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 int rcu_preempted_readers_exp(void);
136 static void rcu_report_exp_done(void);
139 * Return true if the CPU has not yet responded to the current grace period.
141 static int rcu_cpu_blocking_cur_gp(void)
143 return rcu_preempt_ctrlblk.gpcpu != rcu_preempt_ctrlblk.gpnum;
147 * Check for a running RCU reader. Because there is only one CPU,
148 * there can be but one running RCU reader at a time. ;-)
150 * Returns zero if there are no running readers. Returns a positive
151 * number if there is at least one reader within its RCU read-side
152 * critical section. Returns a negative number if an outermost reader
153 * is in the midst of exiting from its RCU read-side critical section
155 * Returns zero if there are no running readers. Returns a positive
156 * number if there is at least one reader within its RCU read-side
157 * critical section. Returns a negative number if an outermost reader
158 * is in the midst of exiting from its RCU read-side critical section.
160 static int rcu_preempt_running_reader(void)
162 return current->rcu_read_lock_nesting;
166 * Check for preempted RCU readers blocking any grace period.
167 * If the caller needs a reliable answer, it must disable hard irqs.
169 static int rcu_preempt_blocked_readers_any(void)
171 return !list_empty(&rcu_preempt_ctrlblk.blkd_tasks);
175 * Check for preempted RCU readers blocking the current grace period.
176 * If the caller needs a reliable answer, it must disable hard irqs.
178 static int rcu_preempt_blocked_readers_cgp(void)
180 return rcu_preempt_ctrlblk.gp_tasks != NULL;
184 * Return true if another preemptible-RCU grace period is needed.
186 static int rcu_preempt_needs_another_gp(void)
188 return *rcu_preempt_ctrlblk.rcb.curtail != NULL;
192 * Return true if a preemptible-RCU grace period is in progress.
193 * The caller must disable hardirqs.
195 static int rcu_preempt_gp_in_progress(void)
197 return rcu_preempt_ctrlblk.completed != rcu_preempt_ctrlblk.gpnum;
201 * Advance a ->blkd_tasks-list pointer to the next entry, instead
202 * returning NULL if at the end of the list.
204 static struct list_head *rcu_next_node_entry(struct task_struct *t)
206 struct list_head *np;
208 np = t->rcu_node_entry.next;
209 if (np == &rcu_preempt_ctrlblk.blkd_tasks)
214 #ifdef CONFIG_RCU_TRACE
216 #ifdef CONFIG_RCU_BOOST
217 static void rcu_initiate_boost_trace(void);
218 #endif /* #ifdef CONFIG_RCU_BOOST */
221 * Dump additional statistice for TINY_PREEMPT_RCU.
223 static void show_tiny_preempt_stats(struct seq_file *m)
225 seq_printf(m, "rcu_preempt: qlen=%ld gp=%lu g%u/p%u/c%u tasks=%c%c%c\n",
226 rcu_preempt_ctrlblk.rcb.qlen,
227 rcu_preempt_ctrlblk.n_grace_periods,
228 rcu_preempt_ctrlblk.gpnum,
229 rcu_preempt_ctrlblk.gpcpu,
230 rcu_preempt_ctrlblk.completed,
231 "T."[list_empty(&rcu_preempt_ctrlblk.blkd_tasks)],
232 "N."[!rcu_preempt_ctrlblk.gp_tasks],
233 "E."[!rcu_preempt_ctrlblk.exp_tasks]);
234 #ifdef CONFIG_RCU_BOOST
235 seq_printf(m, "%sttb=%c ntb=%lu neb=%lu nnb=%lu j=%04x bt=%04x\n",
237 "B."[!rcu_preempt_ctrlblk.boost_tasks],
238 rcu_preempt_ctrlblk.n_tasks_boosted,
239 rcu_preempt_ctrlblk.n_exp_boosts,
240 rcu_preempt_ctrlblk.n_normal_boosts,
241 (int)(jiffies & 0xffff),
242 (int)(rcu_preempt_ctrlblk.boost_time & 0xffff));
243 seq_printf(m, "%s: nt=%lu egt=%lu bt=%lu ny=%lu nos=%lu\n",
245 rcu_preempt_ctrlblk.n_balk_blkd_tasks,
246 rcu_preempt_ctrlblk.n_balk_exp_gp_tasks,
247 rcu_preempt_ctrlblk.n_balk_boost_tasks,
248 rcu_preempt_ctrlblk.n_balk_notyet,
249 rcu_preempt_ctrlblk.n_balk_nos);
250 #endif /* #ifdef CONFIG_RCU_BOOST */
253 #endif /* #ifdef CONFIG_RCU_TRACE */
255 #ifdef CONFIG_RCU_BOOST
257 #include "rtmutex_common.h"
259 #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO
261 /* Controls for rcu_kthread() kthread. */
262 static struct task_struct *rcu_kthread_task;
263 static DECLARE_WAIT_QUEUE_HEAD(rcu_kthread_wq);
264 static unsigned long have_rcu_kthread_work;
267 * Carry out RCU priority boosting on the task indicated by ->boost_tasks,
268 * and advance ->boost_tasks to the next task in the ->blkd_tasks list.
