1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
2 "http://www.w3.org/TR/html4/loose.dtd">
4 <head><title>A Tour Through TREE_RCU's Expedited Grace Periods</title>
5 <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
9 This document describes RCU's expedited grace periods.
10 Unlike RCU's normal grace periods, which accept long latencies to attain
11 high efficiency and minimal disturbance, expedited grace periods accept
12 lower efficiency and significant disturbance to attain shorter latencies.
15 There are two flavors of RCU (RCU-preempt and RCU-sched), with an earlier
16 third RCU-bh flavor having been implemented in terms of the other two.
17 Each of the two implementations is covered in its own section.
20 <li> <a href="#Expedited Grace Period Design">
21 Expedited Grace Period Design</a>
22 <li> <a href="#RCU-preempt Expedited Grace Periods">
23 RCU-preempt Expedited Grace Periods</a>
24 <li> <a href="#RCU-sched Expedited Grace Periods">
25 RCU-sched Expedited Grace Periods</a>
26 <li> <a href="#Expedited Grace Period and CPU Hotplug">
27 Expedited Grace Period and CPU Hotplug</a>
28 <li> <a href="#Expedited Grace Period Refinements">
29 Expedited Grace Period Refinements</a>
32 <h2><a name="Expedited Grace Period Design">
33 Expedited Grace Period Design</a></h2>
36 The expedited RCU grace periods cannot be accused of being subtle,
37 given that they for all intents and purposes hammer every CPU that
38 has not yet provided a quiescent state for the current expedited
40 The one saving grace is that the hammer has grown a bit smaller
41 over time: The old call to <tt>try_stop_cpus()</tt> has been
42 replaced with a set of calls to <tt>smp_call_function_single()</tt>,
43 each of which results in an IPI to the target CPU.
44 The corresponding handler function checks the CPU's state, motivating
45 a faster quiescent state where possible, and triggering a report
46 of that quiescent state.
47 As always for RCU, once everything has spent some time in a quiescent
48 state, the expedited grace period has completed.
51 The details of the <tt>smp_call_function_single()</tt> handler's
52 operation depend on the RCU flavor, as described in the following
55 <h2><a name="RCU-preempt Expedited Grace Periods">
56 RCU-preempt Expedited Grace Periods</a></h2>
59 The overall flow of the handling of a given CPU by an RCU-preempt
60 expedited grace period is shown in the following diagram:
62 <p><img src="ExpRCUFlow.svg" alt="ExpRCUFlow.svg" width="55%">
65 The solid arrows denote direct action, for example, a function call.
66 The dotted arrows denote indirect action, for example, an IPI
67 or a state that is reached after some time.
70 If a given CPU is offline or idle, <tt>synchronize_rcu_expedited()</tt>
71 will ignore it because idle and offline CPUs are already residing
73 Otherwise, the expedited grace period will use
74 <tt>smp_call_function_single()</tt> to send the CPU an IPI, which
75 is handled by <tt>rcu_exp_handler()</tt>.
78 However, because this is preemptible RCU, <tt>rcu_exp_handler()</tt>
79 can check to see if the CPU is currently running in an RCU read-side
81 If not, the handler can immediately report a quiescent state.
82 Otherwise, it sets flags so that the outermost <tt>rcu_read_unlock()</tt>
83 invocation will provide the needed quiescent-state report.
84 This flag-setting avoids the previous forced preemption of all
85 CPUs that might have RCU read-side critical sections.
86 In addition, this flag-setting is done so as to avoid increasing
87 the overhead of the common-case fastpath through the scheduler.
90 Again because this is preemptible RCU, an RCU read-side critical section
92 When that happens, RCU will enqueue the task, which will the continue to
93 block the current expedited grace period until it resumes and finds its
94 outermost <tt>rcu_read_unlock()</tt>.
95 The CPU will report a quiescent state just after enqueuing the task because
96 the CPU is no longer blocking the grace period.
