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 RCU's Requirements [LWN.net]</title>
5 <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">
7 <h1>A Tour Through RCU's Requirements</h1>
9 <p>Copyright IBM Corporation, 2015</p>
10 <p>Author: Paul E. McKenney</p>
11 <p><i>The initial version of this document appeared in the
12 <a href="https://lwn.net/">LWN</a> articles
13 <a href="https://lwn.net/Articles/652156/">here</a>,
14 <a href="https://lwn.net/Articles/652677/">here</a>, and
15 <a href="https://lwn.net/Articles/653326/">here</a>.</i></p>
20 Read-copy update (RCU) is a synchronization mechanism that is often
21 used as a replacement for reader-writer locking.
22 RCU is unusual in that updaters do not block readers,
23 which means that RCU's read-side primitives can be exceedingly fast
25 In addition, updaters can make useful forward progress concurrently
27 However, all this concurrency between RCU readers and updaters does raise
28 the question of exactly what RCU readers are doing, which in turn
29 raises the question of exactly what RCU's requirements are.
32 This document therefore summarizes RCU's requirements, and can be thought
33 of as an informal, high-level specification for RCU.
34 It is important to understand that RCU's specification is primarily
36 in fact, I learned about many of these requirements the hard way.
37 This situation might cause some consternation, however, not only
38 has this learning process been a lot of fun, but it has also been
39 a great privilege to work with so many people willing to apply
40 technologies in interesting new ways.
43 All that aside, here are the categories of currently known RCU requirements:
47 <li> <a href="#Fundamental Requirements">
48 Fundamental Requirements</a>
49 <li> <a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
50 <li> <a href="#Parallelism Facts of Life">
51 Parallelism Facts of Life</a>
52 <li> <a href="#Quality-of-Implementation Requirements">
53 Quality-of-Implementation Requirements</a>
54 <li> <a href="#Linux Kernel Complications">
55 Linux Kernel Complications</a>
56 <li> <a href="#Software-Engineering Requirements">
57 Software-Engineering Requirements</a>
58 <li> <a href="#Other RCU Flavors">
60 <li> <a href="#Possible Future Changes">
61 Possible Future Changes</a>
65 This is followed by a <a href="#Summary">summary</a>,
66 however, the answers to each quick quiz immediately follows the quiz.
67 Select the big white space with your mouse to see the answer.
69 <h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>
72 RCU's fundamental requirements are the closest thing RCU has to hard
73 mathematical requirements.
77 <li> <a href="#Grace-Period Guarantee">
78 Grace-Period Guarantee</a>
79 <li> <a href="#Publish-Subscribe Guarantee">
80 Publish-Subscribe Guarantee</a>
81 <li> <a href="#Memory-Barrier Guarantees">
82 Memory-Barrier Guarantees</a>
83 <li> <a href="#RCU Primitives Guaranteed to Execute Unconditionally">
84 RCU Primitives Guaranteed to Execute Unconditionally</a>
85 <li> <a href="#Guaranteed Read-to-Write Upgrade">
86 Guaranteed Read-to-Write Upgrade</a>
89 <h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>
92 RCU's grace-period guarantee is unusual in being premeditated:
93 Jack Slingwine and I had this guarantee firmly in mind when we started
94 work on RCU (then called “rclock”) in the early 1990s.
95 That said, the past two decades of experience with RCU have produced
96 a much more detailed understanding of this guarantee.
99 RCU's grace-period guarantee allows updaters to wait for the completion
100 of all pre-existing RCU read-side critical sections.
101 An RCU read-side critical section
102 begins with the marker <tt>rcu_read_lock()</tt> and ends with
103 the marker <tt>rcu_read_unlock()</tt>.
104 These markers may be nested, and RCU treats a nested set as one
105 big RCU read-side critical section.
106 Production-quality implementations of <tt>rcu_read_lock()</tt> and
107 <tt>rcu_read_unlock()</tt> are extremely lightweight, and in
108 fact have exactly zero overhead in Linux kernels built for production
109 use with <tt>CONFIG_PREEMPT=n</tt>.
112 This guarantee allows ordering to be enforced with extremely low
113 overhead to readers, for example:
127 11 void thread1(void)
130 14 synchronize_rcu();
137 Because the <tt>synchronize_rcu()</tt> on line 14 waits for
138 all pre-existing readers, any instance of <tt>thread0()</tt> that
139 loads a value of zero from <tt>x</tt> must complete before
140 <tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
141 also load a value of zero from <tt>y</tt>.
142 Similarly, any instance of <tt>thread0()</tt> that loads a value of
143 one from <tt>y</tt> must have started after the
144 <tt>synchronize_rcu()</tt> started, and must therefore also load
145 a value of one from <tt>x</tt>.
146 Therefore, the outcome:
149 (r1 == 0 && r2 == 1)
155 <tr><th> </th></tr>
156 <tr><th align="left">Quick Quiz:</th></tr>
159 You said that updaters can make useful forward progress concurrently
160 with readers, but pre-existing readers will block
161 <tt>synchronize_rcu()</tt>!!!
162 Just who are you trying to fool???
164 <tr><th align="left">Answer:</th></tr>
165 <tr><td bgcolor="#ffffff"><font color="ffffff">
166 First, if updaters do not wish to be blocked by readers, they can use
167 <tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
169 Second, even when using <tt>synchronize_rcu()</tt>, the other
170 update-side code does run concurrently with readers, whether
173 <tr><td> </td></tr>
177 This scenario resembles one of the first uses of RCU in
178 <a href="https://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
179 which managed a distributed lock manager's transition into
180 a state suitable for handling recovery from node failure,
181 more or less as follows:
185 1 #define STATE_NORMAL 0
186 2 #define STATE_WANT_RECOVERY 1
187 3 #define STATE_RECOVERING 2
188 4 #define STATE_WANT_NORMAL 3
190 6 int state = STATE_NORMAL;
192 8 void do_something_dlm(void)
197 13 state_snap = READ_ONCE(state);
198 14 if (state_snap == STATE_NORMAL)
201 17 do_something_carefully();
202 18 rcu_read_unlock();
205 21 void start_recovery(void)
207 23 WRITE_ONCE(state, STATE_WANT_RECOVERY);
208 24 synchronize_rcu();
209 25 WRITE_ONCE(state, STATE_RECOVERING);
211 27 WRITE_ONCE(state, STATE_WANT_NORMAL);
212 28 synchronize_rcu();
213 29 WRITE_ONCE(state, STATE_NORMAL);
219 The RCU read-side critical section in <tt>do_something_dlm()</tt>
220 works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
221 to guarantee that <tt>do_something()</tt> never runs concurrently
222 with <tt>recovery()</tt>, but with little or no synchronization
223 overhead in <tt>do_something_dlm()</tt>.
226 <tr><th> </th></tr>
227 <tr><th align="left">Quick Quiz:</th></tr>
229 Why is the <tt>synchronize_rcu()</tt> on line 28 needed?
231 <tr><th align="left">Answer:</th></tr>
232 <tr><td bgcolor="#ffffff"><font color="ffffff">
233 Without that extra grace period, memory reordering could result in
234 <tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
235 concurrently with the last bits of <tt>recovery()</tt>.
237 <tr><td> </td></tr>
241 In order to avoid fatal problems such as deadlocks,
242 an RCU read-side critical section must not contain calls to
243 <tt>synchronize_rcu()</tt>.
244 Similarly, an RCU read-side critical section must not
245 contain anything that waits, directly or indirectly, on completion of
246 an invocation of <tt>synchronize_rcu()</tt>.
249 Although RCU's grace-period guarantee is useful in and of itself, with
250 <a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
251 it would be good to be able to use RCU to coordinate read-side
252 access to linked data structures.
253 For this, the grace-period guarantee is not sufficient, as can
254 be seen in function <tt>add_gp_buggy()</tt> below.
255 We will look at the reader's code later, but in the meantime, just think of
256 the reader as locklessly picking up the <tt>gp</tt> pointer,
257 and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
258 <tt>->a</tt> and <tt>->b</tt> fields.
262 1 bool add_gp_buggy(int a, int b)
264 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
267 6 spin_lock(&gp_lock);
268 7 if (rcu_access_pointer(gp)) {
269 8 spin_unlock(&gp_lock);
274 13 gp = p; /* ORDERING BUG */
275 14 spin_unlock(&gp_lock);
282 The problem is that both the compiler and weakly ordered CPUs are within
283 their rights to reorder this code as follows:
287 1 bool add_gp_buggy_optimized(int a, int b)
289 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
292 6 spin_lock(&gp_lock);
293 7 if (rcu_access_pointer(gp)) {
294 8 spin_unlock(&gp_lock);
297 <b>11 gp = p; /* ORDERING BUG */
300 14 spin_unlock(&gp_lock);
307 If an RCU reader fetches <tt>gp</tt> just after
308 <tt>add_gp_buggy_optimized</tt> executes line 11,
309 it will see garbage in the <tt>->a</tt> and <tt>->b</tt>
311 And this is but one of many ways in which compiler and hardware optimizations
313 Therefore, we clearly need some way to prevent the compiler and the CPU from
314 reordering in this manner, which brings us to the publish-subscribe
315 guarantee discussed in the next section.
317 <h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>
320 RCU's publish-subscribe guarantee allows data to be inserted
321 into a linked data structure without disrupting RCU readers.
322 The updater uses <tt>rcu_assign_pointer()</tt> to insert the
323 new data, and readers use <tt>rcu_dereference()</tt> to
324 access data, whether new or old.
325 The following shows an example of insertion:
329 1 bool add_gp(int a, int b)
331 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
334 6 spin_lock(&gp_lock);
335 7 if (rcu_access_pointer(gp)) {
336 8 spin_unlock(&gp_lock);
341 13 rcu_assign_pointer(gp, p);
342 14 spin_unlock(&gp_lock);
349 The <tt>rcu_assign_pointer()</tt> on line 13 is conceptually
350 equivalent to a simple assignment statement, but also guarantees
351 that its assignment will
352 happen after the two assignments in lines 11 and 12,
353 similar to the C11 <tt>memory_order_release</tt> store operation.
354 It also prevents any number of “interesting” compiler
355 optimizations, for example, the use of <tt>gp</tt> as a scratch
356 location immediately preceding the assignment.
359 <tr><th> </th></tr>
360 <tr><th align="left">Quick Quiz:</th></tr>
362 But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
363 two assignments to <tt>p->a</tt> and <tt>p->b</tt>
364 from being reordered.
365 Can't that also cause problems?
367 <tr><th align="left">Answer:</th></tr>
368 <tr><td bgcolor="#ffffff"><font color="ffffff">
370 The readers cannot see either of these two fields until
371 the assignment to <tt>gp</tt>, by which time both fields are
373 So reordering the assignments
374 to <tt>p->a</tt> and <tt>p->b</tt> cannot possibly
377 <tr><td> </td></tr>
381 It is tempting to assume that the reader need not do anything special
382 to control its accesses to the RCU-protected data,
383 as shown in <tt>do_something_gp_buggy()</tt> below:
387 1 bool do_something_gp_buggy(void)
390 4 p = gp; /* OPTIMIZATIONS GALORE!!! */
392 6 do_something(p->a, p->b);
396 10 rcu_read_unlock();
403 However, this temptation must be resisted because there are a
404 surprisingly large number of ways that the compiler
406 <a href="https://h71000.www7.hp.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
407 can trip this code up.
408 For but one example, if the compiler were short of registers, it
409 might choose to refetch from <tt>gp</tt> rather than keeping
410 a separate copy in <tt>p</tt> as follows:
414 1 bool do_something_gp_buggy_optimized(void)
417 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */
418 <b> 5 do_something(gp->a, gp->b);</b>
429 If this function ran concurrently with a series of updates that
430 replaced the current structure with a new one,
431 the fetches of <tt>gp->a</tt>
432 and <tt>gp->b</tt> might well come from two different structures,
433 which could cause serious confusion.
434 To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
435 <tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:
439 1 bool do_something_gp(void)
442 4 p = rcu_dereference(gp);
444 6 do_something(p->a, p->b);
448 10 rcu_read_unlock();
455 The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
456 memory barriers in the Linux kernel.
458 <a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
459 ever appear, then <tt>rcu_dereference()</tt> could be implemented
460 as a <tt>memory_order_consume</tt> load.
461 Regardless of the exact implementation, a pointer fetched by
462 <tt>rcu_dereference()</tt> may not be used outside of the
463 outermost RCU read-side critical section containing that
464 <tt>rcu_dereference()</tt>, unless protection of
465 the corresponding data element has been passed from RCU to some
466 other synchronization mechanism, most commonly locking or
467 <a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.
470 In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
471 use <tt>rcu_dereference()</tt>, and these two RCU API elements
472 work together to ensure that readers have a consistent view of
473 newly added data elements.
476 Of course, it is also necessary to remove elements from RCU-protected
477 data structures, for example, using the following process:
480 <li> Remove the data element from the enclosing structure.
481 <li> Wait for all pre-existing RCU read-side critical sections
482 to complete (because only pre-existing readers can possibly have
483 a reference to the newly removed data element).
484 <li> At this point, only the updater has a reference to the
485 newly removed data element, so it can safely reclaim
486 the data element, for example, by passing it to <tt>kfree()</tt>.
489 This process is implemented by <tt>remove_gp_synchronous()</tt>:
493 1 bool remove_gp_synchronous(void)
497 5 spin_lock(&gp_lock);
498 6 p = rcu_access_pointer(gp);
500 8 spin_unlock(&gp_lock);
503 11 rcu_assign_pointer(gp, NULL);
504 12 spin_unlock(&gp_lock);
505 13 synchronize_rcu();
513 This function is straightforward, with line 13 waiting for a grace
514 period before line 14 frees the old data element.
515 This waiting ensures that readers will reach line 7 of
516 <tt>do_something_gp()</tt> before the data element referenced by
518 The <tt>rcu_access_pointer()</tt> on line 6 is similar to
519 <tt>rcu_dereference()</tt>, except that:
522 <li> The value returned by <tt>rcu_access_pointer()</tt>
523 cannot be dereferenced.
524 If you want to access the value pointed to as well as
525 the pointer itself, use <tt>rcu_dereference()</tt>
526 instead of <tt>rcu_access_pointer()</tt>.
527 <li> The call to <tt>rcu_access_pointer()</tt> need not be
529 In contrast, <tt>rcu_dereference()</tt> must either be
530 within an RCU read-side critical section or in a code
531 segment where the pointer cannot change, for example, in
532 code protected by the corresponding update-side lock.
536 <tr><th> </th></tr>
537 <tr><th align="left">Quick Quiz:</th></tr>
539 Without the <tt>rcu_dereference()</tt> or the
540 <tt>rcu_access_pointer()</tt>, what destructive optimizations
541 might the compiler make use of?
543 <tr><th align="left">Answer:</th></tr>
544 <tr><td bgcolor="#ffffff"><font color="ffffff">
545 Let's start with what happens to <tt>do_something_gp()</tt>
546 if it fails to use <tt>rcu_dereference()</tt>.