270 static int rcu_boost(void)
274 struct task_struct *t;
275 struct list_head *tb;
277 if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
278 rcu_preempt_ctrlblk.exp_tasks == NULL)
279 return 0; /* Nothing to boost. */
281 local_irq_save(flags);
284 * Recheck with irqs disabled: all tasks in need of boosting
285 * might exit their RCU read-side critical sections on their own
286 * if we are preempted just before disabling irqs.
288 if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
289 rcu_preempt_ctrlblk.exp_tasks == NULL) {
290 local_irq_restore(flags);
295 * Preferentially boost tasks blocking expedited grace periods.
296 * This cannot starve the normal grace periods because a second
297 * expedited grace period must boost all blocked tasks, including
298 * those blocking the pre-existing normal grace period.
300 if (rcu_preempt_ctrlblk.exp_tasks != NULL) {
301 tb = rcu_preempt_ctrlblk.exp_tasks;
302 RCU_TRACE(rcu_preempt_ctrlblk.n_exp_boosts++);
304 tb = rcu_preempt_ctrlblk.boost_tasks;
305 RCU_TRACE(rcu_preempt_ctrlblk.n_normal_boosts++);
307 RCU_TRACE(rcu_preempt_ctrlblk.n_tasks_boosted++);
310 * We boost task t by manufacturing an rt_mutex that appears to
311 * be held by task t. We leave a pointer to that rt_mutex where
312 * task t can find it, and task t will release the mutex when it
313 * exits its outermost RCU read-side critical section. Then
314 * simply acquiring this artificial rt_mutex will boost task
315 * t's priority. (Thanks to tglx for suggesting this approach!)
317 t = container_of(tb, struct task_struct, rcu_node_entry);
318 rt_mutex_init_proxy_locked(&mtx, t);
319 t->rcu_boost_mutex = &mtx;
320 local_irq_restore(flags);
322 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
324 return ACCESS_ONCE(rcu_preempt_ctrlblk.boost_tasks) != NULL ||
325 ACCESS_ONCE(rcu_preempt_ctrlblk.exp_tasks) != NULL;
329 * Check to see if it is now time to start boosting RCU readers blocking
330 * the current grace period, and, if so, tell the rcu_kthread_task to
331 * start boosting them. If there is an expedited boost in progress,
332 * we wait for it to complete.
334 * If there are no blocked readers blocking the current grace period,
335 * return 0 to let the caller know, otherwise return 1. Note that this
336 * return value is independent of whether or not boosting was done.
338 static int rcu_initiate_boost(void)
340 if (!rcu_preempt_blocked_readers_cgp() &&
341 rcu_preempt_ctrlblk.exp_tasks == NULL) {
342 RCU_TRACE(rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++);
345 if (rcu_preempt_ctrlblk.exp_tasks != NULL ||
346 (rcu_preempt_ctrlblk.gp_tasks != NULL &&
347 rcu_preempt_ctrlblk.boost_tasks == NULL &&
348 ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))) {
349 if (rcu_preempt_ctrlblk.exp_tasks == NULL)
350 rcu_preempt_ctrlblk.boost_tasks =
351 rcu_preempt_ctrlblk.gp_tasks;
352 invoke_rcu_callbacks();
354 RCU_TRACE(rcu_initiate_boost_trace());
359 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
362 * Do priority-boost accounting for the start of a new grace period.
364 static void rcu_preempt_boost_start_gp(void)
366 rcu_preempt_ctrlblk.boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
369 #else /* #ifdef CONFIG_RCU_BOOST */
372 * If there is no RCU priority boosting, we don't initiate boosting,
373 * but we do indicate whether there are blocked readers blocking the
374 * current grace period.