97 It is instead the preempted task doing the blocking.
98 The list of blocked tasks is managed by <tt>rcu_preempt_ctxt_queue()</tt>,
99 which is called from <tt>rcu_preempt_note_context_switch()</tt>, which
100 in turn is called from <tt>rcu_note_context_switch()</tt>, which in
101 turn is called from the scheduler.
104 <tr><th> </th></tr>
105 <tr><th align="left">Quick Quiz:</th></tr>
107 Why not just have the expedited grace period check the
108 state of all the CPUs?
109 After all, that would avoid all those real-time-unfriendly IPIs.
111 <tr><th align="left">Answer:</th></tr>
112 <tr><td bgcolor="#ffffff"><font color="ffffff">
113 Because we want the RCU read-side critical sections to run fast,
114 which means no memory barriers.
115 Therefore, it is not possible to safely check the state from some
117 And even if it was possible to safely check the state, it would
118 still be necessary to IPI the CPU to safely interact with the
119 upcoming <tt>rcu_read_unlock()</tt> invocation, which means that
120 the remote state testing would not help the worst-case
121 latency that real-time applications care about.
123 <p><font color="ffffff">One way to prevent your real-time
124 application from getting hit with these IPIs is to
125 build your kernel with <tt>CONFIG_NO_HZ_FULL=y</tt>.
126 RCU would then perceive the CPU running your application
127 as being idle, and it would be able to safely detect that
128 state without needing to IPI the CPU.
130 <tr><td> </td></tr>
134 Please note that this is just the overall flow:
135 Additional complications can arise due to races with CPUs going idle
136 or offline, among other things.
138 <h2><a name="RCU-sched Expedited Grace Periods">
139 RCU-sched Expedited Grace Periods</a></h2>
142 The overall flow of the handling of a given CPU by an RCU-sched
143 expedited grace period is shown in the following diagram:
145 <p><img src="ExpSchedFlow.svg" alt="ExpSchedFlow.svg" width="55%">
148 As with RCU-preempt, RCU-sched's
149 <tt>synchronize_sched_expedited()</tt> ignores offline and
150 idle CPUs, again because they are in remotely detectable
153 <tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
154 leave no trace of their invocation, in general it is not possible to tell
155 whether or not the current CPU is in an RCU read-side critical section.
156 The best that RCU-sched's <tt>rcu_exp_handler()</tt> can do is to check
157 for idle, on the off-chance that the CPU went idle while the IPI
159 If the CPU is idle, then <tt>rcu_exp_handler()</tt> reports
162 <p> Otherwise, the handler forces a future context switch by setting the
163 NEED_RESCHED flag of the current task's thread flag and the CPU preempt
165 At the time of the context switch, the CPU reports the quiescent state.
166 Should the CPU go offline first, it will report the quiescent state
169 <h2><a name="Expedited Grace Period and CPU Hotplug">
170 Expedited Grace Period and CPU Hotplug</a></h2>
173 The expedited nature of expedited grace periods require a much tighter
174 interaction with CPU hotplug operations than is required for normal
176 In addition, attempting to IPI offline CPUs will result in splats, but
177 failing to IPI online CPUs can result in too-short grace periods.
178 Neither option is acceptable in production kernels.
181 The interaction between expedited grace periods and CPU hotplug operations
182 is carried out at several levels:
185 <li> The number of CPUs that have ever been online is tracked
186 by the <tt>rcu_state</tt> structure's <tt>->ncpus</tt>
188 The <tt>rcu_state</tt> structure's <tt>->ncpus_snap</tt>
189 field tracks the number of CPUs that have ever been online
190 at the beginning of an RCU expedited grace period.
191 Note that this number never decreases, at least in the absence
193 <li> The identities of the CPUs that have ever been online is
194 tracked by the <tt>rcu_node</tt> structure's
195 <tt>->expmaskinitnext</tt> field.
196 The <tt>rcu_node</tt> structure's <tt>->expmaskinit</tt>
197 field tracks the identities of the CPUs that were online
198 at least once at the beginning of the most recent RCU
199 expedited grace period.
