2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
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, you can access it online at
18 * http://www.gnu.org/licenses/gpl-2.0.html.
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
27 #include <linux/delay.h>
28 #include <linux/gfp.h>
29 #include <linux/oom.h>
30 #include <linux/sched/debug.h>
31 #include <linux/smpboot.h>
32 #include <linux/sched/isolation.h>
33 #include <uapi/linux/sched/types.h>
34 #include "../time/tick-internal.h"
36 #ifdef CONFIG_RCU_BOOST
38 #include "../locking/rtmutex_common.h"
41 * Control variables for per-CPU and per-rcu_node kthreads. These
42 * handle all flavors of RCU.
44 static DEFINE_PER_CPU(struct task_struct
*, rcu_cpu_kthread_task
);
45 DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status
);
46 DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops
);
47 DEFINE_PER_CPU(char, rcu_cpu_has_work
);
49 #else /* #ifdef CONFIG_RCU_BOOST */
52 * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST,
53 * all uses are in dead code. Provide a definition to keep the compiler
54 * happy, but add WARN_ON_ONCE() to complain if used in the wrong place.
55 * This probably needs to be excluded from -rt builds.
57 #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; })
58 #define rt_mutex_futex_unlock(x) WARN_ON_ONCE(1)
60 #endif /* #else #ifdef CONFIG_RCU_BOOST */
62 #ifdef CONFIG_RCU_NOCB_CPU
63 static cpumask_var_t rcu_nocb_mask
; /* CPUs to have callbacks offloaded. */
64 static bool __read_mostly rcu_nocb_poll
; /* Offload kthread are to poll. */
65 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
68 * Check the RCU kernel configuration parameters and print informative
69 * messages about anything out of the ordinary.
71 static void __init
rcu_bootup_announce_oddness(void)
73 if (IS_ENABLED(CONFIG_RCU_TRACE
))
74 pr_info("\tRCU event tracing is enabled.\n");
75 if ((IS_ENABLED(CONFIG_64BIT
) && RCU_FANOUT
!= 64) ||
76 (!IS_ENABLED(CONFIG_64BIT
) && RCU_FANOUT
!= 32))
77 pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n",
80 pr_info("\tHierarchical RCU autobalancing is disabled.\n");
81 if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ
))
82 pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n");
83 if (IS_ENABLED(CONFIG_PROVE_RCU
))
84 pr_info("\tRCU lockdep checking is enabled.\n");
85 if (RCU_NUM_LVLS
>= 4)
86 pr_info("\tFour(or more)-level hierarchy is enabled.\n");
87 if (RCU_FANOUT_LEAF
!= 16)
88 pr_info("\tBuild-time adjustment of leaf fanout to %d.\n",
90 if (rcu_fanout_leaf
!= RCU_FANOUT_LEAF
)
91 pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf
);
92 if (nr_cpu_ids
!= NR_CPUS
)
93 pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS
, nr_cpu_ids
);
94 #ifdef CONFIG_RCU_BOOST
95 pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio
, CONFIG_RCU_BOOST_DELAY
);
97 if (blimit
!= DEFAULT_RCU_BLIMIT
)
98 pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit
);
99 if (qhimark
!= DEFAULT_RCU_QHIMARK
)
100 pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark
);
101 if (qlowmark
!= DEFAULT_RCU_QLOMARK
)
102 pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark
);
103 if (jiffies_till_first_fqs
!= ULONG_MAX
)
104 pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs
);
105 if (jiffies_till_next_fqs
!= ULONG_MAX
)
106 pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs
);
107 if (rcu_kick_kthreads
)
108 pr_info("\tKick kthreads if too-long grace period.\n");
109 if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD
))
110 pr_info("\tRCU callback double-/use-after-free debug enabled.\n");
111 if (gp_preinit_delay
)
112 pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay
);
114 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay
);
115 if (gp_cleanup_delay
)
116 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay
);
117 if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG
))
118 pr_info("\tRCU debug extended QS entry/exit.\n");
119 rcupdate_announce_bootup_oddness();
122 #ifdef CONFIG_PREEMPT_RCU
124 RCU_STATE_INITIALIZER(rcu_preempt
, 'p', call_rcu
);
125 static struct rcu_state
*const rcu_state_p
= &rcu_preempt_state
;
126 static struct rcu_data __percpu
*const rcu_data_p
= &rcu_preempt_data
;
128 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
,
132 * Tell them what RCU they are running.
134 static void __init
rcu_bootup_announce(void)
136 pr_info("Preemptible hierarchical RCU implementation.\n");
137 rcu_bootup_announce_oddness();
140 /* Flags for rcu_preempt_ctxt_queue() decision table. */
141 #define RCU_GP_TASKS 0x8
142 #define RCU_EXP_TASKS 0x4
143 #define RCU_GP_BLKD 0x2
144 #define RCU_EXP_BLKD 0x1
147 * Queues a task preempted within an RCU-preempt read-side critical
148 * section into the appropriate location within the ->blkd_tasks list,
149 * depending on the states of any ongoing normal and expedited grace
150 * periods. The ->gp_tasks pointer indicates which element the normal
151 * grace period is waiting on (NULL if none), and the ->exp_tasks pointer
152 * indicates which element the expedited grace period is waiting on (again,
153 * NULL if none). If a grace period is waiting on a given element in the
154 * ->blkd_tasks list, it also waits on all subsequent elements. Thus,
155 * adding a task to the tail of the list blocks any grace period that is
156 * already waiting on one of the elements. In contrast, adding a task
157 * to the head of the list won't block any grace period that is already
158 * waiting on one of the elements.
160 * This queuing is imprecise, and can sometimes make an ongoing grace
161 * period wait for a task that is not strictly speaking blocking it.
162 * Given the choice, we needlessly block a normal grace period rather than
163 * blocking an expedited grace period.
165 * Note that an endless sequence of expedited grace periods still cannot
166 * indefinitely postpone a normal grace period. Eventually, all of the
167 * fixed number of preempted tasks blocking the normal grace period that are
168 * not also blocking the expedited grace period will resume and complete
169 * their RCU read-side critical sections. At that point, the ->gp_tasks
170 * pointer will equal the ->exp_tasks pointer, at which point the end of
171 * the corresponding expedited grace period will also be the end of the
172 * normal grace period.
174 static void rcu_preempt_ctxt_queue(struct rcu_node
*rnp
, struct rcu_data
*rdp
)
175 __releases(rnp
->lock
) /* But leaves rrupts disabled. */
177 int blkd_state
= (rnp
->gp_tasks
? RCU_GP_TASKS
: 0) +
178 (rnp
->exp_tasks
? RCU_EXP_TASKS
: 0) +
179 (rnp
->qsmask
& rdp
->grpmask
? RCU_GP_BLKD
: 0) +
180 (rnp
->expmask
& rdp
->grpmask
? RCU_EXP_BLKD
: 0);
181 struct task_struct
*t
= current
;
183 raw_lockdep_assert_held_rcu_node(rnp
);
184 WARN_ON_ONCE(rdp
->mynode
!= rnp
);
185 WARN_ON_ONCE(!rcu_is_leaf_node(rnp
));
188 * Decide where to queue the newly blocked task. In theory,
189 * this could be an if-statement. In practice, when I tried
190 * that, it was quite messy.
192 switch (blkd_state
) {
195 case RCU_EXP_TASKS
+ RCU_GP_BLKD
:
197 case RCU_GP_TASKS
+ RCU_EXP_TASKS
:
200 * Blocking neither GP, or first task blocking the normal
201 * GP but not blocking the already-waiting expedited GP.
202 * Queue at the head of the list to avoid unnecessarily
203 * blocking the already-waiting GPs.
205 list_add(&t
->rcu_node_entry
, &rnp
->blkd_tasks
);
210 case RCU_GP_BLKD
+ RCU_EXP_BLKD
:
211 case RCU_GP_TASKS
+ RCU_EXP_BLKD
:
212 case RCU_GP_TASKS
+ RCU_GP_BLKD
+ RCU_EXP_BLKD
:
213 case RCU_GP_TASKS
+ RCU_EXP_TASKS
+ RCU_GP_BLKD
+ RCU_EXP_BLKD
:
216 * First task arriving that blocks either GP, or first task
217 * arriving that blocks the expedited GP (with the normal
218 * GP already waiting), or a task arriving that blocks
219 * both GPs with both GPs already waiting. Queue at the
220 * tail of the list to avoid any GP waiting on any of the
221 * already queued tasks that are not blocking it.
223 list_add_tail(&t
->rcu_node_entry
, &rnp
->blkd_tasks
);
226 case RCU_EXP_TASKS
+ RCU_EXP_BLKD
:
227 case RCU_EXP_TASKS
+ RCU_GP_BLKD
+ RCU_EXP_BLKD
:
228 case RCU_GP_TASKS
+ RCU_EXP_TASKS
+ RCU_EXP_BLKD
:
231 * Second or subsequent task blocking the expedited GP.
232 * The task either does not block the normal GP, or is the
233 * first task blocking the normal GP. Queue just after
234 * the first task blocking the expedited GP.
236 list_add(&t
->rcu_node_entry
, rnp
->exp_tasks
);
239 case RCU_GP_TASKS
+ RCU_GP_BLKD
:
240 case RCU_GP_TASKS
+ RCU_EXP_TASKS
+ RCU_GP_BLKD
:
243 * Second or subsequent task blocking the normal GP.
244 * The task does not block the expedited GP. Queue just
245 * after the first task blocking the normal GP.
247 list_add(&t
->rcu_node_entry
, rnp
->gp_tasks
);
252 /* Yet another exercise in excessive paranoia. */
258 * We have now queued the task. If it was the first one to
259 * block either grace period, update the ->gp_tasks and/or
260 * ->exp_tasks pointers, respectively, to reference the newly
263 if (!rnp
->gp_tasks
&& (blkd_state
& RCU_GP_BLKD
))
264 rnp
->gp_tasks
= &t
->rcu_node_entry
;
265 if (!rnp
->exp_tasks
&& (blkd_state
& RCU_EXP_BLKD
))
266 rnp
->exp_tasks
= &t
->rcu_node_entry
;
267 WARN_ON_ONCE(!(blkd_state
& RCU_GP_BLKD
) !=
268 !(rnp
->qsmask
& rdp
->grpmask
));
269 WARN_ON_ONCE(!(blkd_state
& RCU_EXP_BLKD
) !=
270 !(rnp
->expmask
& rdp
->grpmask
));
271 raw_spin_unlock_rcu_node(rnp
); /* interrupts remain disabled. */
274 * Report the quiescent state for the expedited GP. This expedited
275 * GP should not be able to end until we report, so there should be
276 * no need to check for a subsequent expedited GP. (Though we are
277 * still in a quiescent state in any case.)
279 if (blkd_state
& RCU_EXP_BLKD
&&
280 t
->rcu_read_unlock_special
.b
.exp_need_qs
) {
281 t
->rcu_read_unlock_special
.b
.exp_need_qs
= false;
282 rcu_report_exp_rdp(rdp
->rsp
, rdp
, true);
284 WARN_ON_ONCE(t
->rcu_read_unlock_special
.b
.exp_need_qs
);
289 * Record a preemptible-RCU quiescent state for the specified CPU. Note
290 * that this just means that the task currently running on the CPU is
291 * not in a quiescent state. There might be any number of tasks blocked
292 * while in an RCU read-side critical section.
