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 distince 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.
563 with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
564 <tt>sparse</tt> will complain if you
565 define <tt>gp</tt> with <tt>__rcu</tt> and then
566 access it without using
567 either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
569 <tr><td> </td></tr>
573 In short, RCU's publish-subscribe guarantee is provided by the combination
574 of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
575 This guarantee allows data elements to be safely added to RCU-protected
576 linked data structures without disrupting RCU readers.
577 This guarantee can be used in combination with the grace-period
578 guarantee to also allow data elements to be removed from RCU-protected
579 linked data structures, again without disrupting RCU readers.
582 This guarantee was only partially premeditated.
583 DYNIX/ptx used an explicit memory barrier for publication, but had nothing
584 resembling <tt>rcu_dereference()</tt> for subscription, nor did it
585 have anything resembling the <tt>smp_read_barrier_depends()</tt>
586 that was later subsumed into <tt>rcu_dereference()</tt>.
587 The need for these operations made itself known quite suddenly at a
588 late-1990s meeting with the DEC Alpha architects, back in the days when
589 DEC was still a free-standing company.
590 It took the Alpha architects a good hour to convince me that any sort
591 of barrier would ever be needed, and it then took me a good <i>two</i> hours
592 to convince them that their documentation did not make this point clear.
593 More recent work with the C and C++ standards committees have provided
594 much education on tricks and traps from the compiler.
595 In short, compilers were much less tricky in the early 1990s, but in
596 2015, don't even think about omitting <tt>rcu_dereference()</tt>!
598 <h3><a name="Memory-Barrier Guarantees">Memory-Barrier Guarantees</a></h3>
601 The previous section's simple linked-data-structure scenario clearly
602 demonstrates the need for RCU's stringent memory-ordering guarantees on
603 systems with more than one CPU:
606 <li> Each CPU that has an RCU read-side critical section that
607 begins before <tt>synchronize_rcu()</tt> starts is
608 guaranteed to execute a full memory barrier between the time
609 that the RCU read-side critical section ends and the time that
610 <tt>synchronize_rcu()</tt> returns.
611 Without this guarantee, a pre-existing RCU read-side critical section
612 might hold a reference to the newly removed <tt>struct foo</tt>
613 after the <tt>kfree()</tt> on line 14 of
614 <tt>remove_gp_synchronous()</tt>.
615 <li> Each CPU that has an RCU read-side critical section that ends
616 after <tt>synchronize_rcu()</tt> returns is guaranteed
617 to execute a full memory barrier between the time that
618 <tt>synchronize_rcu()</tt> begins and the time that the RCU
619 read-side critical section begins.
620 Without this guarantee, a later RCU read-side critical section
621 running after the <tt>kfree()</tt> on line 14 of
622 <tt>remove_gp_synchronous()</tt> might
623 later run <tt>do_something_gp()</tt> and find the
624 newly deleted <tt>struct foo</tt>.
625 <li> If the task invoking <tt>synchronize_rcu()</tt> remains
626 on a given CPU, then that CPU is guaranteed to execute a full
627 memory barrier sometime during the execution of
628 <tt>synchronize_rcu()</tt>.
629 This guarantee ensures that the <tt>kfree()</tt> on
630 line 14 of <tt>remove_gp_synchronous()</tt> really does
631 execute after the removal on line 11.
632 <li> If the task invoking <tt>synchronize_rcu()</tt> migrates
633 among a group of CPUs during that invocation, then each of the
634 CPUs in that group is guaranteed to execute a full memory barrier
635 sometime during the execution of <tt>synchronize_rcu()</tt>.
636 This guarantee also ensures that the <tt>kfree()</tt> on
637 line 14 of <tt>remove_gp_synchronous()</tt> really does
638 execute after the removal on
639 line 11, but also in the case where the thread executing the
640 <tt>synchronize_rcu()</tt> migrates in the meantime.
644 <tr><th> </th></tr>
645 <tr><th align="left">Quick Quiz:</th></tr>
647 Given that multiple CPUs can start RCU read-side critical sections
648 at any time without any ordering whatsoever, how can RCU possibly
649 tell whether or not a given RCU read-side critical section starts
650 before a given instance of <tt>synchronize_rcu()</tt>?
652 <tr><th align="left">Answer:</th></tr>
653 <tr><td bgcolor="#ffffff"><font color="ffffff">
654 If RCU cannot tell whether or not a given
655 RCU read-side critical section starts before a
656 given instance of <tt>synchronize_rcu()</tt>,
657 then it must assume that the RCU read-side critical section
659 In other words, a given instance of <tt>synchronize_rcu()</tt>
660 can avoid waiting on a given RCU read-side critical section only
661 if it can prove that <tt>synchronize_rcu()</tt> started first.
663 <tr><td> </td></tr>
667 <tr><th> </th></tr>
668 <tr><th align="left">Quick Quiz:</th></tr>
670 The first and second guarantees require unbelievably strict ordering!
671 Are all these memory barriers <i> really</i> required?
673 <tr><th align="left">Answer:</th></tr>
674 <tr><td bgcolor="#ffffff"><font color="ffffff">
675 Yes, they really are required.
676 To see why the first guarantee is required, consider the following
681 <li> <font color="ffffff">
682 CPU 1: <tt>rcu_read_lock()</tt>
684 <li> <font color="ffffff">
685 CPU 1: <tt>q = rcu_dereference(gp);
686 /* Very likely to return p. */</tt>
688 <li> <font color="ffffff">
689 CPU 0: <tt>list_del_rcu(p);</tt>
691 <li> <font color="ffffff">
692 CPU 0: <tt>synchronize_rcu()</tt> starts.
694 <li> <font color="ffffff">
695 CPU 1: <tt>do_something_with(q->a);
696 /* No smp_mb(), so might happen after kfree(). */</tt>
698 <li> <font color="ffffff">
699 CPU 1: <tt>rcu_read_unlock()</tt>
701 <li> <font color="ffffff">
702 CPU 0: <tt>synchronize_rcu()</tt> returns.
704 <li> <font color="ffffff">
705 CPU 0: <tt>kfree(p);</tt>
709 <p><font color="ffffff">
710 Therefore, there absolutely must be a full memory barrier between the
711 end of the RCU read-side critical section and the end of the
715 <p><font color="ffffff">
716 The sequence of events demonstrating the necessity of the second rule
721 <li> <font color="ffffff">CPU 0: <tt>list_del_rcu(p);</tt>
723 <li> <font color="ffffff">CPU 0: <tt>synchronize_rcu()</tt> starts.
725 <li> <font color="ffffff">CPU 1: <tt>rcu_read_lock()</tt>
727 <li> <font color="ffffff">CPU 1: <tt>q = rcu_dereference(gp);
728 /* Might return p if no memory barrier. */</tt>
730 <li> <font color="ffffff">CPU 0: <tt>synchronize_rcu()</tt> returns.
732 <li> <font color="ffffff">CPU 0: <tt>kfree(p);</tt>
734 <li> <font color="ffffff">
735 CPU 1: <tt>do_something_with(q->a); /* Boom!!! */</tt>
737 <li> <font color="ffffff">CPU 1: <tt>rcu_read_unlock()</tt>
741 <p><font color="ffffff">
742 And similarly, without a memory barrier between the beginning of the
743 grace period and the beginning of the RCU read-side critical section,
744 CPU 1 might end up accessing the freelist.
747 <p><font color="ffffff">
748 The “as if” rule of course applies, so that any
749 implementation that acts as if the appropriate memory barriers
750 were in place is a correct implementation.
751 That said, it is much easier to fool yourself into believing
752 that you have adhered to the as-if rule than it is to actually
755 <tr><td> </td></tr>
759 <tr><th> </th></tr>
760 <tr><th align="left">Quick Quiz:</th></tr>
762 You claim that <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>
763 generate absolutely no code in some kernel builds.
764 This means that the compiler might arbitrarily rearrange consecutive
765 RCU read-side critical sections.
766 Given such rearrangement, if a given RCU read-side critical section
767 is done, how can you be sure that all prior RCU read-side critical
769 Won't the compiler rearrangements make that impossible to determine?
771 <tr><th align="left">Answer:</th></tr>
772 <tr><td bgcolor="#ffffff"><font color="ffffff">
773 In cases where <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>
774 generate absolutely no code, RCU infers quiescent states only at
775 special locations, for example, within the scheduler.
776 Because calls to <tt>schedule()</tt> had better prevent calling-code
777 accesses to shared variables from being rearranged across the call to
778 <tt>schedule()</tt>, if RCU detects the end of a given RCU read-side
779 critical section, it will necessarily detect the end of all prior
780 RCU read-side critical sections, no matter how aggressively the
781 compiler scrambles the code.
784 <p><font color="ffffff">
785 Again, this all assumes that the compiler cannot scramble code across
786 calls to the scheduler, out of interrupt handlers, into the idle loop,
787 into user-mode code, and so on.
788 But if your kernel build allows that sort of scrambling, you have broken
789 far more than just RCU!
791 <tr><td> </td></tr>
795 Note that these memory-barrier requirements do not replace the fundamental
796 RCU requirement that a grace period wait for all pre-existing readers.
797 On the contrary, the memory barriers called out in this section must operate in
798 such a way as to <i>enforce</i> this fundamental requirement.
799 Of course, different implementations enforce this requirement in different
800 ways, but enforce it they must.
802 <h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
805 The common-case RCU primitives are unconditional.
806 They are invoked, they do their job, and they return, with no possibility
807 of error, and no need to retry.
808 This is a key RCU design philosophy.
811 However, this philosophy is pragmatic rather than pigheaded.
812 If someone comes up with a good justification for a particular conditional
813 RCU primitive, it might well be implemented and added.
814 After all, this guarantee was reverse-engineered, not premeditated.
815 The unconditional nature of the RCU primitives was initially an
816 accident of implementation, and later experience with synchronization
817 primitives with conditional primitives caused me to elevate this
818 accident to a guarantee.
819 Therefore, the justification for adding a conditional primitive to
820 RCU would need to be based on detailed and compelling use cases.
822 <h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
825 As far as RCU is concerned, it is always possible to carry out an
826 update within an RCU read-side critical section.
827 For example, that RCU read-side critical section might search for
828 a given data element, and then might acquire the update-side
829 spinlock in order to update that element, all while remaining
830 in that RCU read-side critical section.
831 Of course, it is necessary to exit the RCU read-side critical section
832 before invoking <tt>synchronize_rcu()</tt>, however, this
833 inconvenience can be avoided through use of the
834 <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
835 described later in this document.
838 <tr><th> </th></tr>
839 <tr><th align="left">Quick Quiz:</th></tr>
841 But how does the upgrade-to-write operation exclude other readers?
843 <tr><th align="left">Answer:</th></tr>
844 <tr><td bgcolor="#ffffff"><font color="ffffff">
845 It doesn't, just like normal RCU updates, which also do not exclude
848 <tr><td> </td></tr>
852 This guarantee allows lookup code to be shared between read-side
853 and update-side code, and was premeditated, appearing in the earliest
854 DYNIX/ptx RCU documentation.
856 <h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
859 RCU provides extremely lightweight readers, and its read-side guarantees,
860 though quite useful, are correspondingly lightweight.
861 It is therefore all too easy to assume that RCU is guaranteeing more
863 Of course, the list of things that RCU does not guarantee is infinitely
864 long, however, the following sections list a few non-guarantees that
865 have caused confusion.