200 The <tt>rcu_state</tt> structure's <tt>->ncpus</tt> and
201 <tt>->ncpus_snap</tt> fields are used to detect when
202 new CPUs have come online for the first time, that is,
203 when the <tt>rcu_node</tt> structure's <tt>->expmaskinitnext</tt>
204 field has changed since the beginning of the last RCU
205 expedited grace period, which triggers an update of each
206 <tt>rcu_node</tt> structure's <tt>->expmaskinit</tt>
207 field from its <tt>->expmaskinitnext</tt> field.
208 <li> Each <tt>rcu_node</tt> structure's <tt>->expmaskinit</tt>
209 field is used to initialize that structure's
210 <tt>->expmask</tt> at the beginning of each RCU
211 expedited grace period.
212 This means that only those CPUs that have been online at least
213 once will be considered for a given grace period.
214 <li> Any CPU that goes offline will clear its bit in its leaf
215 <tt>rcu_node</tt> structure's <tt>->qsmaskinitnext</tt>
216 field, so any CPU with that bit clear can safely be ignored.
217 However, it is possible for a CPU coming online or going offline
218 to have this bit set for some time while <tt>cpu_online</tt>
219 returns <tt>false</tt>.
220 <li> For each non-idle CPU that RCU believes is currently online, the grace
221 period invokes <tt>smp_call_function_single()</tt>.
222 If this succeeds, the CPU was fully online.
223 Failure indicates that the CPU is in the process of coming online
224 or going offline, in which case it is necessary to wait for a
225 short time period and try again.
226 The purpose of this wait (or series of waits, as the case may be)
227 is to permit a concurrent CPU-hotplug operation to complete.
228 <li> In the case of RCU-sched, one of the last acts of an outgoing CPU
229 is to invoke <tt>rcu_report_dead()</tt>, which
230 reports a quiescent state for that CPU.
231 However, this is likely paranoia-induced redundancy. <!-- @@@ -->
235 <tr><th> </th></tr>
236 <tr><th align="left">Quick Quiz:</th></tr>
238 Why all the dancing around with multiple counters and masks
239 tracking CPUs that were once online?
240 Why not just have a single set of masks tracking the currently
241 online CPUs and be done with it?
243 <tr><th align="left">Answer:</th></tr>
244 <tr><td bgcolor="#ffffff"><font color="ffffff">
245 Maintaining single set of masks tracking the online CPUs <i>sounds</i>
246 easier, at least until you try working out all the race conditions
247 between grace-period initialization and CPU-hotplug operations.
248 For example, suppose initialization is progressing down the
249 tree while a CPU-offline operation is progressing up the tree.
250 This situation can result in bits set at the top of the tree
251 that have no counterparts at the bottom of the tree.
252 Those bits will never be cleared, which will result in
254 In short, that way lies madness, to say nothing of a great many
255 bugs, hangs, and deadlocks.
257 <p><font color="ffffff">
258 In contrast, the current multi-mask multi-counter scheme ensures
259 that grace-period initialization will always see consistent masks
260 up and down the tree, which brings significant simplifications
261 over the single-mask method.
263 <p><font color="ffffff">
264 This is an instance of
265 <a href="http://www.cs.columbia.edu/~library/TR-repository/reports/reports-1992/cucs-039-92.ps.gz"><font color="ffffff">
266 deferring work in order to avoid synchronization</a>.
267 Lazily recording CPU-hotplug events at the beginning of the next
268 grace period greatly simplifies maintenance of the CPU-tracking
269 bitmasks in the <tt>rcu_node</tt> tree.
271 <tr><td> </td></tr>
274 <h2><a name="Expedited Grace Period Refinements">
275 Expedited Grace Period Refinements</a></h2>
278 <li> <a href="#Idle-CPU Checks">Idle-CPU checks</a>.