376 static int rcu_initiate_boost(void)
378 return rcu_preempt_blocked_readers_cgp();
382 * If there is no RCU priority boosting, nothing to do at grace-period start.
384 static void rcu_preempt_boost_start_gp(void)
388 #endif /* else #ifdef CONFIG_RCU_BOOST */
391 * Record a preemptible-RCU quiescent state for the specified CPU. Note
392 * that this just means that the task currently running on the CPU is
393 * in a quiescent state. There might be any number of tasks blocked
394 * while in an RCU read-side critical section.
396 * Unlike the other rcu_*_qs() functions, callers to this function
397 * must disable irqs in order to protect the assignment to
398 * ->rcu_read_unlock_special.
400 * Because this is a single-CPU implementation, the only way a grace
401 * period can end is if the CPU is in a quiescent state. The reason is
402 * that a blocked preemptible-RCU reader can exit its critical section
403 * only if the CPU is running it at the time. Therefore, when the
404 * last task blocking the current grace period exits its RCU read-side
405 * critical section, neither the CPU nor blocked tasks will be stopping
406 * the current grace period. (In contrast, SMP implementations
407 * might have CPUs running in RCU read-side critical sections that
408 * block later grace periods -- but this is not possible given only
411 static void rcu_preempt_cpu_qs(void)
413 /* Record both CPU and task as having responded to current GP. */
414 rcu_preempt_ctrlblk.gpcpu = rcu_preempt_ctrlblk.gpnum;
415 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
417 /* If there is no GP then there is nothing more to do. */
418 if (!rcu_preempt_gp_in_progress())
421 * Check up on boosting. If there are readers blocking the
422 * current grace period, leave.
424 if (rcu_initiate_boost())
427 /* Advance callbacks. */
428 rcu_preempt_ctrlblk.completed = rcu_preempt_ctrlblk.gpnum;
429 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.rcb.curtail;
430 rcu_preempt_ctrlblk.rcb.curtail = rcu_preempt_ctrlblk.nexttail;
432 /* If there are no blocked readers, next GP is done instantly. */
433 if (!rcu_preempt_blocked_readers_any())
434 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.nexttail;
436 /* If there are done callbacks, cause them to be invoked. */
437 if (*rcu_preempt_ctrlblk.rcb.donetail != NULL)
438 invoke_rcu_callbacks();
442 * Start a new RCU grace period if warranted. Hard irqs must be disabled.
444 static void rcu_preempt_start_gp(void)
446 if (!rcu_preempt_gp_in_progress() && rcu_preempt_needs_another_gp()) {
448 /* Official start of GP. */
449 rcu_preempt_ctrlblk.gpnum++;
450 RCU_TRACE(rcu_preempt_ctrlblk.n_grace_periods++);
452 /* Any blocked RCU readers block new GP. */
453 if (rcu_preempt_blocked_readers_any())
454 rcu_preempt_ctrlblk.gp_tasks =
455 rcu_preempt_ctrlblk.blkd_tasks.next;
457 /* Set up for RCU priority boosting. */
458 rcu_preempt_boost_start_gp();
460 /* If there is no running reader, CPU is done with GP. */
461 if (!rcu_preempt_running_reader())
462 rcu_preempt_cpu_qs();
467 * We have entered the scheduler, and the current task might soon be
468 * context-switched away from. If this task is in an RCU read-side
469 * critical section, we will no longer be able to rely on the CPU to
470 * record that fact, so we enqueue the task on the blkd_tasks list.
471 * If the task started after the current grace period began, as recorded
472 * by ->gpcpu, we enqueue at the beginning of the list. Otherwise
473 * before the element referenced by ->gp_tasks (or at the tail if
474 * ->gp_tasks is NULL) and point ->gp_tasks at the newly added element.
475 * The task will dequeue itself when it exits the outermost enclosing
476 * RCU read-side critical section. Therefore, the current grace period
477 * cannot be permitted to complete until the ->gp_tasks pointer becomes
480 * Caller must disable preemption.
482 void rcu_preempt_note_context_switch(void)
484 struct task_struct *t = current;
487 local_irq_save(flags); /* must exclude scheduler_tick(). */
488 if (rcu_preempt_running_reader() > 0 &&
489 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
491 /* Possibly blocking in an RCU read-side critical section. */
492 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
495 * If this CPU has already checked in, then this task
496 * will hold up the next grace period rather than the
497 * current grace period. Queue the task accordingly.
