cannot block and cannot safely manipulate state.
557 * They therefore cannot possibly be special, so just leave.
562 local_irq_save(flags);
565 * If RCU core is waiting for this CPU to exit critical section,
566 * let it know that we have done so.
568 special = t->rcu_read_unlock_special;
569 if (special & RCU_READ_UNLOCK_NEED_QS)
570 rcu_preempt_cpu_qs();
572 /* Hardware IRQ handlers cannot block. */
574 local_irq_restore(flags);
578 /* Clean up if blocked during RCU read-side critical section. */
579 if (special & RCU_READ_UNLOCK_BLOCKED) {
580 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
583 * Remove this task from the ->blkd_tasks list and adjust
584 * any pointers that might have been referencing it.
586 empty = !rcu_preempt_blocked_readers_cgp();
587 empty_exp = rcu_preempt_ctrlblk.exp_tasks == NULL;
588 np = rcu_next_node_entry(t);
589 list_del_init(&t->rcu_node_entry);
590 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.gp_tasks)
591 rcu_preempt_ctrlblk.gp_tasks = np;
592 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.exp_tasks)
593 rcu_preempt_ctrlblk.exp_tasks = np;
594 #ifdef CONFIG_RCU_BOOST
595 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.boost_tasks)
596 rcu_preempt_ctrlblk.boost_tasks = np;
597 #endif /* #ifdef CONFIG_RCU_BOOST */
600 * If this was the last task on the current list, and if
601 * we aren't waiting on the CPU, report the quiescent state
602 * and start a new grace period if needed.
604 if (!empty && !rcu_preempt_blocked_readers_cgp()) {
605 rcu_preempt_cpu_qs();
606 rcu_preempt_start_gp();
610 * If this was the last task on the expedited lists,
611 * then we need wake up the waiting task.
613 if (!empty_exp && rcu_preempt_ctrlblk.exp_tasks == NULL)
614 rcu_report_exp_done();
616 #ifdef CONFIG_RCU_BOOST
617 /* Unboost self if was boosted. */
618 if (special & RCU_READ_UNLOCK_BOOSTED) {
619 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BOOSTED;
620 rt_mutex_unlock(t->rcu_boost_mutex);
621 t->rcu_boost_mutex = NULL;
623 #endif /* #ifdef CONFIG_RCU_BOOST */
624 local_irq_restore(flags);
628 * Tiny-preemptible RCU implementation for rcu_read_unlock().
629 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
630 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
631 * invoke rcu_read_unlock_special() to clean up after a context switch
632 * in an RCU read-side critical section and other special cases.
634 void __rcu_read_unlock(void)
636 struct task_struct *t = current;
638 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutiny.c */
639 if (t->rcu_read_lock_nesting != 1)
640 --t->rcu_read_lock_nesting;
642 t->rcu_read_lock_nesting = INT_MIN;
643 barrier(); /* assign before ->rcu_read_unlock_special load */
644 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
645 rcu_read_unlock_special(t);
646 barrier(); /* ->rcu_read_unlock_special load before assign */
647 t->rcu_read_lock_nesting = 0;
649 #ifdef CONFIG_PROVE_LOCKING
651 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
653 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
655 #endif /* #ifdef CONFIG_PROVE_LOCKING */
657 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
660 * Check for a quiescent state from the current CPU. When a task blocks,
661 * the task is recorded in the rcu_preempt_ctrlblk structure, which is
662 * checked elsewhere. This is called from the scheduling-clock interrupt.
664 * Caller must disable hard irqs.
666 static void rcu_preempt_check_callbacks(void)
668 struct task_struct *t = current;
670 if (rcu_preempt_gp_in_progress() &&
671 (!rcu_preempt_running_reader() ||
672 !rcu_cpu_blocking_cur_gp()))
673 rcu_preempt_cpu_qs();
674 if (&rcu_preempt_ctrlblk.rcb.rcucblist !=
675 rcu_preempt_ctrlblk.rcb.donetail)
676 invoke_rcu_callbacks();
677 if (rcu_preempt_gp_in_progress() &&
678 rcu_cpu_blocking_cur_gp() &&
679 rcu_preempt_running_reader() > 0)
680 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
684 * TINY_PREEMPT_RCU has an extra callback-list tail pointer to
685 * update, so this is invoked from rcu_process_callbacks() to
686 * handle that case. Of course, it is invoked for all flavors of
687 * RCU, but RCU callbacks can appear only on one of the lists, and
688 * neither ->nexttail nor ->donetail can possibly be NULL, so there
689 * is no need for an explicit check.