294 * As with the other rcu_*_qs() functions, callers to this function
295 * must disable preemption.
297 static void rcu_preempt_qs(void)
299 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n");
300 if (__this_cpu_read(rcu_data_p
->cpu_no_qs
.s
)) {
301 trace_rcu_grace_period(TPS("rcu_preempt"),
302 __this_cpu_read(rcu_data_p
->gpnum
),
304 __this_cpu_write(rcu_data_p
->cpu_no_qs
.b
.norm
, false);
305 barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */
306 current
->rcu_read_unlock_special
.b
.need_qs
= false;
311 * We have entered the scheduler, and the current task might soon be
312 * context-switched away from. If this task is in an RCU read-side
313 * critical section, we will no longer be able to rely on the CPU to
314 * record that fact, so we enqueue the task on the blkd_tasks list.
315 * The task will dequeue itself when it exits the outermost enclosing
316 * RCU read-side critical section. Therefore, the current grace period
317 * cannot be permitted to complete until the blkd_tasks list entries
318 * predating the current grace period drain, in other words, until
319 * rnp->gp_tasks becomes NULL.
321 * Caller must disable interrupts.
323 static void rcu_preempt_note_context_switch(bool preempt
)
325 struct task_struct
*t
= current
;
326 struct rcu_data
*rdp
;
327 struct rcu_node
*rnp
;
329 lockdep_assert_irqs_disabled();
330 WARN_ON_ONCE(!preempt
&& t
->rcu_read_lock_nesting
> 0);
331 if (t
->rcu_read_lock_nesting
> 0 &&
332 !t
->rcu_read_unlock_special
.b
.blocked
) {
334 /* Possibly blocking in an RCU read-side critical section. */
335 rdp
= this_cpu_ptr(rcu_state_p
->rda
);
337 raw_spin_lock_rcu_node(rnp
);
338 t
->rcu_read_unlock_special
.b
.blocked
= true;
339 t
->rcu_blocked_node
= rnp
;
342 * Verify the CPU's sanity, trace the preemption, and
343 * then queue the task as required based on the states
344 * of any ongoing and expedited grace periods.
346 WARN_ON_ONCE((rdp
->grpmask
& rcu_rnp_online_cpus(rnp
)) == 0);
347 WARN_ON_ONCE(!list_empty(&t
->rcu_node_entry
));
348 trace_rcu_preempt_task(rdp
->rsp
->name
,
350 (rnp
->qsmask
& rdp
->grpmask
)
353 rcu_preempt_ctxt_queue(rnp
, rdp
);
354 } else if (t
->rcu_read_lock_nesting
< 0 &&
355 t
->rcu_read_unlock_special
.s
) {
358 * Complete exit from RCU read-side critical section on
359 * behalf of preempted instance of __rcu_read_unlock().
361 rcu_read_unlock_special(t
);
365 * Either we were not in an RCU read-side critical section to
366 * begin with, or we have now recorded that critical section
367 * globally. Either way, we can now note a quiescent state
368 * for this CPU. Again, if we were in an RCU read-side critical
369 * section, and if that critical section was blocking the current
370 * grace period, then the fact that the task has been enqueued
371 * means that we continue to block the current grace period.
377 * Check for preempted RCU readers blocking the current grace period
378 * for the specified rcu_node structure. If the caller needs a reliable
379 * answer, it must hold the rcu_node's ->lock.
381 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
)
383 return rnp
->gp_tasks
!= NULL
;
387 * Preemptible RCU implementation for rcu_read_lock().
388 * Just increment ->rcu_read_lock_nesting, shared state will be updated
391 void __rcu_read_lock(void)
393 current
->rcu_read_lock_nesting
++;
394 barrier(); /* critical section after entry code. */
396 EXPORT_SYMBOL_GPL(__rcu_read_lock
);
399 * Preemptible RCU implementation for rcu_read_unlock().
400 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
401 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
402 * invoke rcu_read_unlock_special() to clean up after a context switch
403 * in an RCU read-side critical section and other special cases.
405 void __rcu_read_unlock(void)
407 struct task_struct
*t
= current
;
409 if (t
->rcu_read_lock_nesting
!= 1) {
410 --t
->rcu_read_lock_nesting
;
412 barrier(); /* critical section before exit code. */
413 t
->rcu_read_lock_nesting
= INT_MIN
;
414 barrier(); /* assign before ->rcu_read_unlock_special load */
415 if (unlikely(READ_ONCE(t
->rcu_read_unlock_special
.s
)))
416 rcu_read_unlock_special(t
);
417 barrier(); /* ->rcu_read_unlock_special load before assign */
418 t
->rcu_read_lock_nesting
= 0;
420 #ifdef CONFIG_PROVE_LOCKING
422 int rrln
= READ_ONCE(t
->rcu_read_lock_nesting
);
424 WARN_ON_ONCE(rrln
< 0 && rrln
> INT_MIN
/ 2);
426 #endif /* #ifdef CONFIG_PROVE_LOCKING */
428 EXPORT_SYMBOL_GPL(__rcu_read_unlock
);
431 * Advance a ->blkd_tasks-list pointer to the next entry, instead
432 * returning NULL if at the end of the list.
434 static struct list_head
*rcu_next_node_entry(struct task_struct
*t
,
435 struct rcu_node
*rnp
)
437 struct list_head
*np
;
439 np
= t
->rcu_node_entry
.next
;
440 if (np
== &rnp
->blkd_tasks
)
446 * Return true if the specified rcu_node structure has tasks that were
447 * preempted within an RCU read-side critical section.
449 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
)
451 return !list_empty(&rnp
->blkd_tasks
);
455 * Handle special cases during rcu_read_unlock(), such as needing to
456 * notify RCU core processing or task having blocked during the RCU
457 * read-side critical section.
459 void rcu_read_unlock_special(struct task_struct
*t
)
465 struct list_head
*np
;
466 bool drop_boost_mutex
= false;
467 struct rcu_data
*rdp
;
468 struct rcu_node
*rnp
;
469 union rcu_special special
;
471 /* NMI handlers cannot block and cannot safely manipulate state. */
475 local_irq_save(flags
);
478 * If RCU core is waiting for this CPU to exit its critical section,
479 * report the fact that it has exited. Because irqs are disabled,
480 * t->rcu_read_unlock_special cannot change.
482 special
= t
->rcu_read_unlock_special
;
483 if (special
.b
.need_qs
) {
485 t
->rcu_read_unlock_special
.b
.need_qs
= false;
486 if (!t
->rcu_read_unlock_special
.s
) {
487 local_irq_restore(flags
);
493 * Respond to a request for an expedited grace period, but only if
494 * we were not preempted, meaning that we were running on the same
495 * CPU throughout. If we were preempted, the exp_need_qs flag
496 * would have been cleared at the time of the first preemption,
497 * and the quiescent state would be reported when we were dequeued.
499 if (special
.b
.exp_need_qs
) {
500 WARN_ON_ONCE(special
.b
.blocked
);
501 t
->rcu_read_unlock_special
.b
.exp_need_qs
= false;
502 rdp
= this_cpu_ptr(rcu_state_p
->rda
);
503 rcu_report_exp_rdp(rcu_state_p
, rdp
, true);
504 if (!t
->rcu_read_unlock_special
.s
) {
505 local_irq_restore(flags
);
510 /* Hardware IRQ handlers cannot block, complain if they get here. */
511 if (in_irq() || in_serving_softirq()) {
512 lockdep_rcu_suspicious(__FILE__
, __LINE__
,
513 "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n");
514 pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n",
515 t
->rcu_read_unlock_special
.s
,
516 t
->rcu_read_unlock_special
.b
.blocked
,
517 t
->rcu_read_unlock_special
.b
.exp_need_qs
,
518 t
->rcu_read_unlock_special
.b
.need_qs
);
519 local_irq_restore(flags
);
523 /* Clean up if blocked during RCU read-side critical section. */
524 if (special
.b
.blocked
) {
525 t
->rcu_read_unlock_special
.b
.blocked
= false;
528 * Remove this task from the list it blocked on. The task
529 * now remains queued on the rcu_node corresponding to the
530 * CPU it first blocked on, so there is no longer any need
531 * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia.
533 rnp
= t
->rcu_blocked_node
;
534 raw_spin_lock_rcu_node(rnp
); /* irqs already disabled. */
535 WARN_ON_ONCE(rnp
!= t
->rcu_blocked_node
);
536 WARN_ON_ONCE(!rcu_is_leaf_node(rnp
));
537 empty_norm
= !rcu_preempt_blocked_readers_cgp(rnp
);
538 empty_exp
= sync_rcu_preempt_exp_done(rnp
);
539 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
540 np
= rcu_next_node_entry(t
, rnp
);
541 list_del_init(&t
->rcu_node_entry
);
542 t
->rcu_blocked_node
= NULL
;
543 trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
545 if (&t
->rcu_node_entry
== rnp
->gp_tasks
)
547 if (&t
->rcu_node_entry
== rnp
->exp_tasks
)
549 if (IS_ENABLED(CONFIG_RCU_BOOST
)) {
550 /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
551 drop_boost_mutex
= rt_mutex_owner(&rnp
->boost_mtx
) == t
;
552 if (&t
->rcu_node_entry
== rnp
->boost_tasks
)
553 rnp
->boost_tasks
= np
;
557 * If this was the last task on the current list, and if
558 * we aren't waiting on any CPUs, report the quiescent state.
559 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
560 * so we must take a snapshot of the expedited state.
562 empty_exp_now
= sync_rcu_preempt_exp_done(rnp
);
563 if (!empty_norm
&& !rcu_preempt_blocked_readers_cgp(rnp
)) {
564 trace_rcu_quiescent_state_report(TPS("preempt_rcu"),
571 rcu_report_unblock_qs_rnp(rcu_state_p
, rnp
, flags
);
573 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
576 /* Unboost if we were boosted. */
577 if (IS_ENABLED(CONFIG_RCU_BOOST
) && drop_boost_mutex
)
578 rt_mutex_futex_unlock(&rnp
->boost_mtx
);
581 * If this was the last task on the expedited lists,
582 * then we need to report up the rcu_node hierarchy.
584 if (!empty_exp
&& empty_exp_now
)
585 rcu_report_exp_rnp(rcu_state_p
, rnp
, true);
587 local_irq_restore(flags
);
592 * Dump detailed information for all tasks blocking the current RCU
593 * grace period on the specified rcu_node structure.
595 static void rcu_print_detail_task_stall_rnp(struct rcu_node
*rnp
)
598 struct task_struct
*t
;
600 raw_spin_lock_irqsave_rcu_node(rnp
, flags
);
601 if (!rcu_preempt_blocked_readers_cgp(rnp
)) {
602 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
605 t
= list_entry(rnp
->gp_tasks
->prev
,
606 struct task_struct
, rcu_node_entry
);
607 list_for_each_entry_continue(t
, &rnp
->blkd_tasks
, rcu_node_entry
) {
609 * We could be printing a lot while holding a spinlock.
610 * Avoid triggering hard lockup.