279 <li> <a href="#Batching via Sequence Counter">
280 Batching via sequence counter</a>.
281 <li> <a href="#Funnel Locking and Wait/Wakeup">
282 Funnel locking and wait/wakeup</a>.
283 <li> <a href="#Use of Workqueues">Use of Workqueues</a>.
284 <li> <a href="#Stall Warnings">Stall warnings</a>.
285 <li> <a href="#Mid-Boot Operation">Mid-boot operation</a>.
288 <h3><a name="Idle-CPU Checks">Idle-CPU Checks</a></h3>
291 Each expedited grace period checks for idle CPUs when initially forming
292 the mask of CPUs to be IPIed and again just before IPIing a CPU
293 (both checks are carried out by <tt>sync_rcu_exp_select_cpus()</tt>).
294 If the CPU is idle at any time between those two times, the CPU will
296 Instead, the task pushing the grace period forward will include the
297 idle CPUs in the mask passed to <tt>rcu_report_exp_cpu_mult()</tt>.
300 For RCU-sched, there is an additional check:
301 If the IPI has interrupted the idle loop, then
302 <tt>rcu_exp_handler()</tt> invokes <tt>rcu_report_exp_rdp()</tt>
303 to report the corresponding quiescent state.
306 For RCU-preempt, there is no specific check for idle in the
307 IPI handler (<tt>rcu_exp_handler()</tt>), but because
308 RCU read-side critical sections are not permitted within the
309 idle loop, if <tt>rcu_exp_handler()</tt> sees that the CPU is within
310 RCU read-side critical section, the CPU cannot possibly be idle.
311 Otherwise, <tt>rcu_exp_handler()</tt> invokes
312 <tt>rcu_report_exp_rdp()</tt> to report the corresponding quiescent
313 state, regardless of whether or not that quiescent state was due to
317 In summary, RCU expedited grace periods check for idle when building
318 the bitmask of CPUs that must be IPIed, just before sending each IPI,
319 and (either explicitly or implicitly) within the IPI handler.
321 <h3><a name="Batching via Sequence Counter">
322 Batching via Sequence Counter</a></h3>
325 If each grace-period request was carried out separately, expedited
326 grace periods would have abysmal scalability and
327 problematic high-load characteristics.
328 Because each grace-period operation can serve an unlimited number of
329 updates, it is important to <i>batch</i> requests, so that a single
330 expedited grace-period operation will cover all requests in the
334 This batching is controlled by a sequence counter named
335 <tt>->expedited_sequence</tt> in the <tt>rcu_state</tt> structure.
336 This counter has an odd value when there is an expedited grace period
337 in progress and an even value otherwise, so that dividing the counter
338 value by two gives the number of completed grace periods.
339 During any given update request, the counter must transition from
340 even to odd and then back to even, thus indicating that a grace
342 Therefore, if the initial value of the counter is <tt>s</tt>,
343 the updater must wait until the counter reaches at least the
344 value <tt>(s+3)&~0x1</tt>.
345 This counter is managed by the following access functions:
348 <li> <tt>rcu_exp_gp_seq_start()</tt>, which marks the start of
349 an expedited grace period.
350 <li> <tt>rcu_exp_gp_seq_end()</tt>, which marks the end of an
351 expedited grace period.
352 <li> <tt>rcu_exp_gp_seq_snap()</tt>, which obtains a snapshot of
354 <li> <tt>rcu_exp_gp_seq_done()</tt>, which returns <tt>true</tt>
355 if a full expedited grace period has elapsed since the
356 corresponding call to <tt>rcu_exp_gp_seq_snap()</tt>.
360 Again, only one request in a given batch need actually carry out
361 a grace-period operation, which means there must be an efficient
362 way to identify which of many concurrent reqeusts will initiate
363 the grace period, and that there be an efficient way for the
364 remaining requests to wait for that grace period to complete.