547 It could reuse a value formerly fetched from this same pointer.
548 It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
549 manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
550 mash-up of two distinct pointer values.
551 It might even use value-speculation optimizations, where it makes
552 a wrong guess, but by the time it gets around to checking the
553 value, an update has changed the pointer to match the wrong guess.
554 Too bad about any dereferences that returned pre-initialization garbage
558 <p><font color="ffffff">
559 For <tt>remove_gp_synchronous()</tt>, as long as all modifications
560 to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
561 the above optimizations are harmless.
562 However, <tt>sparse</tt> will complain if you
563 define <tt>gp</tt> with <tt>__rcu</tt> and then
564 access it without using
565 either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
567 <tr><td> </td></tr>
571 In short, RCU's publish-subscribe guarantee is provided by the combination
572 of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
573 This guarantee allows data elements to be safely added to RCU-protected
574 linked data structures without disrupting RCU readers.
575 This guarantee can be used in combination with the grace-period
576 guarantee to also allow data elements to be removed from RCU-protected
577 linked data structures, again without disrupting RCU readers.
580 This guarantee was only partially premeditated.
581 DYNIX/ptx used an explicit memory barrier for publication, but had nothing
582 resembling <tt>rcu_dereference()</tt> for subscription, nor did it
583 have anything resembling the <tt>smp_read_barrier_depends()</tt>
584 that was later subsumed into <tt>rcu_dereference()</tt> and later
585 still into <tt>READ_ONCE()</tt>.
586 The need for these operations made itself known quite suddenly at a
587 late-1990s meeting with the DEC Alpha architects, back in the days when
588 DEC was still a free-standing company.
589 It took the Alpha architects a good hour to convince me that any sort
590 of barrier would ever be needed, and it then took me a good <i>two</i> hours
591 to convince them that their documentation did not make this point clear.
592 More recent work with the C and C++ standards committees have provided
593 much education on tricks and traps from the compiler.
594 In short, compilers were much less tricky in the early 1990s, but in
595 2015, don't even think about omitting <tt>rcu_dereference()</tt>!
597 <h3><a name="Memory-Barrier Guarantees">Memory-Barrier Guarantees</a></h3>
600 The previous section's simple linked-data-structure scenario clearly
601 demonstrates the need for RCU's stringent memory-ordering guarantees on
602 systems with more than one CPU:
605 <li> Each CPU that has an RCU read-side critical section that
606 begins before <tt>synchronize_rcu()</tt> starts is
607 guaranteed to execute a full memory barrier between the time
608 that the RCU read-side critical section ends and the time that
609 <tt>synchronize_rcu()</tt> returns.
610 Without this guarantee, a pre-existing RCU read-side critical section
611 might hold a reference to the newly removed <tt>struct foo</tt>
612 after the <tt>kfree()</tt> on line 14 of
613 <tt>remove_gp_synchronous()</tt>.
614 <li> Each CPU that has an RCU read-side critical section that ends
615 after <tt>synchronize_rcu()</tt> returns is guaranteed
616 to execute a full memory barrier between the time that
617 <tt>synchronize_rcu()</tt> begins and the time that the RCU
618 read-side critical section begins.
619 Without this guarantee, a later RCU read-side critical section
620 running after the <tt>kfree()</tt> on line 14 of
621 <tt>remove_gp_synchronous()</tt> might
622 later run <tt>do_something_gp()</tt> and find the
623 newly deleted <tt>struct foo</tt>.
624 <li> If the task invoking <tt>synchronize_rcu()</tt> remains
625 on a given CPU, then that CPU is guaranteed to execute a full
626 memory barrier sometime during the execution of
627 <tt>synchronize_rcu()</tt>.
628 This guarantee ensures that the <tt>kfree()</tt> on
629 line 14 of <tt>remove_gp_synchronous()</tt> really does
630 execute after the removal on line 11.
631 <li> If the task invoking <tt>synchronize_rcu()</tt> migrates
632 among a group of CPUs during that invocation, then each of the
633 CPUs in that group is guaranteed to execute a full memory barrier
634 sometime during the execution of <tt>synchronize_rcu()</tt>.
635 This guarantee also ensures that the <tt>kfree()</tt> on
636 line 14 of <tt>remove_gp_synchronous()</tt> really does
637 execute after the removal on
638 line 11, but also in the case where the thread executing the
639 <tt>synchronize_rcu()</tt> migrates in the meantime.
643 <tr><th> </th></tr>
644 <tr><th align="left">Quick Quiz:</th></tr>
646 Given that multiple CPUs can start RCU read-side critical sections
647 at any time without any ordering whatsoever, how can RCU possibly
648 tell whether or not a given RCU read-side critical section starts
649 before a given instance of <tt>synchronize_rcu()</tt>?
651 <tr><th align="left">Answer:</th></tr>
652 <tr><td bgcolor="#ffffff"><font color="ffffff">
653 If RCU cannot tell whether or not a given
654 RCU read-side critical section starts before a
655 given instance of <tt>synchronize_rcu()</tt>,
656 then it must assume that the RCU read-side critical section
658 In other words, a given instance of <tt>synchronize_rcu()</tt>
659 can avoid waiting on a given RCU read-side critical section only
660 if it can prove that <tt>synchronize_rcu()</tt> started first.
663 <p><font color="ffffff">
664 A related question is “When <tt>rcu_read_lock()</tt>
665 doesn't generate any code, why does it matter how it relates
666 to a grace period?”
667 The answer is that it is not the relationship of
668 <tt>rcu_read_lock()</tt> itself that is important, but rather
669 the relationship of the code within the enclosed RCU read-side
670 critical section to the code preceding and following the
672 If we take this viewpoint, then a given RCU read-side critical
673 section begins before a given grace period when some access
674 preceding the grace period observes the effect of some access
675 within the critical section, in which case none of the accesses
676 within the critical section may observe the effects of any
677 access following the grace period.
680 <p><font color="ffffff">
681 As of late 2016, mathematical models of RCU take this
682 viewpoint, for example, see slides 62 and 63
684 <a href="http://www2.rdrop.com/users/paulmck/scalability/paper/LinuxMM.2016.10.04c.LCE.pdf">2016 LinuxCon EU</a>
687 <tr><td> </td></tr>
691 <tr><th> </th></tr>
692 <tr><th align="left">Quick Quiz:</th></tr>
694 The first and second guarantees require unbelievably strict ordering!
695 Are all these memory barriers <i> really</i> required?
697 <tr><th align="left">Answer:</th></tr>
698 <tr><td bgcolor="#ffffff"><font color="ffffff">
699 Yes, they really are required.
700 To see why the first guarantee is required, consider the following
705 <li> <font color="ffffff">
706 CPU 1: <tt>rcu_read_lock()</tt>
708 <li> <font color="ffffff">
709 CPU 1: <tt>q = rcu_dereference(gp);
710 /* Very likely to return p. */</tt>
712 <li> <font color="ffffff">
713 CPU 0: <tt>list_del_rcu(p);</tt>
715 <li> <font color="ffffff">
716 CPU 0: <tt>synchronize_rcu()</tt> starts.
718 <li> <font color="ffffff">
719 CPU 1: <tt>do_something_with(q->a);
720 /* No smp_mb(), so might happen after kfree(). */</tt>
722 <li> <font color="ffffff">
723 CPU 1: <tt>rcu_read_unlock()</tt>
725 <li> <font color="ffffff">
726 CPU 0: <tt>synchronize_rcu()</tt> returns.
728 <li> <font color="ffffff">
729 CPU 0: <tt>kfree(p);</tt>
733 <p><font color="ffffff">
734 Therefore, there absolutely must be a full memory barrier between the
735 end of the RCU read-side critical section and the end of the
739 <p><font color="ffffff">
740 The sequence of events demonstrating the necessity of the second rule
745 <li> <font color="ffffff">CPU 0: <tt>list_del_rcu(p);</tt>
747 <li> <font color="ffffff">CPU 0: <tt>synchronize_rcu()</tt> starts.
749 <li> <font color="ffffff">CPU 1: <tt>rcu_read_lock()</tt>
751 <li> <font color="ffffff">CPU 1: <tt>q = rcu_dereference(gp);
752 /* Might return p if no memory barrier. */</tt>
754 <li> <font color="ffffff">CPU 0: <tt>synchronize_rcu()</tt> returns.
756 <li> <font color="ffffff">CPU 0: <tt>kfree(p);</tt>
758 <li> <font color="ffffff">
759 CPU 1: <tt>do_something_with(q->a); /* Boom!!! */</tt>
761 <li> <font color="ffffff">CPU 1: <tt>rcu_read_unlock()</tt>
765 <p><font color="ffffff">
766 And similarly, without a memory barrier between the beginning of the
767 grace period and the beginning of the RCU read-side critical section,
768 CPU 1 might end up accessing the freelist.
771 <p><font color="ffffff">
772 The “as if” rule of course applies, so that any
773 implementation that acts as if the appropriate memory barriers
774 were in place is a correct implementation.
775 That said, it is much easier to fool yourself into believing
776 that you have adhered to the as-if rule than it is to actually
779 <tr><td> </td></tr>
783 <tr><th> </th></tr>
784 <tr><th align="left">Quick Quiz:</th></tr>
786 You claim that <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>
787 generate absolutely no code in some kernel builds.
788 This means that the compiler might arbitrarily rearrange consecutive
789 RCU read-side critical sections.
790 Given such rearrangement, if a given RCU read-side critical section
791 is done, how can you be sure that all prior RCU read-side critical
793 Won't the compiler rearrangements make that impossible to determine?
795 <tr><th align="left">Answer:</th></tr>
796 <tr><td bgcolor="#ffffff"><font color="ffffff">
797 In cases where <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>
798 generate absolutely no code, RCU infers quiescent states only at
799 special locations, for example, within the scheduler.
800 Because calls to <tt>schedule()</tt> had better prevent calling-code
801 accesses to shared variables from being rearranged across the call to
802 <tt>schedule()</tt>, if RCU detects the end of a given RCU read-side
803 critical section, it will necessarily detect the end of all prior
804 RCU read-side critical sections, no matter how aggressively the
805 compiler scrambles the code.
808 <p><font color="ffffff">
809 Again, this all assumes that the compiler cannot scramble code across
810 calls to the scheduler, out of interrupt handlers, into the idle loop,
811 into user-mode code, and so on.
812 But if your kernel build allows that sort of scrambling, you have broken
813 far more than just RCU!
815 <tr><td> </td></tr>
819 Note that these memory-barrier requirements do not replace the fundamental
820 RCU requirement that a grace period wait for all pre-existing readers.
821 On the contrary, the memory barriers called out in this section must operate in
822 such a way as to <i>enforce</i> this fundamental requirement.
823 Of course, different implementations enforce this requirement in different
824 ways, but enforce it they must.
826 <h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
829 The common-case RCU primitives are unconditional.
830 They are invoked, they do their job, and they return, with no possibility
831 of error, and no need to retry.
832 This is a key RCU design philosophy.
835 However, this philosophy is pragmatic rather than pigheaded.
836 If someone comes up with a good justification for a particular conditional
837 RCU primitive, it might well be implemented and added.
838 After all, this guarantee was reverse-engineered, not premeditated.
839 The unconditional nature of the RCU primitives was initially an
840 accident of implementation, and later experience with synchronization
841 primitives with conditional primitives caused me to elevate this
842 accident to a guarantee.
843 Therefore, the justification for adding a conditional primitive to
844 RCU would need to be based on detailed and compelling use cases.
846 <h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
849 As far as RCU is concerned, it is always possible to carry out an
850 update within an RCU read-side critical section.
851 For example, that RCU read-side critical section might search for
852 a given data element, and then might acquire the update-side
853 spinlock in order to update that element, all while remaining
854 in that RCU read-side critical section.
855 Of course, it is necessary to exit the RCU read-side critical section
856 before invoking <tt>synchronize_rcu()</tt>, however, this
857 inconvenience can be avoided through use of the
858 <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
859 described later in this document.
862 <tr><th> </th></tr>
863 <tr><th align="left">Quick Quiz:</th></tr>
865 But how does the upgrade-to-write operation exclude other readers?
867 <tr><th align="left">Answer:</th></tr>
868 <tr><td bgcolor="#ffffff"><font color="ffffff">
869 It doesn't, just like normal RCU updates, which also do not exclude
872 <tr><td> </td></tr>
876 This guarantee allows lookup code to be shared between read-side
877 and update-side code, and was premeditated, appearing in the earliest
878 DYNIX/ptx RCU documentation.
880 <h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
883 RCU provides extremely lightweight readers, and its read-side guarantees,
884 though quite useful, are correspondingly lightweight.
885 It is therefore all too easy to assume that RCU is guaranteeing more
887 Of course, the list of things that RCU does not guarantee is infinitely
888 long, however, the following sections list a few non-guarantees that
889 have caused confusion.
890 Except where otherwise noted, these non-guarantees were premeditated.
893 <li> <a href="#Readers Impose Minimal Ordering">
894 Readers Impose Minimal Ordering</a>
895 <li> <a href="#Readers Do Not Exclude Updaters">
896 Readers Do Not Exclude Updaters</a>
897 <li> <a href="#Updaters Only Wait For Old Readers">
898 Updaters Only Wait For Old Readers</a>
899 <li> <a href="#Grace Periods Don't Partition Read-Side Critical Sections">
900 Grace Periods Don't Partition Read-Side Critical Sections</a>
901 <li> <a href="#Read-Side Critical Sections Don't Partition Grace Periods">
902 Read-Side Critical Sections Don't Partition Grace Periods</a>
905 <h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
908 Reader-side markers such as <tt>rcu_read_lock()</tt> and
909 <tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
910 except through their interaction with the grace-period APIs such as
911 <tt>synchronize_rcu()</tt>.
912 To see this, consider the following pair of threads:
926 11 void thread1(void)
929 14 r1 = READ_ONCE(y);
930 15 rcu_read_unlock();
932 17 r2 = READ_ONCE(x);
933 18 rcu_read_unlock();
939 After <tt>thread0()</tt> and <tt>thread1()</tt> execute
940 concurrently, it is quite possible to have
944 (r1 == 1 && r2 == 0)
948 (that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
949 which would not be possible if <tt>rcu_read_lock()</tt> and
950 <tt>rcu_read_unlock()</tt> had much in the way of ordering
952 But they do not, so the CPU is within its rights
953 to do significant reordering.
954 This is by design: Any significant ordering constraints would slow down
955 these fast-path APIs.
958 <tr><th> </th></tr>
959 <tr><th align="left">Quick Quiz:</th></tr>
961 Can't the compiler also reorder this code?
963 <tr><th align="left">Answer:</th></tr>
964 <tr><td bgcolor="#ffffff"><font color="ffffff">
965 No, the volatile casts in <tt>READ_ONCE()</tt> and
966 <tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
967 this particular case.
969 <tr><td> </td></tr>
972 <h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
975 Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
977 All they do is to prevent grace periods from ending.
978 The following example illustrates this:
987 6 do_something_with_nonzero_x();
989 8 WARN_ON(!r2); /* BUG!!! */
991 10 rcu_read_unlock();
994 13 void thread1(void)
996 15 spin_lock(&my_lock);
999 18 spin_unlock(&my_lock);
1005 If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
1006 excluded the <tt>thread1()</tt> function's update,
1007 the <tt>WARN_ON()</tt> could never fire.