612 touch_nmi_watchdog();
615 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
619 * Dump detailed information for all tasks blocking the current RCU
622 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
624 struct rcu_node
*rnp
= rcu_get_root(rsp
);
626 rcu_print_detail_task_stall_rnp(rnp
);
627 rcu_for_each_leaf_node(rsp
, rnp
)
628 rcu_print_detail_task_stall_rnp(rnp
);
631 static void rcu_print_task_stall_begin(struct rcu_node
*rnp
)
633 pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
634 rnp
->level
, rnp
->grplo
, rnp
->grphi
);
637 static void rcu_print_task_stall_end(void)
643 * Scan the current list of tasks blocked within RCU read-side critical
644 * sections, printing out the tid of each.
646 static int rcu_print_task_stall(struct rcu_node
*rnp
)
648 struct task_struct
*t
;
651 if (!rcu_preempt_blocked_readers_cgp(rnp
))
653 rcu_print_task_stall_begin(rnp
);
654 t
= list_entry(rnp
->gp_tasks
->prev
,
655 struct task_struct
, rcu_node_entry
);
656 list_for_each_entry_continue(t
, &rnp
->blkd_tasks
, rcu_node_entry
) {
657 pr_cont(" P%d", t
->pid
);
660 rcu_print_task_stall_end();
665 * Scan the current list of tasks blocked within RCU read-side critical
666 * sections, printing out the tid of each that is blocking the current
667 * expedited grace period.
669 static int rcu_print_task_exp_stall(struct rcu_node
*rnp
)
671 struct task_struct
*t
;
676 t
= list_entry(rnp
->exp_tasks
->prev
,
677 struct task_struct
, rcu_node_entry
);
678 list_for_each_entry_continue(t
, &rnp
->blkd_tasks
, rcu_node_entry
) {
679 pr_cont(" P%d", t
->pid
);
686 * Check that the list of blocked tasks for the newly completed grace
687 * period is in fact empty. It is a serious bug to complete a grace
688 * period that still has RCU readers blocked! This function must be
689 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
690 * must be held by the caller.
692 * Also, if there are blocked tasks on the list, they automatically
693 * block the newly created grace period, so set up ->gp_tasks accordingly.
695 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
697 struct task_struct
*t
;
699 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
700 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp
));
701 if (rcu_preempt_has_tasks(rnp
)) {
702 rnp
->gp_tasks
= rnp
->blkd_tasks
.next
;
703 t
= container_of(rnp
->gp_tasks
, struct task_struct
,
705 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
708 WARN_ON_ONCE(rnp
->qsmask
);
712 * Check for a quiescent state from the current CPU. When a task blocks,
713 * the task is recorded in the corresponding CPU's rcu_node structure,
714 * which is checked elsewhere.
716 * Caller must disable hard irqs.
718 static void rcu_preempt_check_callbacks(void)
720 struct task_struct
*t
= current
;
722 if (t
->rcu_read_lock_nesting
== 0) {
726 if (t
->rcu_read_lock_nesting
> 0 &&
727 __this_cpu_read(rcu_data_p
->core_needs_qs
) &&
728 __this_cpu_read(rcu_data_p
->cpu_no_qs
.b
.norm
))
729 t
->rcu_read_unlock_special
.b
.need_qs
= true;
733 * call_rcu() - Queue an RCU callback for invocation after a grace period.
734 * @head: structure to be used for queueing the RCU updates.
735 * @func: actual callback function to be invoked after the grace period
737 * The callback function will be invoked some time after a full grace
738 * period elapses, in other words after all pre-existing RCU read-side
739 * critical sections have completed. However, the callback function
740 * might well execute concurrently with RCU read-side critical sections
741 * that started after call_rcu() was invoked. RCU read-side critical
742 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
745 * Note that all CPUs must agree that the grace period extended beyond
746 * all pre-existing RCU read-side critical section. On systems with more
747 * than one CPU, this means that when "func()" is invoked, each CPU is
748 * guaranteed to have executed a full memory barrier since the end of its
749 * last RCU read-side critical section whose beginning preceded the call
750 * to call_rcu(). It also means that each CPU executing an RCU read-side
751 * critical section that continues beyond the start of "func()" must have
752 * executed a memory barrier after the call_rcu() but before the beginning
753 * of that RCU read-side critical section. Note that these guarantees
754 * include CPUs that are offline, idle, or executing in user mode, as
755 * well as CPUs that are executing in the kernel.
757 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
758 * resulting RCU callback function "func()", then both CPU A and CPU B are
759 * guaranteed to execute a full memory barrier during the time interval
760 * between the call to call_rcu() and the invocation of "func()" -- even
761 * if CPU A and CPU B are the same CPU (but again only if the system has
762 * more than one CPU).
764 void call_rcu(struct rcu_head
*head
, rcu_callback_t func
)
766 __call_rcu(head
, func
, rcu_state_p
, -1, 0);
768 EXPORT_SYMBOL_GPL(call_rcu
);
771 * synchronize_rcu - wait until a grace period has elapsed.
773 * Control will return to the caller some time after a full grace
774 * period has elapsed, in other words after all currently executing RCU
775 * read-side critical sections have completed. Note, however, that
776 * upon return from synchronize_rcu(), the caller might well be executing
777 * concurrently with new RCU read-side critical sections that began while
778 * synchronize_rcu() was waiting. RCU read-side critical sections are
779 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
781 * See the description of synchronize_sched() for more detailed
782 * information on memory-ordering guarantees. However, please note
783 * that -only- the memory-ordering guarantees apply. For example,
784 * synchronize_rcu() is -not- guaranteed to wait on things like code
785 * protected by preempt_disable(), instead, synchronize_rcu() is -only-
786 * guaranteed to wait on RCU read-side critical sections, that is, sections
787 * of code protected by rcu_read_lock().
789 void synchronize_rcu(void)
791 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map
) ||
792 lock_is_held(&rcu_lock_map
) ||
793 lock_is_held(&rcu_sched_lock_map
),
794 "Illegal synchronize_rcu() in RCU read-side critical section");
795 if (rcu_scheduler_active
== RCU_SCHEDULER_INACTIVE
)
797 if (rcu_gp_is_expedited())
798 synchronize_rcu_expedited();
800 wait_rcu_gp(call_rcu
);
802 EXPORT_SYMBOL_GPL(synchronize_rcu
);
805 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
807 * Note that this primitive does not necessarily wait for an RCU grace period
808 * to complete. For example, if there are no RCU callbacks queued anywhere
809 * in the system, then rcu_barrier() is within its rights to return
810 * immediately, without waiting for anything, much less an RCU grace period.
812 void rcu_barrier(void)
814 _rcu_barrier(rcu_state_p
);
816 EXPORT_SYMBOL_GPL(rcu_barrier
);
819 * Initialize preemptible RCU's state structures.
821 static void __init
__rcu_init_preempt(void)
823 rcu_init_one(rcu_state_p
);
827 * Check for a task exiting while in a preemptible-RCU read-side
828 * critical section, clean up if so. No need to issue warnings,
829 * as debug_check_no_locks_held() already does this if lockdep
834 struct task_struct
*t
= current
;
836 if (likely(list_empty(¤t
->rcu_node_entry
)))
838 t
->rcu_read_lock_nesting
= 1;
840 t
->rcu_read_unlock_special
.b
.blocked
= true;
844 #else /* #ifdef CONFIG_PREEMPT_RCU */
846 static struct rcu_state
*const rcu_state_p
= &rcu_sched_state
;
849 * Tell them what RCU they are running.
851 static void __init
rcu_bootup_announce(void)
853 pr_info("Hierarchical RCU implementation.\n");
854 rcu_bootup_announce_oddness();
858 * Because preemptible RCU does not exist, we never have to check for
859 * CPUs being in quiescent states.
861 static void rcu_preempt_note_context_switch(bool preempt
)
866 * Because preemptible RCU does not exist, there are never any preempted
869 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
)
875 * Because there is no preemptible RCU, there can be no readers blocked.
877 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
)
883 * Because preemptible RCU does not exist, we never have to check for
884 * tasks blocked within RCU read-side critical sections.
886 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
891 * Because preemptible RCU does not exist, we never have to check for
892 * tasks blocked within RCU read-side critical sections.
894 static int rcu_print_task_stall(struct rcu_node
*rnp
)
900 * Because preemptible RCU does not exist, we never have to check for
901 * tasks blocked within RCU read-side critical sections that are
902 * blocking the current expedited grace period.
904 static int rcu_print_task_exp_stall(struct rcu_node
*rnp
)
910 * Because there is no preemptible RCU, there can be no readers blocked,
911 * so there is no need to check for blocked tasks. So check only for
912 * bogus qsmask values.
914 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
916 WARN_ON_ONCE(rnp
->qsmask
);
920 * Because preemptible RCU does not exist, it never has any callbacks
923 static void rcu_preempt_check_callbacks(void)
928 * Because preemptible RCU does not exist, rcu_barrier() is just
929 * another name for rcu_barrier_sched().
931 void rcu_barrier(void)
935 EXPORT_SYMBOL_GPL(rcu_barrier
);
938 * Because preemptible RCU does not exist, it need not be initialized.
940 static void __init
__rcu_init_preempt(void)
945 * Because preemptible RCU does not exist, tasks cannot possibly exit
946 * while in preemptible RCU read-side critical sections.
952 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
954 #ifdef CONFIG_RCU_BOOST
956 static void rcu_wake_cond(struct task_struct
*t
, int status
)
959 * If the thread is yielding, only wake it when this
960 * is invoked from idle
962 if (status
!= RCU_KTHREAD_YIELDING
|| is_idle_task(current
))
967 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
968 * or ->boost_tasks, advancing the pointer to the next task in the
971 * Note that irqs must be enabled: boosting the task can block.
972 * Returns 1 if there are more tasks needing to be boosted.
974 static int rcu_boost(struct rcu_node
*rnp
)
977 struct task_struct
*t
;
978 struct list_head
*tb
;
980 if (READ_ONCE(rnp
->exp_tasks
) == NULL
&&
981 READ_ONCE(rnp
->boost_tasks
) == NULL
)
982 return 0; /* Nothing left to boost. */
984 raw_spin_lock_irqsave_rcu_node(rnp
, flags
);
987 * Recheck under the lock: all tasks in need of boosting
988 * might exit their RCU read-side critical sections on their own.
990 if (rnp
->exp_tasks
== NULL
&& rnp
->boost_tasks
== NULL
) {
991 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
996 * Preferentially boost tasks blocking expedited grace periods.
997 * This cannot starve the normal grace periods because a second
998 * expedited grace period must boost all blocked tasks, including
999 * those blocking the pre-existing normal grace period.
1001 if (rnp
->exp_tasks
!= NULL
)
1002 tb
= rnp
->exp_tasks
;
1004 tb
= rnp
->boost_tasks
;
1007 * We boost task t by manufacturing an rt_mutex that appears to
1008 * be held by task t. We leave a pointer to that rt_mutex where
1009 * task t can find it, and task t will release the mutex when it
1010 * exits its outermost RCU read-side critical section. Then
1011 * simply acquiring this artificial rt_mutex will boost task
1012 * t's priority. (Thanks to tglx for suggesting this approach!)