498 * If the task is queued for the current grace period
499 * (i.e., this CPU has not yet passed through a quiescent
500 * state for the current grace period), then as long
501 * as that task remains queued, the current grace period
504 list_add(&t->rcu_node_entry, &rcu_preempt_ctrlblk.blkd_tasks);
505 if (rcu_cpu_blocking_cur_gp())
506 rcu_preempt_ctrlblk.gp_tasks = &t->rcu_node_entry;
507 } else if (rcu_preempt_running_reader() < 0 &&
508 t->rcu_read_unlock_special) {
510 * Complete exit from RCU read-side critical section on
511 * behalf of preempted instance of __rcu_read_unlock().
513 rcu_read_unlock_special(t);
517 * Either we were not in an RCU read-side critical section to
518 * begin with, or we have now recorded that critical section
519 * globally. Either way, we can now note a quiescent state
520 * for this CPU. Again, if we were in an RCU read-side critical
521 * section, and if that critical section was blocking the current
522 * grace period, then the fact that the task has been enqueued
523 * means that current grace period continues to be blocked.
525 rcu_preempt_cpu_qs();
526 local_irq_restore(flags);
530 * Handle special cases during rcu_read_unlock(), such as needing to
531 * notify RCU core processing or task having blocked during the RCU
532 * read-side critical section.
534 void rcu_read_unlock_special(struct task_struct *t)
539 struct list_head *np;
540 #ifdef CONFIG_RCU_BOOST
541 struct rt_mutex *rbmp = NULL;
542 #endif /* #ifdef CONFIG_RCU_BOOST */
546 * NMI handlers cannot block and cannot safely manipulate state.
547 * They therefore cannot possibly be special, so just leave.
552 local_irq_save(flags);
555 * If RCU core is waiting for this CPU to exit critical section,
556 * let it know that we have done so.
558 special = t->rcu_read_unlock_special;
559 if (special & RCU_READ_UNLOCK_NEED_QS)
560 rcu_preempt_cpu_qs();
562 /* Hardware IRQ handlers cannot block. */
563 if (in_irq() || in_serving_softirq()) {
564 local_irq_restore(flags);
568 /* Clean up if blocked during RCU read-side critical section. */
569 if (special & RCU_READ_UNLOCK_BLOCKED) {
570 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
573 * Remove this task from the ->blkd_tasks list and adjust
574 * any pointers that might have been referencing it.
576 empty = !rcu_preempt_blocked_readers_cgp();
577 empty_exp = rcu_preempt_ctrlblk.exp_tasks == NULL;
578 np = rcu_next_node_entry(t);
579 list_del_init(&t->rcu_node_entry);
580 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.gp_tasks)
581 rcu_preempt_ctrlblk.gp_tasks = np;
582 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.exp_tasks)
583 rcu_preempt_ctrlblk.exp_tasks = np;
584 #ifdef CONFIG_RCU_BOOST
585 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.boost_tasks)
586 rcu_preempt_ctrlblk.boost_tasks = np;
587 #endif /* #ifdef CONFIG_RCU_BOOST */
590 * If this was the last task on the current list, and if
591 * we aren't waiting on the CPU, report the quiescent state
592 * and start a new grace period if needed.
594 if (!empty && !rcu_preempt_blocked_readers_cgp()) {
595 rcu_preempt_cpu_qs();
596 rcu_preempt_start_gp();
600 * If this was the last task on the expedited lists,
601 * then we need wake up the waiting task.
603 if (!empty_exp && rcu_preempt_ctrlblk.exp_tasks == NULL)
604 rcu_report_exp_done();
606 #ifdef CONFIG_RCU_BOOST
607 /* Unboost self if was boosted. */
608 if (t->rcu_boost_mutex != NULL) {
609 rbmp = t->rcu_boost_mutex;
610 t->rcu_boost_mutex = NULL;
611 rt_mutex_unlock(rbmp);
613 #endif /* #ifdef CONFIG_RCU_BOOST */
614 local_irq_restore(flags);
618 * Check for a quiescent state from the current CPU. When a task blocks,
619 * the task is recorded in the rcu_preempt_ctrlblk structure, which is
620 * checked elsewhere. This is called from the scheduling-clock interrupt.