1008 But the fact is that <tt>rcu_read_lock()</tt> does not exclude
1009 much of anything aside from subsequent grace periods, of which
1010 <tt>thread1()</tt> has none, so the
1011 <tt>WARN_ON()</tt> can and does fire.
1013 <h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
1016 It might be tempting to assume that after <tt>synchronize_rcu()</tt>
1017 completes, there are no readers executing.
1018 This temptation must be avoided because
1019 new readers can start immediately after <tt>synchronize_rcu()</tt>
1020 starts, and <tt>synchronize_rcu()</tt> is under no
1021 obligation to wait for these new readers.
1024 <tr><th> </th></tr>
1025 <tr><th align="left">Quick Quiz:</th></tr>
1027 Suppose that synchronize_rcu() did wait until <i>all</i>
1028 readers had completed instead of waiting only on
1029 pre-existing readers.
1030 For how long would the updater be able to rely on there
1033 <tr><th align="left">Answer:</th></tr>
1034 <tr><td bgcolor="#ffffff"><font color="ffffff">
1036 Even if <tt>synchronize_rcu()</tt> were to wait until
1037 all readers had completed, a new reader might start immediately after
1038 <tt>synchronize_rcu()</tt> completed.
1039 Therefore, the code following
1040 <tt>synchronize_rcu()</tt> can <i>never</i> rely on there being
1043 <tr><td> </td></tr>
1046 <h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
1047 Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
1050 It is tempting to assume that if any part of one RCU read-side critical
1051 section precedes a given grace period, and if any part of another RCU
1052 read-side critical section follows that same grace period, then all of
1053 the first RCU read-side critical section must precede all of the second.
1054 However, this just isn't the case: A single grace period does not
1055 partition the set of RCU read-side critical sections.
1056 An example of this situation can be illustrated as follows, where
1057 <tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
1061 1 void thread0(void)
1066 6 rcu_read_unlock();
1069 9 void thread1(void)
1071 11 r1 = READ_ONCE(a);
1072 12 synchronize_rcu();
1073 13 WRITE_ONCE(c, 1);
1076 16 void thread2(void)
1079 19 r2 = READ_ONCE(b);
1080 20 r3 = READ_ONCE(c);
1081 21 rcu_read_unlock();
1087 It turns out that the outcome:
1091 (r1 == 1 && r2 == 0 && r3 == 1)
1095 is entirely possible.
1096 The following figure show how this can happen, with each circled
1097 <tt>QS</tt> indicating the point at which RCU recorded a
1098 <i>quiescent state</i> for each thread, that is, a state in which
1099 RCU knows that the thread cannot be in the midst of an RCU read-side
1100 critical section that started before the current grace period:
1102 <p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
1105 If it is necessary to partition RCU read-side critical sections in this
1106 manner, it is necessary to use two grace periods, where the first
1107 grace period is known to end before the second grace period starts:
1111 1 void thread0(void)
1116 6 rcu_read_unlock();
1119 9 void thread1(void)
1121 11 r1 = READ_ONCE(a);
1122 12 synchronize_rcu();
1123 13 WRITE_ONCE(c, 1);
1126 16 void thread2(void)
1128 18 r2 = READ_ONCE(c);
1129 19 synchronize_rcu();
1130 20 WRITE_ONCE(d, 1);
1133 23 void thread3(void)
1136 26 r3 = READ_ONCE(b);
1137 27 r4 = READ_ONCE(d);
1138 28 rcu_read_unlock();
1144 Here, if <tt>(r1 == 1)</tt>, then
1145 <tt>thread0()</tt>'s write to <tt>b</tt> must happen
1146 before the end of <tt>thread1()</tt>'s grace period.
1147 If in addition <tt>(r4 == 1)</tt>, then
1148 <tt>thread3()</tt>'s read from <tt>b</tt> must happen
1149 after the beginning of <tt>thread2()</tt>'s grace period.
1150 If it is also the case that <tt>(r2 == 1)</tt>, then the
1151 end of <tt>thread1()</tt>'s grace period must precede the
1152 beginning of <tt>thread2()</tt>'s grace period.
1153 This mean that the two RCU read-side critical sections cannot overlap,
1154 guaranteeing that <tt>(r3 == 1)</tt>.
1155 As a result, the outcome:
1159 (r1 == 1 && r2 == 1 && r3 == 0 && r4 == 1)
1166 This non-requirement was also non-premeditated, but became apparent
1167 when studying RCU's interaction with memory ordering.
1169 <h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
1170 Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
1173 It is also tempting to assume that if an RCU read-side critical section
1174 happens between a pair of grace periods, then those grace periods cannot
1176 However, this temptation leads nowhere good, as can be illustrated by
1177 the following, with all variables initially zero:
1181 1 void thread0(void)
1186 6 rcu_read_unlock();
1189 9 void thread1(void)
1191 11 r1 = READ_ONCE(a);
1192 12 synchronize_rcu();
1193 13 WRITE_ONCE(c, 1);
1196 16 void thread2(void)
1199 19 WRITE_ONCE(d, 1);
1200 20 r2 = READ_ONCE(c);
1201 21 rcu_read_unlock();
1204 24 void thread3(void)
1206 26 r3 = READ_ONCE(d);
1207 27 synchronize_rcu();
1208 28 WRITE_ONCE(e, 1);
1211 31 void thread4(void)
1214 34 r4 = READ_ONCE(b);
1215 35 r5 = READ_ONCE(e);
1216 36 rcu_read_unlock();
1222 In this case, the outcome:
1226 (r1 == 1 && r2 == 1 && r3 == 1 && r4 == 0 && r5 == 1)
1230 is entirely possible, as illustrated below:
1232 <p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
1235 Again, an RCU read-side critical section can overlap almost all of a
1236 given grace period, just so long as it does not overlap the entire
1238 As a result, an RCU read-side critical section cannot partition a pair
1239 of RCU grace periods.
1242 <tr><th> </th></tr>
1243 <tr><th align="left">Quick Quiz:</th></tr>
1245 How long a sequence of grace periods, each separated by an RCU
1246 read-side critical section, would be required to partition the RCU
1247 read-side critical sections at the beginning and end of the chain?
1249 <tr><th align="left">Answer:</th></tr>
1250 <tr><td bgcolor="#ffffff"><font color="ffffff">
1251 In theory, an infinite number.
1252 In practice, an unknown number that is sensitive to both implementation
1253 details and timing considerations.
1254 Therefore, even in practice, RCU users must abide by the
1255 theoretical rather than the practical answer.
1257 <tr><td> </td></tr>
1260 <h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
1263 These parallelism facts of life are by no means specific to RCU, but
1264 the RCU implementation must abide by them.
1265 They therefore bear repeating:
1268 <li> Any CPU or task may be delayed at any time,
1269 and any attempts to avoid these delays by disabling
1270 preemption, interrupts, or whatever are completely futile.
1271 This is most obvious in preemptible user-level
1272 environments and in virtualized environments (where
1273 a given guest OS's VCPUs can be preempted at any time by
1274 the underlying hypervisor), but can also happen in bare-metal
1275 environments due to ECC errors, NMIs, and other hardware
1277 Although a delay of more than about 20 seconds can result
1278 in splats, the RCU implementation is obligated to use
1279 algorithms that can tolerate extremely long delays, but where
1280 “extremely long” is not long enough to allow
1281 wrap-around when incrementing a 64-bit counter.
1282 <li> Both the compiler and the CPU can reorder memory accesses.
1283 Where it matters, RCU must use compiler directives and
1284 memory-barrier instructions to preserve ordering.
1285 <li> Conflicting writes to memory locations in any given cache line
1286 will result in expensive cache misses.
1287 Greater numbers of concurrent writes and more-frequent
1288 concurrent writes will result in more dramatic slowdowns.
1289 RCU is therefore obligated to use algorithms that have
1290 sufficient locality to avoid significant performance and
1291 scalability problems.
1292 <li> As a rough rule of thumb, only one CPU's worth of processing
1293 may be carried out under the protection of any given exclusive
1295 RCU must therefore use scalable locking designs.
1296 <li> Counters are finite, especially on 32-bit systems.
1297 RCU's use of counters must therefore tolerate counter wrap,
1298 or be designed such that counter wrap would take way more
1299 time than a single system is likely to run.
1300 An uptime of ten years is quite possible, a runtime
1301 of a century much less so.
1302 As an example of the latter, RCU's dyntick-idle nesting counter
1303 allows 54 bits for interrupt nesting level (this counter
1304 is 64 bits even on a 32-bit system).
1305 Overflowing this counter requires 2<sup>54</sup>
1306 half-interrupts on a given CPU without that CPU ever going idle.
1307 If a half-interrupt happened every microsecond, it would take
1308 570 years of runtime to overflow this counter, which is currently
1309 believed to be an acceptably long time.
1310 <li> Linux systems can have thousands of CPUs running a single
1311 Linux kernel in a single shared-memory environment.
1312 RCU must therefore pay close attention to high-end scalability.
1316 This last parallelism fact of life means that RCU must pay special
1317 attention to the preceding facts of life.
1318 The idea that Linux might scale to systems with thousands of CPUs would
1319 have been met with some skepticism in the 1990s, but these requirements
1320 would have otherwise have been unsurprising, even in the early 1990s.
1322 <h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
1325 These sections list quality-of-implementation requirements.
1326 Although an RCU implementation that ignores these requirements could
1327 still be used, it would likely be subject to limitations that would
1328 make it inappropriate for industrial-strength production use.
1329 Classes of quality-of-implementation requirements are as follows:
1332 <li> <a href="#Specialization">Specialization</a>
1333 <li> <a href="#Performance and Scalability">Performance and Scalability</a>
1334 <li> <a href="#Forward Progress">Forward Progress</a>
1335 <li> <a href="#Composability">Composability</a>
1336 <li> <a href="#Corner Cases">Corner Cases</a>
1340 These classes is covered in the following sections.
1342 <h3><a name="Specialization">Specialization</a></h3>
1345 RCU is and always has been intended primarily for read-mostly situations,
1346 which means that RCU's read-side primitives are optimized, often at the
1347 expense of its update-side primitives.
1348 Experience thus far is captured by the following list of situations:
1351 <li> Read-mostly data, where stale and inconsistent data is not
1352 a problem: RCU works great!
1353 <li> Read-mostly data, where data must be consistent:
1355 <li> Read-write data, where data must be consistent:
1356 RCU <i>might</i> work OK.
1358 <li> Write-mostly data, where data must be consistent:
1359 RCU is very unlikely to be the right tool for the job,
1360 with the following exceptions, where RCU can provide:
1362 <li> Existence guarantees for update-friendly mechanisms.
1363 <li> Wait-free read-side primitives for real-time use.
1368 This focus on read-mostly situations means that RCU must interoperate
1369 with other synchronization primitives.
1370 For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
1371 examples discussed earlier use RCU to protect readers and locking to
1372 coordinate updaters.
1373 However, the need extends much farther, requiring that a variety of
1374 synchronization primitives be legal within RCU read-side critical sections,
1375 including spinlocks, sequence locks, atomic operations, reference
1376 counters, and memory barriers.
1379 <tr><th> </th></tr>
1380 <tr><th align="left">Quick Quiz:</th></tr>
1382 What about sleeping locks?
1384 <tr><th align="left">Answer:</th></tr>
1385 <tr><td bgcolor="#ffffff"><font color="ffffff">
1386 These are forbidden within Linux-kernel RCU read-side critical
1387 sections because it is not legal to place a quiescent state
1388 (in this case, voluntary context switch) within an RCU read-side
1390 However, sleeping locks may be used within userspace RCU read-side
1391 critical sections, and also within Linux-kernel sleepable RCU
1392 <a href="#Sleepable RCU"><font color="ffffff">(SRCU)</font></a>
1393 read-side critical sections.
1394 In addition, the -rt patchset turns spinlocks into a
1395 sleeping locks so that the corresponding critical sections
1396 can be preempted, which also means that these sleeplockified
1397 spinlocks (but not other sleeping locks!) may be acquire within
1398 -rt-Linux-kernel RCU read-side critical sections.
1401 <p><font color="ffffff">
1402 Note that it <i>is</i> legal for a normal RCU read-side
1403 critical section to conditionally acquire a sleeping locks
1404 (as in <tt>mutex_trylock()</tt>), but only as long as it does
1405 not loop indefinitely attempting to conditionally acquire that
1407 The key point is that things like <tt>mutex_trylock()</tt>
1408 either return with the mutex held, or return an error indication if
1409 the mutex was not immediately available.
1410 Either way, <tt>mutex_trylock()</tt> returns immediately without
1413 <tr><td> </td></tr>
1417 It often comes as a surprise that many algorithms do not require a
1418 consistent view of data, but many can function in that mode,
1419 with network routing being the poster child.
1420 Internet routing algorithms take significant time to propagate
1421 updates, so that by the time an update arrives at a given system,
1422 that system has been sending network traffic the wrong way for
1423 a considerable length of time.
1424 Having a few threads continue to send traffic the wrong way for a
1425 few more milliseconds is clearly not a problem: In the worst case,
1426 TCP retransmissions will eventually get the data where it needs to go.
1427 In general, when tracking the state of the universe outside of the
1428 computer, some level of inconsistency must be tolerated due to
1429 speed-of-light delays if nothing else.
1432 Furthermore, uncertainty about external state is inherent in many cases.
1433 For example, a pair of veterinarians might use heartbeat to determine
1434 whether or not a given cat was alive.
1435 But how long should they wait after the last heartbeat to decide that
1436 the cat is in fact dead?
1437 Waiting less than 400 milliseconds makes no sense because this would
1438 mean that a relaxed cat would be considered to cycle between death
1439 and life more than 100 times per minute.
1440 Moreover, just as with human beings, a cat's heart might stop for
1441 some period of time, so the exact wait period is a judgment call.
1442 One of our pair of veterinarians might wait 30 seconds before pronouncing
1443 the cat dead, while the other might insist on waiting a full minute.
1444 The two veterinarians would then disagree on the state of the cat during
1445 the final 30 seconds of the minute following the last heartbeat.
1448 Interestingly enough, this same situation applies to hardware.
1449 When push comes to shove, how do we tell whether or not some
1450 external server has failed?
1451 We send messages to it periodically, and declare it failed if we
1452 don't receive a response within a given period of time.
1453 Policy decisions can usually tolerate short
1454 periods of inconsistency.
1455 The policy was decided some time ago, and is only now being put into
1456 effect, so a few milliseconds of delay is normally inconsequential.
1459 However, there are algorithms that absolutely must see consistent data.
1460 For example, the translation between a user-level SystemV semaphore
1461 ID to the corresponding in-kernel data structure is protected by RCU,
1462 but it is absolutely forbidden to update a semaphore that has just been
1464 In the Linux kernel, this need for consistency is accommodated by acquiring
1465 spinlocks located in the in-kernel data structure from within
1466 the RCU read-side critical section, and this is indicated by the
1467 green box in the figure above.