866 Except where otherwise noted, these non-guarantees were premeditated.
869 <li> <a href="#Readers Impose Minimal Ordering">
870 Readers Impose Minimal Ordering</a>
871 <li> <a href="#Readers Do Not Exclude Updaters">
872 Readers Do Not Exclude Updaters</a>
873 <li> <a href="#Updaters Only Wait For Old Readers">
874 Updaters Only Wait For Old Readers</a>
875 <li> <a href="#Grace Periods Don't Partition Read-Side Critical Sections">
876 Grace Periods Don't Partition Read-Side Critical Sections</a>
877 <li> <a href="#Read-Side Critical Sections Don't Partition Grace Periods">
878 Read-Side Critical Sections Don't Partition Grace Periods</a>
879 <li> <a href="#Disabling Preemption Does Not Block Grace Periods">
880 Disabling Preemption Does Not Block Grace Periods</a>
883 <h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
886 Reader-side markers such as <tt>rcu_read_lock()</tt> and
887 <tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
888 except through their interaction with the grace-period APIs such as
889 <tt>synchronize_rcu()</tt>.
890 To see this, consider the following pair of threads:
904 11 void thread1(void)
907 14 r1 = READ_ONCE(y);
908 15 rcu_read_unlock();
910 17 r2 = READ_ONCE(x);
911 18 rcu_read_unlock();
917 After <tt>thread0()</tt> and <tt>thread1()</tt> execute
918 concurrently, it is quite possible to have
922 (r1 == 1 && r2 == 0)
926 (that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
927 which would not be possible if <tt>rcu_read_lock()</tt> and
928 <tt>rcu_read_unlock()</tt> had much in the way of ordering
930 But they do not, so the CPU is within its rights
931 to do significant reordering.
932 This is by design: Any significant ordering constraints would slow down
933 these fast-path APIs.
936 <tr><th> </th></tr>
937 <tr><th align="left">Quick Quiz:</th></tr>
939 Can't the compiler also reorder this code?
941 <tr><th align="left">Answer:</th></tr>
942 <tr><td bgcolor="#ffffff"><font color="ffffff">
943 No, the volatile casts in <tt>READ_ONCE()</tt> and
944 <tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
945 this particular case.
947 <tr><td> </td></tr>
950 <h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
953 Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
955 All they do is to prevent grace periods from ending.
956 The following example illustrates this:
965 6 do_something_with_nonzero_x();
967 8 WARN_ON(!r2); /* BUG!!! */
969 10 rcu_read_unlock();
972 13 void thread1(void)
974 15 spin_lock(&my_lock);
977 18 spin_unlock(&my_lock);
983 If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
984 excluded the <tt>thread1()</tt> function's update,
985 the <tt>WARN_ON()</tt> could never fire.
986 But the fact is that <tt>rcu_read_lock()</tt> does not exclude
987 much of anything aside from subsequent grace periods, of which
988 <tt>thread1()</tt> has none, so the
989 <tt>WARN_ON()</tt> can and does fire.
991 <h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
994 It might be tempting to assume that after <tt>synchronize_rcu()</tt>
995 completes, there are no readers executing.
996 This temptation must be avoided because
997 new readers can start immediately after <tt>synchronize_rcu()</tt>
998 starts, and <tt>synchronize_rcu()</tt> is under no
999 obligation to wait for these new readers.
1002 <tr><th> </th></tr>
1003 <tr><th align="left">Quick Quiz:</th></tr>
1005 Suppose that synchronize_rcu() did wait until <i>all</i>
1006 readers had completed instead of waiting only on
1007 pre-existing readers.
1008 For how long would the updater be able to rely on there
1011 <tr><th align="left">Answer:</th></tr>
1012 <tr><td bgcolor="#ffffff"><font color="ffffff">
1014 Even if <tt>synchronize_rcu()</tt> were to wait until
1015 all readers had completed, a new reader might start immediately after
1016 <tt>synchronize_rcu()</tt> completed.
1017 Therefore, the code following
1018 <tt>synchronize_rcu()</tt> can <i>never</i> rely on there being
1021 <tr><td> </td></tr>
1024 <h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
1025 Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
1028 It is tempting to assume that if any part of one RCU read-side critical
1029 section precedes a given grace period, and if any part of another RCU
1030 read-side critical section follows that same grace period, then all of
1031 the first RCU read-side critical section must precede all of the second.
1032 However, this just isn't the case: A single grace period does not
1033 partition the set of RCU read-side critical sections.
1034 An example of this situation can be illustrated as follows, where
1035 <tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
1039 1 void thread0(void)
1044 6 rcu_read_unlock();
1047 9 void thread1(void)
1049 11 r1 = READ_ONCE(a);
1050 12 synchronize_rcu();
1051 13 WRITE_ONCE(c, 1);
1054 16 void thread2(void)
1057 19 r2 = READ_ONCE(b);
1058 20 r3 = READ_ONCE(c);
1059 21 rcu_read_unlock();
1065 It turns out that the outcome:
1069 (r1 == 1 && r2 == 0 && r3 == 1)
1073 is entirely possible.
1074 The following figure show how this can happen, with each circled
1075 <tt>QS</tt> indicating the point at which RCU recorded a
1076 <i>quiescent state</i> for each thread, that is, a state in which
1077 RCU knows that the thread cannot be in the midst of an RCU read-side
1078 critical section that started before the current grace period:
1080 <p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
1083 If it is necessary to partition RCU read-side critical sections in this
1084 manner, it is necessary to use two grace periods, where the first
1085 grace period is known to end before the second grace period starts:
1089 1 void thread0(void)
1094 6 rcu_read_unlock();
1097 9 void thread1(void)
1099 11 r1 = READ_ONCE(a);
1100 12 synchronize_rcu();
1101 13 WRITE_ONCE(c, 1);
1104 16 void thread2(void)
1106 18 r2 = READ_ONCE(c);
1107 19 synchronize_rcu();
1108 20 WRITE_ONCE(d, 1);
1111 23 void thread3(void)
1114 26 r3 = READ_ONCE(b);
1115 27 r4 = READ_ONCE(d);
1116 28 rcu_read_unlock();
1122 Here, if <tt>(r1 == 1)</tt>, then
1123 <tt>thread0()</tt>'s write to <tt>b</tt> must happen
1124 before the end of <tt>thread1()</tt>'s grace period.
1125 If in addition <tt>(r4 == 1)</tt>, then
1126 <tt>thread3()</tt>'s read from <tt>b</tt> must happen
1127 after the beginning of <tt>thread2()</tt>'s grace period.
1128 If it is also the case that <tt>(r2 == 1)</tt>, then the
1129 end of <tt>thread1()</tt>'s grace period must precede the
1130 beginning of <tt>thread2()</tt>'s grace period.
1131 This mean that the two RCU read-side critical sections cannot overlap,
1132 guaranteeing that <tt>(r3 == 1)</tt>.
1133 As a result, the outcome:
1137 (r1 == 1 && r2 == 1 && r3 == 0 && r4 == 1)
1144 This non-requirement was also non-premeditated, but became apparent
1145 when studying RCU's interaction with memory ordering.
1147 <h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
1148 Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
1151 It is also tempting to assume that if an RCU read-side critical section
1152 happens between a pair of grace periods, then those grace periods cannot
1154 However, this temptation leads nowhere good, as can be illustrated by
1155 the following, with all variables initially zero:
1159 1 void thread0(void)
1164 6 rcu_read_unlock();
1167 9 void thread1(void)
1169 11 r1 = READ_ONCE(a);
1170 12 synchronize_rcu();
1171 13 WRITE_ONCE(c, 1);
1174 16 void thread2(void)
1177 19 WRITE_ONCE(d, 1);
1178 20 r2 = READ_ONCE(c);
1179 21 rcu_read_unlock();
1182 24 void thread3(void)
1184 26 r3 = READ_ONCE(d);
1185 27 synchronize_rcu();
1186 28 WRITE_ONCE(e, 1);
1189 31 void thread4(void)
1192 34 r4 = READ_ONCE(b);
1193 35 r5 = READ_ONCE(e);
1194 36 rcu_read_unlock();
1200 In this case, the outcome:
1204 (r1 == 1 && r2 == 1 && r3 == 1 && r4 == 0 && r5 == 1)
1208 is entirely possible, as illustrated below:
1210 <p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
1213 Again, an RCU read-side critical section can overlap almost all of a
1214 given grace period, just so long as it does not overlap the entire
1216 As a result, an RCU read-side critical section cannot partition a pair
1217 of RCU grace periods.
1220 <tr><th> </th></tr>
1221 <tr><th align="left">Quick Quiz:</th></tr>
1223 How long a sequence of grace periods, each separated by an RCU
1224 read-side critical section, would be required to partition the RCU
1225 read-side critical sections at the beginning and end of the chain?
1227 <tr><th align="left">Answer:</th></tr>
1228 <tr><td bgcolor="#ffffff"><font color="ffffff">
1229 In theory, an infinite number.
1230 In practice, an unknown number that is sensitive to both implementation
1231 details and timing considerations.
1232 Therefore, even in practice, RCU users must abide by the
1233 theoretical rather than the practical answer.
1235 <tr><td> </td></tr>
1238 <h3><a name="Disabling Preemption Does Not Block Grace Periods">
1239 Disabling Preemption Does Not Block Grace Periods</a></h3>
1242 There was a time when disabling preemption on any given CPU would block
1243 subsequent grace periods.
1244 However, this was an accident of implementation and is not a requirement.
1245 And in the current Linux-kernel implementation, disabling preemption
1246 on a given CPU in fact does not block grace periods, as Oleg Nesterov
1247 <a href="https://lkml.kernel.org/g/20150614193825.GA19582@redhat.com">demonstrated</a>.
1250 If you need a preempt-disable region to block grace periods, you need to add
1251 <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
1256 1 preempt_disable();
1259 4 rcu_read_unlock();
1262 7 /* Spinlocks implicitly disable preemption. */
1263 8 spin_lock(&mylock);
1266 11 rcu_read_unlock();
1267 12 spin_unlock(&mylock);
1272 In theory, you could enter the RCU read-side critical section first,
1273 but it is more efficient to keep the entire RCU read-side critical
1274 section contained in the preempt-disable region as shown above.
1275 Of course, RCU read-side critical sections that extend outside of
1276 preempt-disable regions will work correctly, but such critical sections
1277 can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
1279 And no, this is <i>not</i> an invitation to enclose all of your RCU
1280 read-side critical sections within preempt-disable regions, because
1281 doing so would degrade real-time response.
1284 This non-requirement appeared with preemptible RCU.
1285 If you need a grace period that waits on non-preemptible code regions, use
1286 <a href="#Sched Flavor">RCU-sched</a>.
1288 <h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
1291 These parallelism facts of life are by no means specific to RCU, but
1292 the RCU implementation must abide by them.
1293 They therefore bear repeating:
1296 <li> Any CPU or task may be delayed at any time,
1297 and any attempts to avoid these delays by disabling
1298 preemption, interrupts, or whatever are completely futile.
1299 This is most obvious in preemptible user-level
1300 environments and in virtualized environments (where
1301 a given guest OS's VCPUs can be preempted at any time by
1302 the underlying hypervisor), but can also happen in bare-metal
1303 environments due to ECC errors, NMIs, and other hardware
1305 Although a delay of more than about 20 seconds can result
1306 in splats, the RCU implementation is obligated to use
1307 algorithms that can tolerate extremely long delays, but where
1308 “extremely long” is not long enough to allow
1309 wrap-around when incrementing a 64-bit counter.