622 * Caller must disable hard irqs.
624 static void rcu_preempt_check_callbacks(void)
626 struct task_struct *t = current;
628 if (rcu_preempt_gp_in_progress() &&
629 (!rcu_preempt_running_reader() ||
630 !rcu_cpu_blocking_cur_gp()))
631 rcu_preempt_cpu_qs();
632 if (&rcu_preempt_ctrlblk.rcb.rcucblist !=
633 rcu_preempt_ctrlblk.rcb.donetail)
634 invoke_rcu_callbacks();
635 if (rcu_preempt_gp_in_progress() &&
636 rcu_cpu_blocking_cur_gp() &&
637 rcu_preempt_running_reader() > 0)
638 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
642 * TINY_PREEMPT_RCU has an extra callback-list tail pointer to
643 * update, so this is invoked from rcu_process_callbacks() to
644 * handle that case. Of course, it is invoked for all flavors of
645 * RCU, but RCU callbacks can appear only on one of the lists, and
646 * neither ->nexttail nor ->donetail can possibly be NULL, so there
647 * is no need for an explicit check.
649 static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
651 if (rcu_preempt_ctrlblk.nexttail == rcp->donetail)
652 rcu_preempt_ctrlblk.nexttail = &rcp->rcucblist;
656 * Process callbacks for preemptible RCU.
658 static void rcu_preempt_process_callbacks(void)
660 __rcu_process_callbacks(&rcu_preempt_ctrlblk.rcb);
664 * Queue a preemptible -RCU callback for invocation after a grace period.
666 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
670 debug_rcu_head_queue(head);
674 local_irq_save(flags);
675 *rcu_preempt_ctrlblk.nexttail = head;
676 rcu_preempt_ctrlblk.nexttail = &head->next;
677 RCU_TRACE(rcu_preempt_ctrlblk.rcb.qlen++);
678 rcu_preempt_start_gp(); /* checks to see if GP needed. */
679 local_irq_restore(flags);
681 EXPORT_SYMBOL_GPL(call_rcu);
684 * synchronize_rcu - wait until a grace period has elapsed.
686 * Control will return to the caller some time after a full grace
687 * period has elapsed, in other words after all currently executing RCU
688 * read-side critical sections have completed. RCU read-side critical
689 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
692 void synchronize_rcu(void)
694 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
695 !lock_is_held(&rcu_lock_map) &&
696 !lock_is_held(&rcu_sched_lock_map),
697 "Illegal synchronize_rcu() in RCU read-side critical section");
699 #ifdef CONFIG_DEBUG_LOCK_ALLOC
700 if (!rcu_scheduler_active)
702 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
704 WARN_ON_ONCE(rcu_preempt_running_reader());
705 if (!rcu_preempt_blocked_readers_any())
708 /* Once we get past the fastpath checks, same code as rcu_barrier(). */
710 synchronize_rcu_expedited();
714 EXPORT_SYMBOL_GPL(synchronize_rcu);
716 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
717 static unsigned long sync_rcu_preempt_exp_count;
718 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
721 * Return non-zero if there are any tasks in RCU read-side critical
722 * sections blocking the current preemptible-RCU expedited grace period.
723 * If there is no preemptible-RCU expedited grace period currently in
724 * progress, returns zero unconditionally.
726 static int rcu_preempted_readers_exp(void)
728 return rcu_preempt_ctrlblk.exp_tasks != NULL;
732 * Report the exit from RCU read-side critical section for the last task
733 * that queued itself during or before the current expedited preemptible-RCU
736 static void rcu_report_exp_done(void)
738 wake_up(&sync_rcu_preempt_exp_wq);
742 * Wait for an rcu-preempt grace period, but expedite it. The basic idea
743 * is to rely in the fact that there is but one CPU, and that it is
744 * illegal for a task to invoke synchronize_rcu_expedited() while in a
745 * preemptible-RCU read-side critical section. Therefore, any such
746 * critical sections must correspond to blocked tasks, which must therefore
747 * be on the ->blkd_tasks list. So just record the current head of the
748 * list in the ->exp_tasks pointer, and wait for all tasks including and
749 * after the task pointed to by ->exp_tasks to drain.