1468 Many other techniques may be used, and are in fact used within the
1472 In short, RCU is not required to maintain consistency, and other
1473 mechanisms may be used in concert with RCU when consistency is required.
1474 RCU's specialization allows it to do its job extremely well, and its
1475 ability to interoperate with other synchronization mechanisms allows
1476 the right mix of synchronization tools to be used for a given job.
1478 <h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
1481 Energy efficiency is a critical component of performance today,
1482 and Linux-kernel RCU implementations must therefore avoid unnecessarily
1483 awakening idle CPUs.
1484 I cannot claim that this requirement was premeditated.
1485 In fact, I learned of it during a telephone conversation in which I
1486 was given “frank and open” feedback on the importance
1487 of energy efficiency in battery-powered systems and on specific
1488 energy-efficiency shortcomings of the Linux-kernel RCU implementation.
1489 In my experience, the battery-powered embedded community will consider
1490 any unnecessary wakeups to be extremely unfriendly acts.
1491 So much so that mere Linux-kernel-mailing-list posts are
1492 insufficient to vent their ire.
1495 Memory consumption is not particularly important for in most
1496 situations, and has become decreasingly
1497 so as memory sizes have expanded and memory
1498 costs have plummeted.
1499 However, as I learned from Matt Mackall's
1500 <a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
1501 efforts, memory footprint is critically important on single-CPU systems with
1502 non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
1503 <a href="https://lkml.kernel.org/g/20090113221724.GA15307@linux.vnet.ibm.com">tiny RCU</a>
1505 Josh Triplett has since taken over the small-memory banner with his
1506 <a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
1507 project, which resulted in
1508 <a href="#Sleepable RCU">SRCU</a>
1509 becoming optional for those kernels not needing it.
1512 The remaining performance requirements are, for the most part,
1514 For example, in keeping with RCU's read-side specialization,
1515 <tt>rcu_dereference()</tt> should have negligible overhead (for
1516 example, suppression of a few minor compiler optimizations).
1517 Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
1518 <tt>rcu_read_unlock()</tt> should have exactly zero overhead.
1521 In preemptible environments, in the case where the RCU read-side
1522 critical section was not preempted (as will be the case for the
1523 highest-priority real-time process), <tt>rcu_read_lock()</tt> and
1524 <tt>rcu_read_unlock()</tt> should have minimal overhead.
1525 In particular, they should not contain atomic read-modify-write
1526 operations, memory-barrier instructions, preemption disabling,
1527 interrupt disabling, or backwards branches.
1528 However, in the case where the RCU read-side critical section was preempted,
1529 <tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
1530 This is why it is better to nest an RCU read-side critical section
1531 within a preempt-disable region than vice versa, at least in cases
1532 where that critical section is short enough to avoid unduly degrading
1533 real-time latencies.
1536 The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
1537 optimized for throughput.
1538 It may therefore incur several milliseconds of latency in addition to
1539 the duration of the longest RCU read-side critical section.
1540 On the other hand, multiple concurrent invocations of
1541 <tt>synchronize_rcu()</tt> are required to use batching optimizations
1542 so that they can be satisfied by a single underlying grace-period-wait
1544 For example, in the Linux kernel, it is not unusual for a single
1545 grace-period-wait operation to serve more than
1546 <a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
1547 of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
1548 overhead down to nearly zero.
1549 However, the grace-period optimization is also required to avoid
1550 measurable degradation of real-time scheduling and interrupt latencies.
1553 In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
1554 latencies are unacceptable.
1555 In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
1556 instead, reducing the grace-period latency down to a few tens of
1557 microseconds on small systems, at least in cases where the RCU read-side
1558 critical sections are short.
1559 There are currently no special latency requirements for
1560 <tt>synchronize_rcu_expedited()</tt> on large systems, but,
1561 consistent with the empirical nature of the RCU specification,
1562 that is subject to change.
1563 However, there most definitely are scalability requirements:
1564 A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
1565 CPUs should at least make reasonable forward progress.
1566 In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
1567 is permitted to impose modest degradation of real-time latency
1568 on non-idle online CPUs.
1569 Here, “modest” means roughly the same latency
1570 degradation as a scheduling-clock interrupt.
1573 There are a number of situations where even
1574 <tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
1575 latency is unacceptable.
1576 In these situations, the asynchronous <tt>call_rcu()</tt> can be
1577 used in place of <tt>synchronize_rcu()</tt> as follows:
1584 4 struct rcu_head rh;
1587 7 static void remove_gp_cb(struct rcu_head *rhp)
1589 9 struct foo *p = container_of(rhp, struct foo, rh);
1594 14 bool remove_gp_asynchronous(void)
1598 18 spin_lock(&gp_lock);
1599 19 p = rcu_access_pointer(gp);
1601 21 spin_unlock(&gp_lock);
1604 24 rcu_assign_pointer(gp, NULL);
1605 25 call_rcu(&p->rh, remove_gp_cb);
1606 26 spin_unlock(&gp_lock);
1613 A definition of <tt>struct foo</tt> is finally needed, and appears
1615 The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
1616 on line 25, and will be invoked after the end of a subsequent
1618 This gets the same effect as <tt>remove_gp_synchronous()</tt>,
1619 but without forcing the updater to wait for a grace period to elapse.
1620 The <tt>call_rcu()</tt> function may be used in a number of
1621 situations where neither <tt>synchronize_rcu()</tt> nor
1622 <tt>synchronize_rcu_expedited()</tt> would be legal,
1623 including within preempt-disable code, <tt>local_bh_disable()</tt> code,
1624 interrupt-disable code, and interrupt handlers.
1625 However, even <tt>call_rcu()</tt> is illegal within NMI handlers
1626 and from idle and offline CPUs.
1627 The callback function (<tt>remove_gp_cb()</tt> in this case) will be
1628 executed within softirq (software interrupt) environment within the
1630 either within a real softirq handler or under the protection
1631 of <tt>local_bh_disable()</tt>.
1632 In both the Linux kernel and in userspace, it is bad practice to
1633 write an RCU callback function that takes too long.
1634 Long-running operations should be relegated to separate threads or
1635 (in the Linux kernel) workqueues.
1638 <tr><th> </th></tr>
1639 <tr><th align="left">Quick Quiz:</th></tr>
1641 Why does line 19 use <tt>rcu_access_pointer()</tt>?
1642 After all, <tt>call_rcu()</tt> on line 25 stores into the
1643 structure, which would interact badly with concurrent insertions.
1644 Doesn't this mean that <tt>rcu_dereference()</tt> is required?
1646 <tr><th align="left">Answer:</th></tr>
1647 <tr><td bgcolor="#ffffff"><font color="ffffff">
1648 Presumably the <tt>->gp_lock</tt> acquired on line 18 excludes
1649 any changes, including any insertions that <tt>rcu_dereference()</tt>
1650 would protect against.
1651 Therefore, any insertions will be delayed until after
1652 <tt>->gp_lock</tt>
1653 is released on line 25, which in turn means that
1654 <tt>rcu_access_pointer()</tt> suffices.
1656 <tr><td> </td></tr>
1660 However, all that <tt>remove_gp_cb()</tt> is doing is
1661 invoking <tt>kfree()</tt> on the data element.
1662 This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
1663 which allows “fire and forget” operation as shown below:
1670 4 struct rcu_head rh;
1673 7 bool remove_gp_faf(void)
1677 11 spin_lock(&gp_lock);
1678 12 p = rcu_dereference(gp);
1680 14 spin_unlock(&gp_lock);
1683 17 rcu_assign_pointer(gp, NULL);
1684 18 kfree_rcu(p, rh);
1685 19 spin_unlock(&gp_lock);
1692 Note that <tt>remove_gp_faf()</tt> simply invokes
1693 <tt>kfree_rcu()</tt> and proceeds, without any need to pay any
1694 further attention to the subsequent grace period and <tt>kfree()</tt>.
1695 It is permissible to invoke <tt>kfree_rcu()</tt> from the same
1696 environments as for <tt>call_rcu()</tt>.
1697 Interestingly enough, DYNIX/ptx had the equivalents of
1698 <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
1699 <tt>synchronize_rcu()</tt>.
1700 This was due to the fact that RCU was not heavily used within DYNIX/ptx,
1701 so the very few places that needed something like
1702 <tt>synchronize_rcu()</tt> simply open-coded it.
1705 <tr><th> </th></tr>
1706 <tr><th align="left">Quick Quiz:</th></tr>
1708 Earlier it was claimed that <tt>call_rcu()</tt> and
1709 <tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
1711 But how can that be correct, given that the invocation of the callback
1712 and the freeing of the memory (respectively) must still wait for
1713 a grace period to elapse?
1715 <tr><th align="left">Answer:</th></tr>
1716 <tr><td bgcolor="#ffffff"><font color="ffffff">
1717 We could define things this way, but keep in mind that this sort of
1718 definition would say that updates in garbage-collected languages
1719 cannot complete until the next time the garbage collector runs,
1720 which does not seem at all reasonable.
1721 The key point is that in most cases, an updater using either
1722 <tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
1723 next update as soon as it has invoked <tt>call_rcu()</tt> or
1724 <tt>kfree_rcu()</tt>, without having to wait for a subsequent
1727 <tr><td> </td></tr>
1731 But what if the updater must wait for the completion of code to be
1732 executed after the end of the grace period, but has other tasks
1733 that can be carried out in the meantime?
1734 The polling-style <tt>get_state_synchronize_rcu()</tt> and
1735 <tt>cond_synchronize_rcu()</tt> functions may be used for this
1736 purpose, as shown below:
1740 1 bool remove_gp_poll(void)
1745 6 spin_lock(&gp_lock);
1746 7 p = rcu_access_pointer(gp);
1748 9 spin_unlock(&gp_lock);
1751 12 rcu_assign_pointer(gp, NULL);
1752 13 spin_unlock(&gp_lock);
1753 14 s = get_state_synchronize_rcu();
1754 15 do_something_while_waiting();
1755 16 cond_synchronize_rcu(s);
1763 On line 14, <tt>get_state_synchronize_rcu()</tt> obtains a
1764 “cookie” from RCU,
1765 then line 15 carries out other tasks,
1766 and finally, line 16 returns immediately if a grace period has
1767 elapsed in the meantime, but otherwise waits as required.
1768 The need for <tt>get_state_synchronize_rcu</tt> and
1769 <tt>cond_synchronize_rcu()</tt> has appeared quite recently,
1770 so it is too early to tell whether they will stand the test of time.
1773 RCU thus provides a range of tools to allow updaters to strike the
1774 required tradeoff between latency, flexibility and CPU overhead.
1776 <h3><a name="Forward Progress">Forward Progress</a></h3>
1779 In theory, delaying grace-period completion and callback invocation
1781 In practice, not only are memory sizes finite but also callbacks sometimes
1782 do wakeups, and sufficiently deferred wakeups can be difficult
1783 to distinguish from system hangs.
1784 Therefore, RCU must provide a number of mechanisms to promote forward
1788 These mechanisms are not foolproof, nor can they be.
1789 For one simple example, an infinite loop in an RCU read-side critical
1790 section must by definition prevent later grace periods from ever completing.
1791 For a more involved example, consider a 64-CPU system built with
1792 <tt>CONFIG_RCU_NOCB_CPU=y</tt> and booted with <tt>rcu_nocbs=1-63</tt>,
1793 where CPUs 1 through 63 spin in tight loops that invoke
1794 <tt>call_rcu()</tt>.
1795 Even if these tight loops also contain calls to <tt>cond_resched()</tt>
1796 (thus allowing grace periods to complete), CPU 0 simply will
1797 not be able to invoke callbacks as fast as the other 63 CPUs can
1798 register them, at least not until the system runs out of memory.
1799 In both of these examples, the Spiderman principle applies: With great
1800 power comes great responsibility.
1801 However, short of this level of abuse, RCU is required to
1802 ensure timely completion of grace periods and timely invocation of
1806 RCU takes the following steps to encourage timely completion of
1810 <li> If a grace period fails to complete within 100 milliseconds,
1811 RCU causes future invocations of <tt>cond_resched()</tt> on
1812 the holdout CPUs to provide an RCU quiescent state.
1813 RCU also causes those CPUs' <tt>need_resched()</tt> invocations
1814 to return <tt>true</tt>, but only after the corresponding CPU's
1815 next scheduling-clock.
1816 <li> CPUs mentioned in the <tt>nohz_full</tt> kernel boot parameter
1817 can run indefinitely in the kernel without scheduling-clock
1818 interrupts, which defeats the above <tt>need_resched()</tt>
1820 RCU will therefore invoke <tt>resched_cpu()</tt> on any
1821 <tt>nohz_full</tt> CPUs still holding out after
1822 109 milliseconds.
1823 <li> In kernels built with <tt>CONFIG_RCU_BOOST=y</tt>, if a given
1824 task that has been preempted within an RCU read-side critical
1825 section is holding out for more than 500 milliseconds,
1826 RCU will resort to priority boosting.
1827 <li> If a CPU is still holding out 10 seconds into the grace
1828 period, RCU will invoke <tt>resched_cpu()</tt> on it regardless
1829 of its <tt>nohz_full</tt> state.
1833 The above values are defaults for systems running with <tt>HZ=1000</tt>.
1834 They will vary as the value of <tt>HZ</tt> varies, and can also be
1835 changed using the relevant Kconfig options and kernel boot parameters.
1836 RCU currently does not do much sanity checking of these
1837 parameters, so please use caution when changing them.
1838 Note that these forward-progress measures are provided only for RCU,
1840 <a href="#Sleepable RCU">SRCU</a> or
1841 <a href="#Tasks RCU">Tasks RCU</a>.
1844 RCU takes the following steps in <tt>call_rcu()</tt> to encourage timely
1845 invocation of callbacks when any given non-<tt>rcu_nocbs</tt> CPU has
1846 10,000 callbacks, or has 10,000 more callbacks than it had the last time
1847 encouragement was provided:
1850 <li> Starts a grace period, if one is not already in progress.
1851 <li> Forces immediate checking for quiescent states, rather than
1852 waiting for three milliseconds to have elapsed since the
1853 beginning of the grace period.
1854 <li> Immediately tags the CPU's callbacks with their grace period
1855 completion numbers, rather than waiting for the <tt>RCU_SOFTIRQ</tt>
1856 handler to get around to it.
1857 <li> Lifts callback-execution batch limits, which speeds up callback
1858 invocation at the expense of degrading realtime response.
1862 Again, these are default values when running at <tt>HZ=1000</tt>,
1863 and can be overridden.
1864 Again, these forward-progress measures are provided only for RCU,
1866 <a href="#Sleepable RCU">SRCU</a> or
1867 <a href="#Tasks RCU">Tasks RCU</a>.
1868 Even for RCU, callback-invocation forward progress for <tt>rcu_nocbs</tt>
1869 CPUs is much less well-developed, in part because workloads benefiting
1870 from <tt>rcu_nocbs</tt> CPUs tend to invoke <tt>call_rcu()</tt>
1871 relatively infrequently.