365 However, that is the topic of the next section.
367 <h3><a name="Funnel Locking and Wait/Wakeup">
368 Funnel Locking and Wait/Wakeup</a></h3>
371 The natural way to sort out which of a batch of updaters will initiate
372 the expedited grace period is to use the <tt>rcu_node</tt> combining
373 tree, as implemented by the <tt>exp_funnel_lock()</tt> function.
374 The first updater corresponding to a given grace period arriving
375 at a given <tt>rcu_node</tt> structure records its desired grace-period
376 sequence number in the <tt>->exp_seq_rq</tt> field and moves up
377 to the next level in the tree.
378 Otherwise, if the <tt>->exp_seq_rq</tt> field already contains
379 the sequence number for the desired grace period or some later one,
380 the updater blocks on one of four wait queues in the
381 <tt>->exp_wq[]</tt> array, using the second-from-bottom
382 and third-from bottom bits as an index.
383 An <tt>->exp_lock</tt> field in the <tt>rcu_node</tt> structure
384 synchronizes access to these fields.
387 An empty <tt>rcu_node</tt> tree is shown in the following diagram,
388 with the white cells representing the <tt>->exp_seq_rq</tt> field
389 and the red cells representing the elements of the
390 <tt>->exp_wq[]</tt> array.
392 <p><img src="Funnel0.svg" alt="Funnel0.svg" width="75%">
395 The next diagram shows the situation after the arrival of Task A
396 and Task B at the leftmost and rightmost leaf <tt>rcu_node</tt>
397 structures, respectively.
398 The current value of the <tt>rcu_state</tt> structure's
399 <tt>->expedited_sequence</tt> field is zero, so adding three and
400 clearing the bottom bit results in the value two, which both tasks
401 record in the <tt>->exp_seq_rq</tt> field of their respective
402 <tt>rcu_node</tt> structures:
404 <p><img src="Funnel1.svg" alt="Funnel1.svg" width="75%">
407 Each of Tasks A and B will move up to the root
408 <tt>rcu_node</tt> structure.
409 Suppose that Task A wins, recording its desired grace-period sequence
410 number and resulting in the state shown below:
412 <p><img src="Funnel2.svg" alt="Funnel2.svg" width="75%">
415 Task A now advances to initiate a new grace period, while Task B
416 moves up to the root <tt>rcu_node</tt> structure, and, seeing that
417 its desired sequence number is already recorded, blocks on
418 <tt>->exp_wq[1]</tt>.
421 <tr><th> </th></tr>
422 <tr><th align="left">Quick Quiz:</th></tr>
424 Why <tt>->exp_wq[1]</tt>?
425 Given that the value of these tasks' desired sequence number is
426 two, so shouldn't they instead block on <tt>->exp_wq[2]</tt>?
428 <tr><th align="left">Answer:</th></tr>
429 <tr><td bgcolor="#ffffff"><font color="ffffff">
432 <p><font color="ffffff">
433 Recall that the bottom bit of the desired sequence number indicates
434 whether or not a grace period is currently in progress.
435 It is therefore necessary to shift the sequence number right one
436 bit position to obtain the number of the grace period.
437 This results in <tt>->exp_wq[1]</tt>.
439 <tr><td> </td></tr>
443 If Tasks C and D also arrive at this point, they will compute the
444 same desired grace-period sequence number, and see that both leaf
445 <tt>rcu_node</tt> structures already have that value recorded.
446 They will therefore block on their respective <tt>rcu_node</tt>
447 structures' <tt>->exp_wq[1]</tt> fields, as shown below:
449 <p><img src="Funnel3.svg" alt="Funnel3.svg" width="75%">
452 Task A now acquires the <tt>rcu_state</tt> structure's
453 <tt>->exp_mutex</tt> and initiates the grace period, which
454 increments <tt>->expedited_sequence</tt>.
455 Therefore, if Tasks E and F arrive, they will compute
456 a desired sequence number of 4 and will record this value as
459 <p><img src="Funnel4.svg" alt="Funnel4.svg" width="75%">
462 Tasks E and F will propagate up the <tt>rcu_node</tt>
463 combining tree, with Task F blocking on the root <tt>rcu_node</tt>
464 structure and Task E wait for Task A to finish so that
465 it can start the next grace period.