751 void synchronize_rcu_expedited(void)
754 struct rcu_preempt_ctrlblk *rpcp = &rcu_preempt_ctrlblk;
757 barrier(); /* ensure prior action seen before grace period. */
759 WARN_ON_ONCE(rcu_preempt_running_reader());
762 * Acquire lock so that there is only one preemptible RCU grace
763 * period in flight. Of course, if someone does the expedited
764 * grace period for us while we are acquiring the lock, just leave.
766 snap = sync_rcu_preempt_exp_count + 1;
767 mutex_lock(&sync_rcu_preempt_exp_mutex);
768 if (ULONG_CMP_LT(snap, sync_rcu_preempt_exp_count))
769 goto unlock_mb_ret; /* Others did our work for us. */
771 local_irq_save(flags);
774 * All RCU readers have to already be on blkd_tasks because
775 * we cannot legally be executing in an RCU read-side critical
779 /* Snapshot current head of ->blkd_tasks list. */
780 rpcp->exp_tasks = rpcp->blkd_tasks.next;
781 if (rpcp->exp_tasks == &rpcp->blkd_tasks)
782 rpcp->exp_tasks = NULL;
784 /* Wait for tail of ->blkd_tasks list to drain. */
785 if (!rcu_preempted_readers_exp()) {
786 local_irq_restore(flags);
788 rcu_initiate_boost();
789 local_irq_restore(flags);
790 wait_event(sync_rcu_preempt_exp_wq,
791 !rcu_preempted_readers_exp());
794 /* Clean up and exit. */
795 barrier(); /* ensure expedited GP seen before counter increment. */
796 sync_rcu_preempt_exp_count++;
798 mutex_unlock(&sync_rcu_preempt_exp_mutex);
799 barrier(); /* ensure subsequent action seen after grace period. */
801 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
804 * Does preemptible RCU need the CPU to stay out of dynticks mode?
806 int rcu_preempt_needs_cpu(void)
808 return rcu_preempt_ctrlblk.rcb.rcucblist != NULL;
811 #else /* #ifdef CONFIG_TINY_PREEMPT_RCU */
813 #ifdef CONFIG_RCU_TRACE
816 * Because preemptible RCU does not exist, it is not necessary to
817 * dump out its statistics.
819 static void show_tiny_preempt_stats(struct seq_file *m)
823 #endif /* #ifdef CONFIG_RCU_TRACE */
826 * Because preemptible RCU does not exist, it never has any callbacks
829 static void rcu_preempt_check_callbacks(void)
834 * Because preemptible RCU does not exist, it never has any callbacks
837 static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
842 * Because preemptible RCU does not exist, it never has any callbacks
845 static void rcu_preempt_process_callbacks(void)
849 #endif /* #else #ifdef CONFIG_TINY_PREEMPT_RCU */
851 #ifdef CONFIG_RCU_BOOST
854 * Wake up rcu_kthread() to process callbacks now eligible for invocation
855 * or to boost readers.
857 static void invoke_rcu_callbacks(void)
859 have_rcu_kthread_work = 1;
860 if (rcu_kthread_task != NULL)
861 wake_up(&rcu_kthread_wq);
864 #ifdef CONFIG_RCU_TRACE
867 * Is the current CPU running the RCU-callbacks kthread?
868 * Caller must have preemption disabled.
870 static bool rcu_is_callbacks_kthread(void)
872 return rcu_kthread_task == current;
875 #endif /* #ifdef CONFIG_RCU_TRACE */
878 * This kthread invokes RCU callbacks whose grace periods have
879 * elapsed. It is awakened as needed, and takes the place of the
880 * RCU_SOFTIRQ that is used for this purpose when boosting is disabled.
881 * This is a kthread, but it is never stopped, at least not until
882 * the system goes down.
884 static int rcu_kthread(void *arg)
887 unsigned long morework;
891 wait_event_interruptible(rcu_kthread_wq,
892 have_rcu_kthread_work != 0);
893 morework = rcu_boost();
894 local_irq_save(flags);
895 work = have_rcu_kthread_work;
896 have_rcu_kthread_work = morework;
897 local_irq_restore(flags);
899 rcu_process_callbacks(NULL);
900 schedule_timeout_interruptible(1); /* Leave CPU for others. */
903 return 0; /* Not reached, but needed to shut gcc up. */
907 * Spawn the kthread that invokes RCU callbacks.