1872 If workloads emerge that need both <tt>rcu_nocbs</tt> CPUs and high
1873 <tt>call_rcu()</tt> invocation rates, then additional forward-progress
1874 work will be required.
1876 <h3><a name="Composability">Composability</a></h3>
1879 Composability has received much attention in recent years, perhaps in part
1880 due to the collision of multicore hardware with object-oriented techniques
1881 designed in single-threaded environments for single-threaded use.
1882 And in theory, RCU read-side critical sections may be composed, and in
1883 fact may be nested arbitrarily deeply.
1884 In practice, as with all real-world implementations of composable
1885 constructs, there are limitations.
1888 Implementations of RCU for which <tt>rcu_read_lock()</tt>
1889 and <tt>rcu_read_unlock()</tt> generate no code, such as
1890 Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
1891 nested arbitrarily deeply.
1892 After all, there is no overhead.
1893 Except that if all these instances of <tt>rcu_read_lock()</tt>
1894 and <tt>rcu_read_unlock()</tt> are visible to the compiler,
1895 compilation will eventually fail due to exhausting memory,
1896 mass storage, or user patience, whichever comes first.
1897 If the nesting is not visible to the compiler, as is the case with
1898 mutually recursive functions each in its own translation unit,
1899 stack overflow will result.
1900 If the nesting takes the form of loops, perhaps in the guise of tail
1901 recursion, either the control variable
1902 will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
1903 Nevertheless, this class of RCU implementations is one
1904 of the most composable constructs in existence.
1907 RCU implementations that explicitly track nesting depth
1908 are limited by the nesting-depth counter.
1909 For example, the Linux kernel's preemptible RCU limits nesting to
1911 This should suffice for almost all practical purposes.
1912 That said, a consecutive pair of RCU read-side critical sections
1913 between which there is an operation that waits for a grace period
1914 cannot be enclosed in another RCU read-side critical section.
1915 This is because it is not legal to wait for a grace period within
1916 an RCU read-side critical section: To do so would result either
1918 in RCU implicitly splitting the enclosing RCU read-side critical
1919 section, neither of which is conducive to a long-lived and prosperous
1923 It is worth noting that RCU is not alone in limiting composability.
1924 For example, many transactional-memory implementations prohibit
1925 composing a pair of transactions separated by an irrevocable
1926 operation (for example, a network receive operation).
1927 For another example, lock-based critical sections can be composed
1928 surprisingly freely, but only if deadlock is avoided.
1931 In short, although RCU read-side critical sections are highly composable,
1932 care is required in some situations, just as is the case for any other
1933 composable synchronization mechanism.
1935 <h3><a name="Corner Cases">Corner Cases</a></h3>
1938 A given RCU workload might have an endless and intense stream of
1939 RCU read-side critical sections, perhaps even so intense that there
1940 was never a point in time during which there was not at least one
1941 RCU read-side critical section in flight.
1942 RCU cannot allow this situation to block grace periods: As long as
1943 all the RCU read-side critical sections are finite, grace periods
1944 must also be finite.
1947 That said, preemptible RCU implementations could potentially result
1948 in RCU read-side critical sections being preempted for long durations,
1949 which has the effect of creating a long-duration RCU read-side
1951 This situation can arise only in heavily loaded systems, but systems using
1952 real-time priorities are of course more vulnerable.
1953 Therefore, RCU priority boosting is provided to help deal with this
1955 That said, the exact requirements on RCU priority boosting will likely
1956 evolve as more experience accumulates.
1959 Other workloads might have very high update rates.
1960 Although one can argue that such workloads should instead use
1961 something other than RCU, the fact remains that RCU must
1962 handle such workloads gracefully.
1963 This requirement is another factor driving batching of grace periods,
1964 but it is also the driving force behind the checks for large numbers
1965 of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
1966 Finally, high update rates should not delay RCU read-side critical
1967 sections, although some small read-side delays can occur when using
1968 <tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
1969 of <tt>smp_call_function_single()</tt>.
1972 Although all three of these corner cases were understood in the early
1973 1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
1975 in the early 2000s suddenly provided a much deeper appreciation of the
1976 high-update-rate corner case.
1977 This test also motivated addition of some RCU code to react to high update
1978 rates, for example, if a given CPU finds itself with more than 10,000
1979 RCU callbacks queued, it will cause RCU to take evasive action by
1980 more aggressively starting grace periods and more aggressively forcing
1981 completion of grace-period processing.
1982 This evasive action causes the grace period to complete more quickly,
1983 but at the cost of restricting RCU's batching optimizations, thus
1984 increasing the CPU overhead incurred by that grace period.
1986 <h2><a name="Software-Engineering Requirements">
1987 Software-Engineering Requirements</a></h2>
1990 Between Murphy's Law and “To err is human”, it is necessary to
1991 guard against mishaps and misuse:
1994 <li> It is all too easy to forget to use <tt>rcu_read_lock()</tt>
1995 everywhere that it is needed, so kernels built with
1996 <tt>CONFIG_PROVE_RCU=y</tt> will splat if
1997 <tt>rcu_dereference()</tt> is used outside of an
1998 RCU read-side critical section.
1999 Update-side code can use <tt>rcu_dereference_protected()</tt>,
2001 <a href="https://lwn.net/Articles/371986/">lockdep expression</a>
2002 to indicate what is providing the protection.
2003 If the indicated protection is not provided, a lockdep splat
2007 Code shared between readers and updaters can use
2008 <tt>rcu_dereference_check()</tt>, which also takes a
2009 lockdep expression, and emits a lockdep splat if neither
2010 <tt>rcu_read_lock()</tt> nor the indicated protection
2012 In addition, <tt>rcu_dereference_raw()</tt> is used in those
2013 (hopefully rare) cases where the required protection cannot
2014 be easily described.
2015 Finally, <tt>rcu_read_lock_held()</tt> is provided to
2016 allow a function to verify that it has been invoked within
2017 an RCU read-side critical section.
2018 I was made aware of this set of requirements shortly after Thomas
2019 Gleixner audited a number of RCU uses.
2020 <li> A given function might wish to check for RCU-related preconditions
2021 upon entry, before using any other RCU API.
2022 The <tt>rcu_lockdep_assert()</tt> does this job,
2023 asserting the expression in kernels having lockdep enabled
2024 and doing nothing otherwise.
2025 <li> It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
2026 and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
2027 substituting a simple assignment.
2028 To catch this sort of error, a given RCU-protected pointer may be
2029 tagged with <tt>__rcu</tt>, after which sparse
2030 will complain about simple-assignment accesses to that pointer.
2031 Arnd Bergmann made me aware of this requirement, and also
2033 <a href="https://lwn.net/Articles/376011/">patch series</a>.
2034 <li> Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
2035 will splat if a data element is passed to <tt>call_rcu()</tt>
2036 twice in a row, without a grace period in between.
2037 (This error is similar to a double free.)
2038 The corresponding <tt>rcu_head</tt> structures that are
2039 dynamically allocated are automatically tracked, but
2040 <tt>rcu_head</tt> structures allocated on the stack
2041 must be initialized with <tt>init_rcu_head_on_stack()</tt>
2042 and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
2043 Similarly, statically allocated non-stack <tt>rcu_head</tt>
2044 structures must be initialized with <tt>init_rcu_head()</tt>
2045 and cleaned up with <tt>destroy_rcu_head()</tt>.
2046 Mathieu Desnoyers made me aware of this requirement, and also
2048 <a href="https://lkml.kernel.org/g/20100319013024.GA28456@Krystal">patch</a>.
2049 <li> An infinite loop in an RCU read-side critical section will
2050 eventually trigger an RCU CPU stall warning splat, with
2051 the duration of “eventually” being controlled by the
2052 <tt>RCU_CPU_STALL_TIMEOUT</tt> <tt>Kconfig</tt> option, or,
2053 alternatively, by the
2054 <tt>rcupdate.rcu_cpu_stall_timeout</tt> boot/sysfs
2056 However, RCU is not obligated to produce this splat
2057 unless there is a grace period waiting on that particular
2058 RCU read-side critical section.
2060 Some extreme workloads might intentionally delay
2061 RCU grace periods, and systems running those workloads can
2062 be booted with <tt>rcupdate.rcu_cpu_stall_suppress</tt>
2063 to suppress the splats.
2064 This kernel parameter may also be set via <tt>sysfs</tt>.
2065 Furthermore, RCU CPU stall warnings are counter-productive
2066 during sysrq dumps and during panics.
2067 RCU therefore supplies the <tt>rcu_sysrq_start()</tt> and
2068 <tt>rcu_sysrq_end()</tt> API members to be called before
2069 and after long sysrq dumps.
2070 RCU also supplies the <tt>rcu_panic()</tt> notifier that is
2071 automatically invoked at the beginning of a panic to suppress
2072 further RCU CPU stall warnings.
2075 This requirement made itself known in the early 1990s, pretty
2076 much the first time that it was necessary to debug a CPU stall.
2077 That said, the initial implementation in DYNIX/ptx was quite
2078 generic in comparison with that of Linux.
2079 <li> Although it would be very good to detect pointers leaking out
2080 of RCU read-side critical sections, there is currently no
2081 good way of doing this.
2082 One complication is the need to distinguish between pointers
2083 leaking and pointers that have been handed off from RCU to
2084 some other synchronization mechanism, for example, reference
2086 <li> In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
2087 information is provided via event tracing.
2088 <li> Open-coded use of <tt>rcu_assign_pointer()</tt> and
2089 <tt>rcu_dereference()</tt> to create typical linked
2090 data structures can be surprisingly error-prone.
2091 Therefore, RCU-protected
2092 <a href="https://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
2093 and, more recently, RCU-protected
2094 <a href="https://lwn.net/Articles/612100/">hash tables</a>
2096 Many other special-purpose RCU-protected data structures are
2097 available in the Linux kernel and the userspace RCU library.
2098 <li> Some linked structures are created at compile time, but still
2099 require <tt>__rcu</tt> checking.
2100 The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
2102 <li> It is not necessary to use <tt>rcu_assign_pointer()</tt>
2103 when creating linked structures that are to be published via
2104 a single external pointer.
2105 The <tt>RCU_INIT_POINTER()</tt> macro is provided for
2106 this task and also for assigning <tt>NULL</tt> pointers
2111 This not a hard-and-fast list: RCU's diagnostic capabilities will
2112 continue to be guided by the number and type of usage bugs found
2113 in real-world RCU usage.
2115 <h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
2118 The Linux kernel provides an interesting environment for all kinds of
2119 software, including RCU.
2120 Some of the relevant points of interest are as follows:
2123 <li> <a href="#Configuration">Configuration</a>.
2124 <li> <a href="#Firmware Interface">Firmware Interface</a>.
2125 <li> <a href="#Early Boot">Early Boot</a>.
2126 <li> <a href="#Interrupts and NMIs">
2127 Interrupts and non-maskable interrupts (NMIs)</a>.
2128 <li> <a href="#Loadable Modules">Loadable Modules</a>.
2129 <li> <a href="#Hotplug CPU">Hotplug CPU</a>.
2130 <li> <a href="#Scheduler and RCU">Scheduler and RCU</a>.
2131 <li> <a href="#Tracing and RCU">Tracing and RCU</a>.
2132 <li> <a href="#Energy Efficiency">Energy Efficiency</a>.
2133 <li> <a href="#Scheduling-Clock Interrupts and RCU">
2134 Scheduling-Clock Interrupts and RCU</a>.
2135 <li> <a href="#Memory Efficiency">Memory Efficiency</a>.
2136 <li> <a href="#Performance, Scalability, Response Time, and Reliability">
2137 Performance, Scalability, Response Time, and Reliability</a>.
2141 This list is probably incomplete, but it does give a feel for the
2142 most notable Linux-kernel complications.
2143 Each of the following sections covers one of the above topics.
2145 <h3><a name="Configuration">Configuration</a></h3>
2148 RCU's goal is automatic configuration, so that almost nobody
2149 needs to worry about RCU's <tt>Kconfig</tt> options.
2150 And for almost all users, RCU does in fact work well
2151 “out of the box.”
2154 However, there are specialized use cases that are handled by
2155 kernel boot parameters and <tt>Kconfig</tt> options.
2156 Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
2157 about new <tt>Kconfig</tt> options, which requires almost all of them
2158 be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
2161 This all should be quite obvious, but the fact remains that
2162 Linus Torvalds recently had to
2163 <a href="https://lkml.kernel.org/g/CA+55aFy4wcCwaL4okTs8wXhGZ5h-ibecy_Meg9C4MNQrUnwMcg@mail.gmail.com">remind</a>
2164 me of this requirement.
2166 <h3><a name="Firmware Interface">Firmware Interface</a></h3>
2169 In many cases, kernel obtains information about the system from the
2170 firmware, and sometimes things are lost in translation.
2171 Or the translation is accurate, but the original message is bogus.
2174 For example, some systems' firmware overreports the number of CPUs,
2175 sometimes by a large factor.
2176 If RCU naively believed the firmware, as it used to do,
2177 it would create too many per-CPU kthreads.
2178 Although the resulting system will still run correctly, the extra
2179 kthreads needlessly consume memory and can cause confusion
2180 when they show up in <tt>ps</tt> listings.
2183 RCU must therefore wait for a given CPU to actually come online before
2184 it can allow itself to believe that the CPU actually exists.
2185 The resulting “ghost CPUs” (which are never going to
2186 come online) cause a number of
2187 <a href="https://paulmck.livejournal.com/37494.html">interesting complications</a>.
2189 <h3><a name="Early Boot">Early Boot</a></h3>
2192 The Linux kernel's boot sequence is an interesting process,
2193 and RCU is used early, even before <tt>rcu_init()</tt>
2195 In fact, a number of RCU's primitives can be used as soon as the
2196 initial task's <tt>task_struct</tt> is available and the
2197 boot CPU's per-CPU variables are set up.
2198 The read-side primitives (<tt>rcu_read_lock()</tt>,
2199 <tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
2200 and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
2201 as will <tt>rcu_assign_pointer()</tt>.
2204 Although <tt>call_rcu()</tt> may be invoked at any
2205 time during boot, callbacks are not guaranteed to be invoked until after
2206 all of RCU's kthreads have been spawned, which occurs at
2207 <tt>early_initcall()</tt> time.
2208 This delay in callback invocation is due to the fact that RCU does not
2209 invoke callbacks until it is fully initialized, and this full initialization
2210 cannot occur until after the scheduler has initialized itself to the
2211 point where RCU can spawn and run its kthreads.
2212 In theory, it would be possible to invoke callbacks earlier,
2213 however, this is not a panacea because there would be severe restrictions
2214 on what operations those callbacks could invoke.
2217 Perhaps surprisingly, <tt>synchronize_rcu()</tt> and
2218 <tt>synchronize_rcu_expedited()</tt>,
2219 will operate normally
2220 during very early boot, the reason being that there is only one CPU
2221 and preemption is disabled.