466 The resulting state is as shown below:
468 <p><img src="Funnel5.svg" alt="Funnel5.svg" width="75%">
471 Once the grace period completes, Task A
472 starts waking up the tasks waiting for this grace period to complete,
473 increments the <tt>->expedited_sequence</tt>,
474 acquires the <tt>->exp_wake_mutex</tt> and then releases the
475 <tt>->exp_mutex</tt>.
476 This results in the following state:
478 <p><img src="Funnel6.svg" alt="Funnel6.svg" width="75%">
481 Task E can then acquire <tt>->exp_mutex</tt> and increment
482 <tt>->expedited_sequence</tt> to the value three.
483 If new tasks G and H arrive and moves up the combining tree at the
484 same time, the state will be as follows:
486 <p><img src="Funnel7.svg" alt="Funnel7.svg" width="75%">
489 Note that three of the root <tt>rcu_node</tt> structure's
490 waitqueues are now occupied.
491 However, at some point, Task A will wake up the
492 tasks blocked on the <tt>->exp_wq</tt> waitqueues, resulting
493 in the following state:
495 <p><img src="Funnel8.svg" alt="Funnel8.svg" width="75%">
498 Execution will continue with Tasks E and H completing
499 their grace periods and carrying out their wakeups.
502 <tr><th> </th></tr>
503 <tr><th align="left">Quick Quiz:</th></tr>
505 What happens if Task A takes so long to do its wakeups
506 that Task E's grace period completes?
508 <tr><th align="left">Answer:</th></tr>
509 <tr><td bgcolor="#ffffff"><font color="ffffff">
510 Then Task E will block on the <tt>->exp_wake_mutex</tt>,
511 which will also prevent it from releasing <tt>->exp_mutex</tt>,
512 which in turn will prevent the next grace period from starting.
513 This last is important in preventing overflow of the
514 <tt>->exp_wq[]</tt> array.
516 <tr><td> </td></tr>
519 <h3><a name="Use of Workqueues">Use of Workqueues</a></h3>
522 In earlier implementations, the task requesting the expedited
523 grace period also drove it to completion.
524 This straightforward approach had the disadvantage of needing to
525 account for POSIX signals sent to user tasks,
526 so more recent implemementations use the Linux kernel's
527 <a href="https://www.kernel.org/doc/Documentation/core-api/workqueue.rst">workqueues</a>.
530 The requesting task still does counter snapshotting and funnel-lock
531 processing, but the task reaching the top of the funnel lock
532 does a <tt>schedule_work()</tt> (from <tt>_synchronize_rcu_expedited()</tt>
533 so that a workqueue kthread does the actual grace-period processing.
534 Because workqueue kthreads do not accept POSIX signals, grace-period-wait
535 processing need not allow for POSIX signals.
537 In addition, this approach allows wakeups for the previous expedited
538 grace period to be overlapped with processing for the next expedited
540 Because there are only four sets of waitqueues, it is necessary to
541 ensure that the previous grace period's wakeups complete before the
542 next grace period's wakeups start.
543 This is handled by having the <tt>->exp_mutex</tt>
544 guard expedited grace-period processing and the
545 <tt>->exp_wake_mutex</tt> guard wakeups.
546 The key point is that the <tt>->exp_mutex</tt> is not released
547 until the first wakeup is complete, which means that the
548 <tt>->exp_wake_mutex</tt> has already been acquired at that point.
549 This approach ensures that the previous grace period's wakeups can
550 be carried out while the current grace period is in process, but
551 that these wakeups will complete before the next grace period starts.
552 This means that only three waitqueues are required, guaranteeing that
553 the four that are provided are sufficient.
555 <h3><a name="Stall Warnings">Stall Warnings</a></h3>
558 Expediting grace periods does nothing to speed things up when RCU
559 readers take too long, and therefore expedited grace periods check
560 for stalls just as normal grace periods do.