691 static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
693 if (rcu_preempt_ctrlblk.nexttail == rcp->donetail)
694 rcu_preempt_ctrlblk.nexttail = &rcp->rcucblist;
698 * Process callbacks for preemptible RCU.
700 static void rcu_preempt_process_callbacks(void)
702 __rcu_process_callbacks(&rcu_preempt_ctrlblk.rcb);
706 * Queue a preemptible -RCU callback for invocation after a grace period.
708 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
712 debug_rcu_head_queue(head);
716 local_irq_save(flags);
717 *rcu_preempt_ctrlblk.nexttail = head;
718 rcu_preempt_ctrlblk.nexttail = &head->next;
719 RCU_TRACE(rcu_preempt_ctrlblk.rcb.qlen++);
720 rcu_preempt_start_gp(); /* checks to see if GP needed. */
721 local_irq_restore(flags);
723 EXPORT_SYMBOL_GPL(call_rcu);
726 * synchronize_rcu - wait until a grace period has elapsed.
728 * Control will return to the caller some time after a full grace
729 * period has elapsed, in other words after all currently executing RCU
730 * read-side critical sections have completed. RCU read-side critical
731 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
734 void synchronize_rcu(void)
736 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
737 !lock_is_held(&rcu_lock_map) &&
738 !lock_is_held(&rcu_sched_lock_map),
739 "Illegal synchronize_rcu() in RCU read-side critical section");
741 #ifdef CONFIG_DEBUG_LOCK_ALLOC
742 if (!rcu_scheduler_active)
744 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
746 WARN_ON_ONCE(rcu_preempt_running_reader());
747 if (!rcu_preempt_blocked_readers_any())
750 /* Once we get past the fastpath checks, same code as rcu_barrier(). */
753 EXPORT_SYMBOL_GPL(synchronize_rcu);
755 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
756 static unsigned long sync_rcu_preempt_exp_count;
757 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
760 * Return non-zero if there are any tasks in RCU read-side critical
761 * sections blocking the current preemptible-RCU expedited grace period.
762 * If there is no preemptible-RCU expedited grace period currently in
763 * progress, returns zero unconditionally.
765 static int rcu_preempted_readers_exp(void)
767 return rcu_preempt_ctrlblk.exp_tasks != NULL;
771 * Report the exit from RCU read-side critical section for the last task
772 * that queued itself during or before the current expedited preemptible-RCU
775 static void rcu_report_exp_done(void)
777 wake_up(&sync_rcu_preempt_exp_wq);
781 * Wait for an rcu-preempt grace period, but expedite it. The basic idea
782 * is to rely in the fact that there is but one CPU, and that it is
783 * illegal for a task to invoke synchronize_rcu_expedited() while in a
784 * preemptible-RCU read-side critical section. Therefore, any such
785 * critical sections must correspond to blocked tasks, which must therefore
786 * be on the ->blkd_tasks list. So just record the current head of the
787 * list in the ->exp_tasks pointer, and wait for all tasks including and
788 * after the task pointed to by ->exp_tasks to drain.
790 void synchronize_rcu_expedited(void)
793 struct rcu_preempt_ctrlblk *rpcp = &rcu_preempt_ctrlblk;
796 barrier(); /* ensure prior action seen before grace period. */
798 WARN_ON_ONCE(rcu_preempt_running_reader());
801 * Acquire lock so that there is only one preemptible RCU grace
802 * period in flight. Of course, if someone does the expedited
803 * grace period for us while we are acquiring the lock, just leave.