2222 This means that the call <tt>synchronize_rcu()</tt> (or friends)
2223 itself is a quiescent
2224 state and thus a grace period, so the early-boot implementation can
2228 However, once the scheduler has spawned its first kthread, this early
2229 boot trick fails for <tt>synchronize_rcu()</tt> (as well as for
2230 <tt>synchronize_rcu_expedited()</tt>) in <tt>CONFIG_PREEMPT=y</tt>
2232 The reason is that an RCU read-side critical section might be preempted,
2233 which means that a subsequent <tt>synchronize_rcu()</tt> really does have
2234 to wait for something, as opposed to simply returning immediately.
2235 Unfortunately, <tt>synchronize_rcu()</tt> can't do this until all of
2236 its kthreads are spawned, which doesn't happen until some time during
2237 <tt>early_initcalls()</tt> time.
2238 But this is no excuse: RCU is nevertheless required to correctly handle
2239 synchronous grace periods during this time period.
2240 Once all of its kthreads are up and running, RCU starts running
2244 <tr><th> </th></tr>
2245 <tr><th align="left">Quick Quiz:</th></tr>
2247 How can RCU possibly handle grace periods before all of its
2248 kthreads have been spawned???
2250 <tr><th align="left">Answer:</th></tr>
2251 <tr><td bgcolor="#ffffff"><font color="ffffff">
2255 <p><font color="ffffff">
2256 During the “dead zone” between the time that the
2257 scheduler spawns the first task and the time that all of RCU's
2258 kthreads have been spawned, all synchronous grace periods are
2259 handled by the expedited grace-period mechanism.
2260 At runtime, this expedited mechanism relies on workqueues, but
2261 during the dead zone the requesting task itself drives the
2262 desired expedited grace period.
2263 Because dead-zone execution takes place within task context,
2265 Once the dead zone ends, expedited grace periods go back to
2266 using workqueues, as is required to avoid problems that would
2267 otherwise occur when a user task received a POSIX signal while
2268 driving an expedited grace period.
2271 <p><font color="ffffff">
2272 And yes, this does mean that it is unhelpful to send POSIX
2273 signals to random tasks between the time that the scheduler
2274 spawns its first kthread and the time that RCU's kthreads
2275 have all been spawned.
2276 If there ever turns out to be a good reason for sending POSIX
2277 signals during that time, appropriate adjustments will be made.
2278 (If it turns out that POSIX signals are sent during this time for
2279 no good reason, other adjustments will be made, appropriate
2282 <tr><td> </td></tr>
2286 I learned of these boot-time requirements as a result of a series of
2289 <h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
2292 The Linux kernel has interrupts, and RCU read-side critical sections are
2293 legal within interrupt handlers and within interrupt-disabled regions
2294 of code, as are invocations of <tt>call_rcu()</tt>.
2297 Some Linux-kernel architectures can enter an interrupt handler from
2298 non-idle process context, and then just never leave it, instead stealthily
2299 transitioning back to process context.
2300 This trick is sometimes used to invoke system calls from inside the kernel.
2301 These “half-interrupts” mean that RCU has to be very careful
2302 about how it counts interrupt nesting levels.
2303 I learned of this requirement the hard way during a rewrite
2304 of RCU's dyntick-idle code.
2307 The Linux kernel has non-maskable interrupts (NMIs), and
2308 RCU read-side critical sections are legal within NMI handlers.
2309 Thankfully, RCU update-side primitives, including
2310 <tt>call_rcu()</tt>, are prohibited within NMI handlers.
2313 The name notwithstanding, some Linux-kernel architectures
2314 can have nested NMIs, which RCU must handle correctly.
2316 <a href="https://lkml.kernel.org/r/CALCETrXLq1y7e_dKFPgou-FKHB6Pu-r8+t-6Ds+8=va7anBWDA@mail.gmail.com">surprised me</a>
2317 with this requirement;
2318 he also kindly surprised me with
2319 <a href="https://lkml.kernel.org/r/CALCETrXSY9JpW3uE6H8WYk81sg56qasA2aqmjMPsq5dOtzso=g@mail.gmail.com">an algorithm</a>
2320 that meets this requirement.
2323 Furthermore, NMI handlers can be interrupted by what appear to RCU
2324 to be normal interrupts.
2325 One way that this can happen is for code that directly invokes
2326 <tt>rcu_irq_enter()</tt> and <tt>rcu_irq_exit()</tt> to be called
2327 from an NMI handler.
2328 This astonishing fact of life prompted the current code structure,
2329 which has <tt>rcu_irq_enter()</tt> invoking <tt>rcu_nmi_enter()</tt>
2330 and <tt>rcu_irq_exit()</tt> invoking <tt>rcu_nmi_exit()</tt>.
2331 And yes, I also learned of this requirement the hard way.
2333 <h3><a name="Loadable Modules">Loadable Modules</a></h3>
2336 The Linux kernel has loadable modules, and these modules can
2338 After a given module has been unloaded, any attempt to call
2339 one of its functions results in a segmentation fault.
2340 The module-unload functions must therefore cancel any
2341 delayed calls to loadable-module functions, for example,
2342 any outstanding <tt>mod_timer()</tt> must be dealt with
2343 via <tt>del_timer_sync()</tt> or similar.
2346 Unfortunately, there is no way to cancel an RCU callback;
2347 once you invoke <tt>call_rcu()</tt>, the callback function is
2348 eventually going to be invoked, unless the system goes down first.
2349 Because it is normally considered socially irresponsible to crash the system
2350 in response to a module unload request, we need some other way
2351 to deal with in-flight RCU callbacks.
2354 RCU therefore provides
2355 <tt><a href="https://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
2356 which waits until all in-flight RCU callbacks have been invoked.
2357 If a module uses <tt>call_rcu()</tt>, its exit function should therefore
2358 prevent any future invocation of <tt>call_rcu()</tt>, then invoke
2359 <tt>rcu_barrier()</tt>.
2360 In theory, the underlying module-unload code could invoke
2361 <tt>rcu_barrier()</tt> unconditionally, but in practice this would
2362 incur unacceptable latencies.
2365 Nikita Danilov noted this requirement for an analogous filesystem-unmount
2366 situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
2367 The need for <tt>rcu_barrier()</tt> for module unloading became
2371 <b>Important note</b>: The <tt>rcu_barrier()</tt> function is not,
2372 repeat, <i>not</i>, obligated to wait for a grace period.
2373 It is instead only required to wait for RCU callbacks that have
2374 already been posted.
2375 Therefore, if there are no RCU callbacks posted anywhere in the system,
2376 <tt>rcu_barrier()</tt> is within its rights to return immediately.
2377 Even if there are callbacks posted, <tt>rcu_barrier()</tt> does not
2378 necessarily need to wait for a grace period.
2381 <tr><th> </th></tr>
2382 <tr><th align="left">Quick Quiz:</th></tr>
2385 Each RCU callbacks must wait for a grace period to complete,
2386 and <tt>rcu_barrier()</tt> must wait for each pre-existing
2387 callback to be invoked.
2388 Doesn't <tt>rcu_barrier()</tt> therefore need to wait for
2389 a full grace period if there is even one callback posted anywhere
2392 <tr><th align="left">Answer:</th></tr>
2393 <tr><td bgcolor="#ffffff"><font color="ffffff">
2397 <p><font color="ffffff">
2398 Yes, each RCU callbacks must wait for a grace period to complete,
2399 but it might well be partly (or even completely) finished waiting
2400 by the time <tt>rcu_barrier()</tt> is invoked.
2401 In that case, <tt>rcu_barrier()</tt> need only wait for the
2402 remaining portion of the grace period to elapse.
2403 So even if there are quite a few callbacks posted,
2404 <tt>rcu_barrier()</tt> might well return quite quickly.
2407 <p><font color="ffffff">
2408 So if you need to wait for a grace period as well as for all
2409 pre-existing callbacks, you will need to invoke both
2410 <tt>synchronize_rcu()</tt> and <tt>rcu_barrier()</tt>.
2411 If latency is a concern, you can always use workqueues
2412 to invoke them concurrently.
2414 <tr><td> </td></tr>
2417 <h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
2420 The Linux kernel supports CPU hotplug, which means that CPUs
2422 It is of course illegal to use any RCU API member from an offline CPU,
2423 with the exception of <a href="#Sleepable RCU">SRCU</a> read-side
2425 This requirement was present from day one in DYNIX/ptx, but
2426 on the other hand, the Linux kernel's CPU-hotplug implementation
2427 is “interesting.”
2430 The Linux-kernel CPU-hotplug implementation has notifiers that
2431 are used to allow the various kernel subsystems (including RCU)
2432 to respond appropriately to a given CPU-hotplug operation.
2433 Most RCU operations may be invoked from CPU-hotplug notifiers,
2434 including even synchronous grace-period operations such as
2435 <tt>synchronize_rcu()</tt> and <tt>synchronize_rcu_expedited()</tt>.
2438 However, all-callback-wait operations such as
2439 <tt>rcu_barrier()</tt> are also not supported, due to the
2440 fact that there are phases of CPU-hotplug operations where
2441 the outgoing CPU's callbacks will not be invoked until after
2442 the CPU-hotplug operation ends, which could also result in deadlock.
2443 Furthermore, <tt>rcu_barrier()</tt> blocks CPU-hotplug operations
2444 during its execution, which results in another type of deadlock
2445 when invoked from a CPU-hotplug notifier.
2447 <h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
2450 RCU depends on the scheduler, and the scheduler uses RCU to
2451 protect some of its data structures.
2452 The preemptible-RCU <tt>rcu_read_unlock()</tt>
2453 implementation must therefore be written carefully to avoid deadlocks
2454 involving the scheduler's runqueue and priority-inheritance locks.
2455 In particular, <tt>rcu_read_unlock()</tt> must tolerate an
2456 interrupt where the interrupt handler invokes both
2457 <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
2458 This possibility requires <tt>rcu_read_unlock()</tt> to use
2459 negative nesting levels to avoid destructive recursion via
2460 interrupt handler's use of RCU.
2463 This scheduler-RCU requirement came as a
2464 <a href="https://lwn.net/Articles/453002/">complete surprise</a>.
2467 As noted above, RCU makes use of kthreads, and it is necessary to
2468 avoid excessive CPU-time accumulation by these kthreads.
2469 This requirement was no surprise, but RCU's violation of it
2470 when running context-switch-heavy workloads when built with
2471 <tt>CONFIG_NO_HZ_FULL=y</tt>
2472 <a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
2473 RCU has made good progress towards meeting this requirement, even
2474 for context-switch-heavy <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
2475 but there is room for further improvement.
2478 It is forbidden to hold any of scheduler's runqueue or priority-inheritance
2479 spinlocks across an <tt>rcu_read_unlock()</tt> unless interrupts have been
2480 disabled across the entire RCU read-side critical section, that is,
2481 up to and including the matching <tt>rcu_read_lock()</tt>.
2482 Violating this restriction can result in deadlocks involving these
2483 scheduler spinlocks.
2484 There was hope that this restriction might be lifted when interrupt-disabled
2485 calls to <tt>rcu_read_unlock()</tt> started deferring the reporting of
2486 the resulting RCU-preempt quiescent state until the end of the corresponding
2487 interrupts-disabled region.
2488 Unfortunately, timely reporting of the corresponding quiescent state
2489 to expedited grace periods requires a call to <tt>raise_softirq()</tt>,
2490 which can acquire these scheduler spinlocks.
2491 In addition, real-time systems using RCU priority boosting
2492 need this restriction to remain in effect because deferred
2493 quiescent-state reporting would also defer deboosting, which in turn
2494 would degrade real-time latencies.
2497 In theory, if a given RCU read-side critical section could be
2498 guaranteed to be less than one second in duration, holding a scheduler
2499 spinlock across that critical section's <tt>rcu_read_unlock()</tt>
2500 would require only that preemption be disabled across the entire
2501 RCU read-side critical section, not interrupts.
2502 Unfortunately, given the possibility of vCPU preemption, long-running
2503 interrupts, and so on, it is not possible in practice to guarantee
2504 that a given RCU read-side critical section will complete in less than
2506 Therefore, as noted above, if scheduler spinlocks are held across
2507 a given call to <tt>rcu_read_unlock()</tt>, interrupts must be
2508 disabled across the entire RCU read-side critical section.
2510 <h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
2513 It is possible to use tracing on RCU code, but tracing itself
2515 For this reason, <tt>rcu_dereference_raw_notrace()</tt>
2516 is provided for use by tracing, which avoids the destructive
2517 recursion that could otherwise ensue.
2518 This API is also used by virtualization in some architectures,
2519 where RCU readers execute in environments in which tracing
2521 The tracing folks both located the requirement and provided the
2522 needed fix, so this surprise requirement was relatively painless.
2524 <h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
2527 Interrupting idle CPUs is considered socially unacceptable,
2528 especially by people with battery-powered embedded systems.
2529 RCU therefore conserves energy by detecting which CPUs are
2530 idle, including tracking CPUs that have been interrupted from idle.
2531 This is a large part of the energy-efficiency requirement,
2532 so I learned of this via an irate phone call.
2535 Because RCU avoids interrupting idle CPUs, it is illegal to
2536 execute an RCU read-side critical section on an idle CPU.
2537 (Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
2539 The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
2540 event tracing is provided to work around this restriction.
2541 In addition, <tt>rcu_is_watching()</tt> may be used to
2542 test whether or not it is currently legal to run RCU read-side
2543 critical sections on this CPU.
2544 I learned of the need for diagnostics on the one hand
2545 and <tt>RCU_NONIDLE()</tt> on the other while inspecting
2547 Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
2548 which is used quite heavily in the idle loop.
2549 However, there are some restrictions on the code placed within
2550 <tt>RCU_NONIDLE()</tt>:
2553 <li> Blocking is prohibited.
2554 In practice, this is not a serious restriction given that idle
2555 tasks are prohibited from blocking to begin with.
2556 <li> Although nesting <tt>RCU_NONIDLE()</tt> is permitted, they cannot
2557 nest indefinitely deeply.
2558 However, given that they can be nested on the order of a million
2559 deep, even on 32-bit systems, this should not be a serious
2561 This nesting limit would probably be reached long after the
2562 compiler OOMed or the stack overflowed.
2563 <li> Any code path that enters <tt>RCU_NONIDLE()</tt> must sequence
2564 out of that same <tt>RCU_NONIDLE()</tt>.
2565 For example, the following is grossly illegal:
2571 3 goto bad_idea; /* BUG!!! */
2572 4 do_something_else();});
2578 It is just as illegal to transfer control into the middle of
2579 <tt>RCU_NONIDLE()</tt>'s argument.
2580 Yes, in theory, you could transfer in as long as you also
2581 transferred out, but in practice you could also expect to get sharply
2582 worded review comments.
2586 It is similarly socially unacceptable to interrupt an
2587 <tt>nohz_full</tt> CPU running in userspace.
2588 RCU must therefore track <tt>nohz_full</tt> userspace
2590 RCU must therefore be able to sample state at two points in
2591 time, and be able to determine whether or not some other CPU spent
2592 any time idle and/or executing in userspace.