1310 <li> Both the compiler and the CPU can reorder memory accesses.
1311 Where it matters, RCU must use compiler directives and
1312 memory-barrier instructions to preserve ordering.
1313 <li> Conflicting writes to memory locations in any given cache line
1314 will result in expensive cache misses.
1315 Greater numbers of concurrent writes and more-frequent
1316 concurrent writes will result in more dramatic slowdowns.
1317 RCU is therefore obligated to use algorithms that have
1318 sufficient locality to avoid significant performance and
1319 scalability problems.
1320 <li> As a rough rule of thumb, only one CPU's worth of processing
1321 may be carried out under the protection of any given exclusive
1323 RCU must therefore use scalable locking designs.
1324 <li> Counters are finite, especially on 32-bit systems.
1325 RCU's use of counters must therefore tolerate counter wrap,
1326 or be designed such that counter wrap would take way more
1327 time than a single system is likely to run.
1328 An uptime of ten years is quite possible, a runtime
1329 of a century much less so.
1330 As an example of the latter, RCU's dyntick-idle nesting counter
1331 allows 54 bits for interrupt nesting level (this counter
1332 is 64 bits even on a 32-bit system).
1333 Overflowing this counter requires 2<sup>54</sup>
1334 half-interrupts on a given CPU without that CPU ever going idle.
1335 If a half-interrupt happened every microsecond, it would take
1336 570 years of runtime to overflow this counter, which is currently
1337 believed to be an acceptably long time.
1338 <li> Linux systems can have thousands of CPUs running a single
1339 Linux kernel in a single shared-memory environment.
1340 RCU must therefore pay close attention to high-end scalability.
1344 This last parallelism fact of life means that RCU must pay special
1345 attention to the preceding facts of life.
1346 The idea that Linux might scale to systems with thousands of CPUs would
1347 have been met with some skepticism in the 1990s, but these requirements
1348 would have otherwise have been unsurprising, even in the early 1990s.
1350 <h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
1353 These sections list quality-of-implementation requirements.
1354 Although an RCU implementation that ignores these requirements could
1355 still be used, it would likely be subject to limitations that would
1356 make it inappropriate for industrial-strength production use.
1357 Classes of quality-of-implementation requirements are as follows:
1360 <li> <a href="#Specialization">Specialization</a>
1361 <li> <a href="#Performance and Scalability">Performance and Scalability</a>
1362 <li> <a href="#Composability">Composability</a>
1363 <li> <a href="#Corner Cases">Corner Cases</a>
1367 These classes is covered in the following sections.
1369 <h3><a name="Specialization">Specialization</a></h3>
1372 RCU is and always has been intended primarily for read-mostly situations,
1373 which means that RCU's read-side primitives are optimized, often at the
1374 expense of its update-side primitives.
1375 Experience thus far is captured by the following list of situations:
1378 <li> Read-mostly data, where stale and inconsistent data is not
1379 a problem: RCU works great!
1380 <li> Read-mostly data, where data must be consistent:
1382 <li> Read-write data, where data must be consistent:
1383 RCU <i>might</i> work OK.
1385 <li> Write-mostly data, where data must be consistent:
1386 RCU is very unlikely to be the right tool for the job,
1387 with the following exceptions, where RCU can provide:
1389 <li> Existence guarantees for update-friendly mechanisms.
1390 <li> Wait-free read-side primitives for real-time use.
1395 This focus on read-mostly situations means that RCU must interoperate
1396 with other synchronization primitives.
1397 For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
1398 examples discussed earlier use RCU to protect readers and locking to
1399 coordinate updaters.
1400 However, the need extends much farther, requiring that a variety of
1401 synchronization primitives be legal within RCU read-side critical sections,
1402 including spinlocks, sequence locks, atomic operations, reference
1403 counters, and memory barriers.
1406 <tr><th> </th></tr>
1407 <tr><th align="left">Quick Quiz:</th></tr>
1409 What about sleeping locks?
1411 <tr><th align="left">Answer:</th></tr>
1412 <tr><td bgcolor="#ffffff"><font color="ffffff">
1413 These are forbidden within Linux-kernel RCU read-side critical
1414 sections because it is not legal to place a quiescent state
1415 (in this case, voluntary context switch) within an RCU read-side
1417 However, sleeping locks may be used within userspace RCU read-side
1418 critical sections, and also within Linux-kernel sleepable RCU
1419 <a href="#Sleepable RCU"><font color="ffffff">(SRCU)</font></a>
1420 read-side critical sections.
1421 In addition, the -rt patchset turns spinlocks into a
1422 sleeping locks so that the corresponding critical sections
1423 can be preempted, which also means that these sleeplockified
1424 spinlocks (but not other sleeping locks!) may be acquire within
1425 -rt-Linux-kernel RCU read-side critical sections.
1428 <p><font color="ffffff">
1429 Note that it <i>is</i> legal for a normal RCU read-side
1430 critical section to conditionally acquire a sleeping locks
1431 (as in <tt>mutex_trylock()</tt>), but only as long as it does
1432 not loop indefinitely attempting to conditionally acquire that
1434 The key point is that things like <tt>mutex_trylock()</tt>
1435 either return with the mutex held, or return an error indication if
1436 the mutex was not immediately available.
1437 Either way, <tt>mutex_trylock()</tt> returns immediately without
1440 <tr><td> </td></tr>
1444 It often comes as a surprise that many algorithms do not require a
1445 consistent view of data, but many can function in that mode,
1446 with network routing being the poster child.
1447 Internet routing algorithms take significant time to propagate
1448 updates, so that by the time an update arrives at a given system,
1449 that system has been sending network traffic the wrong way for
1450 a considerable length of time.
1451 Having a few threads continue to send traffic the wrong way for a
1452 few more milliseconds is clearly not a problem: In the worst case,
1453 TCP retransmissions will eventually get the data where it needs to go.
1454 In general, when tracking the state of the universe outside of the
1455 computer, some level of inconsistency must be tolerated due to
1456 speed-of-light delays if nothing else.
1459 Furthermore, uncertainty about external state is inherent in many cases.
1460 For example, a pair of veternarians might use heartbeat to determine
1461 whether or not a given cat was alive.
1462 But how long should they wait after the last heartbeat to decide that
1463 the cat is in fact dead?
1464 Waiting less than 400 milliseconds makes no sense because this would
1465 mean that a relaxed cat would be considered to cycle between death
1466 and life more than 100 times per minute.
1467 Moreover, just as with human beings, a cat's heart might stop for
1468 some period of time, so the exact wait period is a judgment call.
1469 One of our pair of veternarians might wait 30 seconds before pronouncing
1470 the cat dead, while the other might insist on waiting a full minute.
1471 The two veternarians would then disagree on the state of the cat during
1472 the final 30 seconds of the minute following the last heartbeat.
1475 Interestingly enough, this same situation applies to hardware.
1476 When push comes to shove, how do we tell whether or not some
1477 external server has failed?
1478 We send messages to it periodically, and declare it failed if we
1479 don't receive a response within a given period of time.
1480 Policy decisions can usually tolerate short
1481 periods of inconsistency.
1482 The policy was decided some time ago, and is only now being put into
1483 effect, so a few milliseconds of delay is normally inconsequential.
1486 However, there are algorithms that absolutely must see consistent data.
1487 For example, the translation between a user-level SystemV semaphore
1488 ID to the corresponding in-kernel data structure is protected by RCU,
1489 but it is absolutely forbidden to update a semaphore that has just been
1491 In the Linux kernel, this need for consistency is accommodated by acquiring
1492 spinlocks located in the in-kernel data structure from within
1493 the RCU read-side critical section, and this is indicated by the
1494 green box in the figure above.
1495 Many other techniques may be used, and are in fact used within the
1499 In short, RCU is not required to maintain consistency, and other
1500 mechanisms may be used in concert with RCU when consistency is required.
1501 RCU's specialization allows it to do its job extremely well, and its
1502 ability to interoperate with other synchronization mechanisms allows
1503 the right mix of synchronization tools to be used for a given job.
1505 <h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
1508 Energy efficiency is a critical component of performance today,
1509 and Linux-kernel RCU implementations must therefore avoid unnecessarily
1510 awakening idle CPUs.
1511 I cannot claim that this requirement was premeditated.
1512 In fact, I learned of it during a telephone conversation in which I
1513 was given “frank and open” feedback on the importance
1514 of energy efficiency in battery-powered systems and on specific
1515 energy-efficiency shortcomings of the Linux-kernel RCU implementation.
1516 In my experience, the battery-powered embedded community will consider
1517 any unnecessary wakeups to be extremely unfriendly acts.
1518 So much so that mere Linux-kernel-mailing-list posts are
1519 insufficient to vent their ire.
1522 Memory consumption is not particularly important for in most
1523 situations, and has become decreasingly
1524 so as memory sizes have expanded and memory
1525 costs have plummeted.
1526 However, as I learned from Matt Mackall's
1527 <a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
1528 efforts, memory footprint is critically important on single-CPU systems with
1529 non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
1530 <a href="https://lkml.kernel.org/g/20090113221724.GA15307@linux.vnet.ibm.com">tiny RCU</a>
1532 Josh Triplett has since taken over the small-memory banner with his
1533 <a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
1534 project, which resulted in
1535 <a href="#Sleepable RCU">SRCU</a>
1536 becoming optional for those kernels not needing it.
1539 The remaining performance requirements are, for the most part,
1541 For example, in keeping with RCU's read-side specialization,
1542 <tt>rcu_dereference()</tt> should have negligible overhead (for
1543 example, suppression of a few minor compiler optimizations).
1544 Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
1545 <tt>rcu_read_unlock()</tt> should have exactly zero overhead.
1548 In preemptible environments, in the case where the RCU read-side
1549 critical section was not preempted (as will be the case for the
1550 highest-priority real-time process), <tt>rcu_read_lock()</tt> and
1551 <tt>rcu_read_unlock()</tt> should have minimal overhead.
1552 In particular, they should not contain atomic read-modify-write
1553 operations, memory-barrier instructions, preemption disabling,
1554 interrupt disabling, or backwards branches.
1555 However, in the case where the RCU read-side critical section was preempted,
1556 <tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
1557 This is why it is better to nest an RCU read-side critical section
1558 within a preempt-disable region than vice versa, at least in cases
1559 where that critical section is short enough to avoid unduly degrading
1560 real-time latencies.
1563 The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
1564 optimized for throughput.
1565 It may therefore incur several milliseconds of latency in addition to
1566 the duration of the longest RCU read-side critical section.
1567 On the other hand, multiple concurrent invocations of
1568 <tt>synchronize_rcu()</tt> are required to use batching optimizations
1569 so that they can be satisfied by a single underlying grace-period-wait
1571 For example, in the Linux kernel, it is not unusual for a single
1572 grace-period-wait operation to serve more than
1573 <a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
1574 of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
1575 overhead down to nearly zero.
1576 However, the grace-period optimization is also required to avoid
1577 measurable degradation of real-time scheduling and interrupt latencies.
1580 In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
1581 latencies are unacceptable.
1582 In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
1583 instead, reducing the grace-period latency down to a few tens of
1584 microseconds on small systems, at least in cases where the RCU read-side
1585 critical sections are short.
1586 There are currently no special latency requirements for
1587 <tt>synchronize_rcu_expedited()</tt> on large systems, but,
1588 consistent with the empirical nature of the RCU specification,
1589 that is subject to change.