805 snap = sync_rcu_preempt_exp_count + 1;
806 mutex_lock(&sync_rcu_preempt_exp_mutex);
807 if (ULONG_CMP_LT(snap, sync_rcu_preempt_exp_count))
808 goto unlock_mb_ret; /* Others did our work for us. */
810 local_irq_save(flags);
813 * All RCU readers have to already be on blkd_tasks because
814 * we cannot legally be executing in an RCU read-side critical
818 /* Snapshot current head of ->blkd_tasks list. */
819 rpcp->exp_tasks = rpcp->blkd_tasks.next;
820 if (rpcp->exp_tasks == &rpcp->blkd_tasks)
821 rpcp->exp_tasks = NULL;
823 /* Wait for tail of ->blkd_tasks list to drain. */
824 if (!rcu_preempted_readers_exp())
825 local_irq_restore(flags);
827 rcu_initiate_boost();
828 local_irq_restore(flags);
829 wait_event(sync_rcu_preempt_exp_wq,
830 !rcu_preempted_readers_exp());
833 /* Clean up and exit. */
834 barrier(); /* ensure expedited GP seen before counter increment. */
835 sync_rcu_preempt_exp_count++;
837 mutex_unlock(&sync_rcu_preempt_exp_mutex);
838 barrier(); /* ensure subsequent action seen after grace period. */
840 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
843 * Does preemptible RCU need the CPU to stay out of dynticks mode?
845 int rcu_preempt_needs_cpu(void)
847 if (!rcu_preempt_running_reader())
848 rcu_preempt_cpu_qs();
849 return rcu_preempt_ctrlblk.rcb.rcucblist != NULL;
853 * Check for a task exiting while in a preemptible -RCU read-side
854 * critical section, clean up if so. No need to issue warnings,
855 * as debug_check_no_locks_held() already does this if lockdep
860 struct task_struct *t = current;
862 if (t->rcu_read_lock_nesting == 0)
864 t->rcu_read_lock_nesting = 1;
868 #else /* #ifdef CONFIG_TINY_PREEMPT_RCU */
870 #ifdef CONFIG_RCU_TRACE
873 * Because preemptible RCU does not exist, it is not necessary to
874 * dump out its statistics.
876 static void show_tiny_preempt_stats(struct seq_file *m)
880 #endif /* #ifdef CONFIG_RCU_TRACE */
883 * Because preemptible RCU does not exist, it never has any callbacks
886 static void rcu_preempt_check_callbacks(void)
891 * Because preemptible RCU does not exist, it never has any callbacks
894 static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
899 * Because preemptible RCU does not exist, it never has any callbacks
902 static void rcu_preempt_process_callbacks(void)
906 #endif /* #else #ifdef CONFIG_TINY_PREEMPT_RCU */
908 #ifdef CONFIG_RCU_BOOST
911 * Wake up rcu_kthread() to process callbacks now eligible for invocation
912 * or to boost readers.
914 static void invoke_rcu_callbacks(void)
916 have_rcu_kthread_work = 1;
917 if (rcu_kthread_task != NULL)
918 wake_up(&rcu_kthread_wq);
921 #ifdef CONFIG_RCU_TRACE
924 * Is the current CPU running the RCU-callbacks kthread?
925 * Caller must have preemption disabled.
927 static bool rcu_is_callbacks_kthread(void)
929 return rcu_kthread_task == current;
932 #endif /* #ifdef CONFIG_RCU_TRACE */
935 * This kthread invokes RCU callbacks whose grace periods have
936 * elapsed. It is awakened as needed, and takes the place of the
937 * RCU_SOFTIRQ that is used for this purpose when boosting is disabled.
938 * This is a kthread, but it is never stopped, at least not until
939 * the system goes down.
941 static int rcu_kthread(void *arg)
944 unsigned long morework;
948 wait_event_interruptible(rcu_kthread_wq,
949 have_rcu_kthread_work != 0);
950 morework = rcu_boost();
951 local_irq_save(flags);
952 work = have_rcu_kthread_work;
953 have_rcu_kthread_work = morework;
954 local_irq_restore(flags);
956 rcu_process_callbacks(NULL);
957 schedule_timeout_interruptible(1); /* Leave CPU for others. */
960 return 0; /* Not reached, but needed to shut gcc up. */
964 * Spawn the kthread that invokes RCU callbacks.