2595 These energy-efficiency requirements have proven quite difficult to
2596 understand and to meet, for example, there have been more than five
2597 clean-sheet rewrites of RCU's energy-efficiency code, the last of
2598 which was finally able to demonstrate
2599 <a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
2601 I learned of many of these requirements via angry phone calls:
2602 Flaming me on the Linux-kernel mailing list was apparently not
2603 sufficient to fully vent their ire at RCU's energy-efficiency bugs!
2605 <h3><a name="Scheduling-Clock Interrupts and RCU">
2606 Scheduling-Clock Interrupts and RCU</a></h3>
2609 The kernel transitions between in-kernel non-idle execution, userspace
2610 execution, and the idle loop.
2611 Depending on kernel configuration, RCU handles these states differently:
2614 <tr><th><tt>HZ</tt> Kconfig</th>
2618 <tr><th align="left"><tt>HZ_PERIODIC</tt></th>
2619 <td>Can rely on scheduling-clock interrupt.</td>
2620 <td>Can rely on scheduling-clock interrupt and its
2621 detection of interrupt from usermode.</td>
2622 <td>Can rely on RCU's dyntick-idle detection.</td></tr>
2623 <tr><th align="left"><tt>NO_HZ_IDLE</tt></th>
2624 <td>Can rely on scheduling-clock interrupt.</td>
2625 <td>Can rely on scheduling-clock interrupt and its
2626 detection of interrupt from usermode.</td>
2627 <td>Can rely on RCU's dyntick-idle detection.</td></tr>
2628 <tr><th align="left"><tt>NO_HZ_FULL</tt></th>
2629 <td>Can only sometimes rely on scheduling-clock interrupt.
2630 In other cases, it is necessary to bound kernel execution
2631 times and/or use IPIs.</td>
2632 <td>Can rely on RCU's dyntick-idle detection.</td>
2633 <td>Can rely on RCU's dyntick-idle detection.</td></tr>
2637 <tr><th> </th></tr>
2638 <tr><th align="left">Quick Quiz:</th></tr>
2640 Why can't <tt>NO_HZ_FULL</tt> in-kernel execution rely on the
2641 scheduling-clock interrupt, just like <tt>HZ_PERIODIC</tt>
2642 and <tt>NO_HZ_IDLE</tt> do?
2644 <tr><th align="left">Answer:</th></tr>
2645 <tr><td bgcolor="#ffffff"><font color="ffffff">
2646 Because, as a performance optimization, <tt>NO_HZ_FULL</tt>
2647 does not necessarily re-enable the scheduling-clock interrupt
2648 on entry to each and every system call.
2650 <tr><td> </td></tr>
2654 However, RCU must be reliably informed as to whether any given
2655 CPU is currently in the idle loop, and, for <tt>NO_HZ_FULL</tt>,
2656 also whether that CPU is executing in usermode, as discussed
2657 <a href="#Energy Efficiency">earlier</a>.
2658 It also requires that the scheduling-clock interrupt be enabled when
2662 <li> If a CPU is either idle or executing in usermode, and RCU believes
2663 it is non-idle, the scheduling-clock tick had better be running.
2664 Otherwise, you will get RCU CPU stall warnings. Or at best,
2665 very long (11-second) grace periods, with a pointless IPI waking
2666 the CPU from time to time.
2667 <li> If a CPU is in a portion of the kernel that executes RCU read-side
2668 critical sections, and RCU believes this CPU to be idle, you will get
2669 random memory corruption. <b>DON'T DO THIS!!!</b>
2671 <br>This is one reason to test with lockdep, which will complain
2672 about this sort of thing.
2673 <li> If a CPU is in a portion of the kernel that is absolutely
2674 positively no-joking guaranteed to never execute any RCU read-side
2675 critical sections, and RCU believes this CPU to to be idle,
2676 no problem. This sort of thing is used by some architectures
2677 for light-weight exception handlers, which can then avoid the
2678 overhead of <tt>rcu_irq_enter()</tt> and <tt>rcu_irq_exit()</tt>
2679 at exception entry and exit, respectively.
2680 Some go further and avoid the entireties of <tt>irq_enter()</tt>
2681 and <tt>irq_exit()</tt>.
2683 <br>Just make very sure you are running some of your tests with
2684 <tt>CONFIG_PROVE_RCU=y</tt>, just in case one of your code paths
2685 was in fact joking about not doing RCU read-side critical sections.
2686 <li> If a CPU is executing in the kernel with the scheduling-clock
2687 interrupt disabled and RCU believes this CPU to be non-idle,
2688 and if the CPU goes idle (from an RCU perspective) every few
2689 jiffies, no problem. It is usually OK for there to be the
2690 occasional gap between idle periods of up to a second or so.
2692 <br>If the gap grows too long, you get RCU CPU stall warnings.
2693 <li> If a CPU is either idle or executing in usermode, and RCU believes
2694 it to be idle, of course no problem.
2695 <li> If a CPU is executing in the kernel, the kernel code
2696 path is passing through quiescent states at a reasonable
2697 frequency (preferably about once per few jiffies, but the
2698 occasional excursion to a second or so is usually OK) and the
2699 scheduling-clock interrupt is enabled, of course no problem.
2701 <br>If the gap between a successive pair of quiescent states grows
2702 too long, you get RCU CPU stall warnings.
2706 <tr><th> </th></tr>
2707 <tr><th align="left">Quick Quiz:</th></tr>
2709 But what if my driver has a hardware interrupt handler
2710 that can run for many seconds?
2711 I cannot invoke <tt>schedule()</tt> from an hardware
2712 interrupt handler, after all!
2714 <tr><th align="left">Answer:</th></tr>
2715 <tr><td bgcolor="#ffffff"><font color="ffffff">
2716 One approach is to do <tt>rcu_irq_exit();rcu_irq_enter();</tt>
2718 But given that long-running interrupt handlers can cause
2719 other problems, not least for response time, shouldn't you
2720 work to keep your interrupt handler's runtime within reasonable
2723 <tr><td> </td></tr>
2727 But as long as RCU is properly informed of kernel state transitions between
2728 in-kernel execution, usermode execution, and idle, and as long as the
2729 scheduling-clock interrupt is enabled when RCU needs it to be, you
2730 can rest assured that the bugs you encounter will be in some other
2731 part of RCU or some other part of the kernel!
2733 <h3><a name="Memory Efficiency">Memory Efficiency</a></h3>
2736 Although small-memory non-realtime systems can simply use Tiny RCU,
2737 code size is only one aspect of memory efficiency.
2738 Another aspect is the size of the <tt>rcu_head</tt> structure
2739 used by <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>.
2740 Although this structure contains nothing more than a pair of pointers,
2741 it does appear in many RCU-protected data structures, including
2742 some that are size critical.
2743 The <tt>page</tt> structure is a case in point, as evidenced by
2744 the many occurrences of the <tt>union</tt> keyword within that structure.
2747 This need for memory efficiency is one reason that RCU uses hand-crafted
2748 singly linked lists to track the <tt>rcu_head</tt> structures that
2749 are waiting for a grace period to elapse.
2750 It is also the reason why <tt>rcu_head</tt> structures do not contain
2751 debug information, such as fields tracking the file and line of the
2752 <tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> that posted them.
2753 Although this information might appear in debug-only kernel builds at some
2754 point, in the meantime, the <tt>->func</tt> field will often provide
2755 the needed debug information.
2758 However, in some cases, the need for memory efficiency leads to even
2759 more extreme measures.
2760 Returning to the <tt>page</tt> structure, the <tt>rcu_head</tt> field
2761 shares storage with a great many other structures that are used at
2762 various points in the corresponding page's lifetime.
2763 In order to correctly resolve certain
2764 <a href="https://lkml.kernel.org/g/1439976106-137226-1-git-send-email-kirill.shutemov@linux.intel.com">race conditions</a>,
2765 the Linux kernel's memory-management subsystem needs a particular bit
2766 to remain zero during all phases of grace-period processing,
2767 and that bit happens to map to the bottom bit of the
2768 <tt>rcu_head</tt> structure's <tt>->next</tt> field.
2769 RCU makes this guarantee as long as <tt>call_rcu()</tt>
2770 is used to post the callback, as opposed to <tt>kfree_rcu()</tt>
2771 or some future “lazy”
2772 variant of <tt>call_rcu()</tt> that might one day be created for
2773 energy-efficiency purposes.
2776 That said, there are limits.
2777 RCU requires that the <tt>rcu_head</tt> structure be aligned to a
2778 two-byte boundary, and passing a misaligned <tt>rcu_head</tt>
2779 structure to one of the <tt>call_rcu()</tt> family of functions
2780 will result in a splat.
2781 It is therefore necessary to exercise caution when packing
2782 structures containing fields of type <tt>rcu_head</tt>.
2783 Why not a four-byte or even eight-byte alignment requirement?
2784 Because the m68k architecture provides only two-byte alignment,
2785 and thus acts as alignment's least common denominator.
2788 The reason for reserving the bottom bit of pointers to
2789 <tt>rcu_head</tt> structures is to leave the door open to
2790 “lazy” callbacks whose invocations can safely be deferred.
2791 Deferring invocation could potentially have energy-efficiency
2792 benefits, but only if the rate of non-lazy callbacks decreases
2793 significantly for some important workload.
2794 In the meantime, reserving the bottom bit keeps this option open
2795 in case it one day becomes useful.
2797 <h3><a name="Performance, Scalability, Response Time, and Reliability">
2798 Performance, Scalability, Response Time, and Reliability</a></h3>
2802 <a href="#Performance and Scalability">earlier discussion</a>,
2803 RCU is used heavily by hot code paths in performance-critical
2804 portions of the Linux kernel's networking, security, virtualization,
2805 and scheduling code paths.
2806 RCU must therefore use efficient implementations, especially in its
2807 read-side primitives.
2808 To that end, it would be good if preemptible RCU's implementation
2809 of <tt>rcu_read_lock()</tt> could be inlined, however, doing
2810 this requires resolving <tt>#include</tt> issues with the
2811 <tt>task_struct</tt> structure.
2814 The Linux kernel supports hardware configurations with up to
2815 4096 CPUs, which means that RCU must be extremely scalable.
2816 Algorithms that involve frequent acquisitions of global locks or
2817 frequent atomic operations on global variables simply cannot be
2818 tolerated within the RCU implementation.
2819 RCU therefore makes heavy use of a combining tree based on the
2820 <tt>rcu_node</tt> structure.
2821 RCU is required to tolerate all CPUs continuously invoking any
2822 combination of RCU's runtime primitives with minimal per-operation
2824 In fact, in many cases, increasing load must <i>decrease</i> the
2825 per-operation overhead, witness the batching optimizations for
2826 <tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
2827 <tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
2828 As a general rule, RCU must cheerfully accept whatever the
2829 rest of the Linux kernel decides to throw at it.
2832 The Linux kernel is used for real-time workloads, especially
2833 in conjunction with the
2834 <a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
2835 The real-time-latency response requirements are such that the
2836 traditional approach of disabling preemption across RCU
2837 read-side critical sections is inappropriate.
2838 Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
2839 use an RCU implementation that allows RCU read-side critical
2840 sections to be preempted.
2841 This requirement made its presence known after users made it
2842 clear that an earlier
2843 <a href="https://lwn.net/Articles/107930/">real-time patch</a>
2844 did not meet their needs, in conjunction with some
2845 <a href="https://lkml.kernel.org/g/20050318002026.GA2693@us.ibm.com">RCU issues</a>
2846 encountered by a very early version of the -rt patchset.
2849 In addition, RCU must make do with a sub-100-microsecond real-time latency
2851 In fact, on smaller systems with the -rt patchset, the Linux kernel
2852 provides sub-20-microsecond real-time latencies for the whole kernel,
2854 RCU's scalability and latency must therefore be sufficient for
2855 these sorts of configurations.
2856 To my surprise, the sub-100-microsecond real-time latency budget
2857 <a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
2858 applies to even the largest systems [PDF]</a>,
2859 up to and including systems with 4096 CPUs.
2860 This real-time requirement motivated the grace-period kthread, which
2861 also simplified handling of a number of race conditions.
2864 RCU must avoid degrading real-time response for CPU-bound threads, whether
2865 executing in usermode (which is one use case for
2866 <tt>CONFIG_NO_HZ_FULL=y</tt>) or in the kernel.
2867 That said, CPU-bound loops in the kernel must execute
2868 <tt>cond_resched()</tt> at least once per few tens of milliseconds
2869 in order to avoid receiving an IPI from RCU.
2872 Finally, RCU's status as a synchronization primitive means that
2873 any RCU failure can result in arbitrary memory corruption that can be
2874 extremely difficult to debug.
2875 This means that RCU must be extremely reliable, which in
2876 practice also means that RCU must have an aggressive stress-test
2878 This stress-test suite is called <tt>rcutorture</tt>.
2881 Although the need for <tt>rcutorture</tt> was no surprise,
2882 the current immense popularity of the Linux kernel is posing
2883 interesting—and perhaps unprecedented—validation
2885 To see this, keep in mind that there are well over one billion
2886 instances of the Linux kernel running today, given Android
2887 smartphones, Linux-powered televisions, and servers.
2888 This number can be expected to increase sharply with the advent of
2889 the celebrated Internet of Things.
2892 Suppose that RCU contains a race condition that manifests on average
2893 once per million years of runtime.
2894 This bug will be occurring about three times per <i>day</i> across
2896 RCU could simply hide behind hardware error rates, given that no one
2897 should really expect their smartphone to last for a million years.
2898 However, anyone taking too much comfort from this thought should
2899 consider the fact that in most jurisdictions, a successful multi-year
2900 test of a given mechanism, which might include a Linux kernel,
2901 suffices for a number of types of safety-critical certifications.
2902 In fact, rumor has it that the Linux kernel is already being used
2903 in production for safety-critical applications.
2904 I don't know about you, but I would feel quite bad if a bug in RCU
2906 Which might explain my recent focus on validation and verification.
2908 <h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
2911 One of the more surprising things about RCU is that there are now
2912 no fewer than five <i>flavors</i>, or API families.
2913 In addition, the primary flavor that has been the sole focus up to
2914 this point has two different implementations, non-preemptible and
2916 The other four flavors are listed below, with requirements for each
2917 described in a separate section.
2920 <li> <a href="#Bottom-Half Flavor">Bottom-Half Flavor (Historical)</a>
2921 <li> <a href="#Sched Flavor">Sched Flavor (Historical)</a>
2922 <li> <a href="#Sleepable RCU">Sleepable RCU</a>
2923 <li> <a href="#Tasks RCU">Tasks RCU</a>
2926 <h3><a name="Bottom-Half Flavor">Bottom-Half Flavor (Historical)</a></h3>
2929 The RCU-bh flavor of RCU has since been expressed in terms of
2930 the other RCU flavors as part of a consolidation of the three
2931 flavors into a single flavor.
2932 The read-side API remains, and continues to disable softirq and to
2933 be accounted for by lockdep.