1590 However, there most definitely are scalability requirements:
1591 A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
1592 CPUs should at least make reasonable forward progress.
1593 In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
1594 is permitted to impose modest degradation of real-time latency
1595 on non-idle online CPUs.
1596 That said, it will likely be necessary to take further steps to reduce this
1597 degradation, hopefully to roughly that of a scheduling-clock interrupt.
1600 There are a number of situations where even
1601 <tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
1602 latency is unacceptable.
1603 In these situations, the asynchronous <tt>call_rcu()</tt> can be
1604 used in place of <tt>synchronize_rcu()</tt> as follows:
1611 4 struct rcu_head rh;
1614 7 static void remove_gp_cb(struct rcu_head *rhp)
1616 9 struct foo *p = container_of(rhp, struct foo, rh);
1621 14 bool remove_gp_asynchronous(void)
1625 18 spin_lock(&gp_lock);
1626 19 p = rcu_dereference(gp);
1628 21 spin_unlock(&gp_lock);
1631 24 rcu_assign_pointer(gp, NULL);
1632 25 call_rcu(&p->rh, remove_gp_cb);
1633 26 spin_unlock(&gp_lock);
1640 A definition of <tt>struct foo</tt> is finally needed, and appears
1642 The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
1643 on line 25, and will be invoked after the end of a subsequent
1645 This gets the same effect as <tt>remove_gp_synchronous()</tt>,
1646 but without forcing the updater to wait for a grace period to elapse.
1647 The <tt>call_rcu()</tt> function may be used in a number of
1648 situations where neither <tt>synchronize_rcu()</tt> nor
1649 <tt>synchronize_rcu_expedited()</tt> would be legal,
1650 including within preempt-disable code, <tt>local_bh_disable()</tt> code,
1651 interrupt-disable code, and interrupt handlers.
1652 However, even <tt>call_rcu()</tt> is illegal within NMI handlers
1653 and from idle and offline CPUs.
1654 The callback function (<tt>remove_gp_cb()</tt> in this case) will be
1655 executed within softirq (software interrupt) environment within the
1657 either within a real softirq handler or under the protection
1658 of <tt>local_bh_disable()</tt>.
1659 In both the Linux kernel and in userspace, it is bad practice to
1660 write an RCU callback function that takes too long.
1661 Long-running operations should be relegated to separate threads or
1662 (in the Linux kernel) workqueues.
1665 <tr><th> </th></tr>
1666 <tr><th align="left">Quick Quiz:</th></tr>
1668 Why does line 19 use <tt>rcu_access_pointer()</tt>?
1669 After all, <tt>call_rcu()</tt> on line 25 stores into the
1670 structure, which would interact badly with concurrent insertions.
1671 Doesn't this mean that <tt>rcu_dereference()</tt> is required?
1673 <tr><th align="left">Answer:</th></tr>
1674 <tr><td bgcolor="#ffffff"><font color="ffffff">
1675 Presumably the <tt>->gp_lock</tt> acquired on line 18 excludes
1676 any changes, including any insertions that <tt>rcu_dereference()</tt>
1677 would protect against.
1678 Therefore, any insertions will be delayed until after
1679 <tt>->gp_lock</tt>
1680 is released on line 25, which in turn means that
1681 <tt>rcu_access_pointer()</tt> suffices.
1683 <tr><td> </td></tr>
1687 However, all that <tt>remove_gp_cb()</tt> is doing is
1688 invoking <tt>kfree()</tt> on the data element.
1689 This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
1690 which allows “fire and forget” operation as shown below:
1697 4 struct rcu_head rh;
1700 7 bool remove_gp_faf(void)
1704 11 spin_lock(&gp_lock);
1705 12 p = rcu_dereference(gp);
1707 14 spin_unlock(&gp_lock);
1710 17 rcu_assign_pointer(gp, NULL);
1711 18 kfree_rcu(p, rh);
1712 19 spin_unlock(&gp_lock);
1719 Note that <tt>remove_gp_faf()</tt> simply invokes
1720 <tt>kfree_rcu()</tt> and proceeds, without any need to pay any
1721 further attention to the subsequent grace period and <tt>kfree()</tt>.
1722 It is permissible to invoke <tt>kfree_rcu()</tt> from the same
1723 environments as for <tt>call_rcu()</tt>.
1724 Interestingly enough, DYNIX/ptx had the equivalents of
1725 <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
1726 <tt>synchronize_rcu()</tt>.
1727 This was due to the fact that RCU was not heavily used within DYNIX/ptx,
1728 so the very few places that needed something like
1729 <tt>synchronize_rcu()</tt> simply open-coded it.
1732 <tr><th> </th></tr>
1733 <tr><th align="left">Quick Quiz:</th></tr>
1735 Earlier it was claimed that <tt>call_rcu()</tt> and
1736 <tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
1738 But how can that be correct, given that the invocation of the callback
1739 and the freeing of the memory (respectively) must still wait for
1740 a grace period to elapse?
1742 <tr><th align="left">Answer:</th></tr>
1743 <tr><td bgcolor="#ffffff"><font color="ffffff">
1744 We could define things this way, but keep in mind that this sort of
1745 definition would say that updates in garbage-collected languages
1746 cannot complete until the next time the garbage collector runs,
1747 which does not seem at all reasonable.
1748 The key point is that in most cases, an updater using either
1749 <tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
1750 next update as soon as it has invoked <tt>call_rcu()</tt> or
1751 <tt>kfree_rcu()</tt>, without having to wait for a subsequent
1754 <tr><td> </td></tr>
1758 But what if the updater must wait for the completion of code to be
1759 executed after the end of the grace period, but has other tasks
1760 that can be carried out in the meantime?
1761 The polling-style <tt>get_state_synchronize_rcu()</tt> and
1762 <tt>cond_synchronize_rcu()</tt> functions may be used for this
1763 purpose, as shown below:
1767 1 bool remove_gp_poll(void)
1772 6 spin_lock(&gp_lock);
1773 7 p = rcu_access_pointer(gp);
1775 9 spin_unlock(&gp_lock);
1778 12 rcu_assign_pointer(gp, NULL);
1779 13 spin_unlock(&gp_lock);
1780 14 s = get_state_synchronize_rcu();
1781 15 do_something_while_waiting();
1782 16 cond_synchronize_rcu(s);
1790 On line 14, <tt>get_state_synchronize_rcu()</tt> obtains a
1791 “cookie” from RCU,
1792 then line 15 carries out other tasks,
1793 and finally, line 16 returns immediately if a grace period has
1794 elapsed in the meantime, but otherwise waits as required.
1795 The need for <tt>get_state_synchronize_rcu</tt> and
1796 <tt>cond_synchronize_rcu()</tt> has appeared quite recently,
1797 so it is too early to tell whether they will stand the test of time.
1800 RCU thus provides a range of tools to allow updaters to strike the
1801 required tradeoff between latency, flexibility and CPU overhead.
1803 <h3><a name="Composability">Composability</a></h3>
1806 Composability has received much attention in recent years, perhaps in part
1807 due to the collision of multicore hardware with object-oriented techniques
1808 designed in single-threaded environments for single-threaded use.
1809 And in theory, RCU read-side critical sections may be composed, and in
1810 fact may be nested arbitrarily deeply.
1811 In practice, as with all real-world implementations of composable
1812 constructs, there are limitations.
1815 Implementations of RCU for which <tt>rcu_read_lock()</tt>
1816 and <tt>rcu_read_unlock()</tt> generate no code, such as
1817 Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
1818 nested arbitrarily deeply.
1819 After all, there is no overhead.
1820 Except that if all these instances of <tt>rcu_read_lock()</tt>
1821 and <tt>rcu_read_unlock()</tt> are visible to the compiler,
1822 compilation will eventually fail due to exhausting memory,
1823 mass storage, or user patience, whichever comes first.
1824 If the nesting is not visible to the compiler, as is the case with
1825 mutually recursive functions each in its own translation unit,
1826 stack overflow will result.
1827 If the nesting takes the form of loops, either the control variable
1828 will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
1829 Nevertheless, this class of RCU implementations is one
1830 of the most composable constructs in existence.
1833 RCU implementations that explicitly track nesting depth
1834 are limited by the nesting-depth counter.
1835 For example, the Linux kernel's preemptible RCU limits nesting to
1837 This should suffice for almost all practical purposes.
1838 That said, a consecutive pair of RCU read-side critical sections
1839 between which there is an operation that waits for a grace period
1840 cannot be enclosed in another RCU read-side critical section.
1841 This is because it is not legal to wait for a grace period within
1842 an RCU read-side critical section: To do so would result either
1844 in RCU implicitly splitting the enclosing RCU read-side critical
1845 section, neither of which is conducive to a long-lived and prosperous
1849 It is worth noting that RCU is not alone in limiting composability.
1850 For example, many transactional-memory implementations prohibit
1851 composing a pair of transactions separated by an irrevocable
1852 operation (for example, a network receive operation).
1853 For another example, lock-based critical sections can be composed
1854 surprisingly freely, but only if deadlock is avoided.
1857 In short, although RCU read-side critical sections are highly composable,
1858 care is required in some situations, just as is the case for any other
1859 composable synchronization mechanism.
1861 <h3><a name="Corner Cases">Corner Cases</a></h3>
1864 A given RCU workload might have an endless and intense stream of
1865 RCU read-side critical sections, perhaps even so intense that there
1866 was never a point in time during which there was not at least one
1867 RCU read-side critical section in flight.
1868 RCU cannot allow this situation to block grace periods: As long as
1869 all the RCU read-side critical sections are finite, grace periods
1870 must also be finite.
1873 That said, preemptible RCU implementations could potentially result
1874 in RCU read-side critical sections being preempted for long durations,
1875 which has the effect of creating a long-duration RCU read-side
1877 This situation can arise only in heavily loaded systems, but systems using
1878 real-time priorities are of course more vulnerable.
1879 Therefore, RCU priority boosting is provided to help deal with this
1881 That said, the exact requirements on RCU priority boosting will likely
1882 evolve as more experience accumulates.
1885 Other workloads might have very high update rates.
1886 Although one can argue that such workloads should instead use
1887 something other than RCU, the fact remains that RCU must
1888 handle such workloads gracefully.
1889 This requirement is another factor driving batching of grace periods,
1890 but it is also the driving force behind the checks for large numbers
1891 of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
1892 Finally, high update rates should not delay RCU read-side critical
1893 sections, although some read-side delays can occur when using
1894 <tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
1895 of <tt>try_stop_cpus()</tt>.
1896 (In the future, <tt>synchronize_rcu_expedited()</tt> will be
1897 converted to use lighter-weight inter-processor interrupts (IPIs),
1898 but this will still disturb readers, though to a much smaller degree.)
1901 Although all three of these corner cases were understood in the early
1902 1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
1904 in the early 2000s suddenly provided a much deeper appreciation of the
1905 high-update-rate corner case.
1906 This test also motivated addition of some RCU code to react to high update
1907 rates, for example, if a given CPU finds itself with more than 10,000
1908 RCU callbacks queued, it will cause RCU to take evasive action by
1909 more aggressively starting grace periods and more aggressively forcing
1910 completion of grace-period processing.
1911 This evasive action causes the grace period to complete more quickly,
1912 but at the cost of restricting RCU's batching optimizations, thus
1913 increasing the CPU overhead incurred by that grace period.