1014 * Note that task t must acquire rnp->lock to remove itself from
1015 * the ->blkd_tasks list, which it will do from exit() if from
1016 * nowhere else. We therefore are guaranteed that task t will
1017 * stay around at least until we drop rnp->lock. Note that
1018 * rnp->lock also resolves races between our priority boosting
1019 * and task t's exiting its outermost RCU read-side critical
1022 t
= container_of(tb
, struct task_struct
, rcu_node_entry
);
1023 rt_mutex_init_proxy_locked(&rnp
->boost_mtx
, t
);
1024 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1025 /* Lock only for side effect: boosts task t's priority. */
1026 rt_mutex_lock(&rnp
->boost_mtx
);
1027 rt_mutex_unlock(&rnp
->boost_mtx
); /* Then keep lockdep happy. */
1029 return READ_ONCE(rnp
->exp_tasks
) != NULL
||
1030 READ_ONCE(rnp
->boost_tasks
) != NULL
;
1034 * Priority-boosting kthread, one per leaf rcu_node.
1036 static int rcu_boost_kthread(void *arg
)
1038 struct rcu_node
*rnp
= (struct rcu_node
*)arg
;
1042 trace_rcu_utilization(TPS("Start boost kthread@init"));
1044 rnp
->boost_kthread_status
= RCU_KTHREAD_WAITING
;
1045 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
1046 rcu_wait(rnp
->boost_tasks
|| rnp
->exp_tasks
);
1047 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
1048 rnp
->boost_kthread_status
= RCU_KTHREAD_RUNNING
;
1049 more2boost
= rcu_boost(rnp
);
1055 rnp
->boost_kthread_status
= RCU_KTHREAD_YIELDING
;
1056 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
1057 schedule_timeout_interruptible(2);
1058 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
1063 trace_rcu_utilization(TPS("End boost kthread@notreached"));
1068 * Check to see if it is time to start boosting RCU readers that are
1069 * blocking the current grace period, and, if so, tell the per-rcu_node
1070 * kthread to start boosting them. If there is an expedited grace
1071 * period in progress, it is always time to boost.
1073 * The caller must hold rnp->lock, which this function releases.
1074 * The ->boost_kthread_task is immortal, so we don't need to worry
1075 * about it going away.
1077 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1078 __releases(rnp
->lock
)
1080 struct task_struct
*t
;
1082 raw_lockdep_assert_held_rcu_node(rnp
);
1083 if (!rcu_preempt_blocked_readers_cgp(rnp
) && rnp
->exp_tasks
== NULL
) {
1084 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1087 if (rnp
->exp_tasks
!= NULL
||
1088 (rnp
->gp_tasks
!= NULL
&&
1089 rnp
->boost_tasks
== NULL
&&
1091 ULONG_CMP_GE(jiffies
, rnp
->boost_time
))) {
1092 if (rnp
->exp_tasks
== NULL
)
1093 rnp
->boost_tasks
= rnp
->gp_tasks
;
1094 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1095 t
= rnp
->boost_kthread_task
;
1097 rcu_wake_cond(t
, rnp
->boost_kthread_status
);
1099 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1104 * Wake up the per-CPU kthread to invoke RCU callbacks.
1106 static void invoke_rcu_callbacks_kthread(void)
1108 unsigned long flags
;
1110 local_irq_save(flags
);
1111 __this_cpu_write(rcu_cpu_has_work
, 1);
1112 if (__this_cpu_read(rcu_cpu_kthread_task
) != NULL
&&
1113 current
!= __this_cpu_read(rcu_cpu_kthread_task
)) {
1114 rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task
),
1115 __this_cpu_read(rcu_cpu_kthread_status
));
1117 local_irq_restore(flags
);
1121 * Is the current CPU running the RCU-callbacks kthread?
1122 * Caller must have preemption disabled.
1124 static bool rcu_is_callbacks_kthread(void)
1126 return __this_cpu_read(rcu_cpu_kthread_task
) == current
;
1129 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1132 * Do priority-boost accounting for the start of a new grace period.
1134 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1136 rnp
->boost_time
= jiffies
+ RCU_BOOST_DELAY_JIFFIES
;
1140 * Create an RCU-boost kthread for the specified node if one does not
1141 * already exist. We only create this kthread for preemptible RCU.
1142 * Returns zero if all is well, a negated errno otherwise.
1144 static int rcu_spawn_one_boost_kthread(struct rcu_state
*rsp
,
1145 struct rcu_node
*rnp
)
1147 int rnp_index
= rnp
- &rsp
->node
[0];
1148 unsigned long flags
;
1149 struct sched_param sp
;
1150 struct task_struct
*t
;
1152 if (rcu_state_p
!= rsp
)
1155 if (!rcu_scheduler_fully_active
|| rcu_rnp_online_cpus(rnp
) == 0)
1159 if (rnp
->boost_kthread_task
!= NULL
)
1161 t
= kthread_create(rcu_boost_kthread
, (void *)rnp
,
1162 "rcub/%d", rnp_index
);
1165 raw_spin_lock_irqsave_rcu_node(rnp
, flags
);
1166 rnp
->boost_kthread_task
= t
;
1167 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1168 sp
.sched_priority
= kthread_prio
;
1169 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1170 wake_up_process(t
); /* get to TASK_INTERRUPTIBLE quickly. */
1174 static void rcu_kthread_do_work(void)
1176 rcu_do_batch(&rcu_sched_state
, this_cpu_ptr(&rcu_sched_data
));
1177 rcu_do_batch(&rcu_bh_state
, this_cpu_ptr(&rcu_bh_data
));
1178 rcu_do_batch(&rcu_preempt_state
, this_cpu_ptr(&rcu_preempt_data
));
1181 static void rcu_cpu_kthread_setup(unsigned int cpu
)
1183 struct sched_param sp
;
1185 sp
.sched_priority
= kthread_prio
;
1186 sched_setscheduler_nocheck(current
, SCHED_FIFO
, &sp
);
1189 static void rcu_cpu_kthread_park(unsigned int cpu
)
1191 per_cpu(rcu_cpu_kthread_status
, cpu
) = RCU_KTHREAD_OFFCPU
;
1194 static int rcu_cpu_kthread_should_run(unsigned int cpu
)
1196 return __this_cpu_read(rcu_cpu_has_work
);
1200 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1201 * RCU softirq used in flavors and configurations of RCU that do not
1202 * support RCU priority boosting.
1204 static void rcu_cpu_kthread(unsigned int cpu
)
1206 unsigned int *statusp
= this_cpu_ptr(&rcu_cpu_kthread_status
);
1207 char work
, *workp
= this_cpu_ptr(&rcu_cpu_has_work
);
1210 for (spincnt
= 0; spincnt
< 10; spincnt
++) {
1211 trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
1213 *statusp
= RCU_KTHREAD_RUNNING
;
1214 this_cpu_inc(rcu_cpu_kthread_loops
);
1215 local_irq_disable();
1220 rcu_kthread_do_work();
1223 trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
1224 *statusp
= RCU_KTHREAD_WAITING
;
1228 *statusp
= RCU_KTHREAD_YIELDING
;
1229 trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
1230 schedule_timeout_interruptible(2);
1231 trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
1232 *statusp
= RCU_KTHREAD_WAITING
;
1236 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1237 * served by the rcu_node in question. The CPU hotplug lock is still
1238 * held, so the value of rnp->qsmaskinit will be stable.
1240 * We don't include outgoingcpu in the affinity set, use -1 if there is
1241 * no outgoing CPU. If there are no CPUs left in the affinity set,
1242 * this function allows the kthread to execute on any CPU.
1244 static void rcu_boost_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1246 struct task_struct
*t
= rnp
->boost_kthread_task
;
1247 unsigned long mask
= rcu_rnp_online_cpus(rnp
);
1253 if (!zalloc_cpumask_var(&cm
, GFP_KERNEL
))
1255 for_each_leaf_node_possible_cpu(rnp
, cpu
)
1256 if ((mask
& leaf_node_cpu_bit(rnp
, cpu
)) &&
1258 cpumask_set_cpu(cpu
, cm
);
1259 if (cpumask_weight(cm
) == 0)
1261 set_cpus_allowed_ptr(t
, cm
);
1262 free_cpumask_var(cm
);
1265 static struct smp_hotplug_thread rcu_cpu_thread_spec
= {
1266 .store
= &rcu_cpu_kthread_task
,
1267 .thread_should_run
= rcu_cpu_kthread_should_run
,
1268 .thread_fn
= rcu_cpu_kthread
,
1269 .thread_comm
= "rcuc/%u",
1270 .setup
= rcu_cpu_kthread_setup
,
1271 .park
= rcu_cpu_kthread_park
,
1275 * Spawn boost kthreads -- called as soon as the scheduler is running.
1277 static void __init
rcu_spawn_boost_kthreads(void)
1279 struct rcu_node
*rnp
;
1282 for_each_possible_cpu(cpu
)
1283 per_cpu(rcu_cpu_has_work
, cpu
) = 0;
1284 BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec
));
1285 rcu_for_each_leaf_node(rcu_state_p
, rnp
)
1286 (void)rcu_spawn_one_boost_kthread(rcu_state_p
, rnp
);
1289 static void rcu_prepare_kthreads(int cpu
)
1291 struct rcu_data
*rdp
= per_cpu_ptr(rcu_state_p
->rda
, cpu
);
1292 struct rcu_node
*rnp
= rdp
->mynode
;
1294 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1295 if (rcu_scheduler_fully_active
)
1296 (void)rcu_spawn_one_boost_kthread(rcu_state_p
, rnp
);
1299 #else /* #ifdef CONFIG_RCU_BOOST */
1301 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1302 __releases(rnp
->lock
)
1304 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
1307 static void invoke_rcu_callbacks_kthread(void)
1312 static bool rcu_is_callbacks_kthread(void)
1317 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1321 static void rcu_boost_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1325 static void __init
rcu_spawn_boost_kthreads(void)
1329 static void rcu_prepare_kthreads(int cpu
)
1333 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1335 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1338 * Check to see if any future RCU-related work will need to be done
1339 * by the current CPU, even if none need be done immediately, returning
1340 * 1 if so. This function is part of the RCU implementation; it is -not-
1341 * an exported member of the RCU API.
1343 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1344 * any flavor of RCU.
1346 int rcu_needs_cpu(u64 basemono
, u64
*nextevt
)
1348 *nextevt
= KTIME_MAX
;
1349 return rcu_cpu_has_callbacks(NULL
);
1353 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1356 static void rcu_cleanup_after_idle(void)
1361 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1364 static void rcu_prepare_for_idle(void)
1369 * Don't bother keeping a running count of the number of RCU callbacks
1370 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1372 static void rcu_idle_count_callbacks_posted(void)
1376 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1379 * This code is invoked when a CPU goes idle, at which point we want
1380 * to have the CPU do everything required for RCU so that it can enter
1381 * the energy-efficient dyntick-idle mode. This is handled by a
1382 * state machine implemented by rcu_prepare_for_idle() below.
1384 * The following three proprocessor symbols control this state machine:
1386 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1387 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1388 * is sized to be roughly one RCU grace period. Those energy-efficiency
1389 * benchmarkers who might otherwise be tempted to set this to a large
1390 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1391 * system. And if you are -that- concerned about energy efficiency,
1392 * just power the system down and be done with it!
1393 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1394 * permitted to sleep in dyntick-idle mode with only lazy RCU
1395 * callbacks pending. Setting this too high can OOM your system.
1397 * The values below work well in practice. If future workloads require
1398 * adjustment, they can be converted into kernel config parameters, though
1399 * making the state machine smarter might be a better option.