909 static int __init rcu_spawn_kthreads(void)
911 struct sched_param sp;
913 rcu_kthread_task = kthread_run(rcu_kthread, NULL, "rcu_kthread");
914 sp.sched_priority = RCU_BOOST_PRIO;
915 sched_setscheduler_nocheck(rcu_kthread_task, SCHED_FIFO, &sp);
918 early_initcall(rcu_spawn_kthreads);
920 #else /* #ifdef CONFIG_RCU_BOOST */
922 /* Hold off callback invocation until early_initcall() time. */
923 static int rcu_scheduler_fully_active __read_mostly;
926 * Start up softirq processing of callbacks.
928 void invoke_rcu_callbacks(void)
930 if (rcu_scheduler_fully_active)
931 raise_softirq(RCU_SOFTIRQ);
934 #ifdef CONFIG_RCU_TRACE
937 * There is no callback kthread, so this thread is never it.
939 static bool rcu_is_callbacks_kthread(void)
944 #endif /* #ifdef CONFIG_RCU_TRACE */
946 static int __init rcu_scheduler_really_started(void)
948 rcu_scheduler_fully_active = 1;
949 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
950 raise_softirq(RCU_SOFTIRQ); /* Invoke any callbacks from early boot. */
953 early_initcall(rcu_scheduler_really_started);
955 #endif /* #else #ifdef CONFIG_RCU_BOOST */
957 #ifdef CONFIG_DEBUG_LOCK_ALLOC
958 #include <linux/kernel_stat.h>
961 * During boot, we forgive RCU lockdep issues. After this function is
962 * invoked, we start taking RCU lockdep issues seriously.
964 void __init rcu_scheduler_starting(void)
966 WARN_ON(nr_context_switches() > 0);
967 rcu_scheduler_active = 1;
970 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
972 #ifdef CONFIG_RCU_TRACE
974 #ifdef CONFIG_RCU_BOOST
976 static void rcu_initiate_boost_trace(void)
978 if (list_empty(&rcu_preempt_ctrlblk.blkd_tasks))
979 rcu_preempt_ctrlblk.n_balk_blkd_tasks++;
980 else if (rcu_preempt_ctrlblk.gp_tasks == NULL &&
981 rcu_preempt_ctrlblk.exp_tasks == NULL)
982 rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++;
983 else if (rcu_preempt_ctrlblk.boost_tasks != NULL)
984 rcu_preempt_ctrlblk.n_balk_boost_tasks++;
985 else if (!ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))
986 rcu_preempt_ctrlblk.n_balk_notyet++;
988 rcu_preempt_ctrlblk.n_balk_nos++;
991 #endif /* #ifdef CONFIG_RCU_BOOST */
993 static void rcu_trace_sub_qlen(struct rcu_ctrlblk *rcp, int n)
997 local_irq_save(flags);
999 local_irq_restore(flags);
1003 * Dump statistics for TINY_RCU, such as they are.
1005 static int show_tiny_stats(struct seq_file *m, void *unused)
1007 show_tiny_preempt_stats(m);
1008 seq_printf(m, "rcu_sched: qlen: %ld\n", rcu_sched_ctrlblk.qlen);
1009 seq_printf(m, "rcu_bh: qlen: %ld\n", rcu_bh_ctrlblk.qlen);
1013 static int show_tiny_stats_open(struct inode *inode, struct file *file)
1015 return single_open(file, show_tiny_stats, NULL);
1018 static const struct file_operations show_tiny_stats_fops = {
1019 .owner = THIS_MODULE,
1020 .open = show_tiny_stats_open,
1022 .llseek = seq_lseek,
1023 .release = single_release,
1026 static struct dentry *rcudir;
1028 static int __init rcutiny_trace_init(void)
1030 struct dentry *retval;
1032 rcudir = debugfs_create_dir("rcu", NULL);
1035 retval = debugfs_create_file("rcudata", 0444, rcudir,
1036 NULL, &show_tiny_stats_fops);
1041 debugfs_remove_recursive(rcudir);
1045 static void __exit rcutiny_trace_cleanup(void)
1047 debugfs_remove_recursive(rcudir);
1050 module_init(rcutiny_trace_init);
1051 module_exit(rcutiny_trace_cleanup);
1053 MODULE_AUTHOR("Paul E. McKenney");
1054 MODULE_DESCRIPTION("Read-Copy Update tracing for tiny implementation");
1055 MODULE_LICENSE("GPL");
1057 #endif /* #ifdef CONFIG_RCU_TRACE */