966 static int __init rcu_spawn_kthreads(void)
968 struct sched_param sp;
970 rcu_kthread_task = kthread_run(rcu_kthread, NULL, "rcu_kthread");
971 sp.sched_priority = RCU_BOOST_PRIO;
972 sched_setscheduler_nocheck(rcu_kthread_task, SCHED_FIFO, &sp);
975 early_initcall(rcu_spawn_kthreads);
977 #else /* #ifdef CONFIG_RCU_BOOST */
979 /* Hold off callback invocation until early_initcall() time. */
980 static int rcu_scheduler_fully_active __read_mostly;
983 * Start up softirq processing of callbacks.
985 void invoke_rcu_callbacks(void)
987 if (rcu_scheduler_fully_active)
988 raise_softirq(RCU_SOFTIRQ);
991 #ifdef CONFIG_RCU_TRACE
994 * There is no callback kthread, so this thread is never it.
996 static bool rcu_is_callbacks_kthread(void)
1001 #endif /* #ifdef CONFIG_RCU_TRACE */
1003 static int __init rcu_scheduler_really_started(void)
1005 rcu_scheduler_fully_active = 1;
1006 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1007 raise_softirq(RCU_SOFTIRQ); /* Invoke any callbacks from early boot. */
1010 early_initcall(rcu_scheduler_really_started);
1012 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1014 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1015 #include <linux/kernel_stat.h>
1018 * During boot, we forgive RCU lockdep issues. After this function is
1019 * invoked, we start taking RCU lockdep issues seriously.
1021 void __init rcu_scheduler_starting(void)
1023 WARN_ON(nr_context_switches() > 0);
1024 rcu_scheduler_active = 1;
1027 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
1029 #ifdef CONFIG_RCU_TRACE
1031 #ifdef CONFIG_RCU_BOOST
1033 static void rcu_initiate_boost_trace(void)
1035 if (list_empty(&rcu_preempt_ctrlblk.blkd_tasks))
1036 rcu_preempt_ctrlblk.n_balk_blkd_tasks++;
1037 else if (rcu_preempt_ctrlblk.gp_tasks == NULL &&
1038 rcu_preempt_ctrlblk.exp_tasks == NULL)
1039 rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++;
1040 else if (rcu_preempt_ctrlblk.boost_tasks != NULL)
1041 rcu_preempt_ctrlblk.n_balk_boost_tasks++;
1042 else if (!ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))
1043 rcu_preempt_ctrlblk.n_balk_notyet++;
1045 rcu_preempt_ctrlblk.n_balk_nos++;
1048 #endif /* #ifdef CONFIG_RCU_BOOST */
1050 static void rcu_trace_sub_qlen(struct rcu_ctrlblk *rcp, int n)
1052 unsigned long flags;
1054 raw_local_irq_save(flags);
1056 raw_local_irq_restore(flags);
1060 * Dump statistics for TINY_RCU, such as they are.
1062 static int show_tiny_stats(struct seq_file *m, void *unused)
1064 show_tiny_preempt_stats(m);
1065 seq_printf(m, "rcu_sched: qlen: %ld\n", rcu_sched_ctrlblk.qlen);
1066 seq_printf(m, "rcu_bh: qlen: %ld\n", rcu_bh_ctrlblk.qlen);
1070 static int show_tiny_stats_open(struct inode *inode, struct file *file)
1072 return single_open(file, show_tiny_stats, NULL);
1075 static const struct file_operations show_tiny_stats_fops = {
1076 .owner = THIS_MODULE,
1077 .open = show_tiny_stats_open,
1079 .llseek = seq_lseek,
1080 .release = single_release,
1083 static struct dentry *rcudir;
1085 static int __init rcutiny_trace_init(void)
1087 struct dentry *retval;
1089 rcudir = debugfs_create_dir("rcu", NULL);
1092 retval = debugfs_create_file("rcudata", 0444, rcudir,
1093 NULL, &show_tiny_stats_fops);
1098 debugfs_remove_recursive(rcudir);
1102 static void __exit rcutiny_trace_cleanup(void)
1104 debugfs_remove_recursive(rcudir);
1107 module_init(rcutiny_trace_init);
1108 module_exit(rcutiny_trace_cleanup);
1110 MODULE_AUTHOR("Paul E. McKenney");
1111 MODULE_DESCRIPTION("Read-Copy Update tracing for tiny implementation");
1112 MODULE_LICENSE("GPL");
1114 #endif /* #ifdef CONFIG_RCU_TRACE */