1915 <h2><a name="Software-Engineering Requirements">
1916 Software-Engineering Requirements</a></h2>
1919 Between Murphy's Law and “To err is human”, it is necessary to
1920 guard against mishaps and misuse:
1923 <li> It is all too easy to forget to use <tt>rcu_read_lock()</tt>
1924 everywhere that it is needed, so kernels built with
1925 <tt>CONFIG_PROVE_RCU=y</tt> will spat if
1926 <tt>rcu_dereference()</tt> is used outside of an
1927 RCU read-side critical section.
1928 Update-side code can use <tt>rcu_dereference_protected()</tt>,
1930 <a href="https://lwn.net/Articles/371986/">lockdep expression</a>
1931 to indicate what is providing the protection.
1932 If the indicated protection is not provided, a lockdep splat
1936 Code shared between readers and updaters can use
1937 <tt>rcu_dereference_check()</tt>, which also takes a
1938 lockdep expression, and emits a lockdep splat if neither
1939 <tt>rcu_read_lock()</tt> nor the indicated protection
1941 In addition, <tt>rcu_dereference_raw()</tt> is used in those
1942 (hopefully rare) cases where the required protection cannot
1943 be easily described.
1944 Finally, <tt>rcu_read_lock_held()</tt> is provided to
1945 allow a function to verify that it has been invoked within
1946 an RCU read-side critical section.
1947 I was made aware of this set of requirements shortly after Thomas
1948 Gleixner audited a number of RCU uses.
1949 <li> A given function might wish to check for RCU-related preconditions
1950 upon entry, before using any other RCU API.
1951 The <tt>rcu_lockdep_assert()</tt> does this job,
1952 asserting the expression in kernels having lockdep enabled
1953 and doing nothing otherwise.
1954 <li> It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
1955 and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
1956 substituting a simple assignment.
1957 To catch this sort of error, a given RCU-protected pointer may be
1958 tagged with <tt>__rcu</tt>, after which running sparse
1959 with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
1960 about simple-assignment accesses to that pointer.
1961 Arnd Bergmann made me aware of this requirement, and also
1963 <a href="https://lwn.net/Articles/376011/">patch series</a>.
1964 <li> Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
1965 will splat if a data element is passed to <tt>call_rcu()</tt>
1966 twice in a row, without a grace period in between.
1967 (This error is similar to a double free.)
1968 The corresponding <tt>rcu_head</tt> structures that are
1969 dynamically allocated are automatically tracked, but
1970 <tt>rcu_head</tt> structures allocated on the stack
1971 must be initialized with <tt>init_rcu_head_on_stack()</tt>
1972 and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
1973 Similarly, statically allocated non-stack <tt>rcu_head</tt>
1974 structures must be initialized with <tt>init_rcu_head()</tt>
1975 and cleaned up with <tt>destroy_rcu_head()</tt>.
1976 Mathieu Desnoyers made me aware of this requirement, and also
1978 <a href="https://lkml.kernel.org/g/20100319013024.GA28456@Krystal">patch</a>.
1979 <li> An infinite loop in an RCU read-side critical section will
1980 eventually trigger an RCU CPU stall warning splat, with
1981 the duration of “eventually” being controlled by the
1982 <tt>RCU_CPU_STALL_TIMEOUT</tt> <tt>Kconfig</tt> option, or,
1983 alternatively, by the
1984 <tt>rcupdate.rcu_cpu_stall_timeout</tt> boot/sysfs
1986 However, RCU is not obligated to produce this splat
1987 unless there is a grace period waiting on that particular
1988 RCU read-side critical section.
1990 Some extreme workloads might intentionally delay
1991 RCU grace periods, and systems running those workloads can
1992 be booted with <tt>rcupdate.rcu_cpu_stall_suppress</tt>
1993 to suppress the splats.
1994 This kernel parameter may also be set via <tt>sysfs</tt>.
1995 Furthermore, RCU CPU stall warnings are counter-productive
1996 during sysrq dumps and during panics.
1997 RCU therefore supplies the <tt>rcu_sysrq_start()</tt> and
1998 <tt>rcu_sysrq_end()</tt> API members to be called before
1999 and after long sysrq dumps.
2000 RCU also supplies the <tt>rcu_panic()</tt> notifier that is
2001 automatically invoked at the beginning of a panic to suppress
2002 further RCU CPU stall warnings.
2005 This requirement made itself known in the early 1990s, pretty
2006 much the first time that it was necessary to debug a CPU stall.
2007 That said, the initial implementation in DYNIX/ptx was quite
2008 generic in comparison with that of Linux.
2009 <li> Although it would be very good to detect pointers leaking out
2010 of RCU read-side critical sections, there is currently no
2011 good way of doing this.
2012 One complication is the need to distinguish between pointers
2013 leaking and pointers that have been handed off from RCU to
2014 some other synchronization mechanism, for example, reference
2016 <li> In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
2017 information is provided via both debugfs and event tracing.
2018 <li> Open-coded use of <tt>rcu_assign_pointer()</tt> and
2019 <tt>rcu_dereference()</tt> to create typical linked
2020 data structures can be surprisingly error-prone.
2021 Therefore, RCU-protected
2022 <a href="https://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
2023 and, more recently, RCU-protected
2024 <a href="https://lwn.net/Articles/612100/">hash tables</a>
2026 Many other special-purpose RCU-protected data structures are
2027 available in the Linux kernel and the userspace RCU library.
2028 <li> Some linked structures are created at compile time, but still
2029 require <tt>__rcu</tt> checking.
2030 The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
2032 <li> It is not necessary to use <tt>rcu_assign_pointer()</tt>
2033 when creating linked structures that are to be published via
2034 a single external pointer.
2035 The <tt>RCU_INIT_POINTER()</tt> macro is provided for
2036 this task and also for assigning <tt>NULL</tt> pointers
2041 This not a hard-and-fast list: RCU's diagnostic capabilities will
2042 continue to be guided by the number and type of usage bugs found
2043 in real-world RCU usage.
2045 <h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
2048 The Linux kernel provides an interesting environment for all kinds of
2049 software, including RCU.
2050 Some of the relevant points of interest are as follows:
2053 <li> <a href="#Configuration">Configuration</a>.
2054 <li> <a href="#Firmware Interface">Firmware Interface</a>.
2055 <li> <a href="#Early Boot">Early Boot</a>.
2056 <li> <a href="#Interrupts and NMIs">
2057 Interrupts and non-maskable interrupts (NMIs)</a>.
2058 <li> <a href="#Loadable Modules">Loadable Modules</a>.
2059 <li> <a href="#Hotplug CPU">Hotplug CPU</a>.
2060 <li> <a href="#Scheduler and RCU">Scheduler and RCU</a>.
2061 <li> <a href="#Tracing and RCU">Tracing and RCU</a>.
2062 <li> <a href="#Energy Efficiency">Energy Efficiency</a>.
2063 <li> <a href="#Memory Efficiency">Memory Efficiency</a>.
2064 <li> <a href="#Performance, Scalability, Response Time, and Reliability">
2065 Performance, Scalability, Response Time, and Reliability</a>.
2069 This list is probably incomplete, but it does give a feel for the
2070 most notable Linux-kernel complications.
2071 Each of the following sections covers one of the above topics.
2073 <h3><a name="Configuration">Configuration</a></h3>
2076 RCU's goal is automatic configuration, so that almost nobody
2077 needs to worry about RCU's <tt>Kconfig</tt> options.
2078 And for almost all users, RCU does in fact work well
2079 “out of the box.”
2082 However, there are specialized use cases that are handled by
2083 kernel boot parameters and <tt>Kconfig</tt> options.
2084 Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
2085 about new <tt>Kconfig</tt> options, which requires almost all of them
2086 be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
2089 This all should be quite obvious, but the fact remains that
2090 Linus Torvalds recently had to
2091 <a href="https://lkml.kernel.org/g/CA+55aFy4wcCwaL4okTs8wXhGZ5h-ibecy_Meg9C4MNQrUnwMcg@mail.gmail.com">remind</a>
2092 me of this requirement.
2094 <h3><a name="Firmware Interface">Firmware Interface</a></h3>
2097 In many cases, kernel obtains information about the system from the
2098 firmware, and sometimes things are lost in translation.
2099 Or the translation is accurate, but the original message is bogus.
2102 For example, some systems' firmware overreports the number of CPUs,
2103 sometimes by a large factor.
2104 If RCU naively believed the firmware, as it used to do,
2105 it would create too many per-CPU kthreads.
2106 Although the resulting system will still run correctly, the extra
2107 kthreads needlessly consume memory and can cause confusion
2108 when they show up in <tt>ps</tt> listings.
2111 RCU must therefore wait for a given CPU to actually come online before
2112 it can allow itself to believe that the CPU actually exists.
2113 The resulting “ghost CPUs” (which are never going to
2114 come online) cause a number of
2115 <a href="https://paulmck.livejournal.com/37494.html">interesting complications</a>.
2117 <h3><a name="Early Boot">Early Boot</a></h3>
2120 The Linux kernel's boot sequence is an interesting process,
2121 and RCU is used early, even before <tt>rcu_init()</tt>
2123 In fact, a number of RCU's primitives can be used as soon as the
2124 initial task's <tt>task_struct</tt> is available and the
2125 boot CPU's per-CPU variables are set up.
2126 The read-side primitives (<tt>rcu_read_lock()</tt>,
2127 <tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
2128 and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
2129 as will <tt>rcu_assign_pointer()</tt>.
2132 Although <tt>call_rcu()</tt> may be invoked at any
2133 time during boot, callbacks are not guaranteed to be invoked until after
2134 the scheduler is fully up and running.
2135 This delay in callback invocation is due to the fact that RCU does not
2136 invoke callbacks until it is fully initialized, and this full initialization
2137 cannot occur until after the scheduler has initialized itself to the
2138 point where RCU can spawn and run its kthreads.
2139 In theory, it would be possible to invoke callbacks earlier,
2140 however, this is not a panacea because there would be severe restrictions
2141 on what operations those callbacks could invoke.
2144 Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
2145 <a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
2146 (<a href="#Bottom-Half Flavor">discussed below</a>),
2148 <a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
2149 will all operate normally
2150 during very early boot, the reason being that there is only one CPU
2151 and preemption is disabled.
2152 This means that the call <tt>synchronize_rcu()</tt> (or friends)
2153 itself is a quiescent
2154 state and thus a grace period, so the early-boot implementation can
2158 Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
2159 continue to operate normally through the remainder of boot, courtesy
2160 of the fact that preemption is disabled across their RCU read-side
2161 critical sections and also courtesy of the fact that there is still
2163 However, once the scheduler starts initializing, preemption is enabled.
2164 There is still only a single CPU, but the fact that preemption is enabled
2165 means that the no-op implementation of <tt>synchronize_rcu()</tt> no
2166 longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
2167 Therefore, as soon as the scheduler starts initializing, the early-boot
2168 fastpath is disabled.
2169 This means that <tt>synchronize_rcu()</tt> switches to its runtime
2170 mode of operation where it posts callbacks, which in turn means that
2171 any call to <tt>synchronize_rcu()</tt> will block until the corresponding
2172 callback is invoked.
2173 Unfortunately, the callback cannot be invoked until RCU's runtime
2174 grace-period machinery is up and running, which cannot happen until
2175 the scheduler has initialized itself sufficiently to allow RCU's
2176 kthreads to be spawned.
2177 Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
2178 initialization can result in deadlock.
2181 <tr><th> </th></tr>
2182 <tr><th align="left">Quick Quiz:</th></tr>
2184 So what happens with <tt>synchronize_rcu()</tt> during
2185 scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
2188 <tr><th align="left">Answer:</th></tr>
2189 <tr><td bgcolor="#ffffff"><font color="ffffff">
2190 In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
2191 maps directly to <tt>synchronize_sched()</tt>.