1401 #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
1402 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1404 static int rcu_idle_gp_delay
= RCU_IDLE_GP_DELAY
;
1405 module_param(rcu_idle_gp_delay
, int, 0644);
1406 static int rcu_idle_lazy_gp_delay
= RCU_IDLE_LAZY_GP_DELAY
;
1407 module_param(rcu_idle_lazy_gp_delay
, int, 0644);
1410 * Try to advance callbacks for all flavors of RCU on the current CPU, but
1411 * only if it has been awhile since the last time we did so. Afterwards,
1412 * if there are any callbacks ready for immediate invocation, return true.
1414 static bool __maybe_unused
rcu_try_advance_all_cbs(void)
1416 bool cbs_ready
= false;
1417 struct rcu_data
*rdp
;
1418 struct rcu_dynticks
*rdtp
= this_cpu_ptr(&rcu_dynticks
);
1419 struct rcu_node
*rnp
;
1420 struct rcu_state
*rsp
;
1422 /* Exit early if we advanced recently. */
1423 if (jiffies
== rdtp
->last_advance_all
)
1425 rdtp
->last_advance_all
= jiffies
;
1427 for_each_rcu_flavor(rsp
) {
1428 rdp
= this_cpu_ptr(rsp
->rda
);
1432 * Don't bother checking unless a grace period has
1433 * completed since we last checked and there are
1434 * callbacks not yet ready to invoke.
1436 if ((rdp
->completed
!= rnp
->completed
||
1437 unlikely(READ_ONCE(rdp
->gpwrap
))) &&
1438 rcu_segcblist_pend_cbs(&rdp
->cblist
))
1439 note_gp_changes(rsp
, rdp
);
1441 if (rcu_segcblist_ready_cbs(&rdp
->cblist
))
1448 * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready
1449 * to invoke. If the CPU has callbacks, try to advance them. Tell the
1450 * caller to set the timeout based on whether or not there are non-lazy
1453 * The caller must have disabled interrupts.
1455 int rcu_needs_cpu(u64 basemono
, u64
*nextevt
)
1457 struct rcu_dynticks
*rdtp
= this_cpu_ptr(&rcu_dynticks
);
1460 lockdep_assert_irqs_disabled();
1462 /* Snapshot to detect later posting of non-lazy callback. */
1463 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
;
1465 /* If no callbacks, RCU doesn't need the CPU. */
1466 if (!rcu_cpu_has_callbacks(&rdtp
->all_lazy
)) {
1467 *nextevt
= KTIME_MAX
;
1471 /* Attempt to advance callbacks. */
1472 if (rcu_try_advance_all_cbs()) {
1473 /* Some ready to invoke, so initiate later invocation. */
1477 rdtp
->last_accelerate
= jiffies
;
1479 /* Request timer delay depending on laziness, and round. */
1480 if (!rdtp
->all_lazy
) {
1481 dj
= round_up(rcu_idle_gp_delay
+ jiffies
,
1482 rcu_idle_gp_delay
) - jiffies
;
1484 dj
= round_jiffies(rcu_idle_lazy_gp_delay
+ jiffies
) - jiffies
;
1486 *nextevt
= basemono
+ dj
* TICK_NSEC
;
1491 * Prepare a CPU for idle from an RCU perspective. The first major task
1492 * is to sense whether nohz mode has been enabled or disabled via sysfs.
1493 * The second major task is to check to see if a non-lazy callback has
1494 * arrived at a CPU that previously had only lazy callbacks. The third
1495 * major task is to accelerate (that is, assign grace-period numbers to)
1496 * any recently arrived callbacks.
1498 * The caller must have disabled interrupts.
1500 static void rcu_prepare_for_idle(void)
1503 struct rcu_data
*rdp
;
1504 struct rcu_dynticks
*rdtp
= this_cpu_ptr(&rcu_dynticks
);
1505 struct rcu_node
*rnp
;
1506 struct rcu_state
*rsp
;
1509 lockdep_assert_irqs_disabled();
1510 if (rcu_is_nocb_cpu(smp_processor_id()))
1513 /* Handle nohz enablement switches conservatively. */
1514 tne
= READ_ONCE(tick_nohz_active
);
1515 if (tne
!= rdtp
->tick_nohz_enabled_snap
) {
1516 if (rcu_cpu_has_callbacks(NULL
))
1517 invoke_rcu_core(); /* force nohz to see update. */
1518 rdtp
->tick_nohz_enabled_snap
= tne
;
1525 * If a non-lazy callback arrived at a CPU having only lazy
1526 * callbacks, invoke RCU core for the side-effect of recalculating
1527 * idle duration on re-entry to idle.
1529 if (rdtp
->all_lazy
&&
1530 rdtp
->nonlazy_posted
!= rdtp
->nonlazy_posted_snap
) {
1531 rdtp
->all_lazy
= false;
1532 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
;
1538 * If we have not yet accelerated this jiffy, accelerate all
1539 * callbacks on this CPU.
1541 if (rdtp
->last_accelerate
== jiffies
)
1543 rdtp
->last_accelerate
= jiffies
;
1544 for_each_rcu_flavor(rsp
) {
1545 rdp
= this_cpu_ptr(rsp
->rda
);
1546 if (!rcu_segcblist_pend_cbs(&rdp
->cblist
))
1549 raw_spin_lock_rcu_node(rnp
); /* irqs already disabled. */
1550 needwake
= rcu_accelerate_cbs(rsp
, rnp
, rdp
);
1551 raw_spin_unlock_rcu_node(rnp
); /* irqs remain disabled. */
1553 rcu_gp_kthread_wake(rsp
);
1558 * Clean up for exit from idle. Attempt to advance callbacks based on
1559 * any grace periods that elapsed while the CPU was idle, and if any
1560 * callbacks are now ready to invoke, initiate invocation.
1562 static void rcu_cleanup_after_idle(void)
1564 lockdep_assert_irqs_disabled();
1565 if (rcu_is_nocb_cpu(smp_processor_id()))
1567 if (rcu_try_advance_all_cbs())
1572 * Keep a running count of the number of non-lazy callbacks posted
1573 * on this CPU. This running counter (which is never decremented) allows
1574 * rcu_prepare_for_idle() to detect when something out of the idle loop
1575 * posts a callback, even if an equal number of callbacks are invoked.
1576 * Of course, callbacks should only be posted from within a trace event
1577 * designed to be called from idle or from within RCU_NONIDLE().
1579 static void rcu_idle_count_callbacks_posted(void)
1581 __this_cpu_add(rcu_dynticks
.nonlazy_posted
, 1);
1585 * Data for flushing lazy RCU callbacks at OOM time.
1587 static atomic_t oom_callback_count
;
1588 static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq
);
1591 * RCU OOM callback -- decrement the outstanding count and deliver the
1592 * wake-up if we are the last one.
1594 static void rcu_oom_callback(struct rcu_head
*rhp
)
1596 if (atomic_dec_and_test(&oom_callback_count
))
1597 wake_up(&oom_callback_wq
);
1601 * Post an rcu_oom_notify callback on the current CPU if it has at
1602 * least one lazy callback. This will unnecessarily post callbacks
1603 * to CPUs that already have a non-lazy callback at the end of their
1604 * callback list, but this is an infrequent operation, so accept some
1605 * extra overhead to keep things simple.
1607 static void rcu_oom_notify_cpu(void *unused
)
1609 struct rcu_state
*rsp
;
1610 struct rcu_data
*rdp
;
1612 for_each_rcu_flavor(rsp
) {
1613 rdp
= raw_cpu_ptr(rsp
->rda
);
1614 if (rcu_segcblist_n_lazy_cbs(&rdp
->cblist
)) {
1615 atomic_inc(&oom_callback_count
);
1616 rsp
->call(&rdp
->oom_head
, rcu_oom_callback
);
1622 * If low on memory, ensure that each CPU has a non-lazy callback.
1623 * This will wake up CPUs that have only lazy callbacks, in turn
1624 * ensuring that they free up the corresponding memory in a timely manner.
1625 * Because an uncertain amount of memory will be freed in some uncertain
1626 * timeframe, we do not claim to have freed anything.
1628 static int rcu_oom_notify(struct notifier_block
*self
,
1629 unsigned long notused
, void *nfreed
)
1633 /* Wait for callbacks from earlier instance to complete. */
1634 wait_event(oom_callback_wq
, atomic_read(&oom_callback_count
) == 0);
1635 smp_mb(); /* Ensure callback reuse happens after callback invocation. */
1638 * Prevent premature wakeup: ensure that all increments happen
1639 * before there is a chance of the counter reaching zero.
1641 atomic_set(&oom_callback_count
, 1);
1643 for_each_online_cpu(cpu
) {
1644 smp_call_function_single(cpu
, rcu_oom_notify_cpu
, NULL
, 1);
1645 cond_resched_tasks_rcu_qs();
1648 /* Unconditionally decrement: no need to wake ourselves up. */
1649 atomic_dec(&oom_callback_count
);
1654 static struct notifier_block rcu_oom_nb
= {
1655 .notifier_call
= rcu_oom_notify
1658 static int __init
rcu_register_oom_notifier(void)
1660 register_oom_notifier(&rcu_oom_nb
);
1663 early_initcall(rcu_register_oom_notifier
);
1665 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1667 #ifdef CONFIG_RCU_FAST_NO_HZ
1669 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
1671 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
1672 unsigned long nlpd
= rdtp
->nonlazy_posted
- rdtp
->nonlazy_posted_snap
;
1674 sprintf(cp
, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c",
1675 rdtp
->last_accelerate
& 0xffff, jiffies
& 0xffff,
1677 rdtp
->all_lazy
? 'L' : '.',
1678 rdtp
->tick_nohz_enabled_snap
? '.' : 'D');
1681 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
1683 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
1688 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
1690 /* Initiate the stall-info list. */
1691 static void print_cpu_stall_info_begin(void)
1697 * Print out diagnostic information for the specified stalled CPU.
1699 * If the specified CPU is aware of the current RCU grace period
1700 * (flavor specified by rsp), then print the number of scheduling
1701 * clock interrupts the CPU has taken during the time that it has
1702 * been aware. Otherwise, print the number of RCU grace periods
1703 * that this CPU is ignorant of, for example, "1" if the CPU was
1704 * aware of the previous grace period.
1706 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
1708 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
)
1710 unsigned long delta
;
1711 char fast_no_hz
[72];
1712 struct rcu_data
*rdp
= per_cpu_ptr(rsp
->rda
, cpu
);
1713 struct rcu_dynticks
*rdtp
= rdp
->dynticks
;
1715 unsigned long ticks_value
;
1718 * We could be printing a lot while holding a spinlock. Avoid
1719 * triggering hard lockup.