2934 Much of the material in this section is therefore strictly historical
2938 The softirq-disable (AKA “bottom-half”,
2939 hence the “_bh” abbreviations)
2940 flavor of RCU, or <i>RCU-bh</i>, was developed by
2941 Dipankar Sarma to provide a flavor of RCU that could withstand the
2942 network-based denial-of-service attacks researched by Robert
2944 These attacks placed so much networking load on the system
2945 that some of the CPUs never exited softirq execution,
2946 which in turn prevented those CPUs from ever executing a context switch,
2947 which, in the RCU implementation of that time, prevented grace periods
2949 The result was an out-of-memory condition and a system hang.
2952 The solution was the creation of RCU-bh, which does
2953 <tt>local_bh_disable()</tt>
2954 across its read-side critical sections, and which uses the transition
2955 from one type of softirq processing to another as a quiescent state
2956 in addition to context switch, idle, user mode, and offline.
2957 This means that RCU-bh grace periods can complete even when some of
2958 the CPUs execute in softirq indefinitely, thus allowing algorithms
2959 based on RCU-bh to withstand network-based denial-of-service attacks.
2963 <tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
2964 disable and re-enable softirq handlers, any attempt to start a softirq
2966 RCU-bh read-side critical section will be deferred.
2967 In this case, <tt>rcu_read_unlock_bh()</tt>
2968 will invoke softirq processing, which can take considerable time.
2969 One can of course argue that this softirq overhead should be associated
2970 with the code following the RCU-bh read-side critical section rather
2971 than <tt>rcu_read_unlock_bh()</tt>, but the fact
2972 is that most profiling tools cannot be expected to make this sort
2973 of fine distinction.
2974 For example, suppose that a three-millisecond-long RCU-bh read-side
2975 critical section executes during a time of heavy networking load.
2976 There will very likely be an attempt to invoke at least one softirq
2977 handler during that three milliseconds, but any such invocation will
2978 be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
2979 This can of course make it appear at first glance as if
2980 <tt>rcu_read_unlock_bh()</tt> was executing very slowly.
2984 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
2986 <tt>rcu_read_lock_bh()</tt>,
2987 <tt>rcu_read_unlock_bh()</tt>,
2988 <tt>rcu_dereference_bh()</tt>,
2989 <tt>rcu_dereference_bh_check()</tt>,
2990 <tt>synchronize_rcu_bh()</tt>,
2991 <tt>synchronize_rcu_bh_expedited()</tt>,
2992 <tt>call_rcu_bh()</tt>,
2993 <tt>rcu_barrier_bh()</tt>, and
2994 <tt>rcu_read_lock_bh_held()</tt>.
2995 However, the update-side APIs are now simple wrappers for other RCU
2996 flavors, namely RCU-sched in CONFIG_PREEMPT=n kernels and RCU-preempt
2999 <h3><a name="Sched Flavor">Sched Flavor (Historical)</a></h3>
3002 The RCU-sched flavor of RCU has since been expressed in terms of
3003 the other RCU flavors as part of a consolidation of the three
3004 flavors into a single flavor.
3005 The read-side API remains, and continues to disable preemption and to
3006 be accounted for by lockdep.
3007 Much of the material in this section is therefore strictly historical
3011 Before preemptible RCU, waiting for an RCU grace period had the
3012 side effect of also waiting for all pre-existing interrupt
3014 However, there are legitimate preemptible-RCU implementations that
3015 do not have this property, given that any point in the code outside
3016 of an RCU read-side critical section can be a quiescent state.
3017 Therefore, <i>RCU-sched</i> was created, which follows “classic”
3018 RCU in that an RCU-sched grace period waits for for pre-existing
3019 interrupt and NMI handlers.
3020 In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
3021 APIs have identical implementations, while kernels built with
3022 <tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
3025 Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
3026 <tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
3027 disable and re-enable preemption, respectively.
3028 This means that if there was a preemption attempt during the
3029 RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
3030 will enter the scheduler, with all the latency and overhead entailed.
3031 Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
3032 as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
3033 However, the highest-priority task won't be preempted, so that task
3034 will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
3038 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
3040 <tt>rcu_read_lock_sched()</tt>,
3041 <tt>rcu_read_unlock_sched()</tt>,
3042 <tt>rcu_read_lock_sched_notrace()</tt>,
3043 <tt>rcu_read_unlock_sched_notrace()</tt>,
3044 <tt>rcu_dereference_sched()</tt>,
3045 <tt>rcu_dereference_sched_check()</tt>,
3046 <tt>synchronize_sched()</tt>,
3047 <tt>synchronize_rcu_sched_expedited()</tt>,
3048 <tt>call_rcu_sched()</tt>,
3049 <tt>rcu_barrier_sched()</tt>, and
3050 <tt>rcu_read_lock_sched_held()</tt>.
3051 However, anything that disables preemption also marks an RCU-sched
3052 read-side critical section, including
3053 <tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
3054 <tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
3057 <h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
3060 For well over a decade, someone saying “I need to block within
3061 an RCU read-side critical section” was a reliable indication
3062 that this someone did not understand RCU.
3063 After all, if you are always blocking in an RCU read-side critical
3064 section, you can probably afford to use a higher-overhead synchronization
3066 However, that changed with the advent of the Linux kernel's notifiers,
3067 whose RCU read-side critical
3068 sections almost never sleep, but sometimes need to.
3069 This resulted in the introduction of
3070 <a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
3074 SRCU allows different domains to be defined, with each such domain
3075 defined by an instance of an <tt>srcu_struct</tt> structure.
3076 A pointer to this structure must be passed in to each SRCU function,
3077 for example, <tt>synchronize_srcu(&ss)</tt>, where
3078 <tt>ss</tt> is the <tt>srcu_struct</tt> structure.
3079 The key benefit of these domains is that a slow SRCU reader in one
3080 domain does not delay an SRCU grace period in some other domain.
3081 That said, one consequence of these domains is that read-side code
3082 must pass a “cookie” from <tt>srcu_read_lock()</tt>
3083 to <tt>srcu_read_unlock()</tt>, for example, as follows:
3089 3 idx = srcu_read_lock(&ss);
3091 5 srcu_read_unlock(&ss, idx);
3096 As noted above, it is legal to block within SRCU read-side critical sections,
3097 however, with great power comes great responsibility.
3098 If you block forever in one of a given domain's SRCU read-side critical
3099 sections, then that domain's grace periods will also be blocked forever.
3100 Of course, one good way to block forever is to deadlock, which can
3101 happen if any operation in a given domain's SRCU read-side critical
3102 section can block waiting, either directly or indirectly, for that domain's
3103 grace period to elapse.
3104 For example, this results in a self-deadlock:
3110 3 idx = srcu_read_lock(&ss);
3112 5 synchronize_srcu(&ss);
3113 6 srcu_read_unlock(&ss, idx);
3118 However, if line 5 acquired a mutex that was held across
3119 a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
3120 deadlock would still be possible.
3121 Furthermore, if line 5 acquired a mutex that was held across
3122 a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
3123 and if an <tt>ss1</tt>-domain SRCU read-side critical section
3124 acquired another mutex that was held across as <tt>ss</tt>-domain
3125 <tt>synchronize_srcu()</tt>,
3126 deadlock would again be possible.
3127 Such a deadlock cycle could extend across an arbitrarily large number
3128 of different SRCU domains.
3129 Again, with great power comes great responsibility.
3132 Unlike the other RCU flavors, SRCU read-side critical sections can
3133 run on idle and even offline CPUs.
3134 This ability requires that <tt>srcu_read_lock()</tt> and
3135 <tt>srcu_read_unlock()</tt> contain memory barriers, which means
3136 that SRCU readers will run a bit slower than would RCU readers.
3137 It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
3138 API, which, in combination with <tt>srcu_read_unlock()</tt>,
3139 guarantees a full memory barrier.
3142 Also unlike other RCU flavors, SRCU's callbacks-wait function
3143 <tt>srcu_barrier()</tt> may be invoked from CPU-hotplug notifiers,
3144 though this is not necessarily a good idea.
3145 The reason that this is possible is that SRCU is insensitive
3146 to whether or not a CPU is online, which means that <tt>srcu_barrier()</tt>
3147 need not exclude CPU-hotplug operations.
3150 SRCU also differs from other RCU flavors in that SRCU's expedited and
3151 non-expedited grace periods are implemented by the same mechanism.
3152 This means that in the current SRCU implementation, expediting a
3153 future grace period has the side effect of expediting all prior
3154 grace periods that have not yet completed.
3155 (But please note that this is a property of the current implementation,
3156 not necessarily of future implementations.)
3157 In addition, if SRCU has been idle for longer than the interval
3158 specified by the <tt>srcutree.exp_holdoff</tt> kernel boot parameter
3159 (25 microseconds by default),
3160 and if a <tt>synchronize_srcu()</tt> invocation ends this idle period,
3161 that invocation will be automatically expedited.
3164 As of v4.12, SRCU's callbacks are maintained per-CPU, eliminating
3165 a locking bottleneck present in prior kernel versions.
3166 Although this will allow users to put much heavier stress on
3167 <tt>call_srcu()</tt>, it is important to note that SRCU does not
3168 yet take any special steps to deal with callback flooding.
3169 So if you are posting (say) 10,000 SRCU callbacks per second per CPU,
3170 you are probably totally OK, but if you intend to post (say) 1,000,000
3171 SRCU callbacks per second per CPU, please run some tests first.
3172 SRCU just might need a few adjustment to deal with that sort of load.
3173 Of course, your mileage may vary based on the speed of your CPUs and
3174 the size of your memory.
3178 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
3180 <tt>srcu_read_lock()</tt>,
3181 <tt>srcu_read_unlock()</tt>,
3182 <tt>srcu_dereference()</tt>,
3183 <tt>srcu_dereference_check()</tt>,
3184 <tt>synchronize_srcu()</tt>,
3185 <tt>synchronize_srcu_expedited()</tt>,
3186 <tt>call_srcu()</tt>,
3187 <tt>srcu_barrier()</tt>, and
3188 <tt>srcu_read_lock_held()</tt>.
3190 <tt>DEFINE_SRCU()</tt>,
3191 <tt>DEFINE_STATIC_SRCU()</tt>, and
3192 <tt>init_srcu_struct()</tt>
3193 APIs for defining and initializing <tt>srcu_struct</tt> structures.
3195 <h3><a name="Tasks RCU">Tasks RCU</a></h3>
3198 Some forms of tracing use “trampolines” to handle the
3199 binary rewriting required to install different types of probes.
3200 It would be good to be able to free old trampolines, which sounds
3201 like a job for some form of RCU.
3202 However, because it is necessary to be able to install a trace
3203 anywhere in the code, it is not possible to use read-side markers
3204 such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
3205 In addition, it does not work to have these markers in the trampoline
3206 itself, because there would need to be instructions following
3207 <tt>rcu_read_unlock()</tt>.
3208 Although <tt>synchronize_rcu()</tt> would guarantee that execution
3209 reached the <tt>rcu_read_unlock()</tt>, it would not be able to
3210 guarantee that execution had completely left the trampoline.
3213 The solution, in the form of
3214 <a href="https://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
3216 read-side critical sections that are delimited by voluntary context
3217 switches, that is, calls to <tt>schedule()</tt>,
3218 <tt>cond_resched()</tt>, and
3219 <tt>synchronize_rcu_tasks()</tt>.
3220 In addition, transitions to and from userspace execution also delimit
3221 tasks-RCU read-side critical sections.
3224 The tasks-RCU API is quite compact, consisting only of
3225 <tt>call_rcu_tasks()</tt>,
3226 <tt>synchronize_rcu_tasks()</tt>, and
3227 <tt>rcu_barrier_tasks()</tt>.
3228 In <tt>CONFIG_PREEMPT=n</tt> kernels, trampolines cannot be preempted,
3229 so these APIs map to
3230 <tt>call_rcu()</tt>,
3231 <tt>synchronize_rcu()</tt>, and
3232 <tt>rcu_barrier()</tt>, respectively.
3233 In <tt>CONFIG_PREEMPT=y</tt> kernels, trampolines can be preempted,
3234 and these three APIs are therefore implemented by separate functions
3235 that check for voluntary context switches.
3237 <h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
3240 One of the tricks that RCU uses to attain update-side scalability is
3241 to increase grace-period latency with increasing numbers of CPUs.
3242 If this becomes a serious problem, it will be necessary to rework the
3243 grace-period state machine so as to avoid the need for the additional
3247 RCU disables CPU hotplug in a few places, perhaps most notably in the
3248 <tt>rcu_barrier()</tt> operations.
3249 If there is a strong reason to use <tt>rcu_barrier()</tt> in CPU-hotplug
3250 notifiers, it will be necessary to avoid disabling CPU hotplug.
3251 This would introduce some complexity, so there had better be a <i>very</i>
3255 The tradeoff between grace-period latency on the one hand and interruptions
3256 of other CPUs on the other hand may need to be re-examined.
3257 The desire is of course for zero grace-period latency as well as zero
3258 interprocessor interrupts undertaken during an expedited grace period
3260 While this ideal is unlikely to be achievable, it is quite possible that
3261 further improvements can be made.
3264 The multiprocessor implementations of RCU use a combining tree that
3265 groups CPUs so as to reduce lock contention and increase cache locality.
3266 However, this combining tree does not spread its memory across NUMA
3267 nodes nor does it align the CPU groups with hardware features such
3268 as sockets or cores.
3269 Such spreading and alignment is currently believed to be unnecessary
3270 because the hotpath read-side primitives do not access the combining
3271 tree, nor does <tt>call_rcu()</tt> in the common case.
3272 If you believe that your architecture needs such spreading and alignment,
3273 then your architecture should also benefit from the
3274 <tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
3275 to the number of CPUs in a socket, NUMA node, or whatever.
3276 If the number of CPUs is too large, use a fraction of the number of
3278 If the number of CPUs is a large prime number, well, that certainly
3279 is an “interesting” architectural choice!
3280 More flexible arrangements might be considered, but only if
3281 <tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
3282 if the inadequacy has been demonstrated by a carefully run and
3283 realistic system-level workload.
3286 Please note that arrangements that require RCU to remap CPU numbers will
3287 require extremely good demonstration of need and full exploration of
3291 RCU's various kthreads are reasonably recent additions.
3292 It is quite likely that adjustments will be required to more gracefully
3293 handle extreme loads.
3294 It might also be necessary to be able to relate CPU utilization by
3295 RCU's kthreads and softirq handlers to the code that instigated this
3297 For example, RCU callback overhead might be charged back to the
3298 originating <tt>call_rcu()</tt> instance, though probably not
3299 in production kernels.
3302 Additional work may be required to provide reasonable forward-progress
3303 guarantees under heavy load for grace periods and for callback
3306 <h2><a name="Summary">Summary</a></h2>
3309 This document has presented more than two decade's worth of RCU
3311 Given that the requirements keep changing, this will not be the last
3312 word on this subject, but at least it serves to get an important
3313 subset of the requirements set forth.
3315 <h2><a name="Acknowledgments">Acknowledgments</a></h2>
3317 I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
3318 Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
3319 Andy Lutomirski for their help in rendering
3320 this article human readable, and to Michelle Rankin for her support
3322 Other contributions are acknowledged in the Linux kernel's git archive.
git.samba.org / sfrench / cifs-2.6.git / blob