2192 Therefore, <tt>synchronize_rcu()</tt> works normally throughout
2193 boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
2194 However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
2195 so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
2196 during scheduler initialization.
2198 <tr><td> </td></tr>
2202 I learned of these boot-time requirements as a result of a series of
2205 <h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
2208 The Linux kernel has interrupts, and RCU read-side critical sections are
2209 legal within interrupt handlers and within interrupt-disabled regions
2210 of code, as are invocations of <tt>call_rcu()</tt>.
2213 Some Linux-kernel architectures can enter an interrupt handler from
2214 non-idle process context, and then just never leave it, instead stealthily
2215 transitioning back to process context.
2216 This trick is sometimes used to invoke system calls from inside the kernel.
2217 These “half-interrupts” mean that RCU has to be very careful
2218 about how it counts interrupt nesting levels.
2219 I learned of this requirement the hard way during a rewrite
2220 of RCU's dyntick-idle code.
2223 The Linux kernel has non-maskable interrupts (NMIs), and
2224 RCU read-side critical sections are legal within NMI handlers.
2225 Thankfully, RCU update-side primitives, including
2226 <tt>call_rcu()</tt>, are prohibited within NMI handlers.
2229 The name notwithstanding, some Linux-kernel architectures
2230 can have nested NMIs, which RCU must handle correctly.
2232 <a href="https://lkml.kernel.org/g/CALCETrXLq1y7e_dKFPgou-FKHB6Pu-r8+t-6Ds+8=va7anBWDA@mail.gmail.com">surprised me</a>
2233 with this requirement;
2234 he also kindly surprised me with
2235 <a href="https://lkml.kernel.org/g/CALCETrXSY9JpW3uE6H8WYk81sg56qasA2aqmjMPsq5dOtzso=g@mail.gmail.com">an algorithm</a>
2236 that meets this requirement.
2238 <h3><a name="Loadable Modules">Loadable Modules</a></h3>
2241 The Linux kernel has loadable modules, and these modules can
2243 After a given module has been unloaded, any attempt to call
2244 one of its functions results in a segmentation fault.
2245 The module-unload functions must therefore cancel any
2246 delayed calls to loadable-module functions, for example,
2247 any outstanding <tt>mod_timer()</tt> must be dealt with
2248 via <tt>del_timer_sync()</tt> or similar.
2251 Unfortunately, there is no way to cancel an RCU callback;
2252 once you invoke <tt>call_rcu()</tt>, the callback function is
2253 going to eventually be invoked, unless the system goes down first.
2254 Because it is normally considered socially irresponsible to crash the system
2255 in response to a module unload request, we need some other way
2256 to deal with in-flight RCU callbacks.
2259 RCU therefore provides
2260 <tt><a href="https://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
2261 which waits until all in-flight RCU callbacks have been invoked.
2262 If a module uses <tt>call_rcu()</tt>, its exit function should therefore
2263 prevent any future invocation of <tt>call_rcu()</tt>, then invoke
2264 <tt>rcu_barrier()</tt>.
2265 In theory, the underlying module-unload code could invoke
2266 <tt>rcu_barrier()</tt> unconditionally, but in practice this would
2267 incur unacceptable latencies.
2270 Nikita Danilov noted this requirement for an analogous filesystem-unmount
2271 situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
2272 The need for <tt>rcu_barrier()</tt> for module unloading became
2275 <h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
2278 The Linux kernel supports CPU hotplug, which means that CPUs
2280 It is of course illegal to use any RCU API member from an offline CPU.
2281 This requirement was present from day one in DYNIX/ptx, but
2282 on the other hand, the Linux kernel's CPU-hotplug implementation
2283 is “interesting.”
2286 The Linux-kernel CPU-hotplug implementation has notifiers that
2287 are used to allow the various kernel subsystems (including RCU)
2288 to respond appropriately to a given CPU-hotplug operation.
2289 Most RCU operations may be invoked from CPU-hotplug notifiers,
2290 including even normal synchronous grace-period operations
2291 such as <tt>synchronize_rcu()</tt>.
2292 However, expedited grace-period operations such as
2293 <tt>synchronize_rcu_expedited()</tt> are not supported,
2294 due to the fact that current implementations block CPU-hotplug
2295 operations, which could result in deadlock.
2298 In addition, all-callback-wait operations such as
2299 <tt>rcu_barrier()</tt> are also not supported, due to the
2300 fact that there are phases of CPU-hotplug operations where
2301 the outgoing CPU's callbacks will not be invoked until after
2302 the CPU-hotplug operation ends, which could also result in deadlock.
2304 <h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
2307 RCU depends on the scheduler, and the scheduler uses RCU to
2308 protect some of its data structures.
2309 This means the scheduler is forbidden from acquiring
2310 the runqueue locks and the priority-inheritance locks
2311 in the middle of an outermost RCU read-side critical section unless either
2312 (1) it releases them before exiting that same
2313 RCU read-side critical section, or
2314 (2) interrupts are disabled across
2315 that entire RCU read-side critical section.
2316 This same prohibition also applies (recursively!) to any lock that is acquired
2317 while holding any lock to which this prohibition applies.
2318 Adhering to this rule prevents preemptible RCU from invoking
2319 <tt>rcu_read_unlock_special()</tt> while either runqueue or
2320 priority-inheritance locks are held, thus avoiding deadlock.
2323 Prior to v4.4, it was only necessary to disable preemption across
2324 RCU read-side critical sections that acquired scheduler locks.
2325 In v4.4, expedited grace periods started using IPIs, and these
2326 IPIs could force a <tt>rcu_read_unlock()</tt> to take the slowpath.
2327 Therefore, this expedited-grace-period change required disabling of
2328 interrupts, not just preemption.
2331 For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
2332 implementation must be written carefully to avoid similar deadlocks.
2333 In particular, <tt>rcu_read_unlock()</tt> must tolerate an
2334 interrupt where the interrupt handler invokes both
2335 <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
2336 This possibility requires <tt>rcu_read_unlock()</tt> to use
2337 negative nesting levels to avoid destructive recursion via
2338 interrupt handler's use of RCU.
2341 This pair of mutual scheduler-RCU requirements came as a
2342 <a href="https://lwn.net/Articles/453002/">complete surprise</a>.
2345 As noted above, RCU makes use of kthreads, and it is necessary to
2346 avoid excessive CPU-time accumulation by these kthreads.
2347 This requirement was no surprise, but RCU's violation of it
2348 when running context-switch-heavy workloads when built with
2349 <tt>CONFIG_NO_HZ_FULL=y</tt>
2350 <a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
2351 RCU has made good progress towards meeting this requirement, even
2352 for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
2353 but there is room for further improvement.
2355 <h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
2358 It is possible to use tracing on RCU code, but tracing itself
2360 For this reason, <tt>rcu_dereference_raw_notrace()</tt>
2361 is provided for use by tracing, which avoids the destructive
2362 recursion that could otherwise ensue.
2363 This API is also used by virtualization in some architectures,
2364 where RCU readers execute in environments in which tracing
2366 The tracing folks both located the requirement and provided the
2367 needed fix, so this surprise requirement was relatively painless.
2369 <h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
2372 Interrupting idle CPUs is considered socially unacceptable,
2373 especially by people with battery-powered embedded systems.
2374 RCU therefore conserves energy by detecting which CPUs are
2375 idle, including tracking CPUs that have been interrupted from idle.
2376 This is a large part of the energy-efficiency requirement,
2377 so I learned of this via an irate phone call.
2380 Because RCU avoids interrupting idle CPUs, it is illegal to
2381 execute an RCU read-side critical section on an idle CPU.
2382 (Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
2384 The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
2385 event tracing is provided to work around this restriction.
2386 In addition, <tt>rcu_is_watching()</tt> may be used to
2387 test whether or not it is currently legal to run RCU read-side
2388 critical sections on this CPU.
2389 I learned of the need for diagnostics on the one hand
2390 and <tt>RCU_NONIDLE()</tt> on the other while inspecting
2392 Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
2393 which is used quite heavily in the idle loop.
2394 However, there are some restrictions on the code placed within
2395 <tt>RCU_NONIDLE()</tt>:
2398 <li> Blocking is prohibited.
2399 In practice, this is not a serious restriction given that idle
2400 tasks are prohibited from blocking to begin with.
2401 <li> Although nesting <tt>RCU_NONIDLE()</tt> is permited, they cannot
2402 nest indefinitely deeply.
2403 However, given that they can be nested on the order of a million
2404 deep, even on 32-bit systems, this should not be a serious
2406 This nesting limit would probably be reached long after the
2407 compiler OOMed or the stack overflowed.
2408 <li> Any code path that enters <tt>RCU_NONIDLE()</tt> must sequence
2409 out of that same <tt>RCU_NONIDLE()</tt>.
2410 For example, the following is grossly illegal:
2416 3 goto bad_idea; /* BUG!!! */
2417 4 do_something_else();});
2423 It is just as illegal to transfer control into the middle of
2424 <tt>RCU_NONIDLE()</tt>'s argument.
2425 Yes, in theory, you could transfer in as long as you also
2426 transferred out, but in practice you could also expect to get sharply
2427 worded review comments.
2431 It is similarly socially unacceptable to interrupt an
2432 <tt>nohz_full</tt> CPU running in userspace.
2433 RCU must therefore track <tt>nohz_full</tt> userspace
2436 <a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
2437 kernels, RCU must separately track idle CPUs on the one hand and
2438 CPUs that are either idle or executing in userspace on the other.
2439 In both cases, RCU must be able to sample state at two points in
2440 time, and be able to determine whether or not some other CPU spent
2441 any time idle and/or executing in userspace.
2444 These energy-efficiency requirements have proven quite difficult to
2445 understand and to meet, for example, there have been more than five
2446 clean-sheet rewrites of RCU's energy-efficiency code, the last of
2447 which was finally able to demonstrate
2448 <a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
2450 I learned of many of these requirements via angry phone calls:
2451 Flaming me on the Linux-kernel mailing list was apparently not
2452 sufficient to fully vent their ire at RCU's energy-efficiency bugs!
2454 <h3><a name="Memory Efficiency">Memory Efficiency</a></h3>
2457 Although small-memory non-realtime systems can simply use Tiny RCU,
2458 code size is only one aspect of memory efficiency.
2459 Another aspect is the size of the <tt>rcu_head</tt> structure
2460 used by <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>.
2461 Although this structure contains nothing more than a pair of pointers,
2462 it does appear in many RCU-protected data structures, including
2463 some that are size critical.
2464 The <tt>page</tt> structure is a case in point, as evidenced by
2465 the many occurrences of the <tt>union</tt> keyword within that structure.
2468 This need for memory efficiency is one reason that RCU uses hand-crafted
2469 singly linked lists to track the <tt>rcu_head</tt> structures that
2470 are waiting for a grace period to elapse.
2471 It is also the reason why <tt>rcu_head</tt> structures do not contain
2472 debug information, such as fields tracking the file and line of the
2473 <tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> that posted them.
2474 Although this information might appear in debug-only kernel builds at some
2475 point, in the meantime, the <tt>->func</tt> field will often provide
2476 the needed debug information.
2479 However, in some cases, the need for memory efficiency leads to even
2480 more extreme measures.
2481 Returning to the <tt>page</tt> structure, the <tt>rcu_head</tt> field
2482 shares storage with a great many other structures that are used at
2483 various points in the corresponding page's lifetime.