1721 touch_nmi_watchdog();
1723 if (rsp
->gpnum
== rdp
->gpnum
) {
1724 ticks_title
= "ticks this GP";
1725 ticks_value
= rdp
->ticks_this_gp
;
1727 ticks_title
= "GPs behind";
1728 ticks_value
= rsp
->gpnum
- rdp
->gpnum
;
1730 print_cpu_stall_fast_no_hz(fast_no_hz
, cpu
);
1731 delta
= rdp
->mynode
->gpnum
- rdp
->rcu_iw_gpnum
;
1732 pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%ld softirq=%u/%u fqs=%ld %s\n",
1734 "O."[!!cpu_online(cpu
)],
1735 "o."[!!(rdp
->grpmask
& rdp
->mynode
->qsmaskinit
)],
1736 "N."[!!(rdp
->grpmask
& rdp
->mynode
->qsmaskinitnext
)],
1737 !IS_ENABLED(CONFIG_IRQ_WORK
) ? '?' :
1738 rdp
->rcu_iw_pending
? (int)min(delta
, 9UL) + '0' :
1740 ticks_value
, ticks_title
,
1741 rcu_dynticks_snap(rdtp
) & 0xfff,
1742 rdtp
->dynticks_nesting
, rdtp
->dynticks_nmi_nesting
,
1743 rdp
->softirq_snap
, kstat_softirqs_cpu(RCU_SOFTIRQ
, cpu
),
1744 READ_ONCE(rsp
->n_force_qs
) - rsp
->n_force_qs_gpstart
,
1748 /* Terminate the stall-info list. */
1749 static void print_cpu_stall_info_end(void)
1754 /* Zero ->ticks_this_gp for all flavors of RCU. */
1755 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
)
1757 rdp
->ticks_this_gp
= 0;
1758 rdp
->softirq_snap
= kstat_softirqs_cpu(RCU_SOFTIRQ
, smp_processor_id());
1761 /* Increment ->ticks_this_gp for all flavors of RCU. */
1762 static void increment_cpu_stall_ticks(void)
1764 struct rcu_state
*rsp
;
1766 for_each_rcu_flavor(rsp
)
1767 raw_cpu_inc(rsp
->rda
->ticks_this_gp
);
1770 #ifdef CONFIG_RCU_NOCB_CPU
1773 * Offload callback processing from the boot-time-specified set of CPUs
1774 * specified by rcu_nocb_mask. For each CPU in the set, there is a
1775 * kthread created that pulls the callbacks from the corresponding CPU,
1776 * waits for a grace period to elapse, and invokes the callbacks.
1777 * The no-CBs CPUs do a wake_up() on their kthread when they insert
1778 * a callback into any empty list, unless the rcu_nocb_poll boot parameter
1779 * has been specified, in which case each kthread actively polls its
1780 * CPU. (Which isn't so great for energy efficiency, but which does
1781 * reduce RCU's overhead on that CPU.)
1783 * This is intended to be used in conjunction with Frederic Weisbecker's
1784 * adaptive-idle work, which would seriously reduce OS jitter on CPUs
1785 * running CPU-bound user-mode computations.
1787 * Offloading of callback processing could also in theory be used as
1788 * an energy-efficiency measure because CPUs with no RCU callbacks
1789 * queued are more aggressive about entering dyntick-idle mode.
1793 /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */
1794 static int __init
rcu_nocb_setup(char *str
)
1796 alloc_bootmem_cpumask_var(&rcu_nocb_mask
);
1797 cpulist_parse(str
, rcu_nocb_mask
);
1800 __setup("rcu_nocbs=", rcu_nocb_setup
);
1802 static int __init
parse_rcu_nocb_poll(char *arg
)
1804 rcu_nocb_poll
= true;
1807 early_param("rcu_nocb_poll", parse_rcu_nocb_poll
);
1810 * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended
1813 static void rcu_nocb_gp_cleanup(struct swait_queue_head
*sq
)
1818 static struct swait_queue_head
*rcu_nocb_gp_get(struct rcu_node
*rnp
)
1820 return &rnp
->nocb_gp_wq
[rnp
->completed
& 0x1];
1823 static void rcu_init_one_nocb(struct rcu_node
*rnp
)
1825 init_swait_queue_head(&rnp
->nocb_gp_wq
[0]);
1826 init_swait_queue_head(&rnp
->nocb_gp_wq
[1]);
1829 /* Is the specified CPU a no-CBs CPU? */
1830 bool rcu_is_nocb_cpu(int cpu
)
1832 if (cpumask_available(rcu_nocb_mask
))
1833 return cpumask_test_cpu(cpu
, rcu_nocb_mask
);
1838 * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock
1839 * and this function releases it.
1841 static void __wake_nocb_leader(struct rcu_data
*rdp
, bool force
,
1842 unsigned long flags
)
1843 __releases(rdp
->nocb_lock
)
1845 struct rcu_data
*rdp_leader
= rdp
->nocb_leader
;
1847 lockdep_assert_held(&rdp
->nocb_lock
);
1848 if (!READ_ONCE(rdp_leader
->nocb_kthread
)) {
1849 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
1852 if (rdp_leader
->nocb_leader_sleep
|| force
) {
1853 /* Prior smp_mb__after_atomic() orders against prior enqueue. */
1854 WRITE_ONCE(rdp_leader
->nocb_leader_sleep
, false);
1855 del_timer(&rdp
->nocb_timer
);
1856 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
1857 smp_mb(); /* ->nocb_leader_sleep before swake_up(). */
1858 swake_up(&rdp_leader
->nocb_wq
);
1860 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
1865 * Kick the leader kthread for this NOCB group, but caller has not
1868 static void wake_nocb_leader(struct rcu_data
*rdp
, bool force
)
1870 unsigned long flags
;
1872 raw_spin_lock_irqsave(&rdp
->nocb_lock
, flags
);
1873 __wake_nocb_leader(rdp
, force
, flags
);
1877 * Arrange to wake the leader kthread for this NOCB group at some
1878 * future time when it is safe to do so.
1880 static void wake_nocb_leader_defer(struct rcu_data
*rdp
, int waketype
,
1883 unsigned long flags
;
1885 raw_spin_lock_irqsave(&rdp
->nocb_lock
, flags
);
1886 if (rdp
->nocb_defer_wakeup
== RCU_NOCB_WAKE_NOT
)
1887 mod_timer(&rdp
->nocb_timer
, jiffies
+ 1);
1888 WRITE_ONCE(rdp
->nocb_defer_wakeup
, waketype
);
1889 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, reason
);
1890 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
1894 * Does the specified CPU need an RCU callback for the specified flavor
1897 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state
*rsp
, int cpu
)
1899 struct rcu_data
*rdp
= per_cpu_ptr(rsp
->rda
, cpu
);
1901 #ifdef CONFIG_PROVE_RCU
1902 struct rcu_head
*rhp
;
1903 #endif /* #ifdef CONFIG_PROVE_RCU */
1906 * Check count of all no-CBs callbacks awaiting invocation.
1907 * There needs to be a barrier before this function is called,
1908 * but associated with a prior determination that no more
1909 * callbacks would be posted. In the worst case, the first
1910 * barrier in _rcu_barrier() suffices (but the caller cannot
1911 * necessarily rely on this, not a substitute for the caller
1912 * getting the concurrency design right!). There must also be
1913 * a barrier between the following load an posting of a callback
1914 * (if a callback is in fact needed). This is associated with an
1915 * atomic_inc() in the caller.
1917 ret
= atomic_long_read(&rdp
->nocb_q_count
);
1919 #ifdef CONFIG_PROVE_RCU
1920 rhp
= READ_ONCE(rdp
->nocb_head
);
1922 rhp
= READ_ONCE(rdp
->nocb_gp_head
);
1924 rhp
= READ_ONCE(rdp
->nocb_follower_head
);
1926 /* Having no rcuo kthread but CBs after scheduler starts is bad! */
1927 if (!READ_ONCE(rdp
->nocb_kthread
) && rhp
&&
1928 rcu_scheduler_fully_active
) {
1929 /* RCU callback enqueued before CPU first came online??? */
1930 pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n",
1934 #endif /* #ifdef CONFIG_PROVE_RCU */
1940 * Enqueue the specified string of rcu_head structures onto the specified
1941 * CPU's no-CBs lists. The CPU is specified by rdp, the head of the
1942 * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy
1943 * counts are supplied by rhcount and rhcount_lazy.
1945 * If warranted, also wake up the kthread servicing this CPUs queues.
1947 static void __call_rcu_nocb_enqueue(struct rcu_data
*rdp
,
1948 struct rcu_head
*rhp
,
1949 struct rcu_head
**rhtp
,
1950 int rhcount
, int rhcount_lazy
,
1951 unsigned long flags
)
1954 struct rcu_head
**old_rhpp
;
1955 struct task_struct
*t
;
1957 /* Enqueue the callback on the nocb list and update counts. */
1958 atomic_long_add(rhcount
, &rdp
->nocb_q_count
);
1959 /* rcu_barrier() relies on ->nocb_q_count add before xchg. */
1960 old_rhpp
= xchg(&rdp
->nocb_tail
, rhtp
);
1961 WRITE_ONCE(*old_rhpp
, rhp
);
1962 atomic_long_add(rhcount_lazy
, &rdp
->nocb_q_count_lazy
);
1963 smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */
1965 /* If we are not being polled and there is a kthread, awaken it ... */
1966 t
= READ_ONCE(rdp
->nocb_kthread
);
1967 if (rcu_nocb_poll
|| !t
) {
1968 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
,
1969 TPS("WakeNotPoll"));
1972 len
= atomic_long_read(&rdp
->nocb_q_count
);
1973 if (old_rhpp
== &rdp
->nocb_head
) {
1974 if (!irqs_disabled_flags(flags
)) {
1975 /* ... if queue was empty ... */
1976 wake_nocb_leader(rdp
, false);
1977 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
,
1980 wake_nocb_leader_defer(rdp
, RCU_NOCB_WAKE
,
1981 TPS("WakeEmptyIsDeferred"));
1983 rdp
->qlen_last_fqs_check
= 0;
1984 } else if (len
> rdp
->qlen_last_fqs_check
+ qhimark
) {
1985 /* ... or if many callbacks queued. */
1986 if (!irqs_disabled_flags(flags
)) {
1987 wake_nocb_leader(rdp
, true);
1988 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
,
1991 wake_nocb_leader_defer(rdp
, RCU_NOCB_WAKE_FORCE
,
1992 TPS("WakeOvfIsDeferred"));
1994 rdp
->qlen_last_fqs_check
= LONG_MAX
/ 2;
1996 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, TPS("WakeNot"));
2002 * This is a helper for __call_rcu(), which invokes this when the normal
2003 * callback queue is inoperable. If this is not a no-CBs CPU, this
2004 * function returns failure back to __call_rcu(), which can complain
2007 * Otherwise, this function queues the callback where the corresponding
2008 * "rcuo" kthread can find it.
2010 static bool __call_rcu_nocb(struct rcu_data
*rdp
, struct rcu_head
*rhp
,
2011 bool lazy
, unsigned long flags
)
2014 if (!rcu_is_nocb_cpu(rdp
->cpu
))
2016 __call_rcu_nocb_enqueue(rdp
, rhp
, &rhp
->next
, 1, lazy
, flags
);
2017 if (__is_kfree_rcu_offset((unsigned long)rhp
->func
))
2018 trace_rcu_kfree_callback(rdp
->rsp
->name
, rhp
,
2019 (unsigned long)rhp
->func
,
2020 -atomic_long_read(&rdp
->nocb_q_count_lazy
),
2021 -atomic_long_read(&rdp
->nocb_q_count
));
2023 trace_rcu_callback(rdp
->rsp
->name
, rhp
,
2024 -atomic_long_read(&rdp
->nocb_q_count_lazy
),
2025 -atomic_long_read(&rdp
->nocb_q_count
));
2028 * If called from an extended quiescent state with interrupts
2029 * disabled, invoke the RCU core in order to allow the idle-entry
2030 * deferred-wakeup check to function.