563 <tr><th> </th></tr>
564 <tr><th align="left">Quick Quiz:</th></tr>
566 But why not just let the normal grace-period machinery
567 detect the stalls, given that a given reader must block
568 both normal and expedited grace periods?
570 <tr><th align="left">Answer:</th></tr>
571 <tr><td bgcolor="#ffffff"><font color="ffffff">
572 Because it is quite possible that at a given time there
573 is no normal grace period in progress, in which case the
574 normal grace period cannot emit a stall warning.
576 <tr><td> </td></tr>
579 The <tt>synchronize_sched_expedited_wait()</tt> function loops waiting
580 for the expedited grace period to end, but with a timeout set to the
581 current RCU CPU stall-warning time.
582 If this time is exceeded, any CPUs or <tt>rcu_node</tt> structures
583 blocking the current grace period are printed.
584 Each stall warning results in another pass through the loop, but the
585 second and subsequent passes use longer stall times.
587 <h3><a name="Mid-Boot Operation">Mid-boot operation</a></h3>
590 The use of workqueues has the advantage that the expedited
591 grace-period code need not worry about POSIX signals.
592 Unfortunately, it has the
593 corresponding disadvantage that workqueues cannot be used until
594 they are initialized, which does not happen until some time after
595 the scheduler spawns the first task.
596 Given that there are parts of the kernel that really do want to
597 execute grace periods during this mid-boot “dead zone”,
598 expedited grace periods must do something else during thie time.
601 What they do is to fall back to the old practice of requiring that the
602 requesting task drive the expedited grace period, as was the case
603 before the use of workqueues.
604 However, the requesting task is only required to drive the grace period
605 during the mid-boot dead zone.
606 Before mid-boot, a synchronous grace period is a no-op.
607 Some time after mid-boot, workqueues are used.
610 Non-expedited non-SRCU synchronous grace periods must also operate
611 normally during mid-boot.
612 This is handled by causing non-expedited grace periods to take the
613 expedited code path during mid-boot.
616 The current code assumes that there are no POSIX signals during
617 the mid-boot dead zone.
618 However, if an overwhelming need for POSIX signals somehow arises,
619 appropriate adjustments can be made to the expedited stall-warning code.
620 One such adjustment would reinstate the pre-workqueue stall-warning
621 checks, but only during the mid-boot dead zone.
624 With this refinement, synchronous grace periods can now be used from
625 task context pretty much any time during the life of the kernel.
626 That is, aside from some points in the suspend, hibernate, or shutdown
629 <h3><a name="Summary">
633 Expedited grace periods use a sequence-number approach to promote
634 batching, so that a single grace-period operation can serve numerous
636 A funnel lock is used to efficiently identify the one task out of
637 a concurrent group that will request the grace period.
638 All members of the group will block on waitqueues provided in
639 the <tt>rcu_node</tt> structure.
640 The actual grace-period processing is carried out by a workqueue.
643 CPU-hotplug operations are noted lazily in order to prevent the need
644 for tight synchronization between expedited grace periods and
645 CPU-hotplug operations.
646 The dyntick-idle counters are used to avoid sending IPIs to idle CPUs,
647 at least in the common case.
648 RCU-preempt and RCU-sched use different IPI handlers and different
649 code to respond to the state changes carried out by those handlers,
650 but otherwise use common code.
653 Quiescent states are tracked using the <tt>rcu_node</tt> tree,
654 and once all necessary quiescent states have been reported,
655 all tasks waiting on this expedited grace period are awakened.
656 A pair of mutexes are used to allow one grace period's wakeups
657 to proceed concurrently with the next grace period's processing.
660 This combination of mechanisms allows expedited grace periods to
661 run reasonably efficiently.
662 However, for non-time-critical tasks, normal grace periods should be
663 used instead because their longer duration permits much higher
664 degrees of batching, and thus much lower per-request overheads.