2484 In order to correctly resolve certain
2485 <a href="https://lkml.kernel.org/g/1439976106-137226-1-git-send-email-kirill.shutemov@linux.intel.com">race conditions</a>,
2486 the Linux kernel's memory-management subsystem needs a particular bit
2487 to remain zero during all phases of grace-period processing,
2488 and that bit happens to map to the bottom bit of the
2489 <tt>rcu_head</tt> structure's <tt>->next</tt> field.
2490 RCU makes this guarantee as long as <tt>call_rcu()</tt>
2491 is used to post the callback, as opposed to <tt>kfree_rcu()</tt>
2492 or some future “lazy”
2493 variant of <tt>call_rcu()</tt> that might one day be created for
2494 energy-efficiency purposes.
2497 That said, there are limits.
2498 RCU requires that the <tt>rcu_head</tt> structure be aligned to a
2499 two-byte boundary, and passing a misaligned <tt>rcu_head</tt>
2500 structure to one of the <tt>call_rcu()</tt> family of functions
2501 will result in a splat.
2502 It is therefore necessary to exercise caution when packing
2503 structures containing fields of type <tt>rcu_head</tt>.
2504 Why not a four-byte or even eight-byte alignment requirement?
2505 Because the m68k architecture provides only two-byte alignment,
2506 and thus acts as alignment's least common denominator.
2509 The reason for reserving the bottom bit of pointers to
2510 <tt>rcu_head</tt> structures is to leave the door open to
2511 “lazy” callbacks whose invocations can safely be deferred.
2512 Deferring invocation could potentially have energy-efficiency
2513 benefits, but only if the rate of non-lazy callbacks decreases
2514 significantly for some important workload.
2515 In the meantime, reserving the bottom bit keeps this option open
2516 in case it one day becomes useful.
2518 <h3><a name="Performance, Scalability, Response Time, and Reliability">
2519 Performance, Scalability, Response Time, and Reliability</a></h3>
2523 <a href="#Performance and Scalability">earlier discussion</a>,
2524 RCU is used heavily by hot code paths in performance-critical
2525 portions of the Linux kernel's networking, security, virtualization,
2526 and scheduling code paths.
2527 RCU must therefore use efficient implementations, especially in its
2528 read-side primitives.
2529 To that end, it would be good if preemptible RCU's implementation
2530 of <tt>rcu_read_lock()</tt> could be inlined, however, doing
2531 this requires resolving <tt>#include</tt> issues with the
2532 <tt>task_struct</tt> structure.
2535 The Linux kernel supports hardware configurations with up to
2536 4096 CPUs, which means that RCU must be extremely scalable.
2537 Algorithms that involve frequent acquisitions of global locks or
2538 frequent atomic operations on global variables simply cannot be
2539 tolerated within the RCU implementation.
2540 RCU therefore makes heavy use of a combining tree based on the
2541 <tt>rcu_node</tt> structure.
2542 RCU is required to tolerate all CPUs continuously invoking any
2543 combination of RCU's runtime primitives with minimal per-operation
2545 In fact, in many cases, increasing load must <i>decrease</i> the
2546 per-operation overhead, witness the batching optimizations for
2547 <tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
2548 <tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
2549 As a general rule, RCU must cheerfully accept whatever the
2550 rest of the Linux kernel decides to throw at it.
2553 The Linux kernel is used for real-time workloads, especially
2554 in conjunction with the
2555 <a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
2556 The real-time-latency response requirements are such that the
2557 traditional approach of disabling preemption across RCU
2558 read-side critical sections is inappropriate.
2559 Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
2560 use an RCU implementation that allows RCU read-side critical
2561 sections to be preempted.
2562 This requirement made its presence known after users made it
2563 clear that an earlier
2564 <a href="https://lwn.net/Articles/107930/">real-time patch</a>
2565 did not meet their needs, in conjunction with some
2566 <a href="https://lkml.kernel.org/g/20050318002026.GA2693@us.ibm.com">RCU issues</a>
2567 encountered by a very early version of the -rt patchset.
2570 In addition, RCU must make do with a sub-100-microsecond real-time latency
2572 In fact, on smaller systems with the -rt patchset, the Linux kernel
2573 provides sub-20-microsecond real-time latencies for the whole kernel,
2575 RCU's scalability and latency must therefore be sufficient for
2576 these sorts of configurations.
2577 To my surprise, the sub-100-microsecond real-time latency budget
2578 <a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
2579 applies to even the largest systems [PDF]</a>,
2580 up to and including systems with 4096 CPUs.
2581 This real-time requirement motivated the grace-period kthread, which
2582 also simplified handling of a number of race conditions.
2585 RCU must avoid degrading real-time response for CPU-bound threads, whether
2586 executing in usermode (which is one use case for
2587 <tt>CONFIG_NO_HZ_FULL=y</tt>) or in the kernel.
2588 That said, CPU-bound loops in the kernel must execute
2589 <tt>cond_resched_rcu_qs()</tt> at least once per few tens of milliseconds
2590 in order to avoid receiving an IPI from RCU.
2593 Finally, RCU's status as a synchronization primitive means that
2594 any RCU failure can result in arbitrary memory corruption that can be
2595 extremely difficult to debug.
2596 This means that RCU must be extremely reliable, which in
2597 practice also means that RCU must have an aggressive stress-test
2599 This stress-test suite is called <tt>rcutorture</tt>.
2602 Although the need for <tt>rcutorture</tt> was no surprise,
2603 the current immense popularity of the Linux kernel is posing
2604 interesting—and perhaps unprecedented—validation
2606 To see this, keep in mind that there are well over one billion
2607 instances of the Linux kernel running today, given Android
2608 smartphones, Linux-powered televisions, and servers.
2609 This number can be expected to increase sharply with the advent of
2610 the celebrated Internet of Things.
2613 Suppose that RCU contains a race condition that manifests on average
2614 once per million years of runtime.
2615 This bug will be occurring about three times per <i>day</i> across
2617 RCU could simply hide behind hardware error rates, given that no one
2618 should really expect their smartphone to last for a million years.
2619 However, anyone taking too much comfort from this thought should
2620 consider the fact that in most jurisdictions, a successful multi-year
2621 test of a given mechanism, which might include a Linux kernel,
2622 suffices for a number of types of safety-critical certifications.
2623 In fact, rumor has it that the Linux kernel is already being used
2624 in production for safety-critical applications.
2625 I don't know about you, but I would feel quite bad if a bug in RCU
2627 Which might explain my recent focus on validation and verification.
2629 <h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
2632 One of the more surprising things about RCU is that there are now
2633 no fewer than five <i>flavors</i>, or API families.
2634 In addition, the primary flavor that has been the sole focus up to
2635 this point has two different implementations, non-preemptible and
2637 The other four flavors are listed below, with requirements for each
2638 described in a separate section.
2641 <li> <a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
2642 <li> <a href="#Sched Flavor">Sched Flavor</a>
2643 <li> <a href="#Sleepable RCU">Sleepable RCU</a>
2644 <li> <a href="#Tasks RCU">Tasks RCU</a>
2645 <li> <a href="#Waiting for Multiple Grace Periods">
2646 Waiting for Multiple Grace Periods</a>
2649 <h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>
2652 The softirq-disable (AKA “bottom-half”,
2653 hence the “_bh” abbreviations)
2654 flavor of RCU, or <i>RCU-bh</i>, was developed by
2655 Dipankar Sarma to provide a flavor of RCU that could withstand the
2656 network-based denial-of-service attacks researched by Robert
2658 These attacks placed so much networking load on the system
2659 that some of the CPUs never exited softirq execution,
2660 which in turn prevented those CPUs from ever executing a context switch,
2661 which, in the RCU implementation of that time, prevented grace periods
2663 The result was an out-of-memory condition and a system hang.
2666 The solution was the creation of RCU-bh, which does
2667 <tt>local_bh_disable()</tt>
2668 across its read-side critical sections, and which uses the transition
2669 from one type of softirq processing to another as a quiescent state
2670 in addition to context switch, idle, user mode, and offline.
2671 This means that RCU-bh grace periods can complete even when some of
2672 the CPUs execute in softirq indefinitely, thus allowing algorithms
2673 based on RCU-bh to withstand network-based denial-of-service attacks.
2677 <tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
2678 disable and re-enable softirq handlers, any attempt to start a softirq
2680 RCU-bh read-side critical section will be deferred.
2681 In this case, <tt>rcu_read_unlock_bh()</tt>
2682 will invoke softirq processing, which can take considerable time.
2683 One can of course argue that this softirq overhead should be associated
2684 with the code following the RCU-bh read-side critical section rather
2685 than <tt>rcu_read_unlock_bh()</tt>, but the fact
2686 is that most profiling tools cannot be expected to make this sort
2687 of fine distinction.
2688 For example, suppose that a three-millisecond-long RCU-bh read-side
2689 critical section executes during a time of heavy networking load.
2690 There will very likely be an attempt to invoke at least one softirq
2691 handler during that three milliseconds, but any such invocation will
2692 be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
2693 This can of course make it appear at first glance as if
2694 <tt>rcu_read_unlock_bh()</tt> was executing very slowly.
2698 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
2700 <tt>rcu_read_lock_bh()</tt>,
2701 <tt>rcu_read_unlock_bh()</tt>,
2702 <tt>rcu_dereference_bh()</tt>,
2703 <tt>rcu_dereference_bh_check()</tt>,
2704 <tt>synchronize_rcu_bh()</tt>,
2705 <tt>synchronize_rcu_bh_expedited()</tt>,
2706 <tt>call_rcu_bh()</tt>,
2707 <tt>rcu_barrier_bh()</tt>, and
2708 <tt>rcu_read_lock_bh_held()</tt>.
2710 <h3><a name="Sched Flavor">Sched Flavor</a></h3>
2713 Before preemptible RCU, waiting for an RCU grace period had the
2714 side effect of also waiting for all pre-existing interrupt
2716 However, there are legitimate preemptible-RCU implementations that
2717 do not have this property, given that any point in the code outside
2718 of an RCU read-side critical section can be a quiescent state.
2719 Therefore, <i>RCU-sched</i> was created, which follows “classic”
2720 RCU in that an RCU-sched grace period waits for for pre-existing
2721 interrupt and NMI handlers.
2722 In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
2723 APIs have identical implementations, while kernels built with
2724 <tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
2727 Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
2728 <tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
2729 disable and re-enable preemption, respectively.
2730 This means that if there was a preemption attempt during the
2731 RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
2732 will enter the scheduler, with all the latency and overhead entailed.
2733 Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
2734 as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
2735 However, the highest-priority task won't be preempted, so that task
2736 will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
2740 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
2742 <tt>rcu_read_lock_sched()</tt>,
2743 <tt>rcu_read_unlock_sched()</tt>,
2744 <tt>rcu_read_lock_sched_notrace()</tt>,
2745 <tt>rcu_read_unlock_sched_notrace()</tt>,
2746 <tt>rcu_dereference_sched()</tt>,
2747 <tt>rcu_dereference_sched_check()</tt>,
2748 <tt>synchronize_sched()</tt>,
2749 <tt>synchronize_rcu_sched_expedited()</tt>,
2750 <tt>call_rcu_sched()</tt>,
2751 <tt>rcu_barrier_sched()</tt>, and
2752 <tt>rcu_read_lock_sched_held()</tt>.