2032 if (irqs_disabled_flags(flags
) &&
2033 !rcu_is_watching() &&
2034 cpu_online(smp_processor_id()))
2041 * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is
2044 static bool __maybe_unused
rcu_nocb_adopt_orphan_cbs(struct rcu_data
*my_rdp
,
2045 struct rcu_data
*rdp
,
2046 unsigned long flags
)
2048 lockdep_assert_irqs_disabled();
2049 if (!rcu_is_nocb_cpu(smp_processor_id()))
2050 return false; /* Not NOCBs CPU, caller must migrate CBs. */
2051 __call_rcu_nocb_enqueue(my_rdp
, rcu_segcblist_head(&rdp
->cblist
),
2052 rcu_segcblist_tail(&rdp
->cblist
),
2053 rcu_segcblist_n_cbs(&rdp
->cblist
),
2054 rcu_segcblist_n_lazy_cbs(&rdp
->cblist
), flags
);
2055 rcu_segcblist_init(&rdp
->cblist
);
2056 rcu_segcblist_disable(&rdp
->cblist
);
2061 * If necessary, kick off a new grace period, and either way wait
2062 * for a subsequent grace period to complete.
2064 static void rcu_nocb_wait_gp(struct rcu_data
*rdp
)
2068 unsigned long flags
;
2070 struct rcu_node
*rnp
= rdp
->mynode
;
2072 raw_spin_lock_irqsave_rcu_node(rnp
, flags
);
2073 c
= rcu_cbs_completed(rdp
->rsp
, rnp
);
2074 needwake
= rcu_start_this_gp(rnp
, rdp
, c
);
2075 raw_spin_unlock_irqrestore_rcu_node(rnp
, flags
);
2077 rcu_gp_kthread_wake(rdp
->rsp
);
2080 * Wait for the grace period. Do so interruptibly to avoid messing
2081 * up the load average.
2083 trace_rcu_this_gp(rnp
, rdp
, c
, TPS("StartWait"));
2085 swait_event_interruptible(
2086 rnp
->nocb_gp_wq
[c
& 0x1],
2087 (d
= ULONG_CMP_GE(READ_ONCE(rnp
->completed
), c
)));
2090 WARN_ON(signal_pending(current
));
2091 trace_rcu_this_gp(rnp
, rdp
, c
, TPS("ResumeWait"));
2093 trace_rcu_this_gp(rnp
, rdp
, c
, TPS("EndWait"));
2094 smp_mb(); /* Ensure that CB invocation happens after GP end. */
2098 * Leaders come here to wait for additional callbacks to show up.
2099 * This function does not return until callbacks appear.
2101 static void nocb_leader_wait(struct rcu_data
*my_rdp
)
2103 bool firsttime
= true;
2104 unsigned long flags
;
2106 struct rcu_data
*rdp
;
2107 struct rcu_head
**tail
;
2111 /* Wait for callbacks to appear. */
2112 if (!rcu_nocb_poll
) {
2113 trace_rcu_nocb_wake(my_rdp
->rsp
->name
, my_rdp
->cpu
, TPS("Sleep"));
2114 swait_event_interruptible(my_rdp
->nocb_wq
,
2115 !READ_ONCE(my_rdp
->nocb_leader_sleep
));
2116 raw_spin_lock_irqsave(&my_rdp
->nocb_lock
, flags
);
2117 my_rdp
->nocb_leader_sleep
= true;
2118 WRITE_ONCE(my_rdp
->nocb_defer_wakeup
, RCU_NOCB_WAKE_NOT
);
2119 del_timer(&my_rdp
->nocb_timer
);
2120 raw_spin_unlock_irqrestore(&my_rdp
->nocb_lock
, flags
);
2121 } else if (firsttime
) {
2122 firsttime
= false; /* Don't drown trace log with "Poll"! */
2123 trace_rcu_nocb_wake(my_rdp
->rsp
->name
, my_rdp
->cpu
, TPS("Poll"));
2127 * Each pass through the following loop checks a follower for CBs.
2128 * We are our own first follower. Any CBs found are moved to
2129 * nocb_gp_head, where they await a grace period.
2132 smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */
2133 for (rdp
= my_rdp
; rdp
; rdp
= rdp
->nocb_next_follower
) {
2134 rdp
->nocb_gp_head
= READ_ONCE(rdp
->nocb_head
);
2135 if (!rdp
->nocb_gp_head
)
2136 continue; /* No CBs here, try next follower. */
2138 /* Move callbacks to wait-for-GP list, which is empty. */
2139 WRITE_ONCE(rdp
->nocb_head
, NULL
);
2140 rdp
->nocb_gp_tail
= xchg(&rdp
->nocb_tail
, &rdp
->nocb_head
);
2144 /* No callbacks? Sleep a bit if polling, and go retry. */
2145 if (unlikely(!gotcbs
)) {
2146 WARN_ON(signal_pending(current
));
2147 if (rcu_nocb_poll
) {
2148 schedule_timeout_interruptible(1);
2150 trace_rcu_nocb_wake(my_rdp
->rsp
->name
, my_rdp
->cpu
,
2156 /* Wait for one grace period. */
2157 rcu_nocb_wait_gp(my_rdp
);
2159 /* Each pass through the following loop wakes a follower, if needed. */
2160 for (rdp
= my_rdp
; rdp
; rdp
= rdp
->nocb_next_follower
) {
2161 if (!rcu_nocb_poll
&&
2162 READ_ONCE(rdp
->nocb_head
) &&
2163 READ_ONCE(my_rdp
->nocb_leader_sleep
)) {
2164 raw_spin_lock_irqsave(&my_rdp
->nocb_lock
, flags
);
2165 my_rdp
->nocb_leader_sleep
= false;/* No need to sleep.*/
2166 raw_spin_unlock_irqrestore(&my_rdp
->nocb_lock
, flags
);
2168 if (!rdp
->nocb_gp_head
)
2169 continue; /* No CBs, so no need to wake follower. */
2171 /* Append callbacks to follower's "done" list. */
2172 raw_spin_lock_irqsave(&rdp
->nocb_lock
, flags
);
2173 tail
= rdp
->nocb_follower_tail
;
2174 rdp
->nocb_follower_tail
= rdp
->nocb_gp_tail
;
2175 *tail
= rdp
->nocb_gp_head
;
2176 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
2177 if (rdp
!= my_rdp
&& tail
== &rdp
->nocb_follower_head
) {
2178 /* List was empty, so wake up the follower. */
2179 swake_up(&rdp
->nocb_wq
);
2183 /* If we (the leader) don't have CBs, go wait some more. */
2184 if (!my_rdp
->nocb_follower_head
)
2189 * Followers come here to wait for additional callbacks to show up.
2190 * This function does not return until callbacks appear.
2192 static void nocb_follower_wait(struct rcu_data
*rdp
)
2195 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, TPS("FollowerSleep"));
2196 swait_event_interruptible(rdp
->nocb_wq
,
2197 READ_ONCE(rdp
->nocb_follower_head
));
2198 if (smp_load_acquire(&rdp
->nocb_follower_head
)) {
2199 /* ^^^ Ensure CB invocation follows _head test. */
2202 WARN_ON(signal_pending(current
));
2203 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, TPS("WokeEmpty"));
2208 * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes
2209 * callbacks queued by the corresponding no-CBs CPU, however, there is
2210 * an optional leader-follower relationship so that the grace-period
2211 * kthreads don't have to do quite so many wakeups.
2213 static int rcu_nocb_kthread(void *arg
)
2216 unsigned long flags
;
2217 struct rcu_head
*list
;
2218 struct rcu_head
*next
;
2219 struct rcu_head
**tail
;
2220 struct rcu_data
*rdp
= arg
;
2222 /* Each pass through this loop invokes one batch of callbacks */
2224 /* Wait for callbacks. */
2225 if (rdp
->nocb_leader
== rdp
)
2226 nocb_leader_wait(rdp
);
2228 nocb_follower_wait(rdp
);
2230 /* Pull the ready-to-invoke callbacks onto local list. */
2231 raw_spin_lock_irqsave(&rdp
->nocb_lock
, flags
);
2232 list
= rdp
->nocb_follower_head
;
2233 rdp
->nocb_follower_head
= NULL
;
2234 tail
= rdp
->nocb_follower_tail
;
2235 rdp
->nocb_follower_tail
= &rdp
->nocb_follower_head
;
2236 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
2238 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, TPS("WokeNonEmpty"));
2240 /* Each pass through the following loop invokes a callback. */
2241 trace_rcu_batch_start(rdp
->rsp
->name
,
2242 atomic_long_read(&rdp
->nocb_q_count_lazy
),
2243 atomic_long_read(&rdp
->nocb_q_count
), -1);
2247 /* Wait for enqueuing to complete, if needed. */
2248 while (next
== NULL
&& &list
->next
!= tail
) {
2249 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
,
2251 schedule_timeout_interruptible(1);
2252 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
,
2256 debug_rcu_head_unqueue(list
);
2258 if (__rcu_reclaim(rdp
->rsp
->name
, list
))
2262 cond_resched_tasks_rcu_qs();
2265 trace_rcu_batch_end(rdp
->rsp
->name
, c
, !!list
, 0, 0, 1);
2266 smp_mb__before_atomic(); /* _add after CB invocation. */
2267 atomic_long_add(-c
, &rdp
->nocb_q_count
);
2268 atomic_long_add(-cl
, &rdp
->nocb_q_count_lazy
);
2273 /* Is a deferred wakeup of rcu_nocb_kthread() required? */
2274 static int rcu_nocb_need_deferred_wakeup(struct rcu_data
*rdp
)
2276 return READ_ONCE(rdp
->nocb_defer_wakeup
);
2279 /* Do a deferred wakeup of rcu_nocb_kthread(). */
2280 static void do_nocb_deferred_wakeup_common(struct rcu_data
*rdp
)
2282 unsigned long flags
;
2285 raw_spin_lock_irqsave(&rdp
->nocb_lock
, flags
);
2286 if (!rcu_nocb_need_deferred_wakeup(rdp
)) {
2287 raw_spin_unlock_irqrestore(&rdp
->nocb_lock
, flags
);
2290 ndw
= READ_ONCE(rdp
->nocb_defer_wakeup
);
2291 WRITE_ONCE(rdp
->nocb_defer_wakeup
, RCU_NOCB_WAKE_NOT
);
2292 __wake_nocb_leader(rdp
, ndw
== RCU_NOCB_WAKE_FORCE
, flags
);
2293 trace_rcu_nocb_wake(rdp
->rsp
->name
, rdp
->cpu
, TPS("DeferredWake"));
2296 /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */
2297 static void do_nocb_deferred_wakeup_timer(struct timer_list
*t
)
2299 struct rcu_data
*rdp
= from_timer(rdp
, t
, nocb_timer
);
2301 do_nocb_deferred_wakeup_common(rdp
);
2305 * Do a deferred wakeup of rcu_nocb_kthread() from fastpath.
2306 * This means we do an inexact common-case check. Note that if
2307 * we miss, ->nocb_timer will eventually clean things up.