2753 However, anything that disables preemption also marks an RCU-sched
2754 read-side critical section, including
2755 <tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
2756 <tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
2759 <h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
2762 For well over a decade, someone saying “I need to block within
2763 an RCU read-side critical section” was a reliable indication
2764 that this someone did not understand RCU.
2765 After all, if you are always blocking in an RCU read-side critical
2766 section, you can probably afford to use a higher-overhead synchronization
2768 However, that changed with the advent of the Linux kernel's notifiers,
2769 whose RCU read-side critical
2770 sections almost never sleep, but sometimes need to.
2771 This resulted in the introduction of
2772 <a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
2776 SRCU allows different domains to be defined, with each such domain
2777 defined by an instance of an <tt>srcu_struct</tt> structure.
2778 A pointer to this structure must be passed in to each SRCU function,
2779 for example, <tt>synchronize_srcu(&ss)</tt>, where
2780 <tt>ss</tt> is the <tt>srcu_struct</tt> structure.
2781 The key benefit of these domains is that a slow SRCU reader in one
2782 domain does not delay an SRCU grace period in some other domain.
2783 That said, one consequence of these domains is that read-side code
2784 must pass a “cookie” from <tt>srcu_read_lock()</tt>
2785 to <tt>srcu_read_unlock()</tt>, for example, as follows:
2791 3 idx = srcu_read_lock(&ss);
2793 5 srcu_read_unlock(&ss, idx);
2798 As noted above, it is legal to block within SRCU read-side critical sections,
2799 however, with great power comes great responsibility.
2800 If you block forever in one of a given domain's SRCU read-side critical
2801 sections, then that domain's grace periods will also be blocked forever.
2802 Of course, one good way to block forever is to deadlock, which can
2803 happen if any operation in a given domain's SRCU read-side critical
2804 section can block waiting, either directly or indirectly, for that domain's
2805 grace period to elapse.
2806 For example, this results in a self-deadlock:
2812 3 idx = srcu_read_lock(&ss);
2814 5 synchronize_srcu(&ss);
2815 6 srcu_read_unlock(&ss, idx);
2820 However, if line 5 acquired a mutex that was held across
2821 a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
2822 deadlock would still be possible.
2823 Furthermore, if line 5 acquired a mutex that was held across
2824 a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
2825 and if an <tt>ss1</tt>-domain SRCU read-side critical section
2826 acquired another mutex that was held across as <tt>ss</tt>-domain
2827 <tt>synchronize_srcu()</tt>,
2828 deadlock would again be possible.
2829 Such a deadlock cycle could extend across an arbitrarily large number
2830 of different SRCU domains.
2831 Again, with great power comes great responsibility.
2834 Unlike the other RCU flavors, SRCU read-side critical sections can
2835 run on idle and even offline CPUs.
2836 This ability requires that <tt>srcu_read_lock()</tt> and
2837 <tt>srcu_read_unlock()</tt> contain memory barriers, which means
2838 that SRCU readers will run a bit slower than would RCU readers.
2839 It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
2840 API, which, in combination with <tt>srcu_read_unlock()</tt>,
2841 guarantees a full memory barrier.
2845 <a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
2847 <tt>srcu_read_lock()</tt>,
2848 <tt>srcu_read_unlock()</tt>,
2849 <tt>srcu_dereference()</tt>,
2850 <tt>srcu_dereference_check()</tt>,
2851 <tt>synchronize_srcu()</tt>,
2852 <tt>synchronize_srcu_expedited()</tt>,
2853 <tt>call_srcu()</tt>,
2854 <tt>srcu_barrier()</tt>, and
2855 <tt>srcu_read_lock_held()</tt>.
2857 <tt>DEFINE_SRCU()</tt>,
2858 <tt>DEFINE_STATIC_SRCU()</tt>, and
2859 <tt>init_srcu_struct()</tt>
2860 APIs for defining and initializing <tt>srcu_struct</tt> structures.
2862 <h3><a name="Tasks RCU">Tasks RCU</a></h3>
2865 Some forms of tracing use “tramopolines” to handle the
2866 binary rewriting required to install different types of probes.
2867 It would be good to be able to free old trampolines, which sounds
2868 like a job for some form of RCU.
2869 However, because it is necessary to be able to install a trace
2870 anywhere in the code, it is not possible to use read-side markers
2871 such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
2872 In addition, it does not work to have these markers in the trampoline
2873 itself, because there would need to be instructions following
2874 <tt>rcu_read_unlock()</tt>.
2875 Although <tt>synchronize_rcu()</tt> would guarantee that execution
2876 reached the <tt>rcu_read_unlock()</tt>, it would not be able to
2877 guarantee that execution had completely left the trampoline.
2880 The solution, in the form of
2881 <a href="https://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
2883 read-side critical sections that are delimited by voluntary context
2884 switches, that is, calls to <tt>schedule()</tt>,
2885 <tt>cond_resched_rcu_qs()</tt>, and
2886 <tt>synchronize_rcu_tasks()</tt>.
2887 In addition, transitions to and from userspace execution also delimit
2888 tasks-RCU read-side critical sections.
2891 The tasks-RCU API is quite compact, consisting only of
2892 <tt>call_rcu_tasks()</tt>,
2893 <tt>synchronize_rcu_tasks()</tt>, and
2894 <tt>rcu_barrier_tasks()</tt>.
2896 <h3><a name="Waiting for Multiple Grace Periods">
2897 Waiting for Multiple Grace Periods</a></h3>
2900 Perhaps you have an RCU protected data structure that is accessed from
2901 RCU read-side critical sections, from softirq handlers, and from
2902 hardware interrupt handlers.
2903 That is three flavors of RCU, the normal flavor, the bottom-half flavor,
2904 and the sched flavor.
2905 How to wait for a compound grace period?
2908 The best approach is usually to “just say no!” and
2909 insert <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>
2910 around each RCU read-side critical section, regardless of what
2911 environment it happens to be in.
2912 But suppose that some of the RCU read-side critical sections are
2913 on extremely hot code paths, and that use of <tt>CONFIG_PREEMPT=n</tt>
2914 is not a viable option, so that <tt>rcu_read_lock()</tt> and
2915 <tt>rcu_read_unlock()</tt> are not free.
2919 You <i>could</i> wait on all three grace periods in succession, as follows:
2923 1 synchronize_rcu();
2924 2 synchronize_rcu_bh();
2925 3 synchronize_sched();
2930 This works, but triples the update-side latency penalty.
2931 In cases where this is not acceptable, <tt>synchronize_rcu_mult()</tt>
2932 may be used to wait on all three flavors of grace period concurrently:
2936 1 synchronize_rcu_mult(call_rcu, call_rcu_bh, call_rcu_sched);
2941 But what if it is necessary to also wait on SRCU?
2942 This can be done as follows:
2946 1 static void call_my_srcu(struct rcu_head *head,
2947 2 void (*func)(struct rcu_head *head))
2949 4 call_srcu(&my_srcu, head, func);
2952 7 synchronize_rcu_mult(call_rcu, call_rcu_bh, call_rcu_sched, call_my_srcu);
2957 If you needed to wait on multiple different flavors of SRCU
2958 (but why???), you would need to create a wrapper function resembling
2959 <tt>call_my_srcu()</tt> for each SRCU flavor.
2962 <tr><th> </th></tr>
2963 <tr><th align="left">Quick Quiz:</th></tr>
2965 But what if I need to wait for multiple RCU flavors, but I also need
2966 the grace periods to be expedited?
2968 <tr><th align="left">Answer:</th></tr>
2969 <tr><td bgcolor="#ffffff"><font color="ffffff">
2970 If you are using expedited grace periods, there should be less penalty
2971 for waiting on them in succession.
2972 But if that is nevertheless a problem, you can use workqueues
2973 or multiple kthreads to wait on the various expedited grace
2974 periods concurrently.
2976 <tr><td> </td></tr>
2980 Again, it is usually better to adjust the RCU read-side critical sections
2981 to use a single flavor of RCU, but when this is not feasible, you can use
2982 <tt>synchronize_rcu_mult()</tt>.
2984 <h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
2987 One of the tricks that RCU uses to attain update-side scalability is
2988 to increase grace-period latency with increasing numbers of CPUs.
2989 If this becomes a serious problem, it will be necessary to rework the
2990 grace-period state machine so as to avoid the need for the additional
2994 Expedited grace periods scan the CPUs, so their latency and overhead
2995 increases with increasing numbers of CPUs.
2996 If this becomes a serious problem on large systems, it will be necessary
2997 to do some redesign to avoid this scalability problem.
3000 RCU disables CPU hotplug in a few places, perhaps most notably in the
3001 expedited grace-period and <tt>rcu_barrier()</tt> operations.
3002 If there is a strong reason to use expedited grace periods in CPU-hotplug
3003 notifiers, it will be necessary to avoid disabling CPU hotplug.
3004 This would introduce some complexity, so there had better be a <i>very</i>
3008 The tradeoff between grace-period latency on the one hand and interruptions
3009 of other CPUs on the other hand may need to be re-examined.
3010 The desire is of course for zero grace-period latency as well as zero
3011 interprocessor interrupts undertaken during an expedited grace period
3013 While this ideal is unlikely to be achievable, it is quite possible that
3014 further improvements can be made.
3017 The multiprocessor implementations of RCU use a combining tree that
3018 groups CPUs so as to reduce lock contention and increase cache locality.
3019 However, this combining tree does not spread its memory across NUMA
3020 nodes nor does it align the CPU groups with hardware features such
3021 as sockets or cores.
3022 Such spreading and alignment is currently believed to be unnecessary
3023 because the hotpath read-side primitives do not access the combining
3024 tree, nor does <tt>call_rcu()</tt> in the common case.
3025 If you believe that your architecture needs such spreading and alignment,
3026 then your architecture should also benefit from the
3027 <tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
3028 to the number of CPUs in a socket, NUMA node, or whatever.
3029 If the number of CPUs is too large, use a fraction of the number of
3031 If the number of CPUs is a large prime number, well, that certainly
3032 is an “interesting” architectural choice!
3033 More flexible arrangements might be considered, but only if
3034 <tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
3035 if the inadequacy has been demonstrated by a carefully run and
3036 realistic system-level workload.
3039 Please note that arrangements that require RCU to remap CPU numbers will
3040 require extremely good demonstration of need and full exploration of
3044 There is an embarrassingly large number of flavors of RCU, and this
3045 number has been increasing over time.
3046 Perhaps it will be possible to combine some at some future date.
3049 RCU's various kthreads are reasonably recent additions.
3050 It is quite likely that adjustments will be required to more gracefully
3051 handle extreme loads.
3052 It might also be necessary to be able to relate CPU utilization by
3053 RCU's kthreads and softirq handlers to the code that instigated this
3055 For example, RCU callback overhead might be charged back to the
3056 originating <tt>call_rcu()</tt> instance, though probably not
3057 in production kernels.
3059 <h2><a name="Summary">Summary</a></h2>
3062 This document has presented more than two decade's worth of RCU
3064 Given that the requirements keep changing, this will not be the last
3065 word on this subject, but at least it serves to get an important
3066 subset of the requirements set forth.
3068 <h2><a name="Acknowledgments">Acknowledgments</a></h2>
3070 I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
3071 Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
3072 Andy Lutomirski for their help in rendering
3073 this article human readable, and to Michelle Rankin for her support
3075 Other contributions are acknowledged in the Linux kernel's git archive.
3076 The cartoon is copyright (c) 2013 by Melissa Broussard,
3078 under the terms of the Creative Commons Attribution-Share Alike 3.0
3079 United States license.
projects / cascardo / linux.git / blob