2309 static void do_nocb_deferred_wakeup(struct rcu_data
*rdp
)
2311 if (rcu_nocb_need_deferred_wakeup(rdp
))
2312 do_nocb_deferred_wakeup_common(rdp
);
2315 void __init
rcu_init_nohz(void)
2318 bool need_rcu_nocb_mask
= false;
2319 struct rcu_state
*rsp
;
2321 #if defined(CONFIG_NO_HZ_FULL)
2322 if (tick_nohz_full_running
&& cpumask_weight(tick_nohz_full_mask
))
2323 need_rcu_nocb_mask
= true;
2324 #endif /* #if defined(CONFIG_NO_HZ_FULL) */
2326 if (!cpumask_available(rcu_nocb_mask
) && need_rcu_nocb_mask
) {
2327 if (!zalloc_cpumask_var(&rcu_nocb_mask
, GFP_KERNEL
)) {
2328 pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n");
2332 if (!cpumask_available(rcu_nocb_mask
))
2335 #if defined(CONFIG_NO_HZ_FULL)
2336 if (tick_nohz_full_running
)
2337 cpumask_or(rcu_nocb_mask
, rcu_nocb_mask
, tick_nohz_full_mask
);
2338 #endif /* #if defined(CONFIG_NO_HZ_FULL) */
2340 if (!cpumask_subset(rcu_nocb_mask
, cpu_possible_mask
)) {
2341 pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n");
2342 cpumask_and(rcu_nocb_mask
, cpu_possible_mask
,
2345 if (cpumask_empty(rcu_nocb_mask
))
2346 pr_info("\tOffload RCU callbacks from CPUs: (none).\n");
2348 pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n",
2349 cpumask_pr_args(rcu_nocb_mask
));
2351 pr_info("\tPoll for callbacks from no-CBs CPUs.\n");
2353 for_each_rcu_flavor(rsp
) {
2354 for_each_cpu(cpu
, rcu_nocb_mask
)
2355 init_nocb_callback_list(per_cpu_ptr(rsp
->rda
, cpu
));
2356 rcu_organize_nocb_kthreads(rsp
);
2360 /* Initialize per-rcu_data variables for no-CBs CPUs. */
2361 static void __init
rcu_boot_init_nocb_percpu_data(struct rcu_data
*rdp
)
2363 rdp
->nocb_tail
= &rdp
->nocb_head
;
2364 init_swait_queue_head(&rdp
->nocb_wq
);
2365 rdp
->nocb_follower_tail
= &rdp
->nocb_follower_head
;
2366 raw_spin_lock_init(&rdp
->nocb_lock
);
2367 timer_setup(&rdp
->nocb_timer
, do_nocb_deferred_wakeup_timer
, 0);
2371 * If the specified CPU is a no-CBs CPU that does not already have its
2372 * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are
2373 * brought online out of order, this can require re-organizing the
2374 * leader-follower relationships.
2376 static void rcu_spawn_one_nocb_kthread(struct rcu_state
*rsp
, int cpu
)
2378 struct rcu_data
*rdp
;
2379 struct rcu_data
*rdp_last
;
2380 struct rcu_data
*rdp_old_leader
;
2381 struct rcu_data
*rdp_spawn
= per_cpu_ptr(rsp
->rda
, cpu
);
2382 struct task_struct
*t
;
2385 * If this isn't a no-CBs CPU or if it already has an rcuo kthread,
2386 * then nothing to do.
2388 if (!rcu_is_nocb_cpu(cpu
) || rdp_spawn
->nocb_kthread
)
2391 /* If we didn't spawn the leader first, reorganize! */
2392 rdp_old_leader
= rdp_spawn
->nocb_leader
;
2393 if (rdp_old_leader
!= rdp_spawn
&& !rdp_old_leader
->nocb_kthread
) {
2395 rdp
= rdp_old_leader
;
2397 rdp
->nocb_leader
= rdp_spawn
;
2398 if (rdp_last
&& rdp
!= rdp_spawn
)
2399 rdp_last
->nocb_next_follower
= rdp
;
2400 if (rdp
== rdp_spawn
) {
2401 rdp
= rdp
->nocb_next_follower
;
2404 rdp
= rdp
->nocb_next_follower
;
2405 rdp_last
->nocb_next_follower
= NULL
;
2408 rdp_spawn
->nocb_next_follower
= rdp_old_leader
;
2411 /* Spawn the kthread for this CPU and RCU flavor. */
2412 t
= kthread_run(rcu_nocb_kthread
, rdp_spawn
,
2413 "rcuo%c/%d", rsp
->abbr
, cpu
);
2415 WRITE_ONCE(rdp_spawn
->nocb_kthread
, t
);
2419 * If the specified CPU is a no-CBs CPU that does not already have its
2420 * rcuo kthreads, spawn them.
2422 static void rcu_spawn_all_nocb_kthreads(int cpu
)
2424 struct rcu_state
*rsp
;
2426 if (rcu_scheduler_fully_active
)
2427 for_each_rcu_flavor(rsp
)
2428 rcu_spawn_one_nocb_kthread(rsp
, cpu
);
2432 * Once the scheduler is running, spawn rcuo kthreads for all online
2433 * no-CBs CPUs. This assumes that the early_initcall()s happen before
2434 * non-boot CPUs come online -- if this changes, we will need to add
2435 * some mutual exclusion.
2437 static void __init
rcu_spawn_nocb_kthreads(void)
2441 for_each_online_cpu(cpu
)
2442 rcu_spawn_all_nocb_kthreads(cpu
);
2445 /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */
2446 static int rcu_nocb_leader_stride
= -1;
2447 module_param(rcu_nocb_leader_stride
, int, 0444);
2450 * Initialize leader-follower relationships for all no-CBs CPU.
2452 static void __init
rcu_organize_nocb_kthreads(struct rcu_state
*rsp
)
2455 int ls
= rcu_nocb_leader_stride
;
2456 int nl
= 0; /* Next leader. */
2457 struct rcu_data
*rdp
;
2458 struct rcu_data
*rdp_leader
= NULL
; /* Suppress misguided gcc warn. */
2459 struct rcu_data
*rdp_prev
= NULL
;
2461 if (!cpumask_available(rcu_nocb_mask
))
2464 ls
= int_sqrt(nr_cpu_ids
);
2465 rcu_nocb_leader_stride
= ls
;
2469 * Each pass through this loop sets up one rcu_data structure.
2470 * Should the corresponding CPU come online in the future, then
2471 * we will spawn the needed set of rcu_nocb_kthread() kthreads.
2473 for_each_cpu(cpu
, rcu_nocb_mask
) {
2474 rdp
= per_cpu_ptr(rsp
->rda
, cpu
);
2475 if (rdp
->cpu
>= nl
) {
2476 /* New leader, set up for followers & next leader. */
2477 nl
= DIV_ROUND_UP(rdp
->cpu
+ 1, ls
) * ls
;
2478 rdp
->nocb_leader
= rdp
;
2481 /* Another follower, link to previous leader. */
2482 rdp
->nocb_leader
= rdp_leader
;
2483 rdp_prev
->nocb_next_follower
= rdp
;
2489 /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */
2490 static bool init_nocb_callback_list(struct rcu_data
*rdp
)
2492 if (!rcu_is_nocb_cpu(rdp
->cpu
))
2495 /* If there are early-boot callbacks, move them to nocb lists. */
2496 if (!rcu_segcblist_empty(&rdp
->cblist
)) {
2497 rdp
->nocb_head
= rcu_segcblist_head(&rdp
->cblist
);
2498 rdp
->nocb_tail
= rcu_segcblist_tail(&rdp
->cblist
);
2499 atomic_long_set(&rdp
->nocb_q_count
,
2500 rcu_segcblist_n_cbs(&rdp
->cblist
));
2501 atomic_long_set(&rdp
->nocb_q_count_lazy
,
2502 rcu_segcblist_n_lazy_cbs(&rdp
->cblist
));
2503 rcu_segcblist_init(&rdp
->cblist
);
2505 rcu_segcblist_disable(&rdp
->cblist
);
2509 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
2511 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state
*rsp
, int cpu
)
2513 WARN_ON_ONCE(1); /* Should be dead code. */
2517 static void rcu_nocb_gp_cleanup(struct swait_queue_head
*sq
)
2521 static struct swait_queue_head
*rcu_nocb_gp_get(struct rcu_node
*rnp
)
2526 static void rcu_init_one_nocb(struct rcu_node
*rnp
)
2530 static bool __call_rcu_nocb(struct rcu_data
*rdp
, struct rcu_head
*rhp
,
2531 bool lazy
, unsigned long flags
)
2536 static bool __maybe_unused
rcu_nocb_adopt_orphan_cbs(struct rcu_data
*my_rdp
,
2537 struct rcu_data
*rdp
,
2538 unsigned long flags
)
2543 static void __init
rcu_boot_init_nocb_percpu_data(struct rcu_data
*rdp
)
2547 static int rcu_nocb_need_deferred_wakeup(struct rcu_data
*rdp
)
2552 static void do_nocb_deferred_wakeup(struct rcu_data
*rdp
)
2556 static void rcu_spawn_all_nocb_kthreads(int cpu
)
2560 static void __init
rcu_spawn_nocb_kthreads(void)
2564 static bool init_nocb_callback_list(struct rcu_data
*rdp
)
2569 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
2572 * An adaptive-ticks CPU can potentially execute in kernel mode for an
2573 * arbitrarily long period of time with the scheduling-clock tick turned
2574 * off. RCU will be paying attention to this CPU because it is in the
2575 * kernel, but the CPU cannot be guaranteed to be executing the RCU state
2576 * machine because the scheduling-clock tick has been disabled. Therefore,
2577 * if an adaptive-ticks CPU is failing to respond to the current grace
2578 * period and has not be idle from an RCU perspective, kick it.
2580 static void __maybe_unused
rcu_kick_nohz_cpu(int cpu
)
2582 #ifdef CONFIG_NO_HZ_FULL
2583 if (tick_nohz_full_cpu(cpu
))
2584 smp_send_reschedule(cpu
);
2585 #endif /* #ifdef CONFIG_NO_HZ_FULL */
2589 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
2590 * grace-period kthread will do force_quiescent_state() processing?
2591 * The idea is to avoid waking up RCU core processing on such a
2592 * CPU unless the grace period has extended for too long.
2594 * This code relies on the fact that all NO_HZ_FULL CPUs are also
2595 * CONFIG_RCU_NOCB_CPU CPUs.
2597 static bool rcu_nohz_full_cpu(struct rcu_state
*rsp
)
2599 #ifdef CONFIG_NO_HZ_FULL
2600 if (tick_nohz_full_cpu(smp_processor_id()) &&
2601 (!rcu_gp_in_progress(rsp
) ||
2602 ULONG_CMP_LT(jiffies
, READ_ONCE(rsp
->gp_start
) + HZ
)))
2604 #endif /* #ifdef CONFIG_NO_HZ_FULL */
2609 * Bind the RCU grace-period kthreads to the housekeeping CPU.
2611 static void rcu_bind_gp_kthread(void)
2613 int __maybe_unused cpu
;
2615 if (!tick_nohz_full_enabled())
2617 housekeeping_affine(current
, HK_FLAG_RCU
);
2620 /* Record the current task on dyntick-idle entry. */
2621 static void rcu_dynticks_task_enter(void)
2623 #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
2624 WRITE_ONCE(current
->rcu_tasks_idle_cpu
, smp_processor_id());
2625 #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
2628 /* Record no current task on dyntick-idle exit. */
2629 static void rcu_dynticks_task_exit(void)
2631 #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
2632 WRITE_ONCE(current
->rcu_tasks_idle_cpu
, -1);
2633 #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */