Commit | Line | Data |
---|---|---|
628c0842 PG |
1 | What is RCU? -- "Read, Copy, Update" |
2 | ||
32300751 PM |
3 | Please note that the "What is RCU?" LWN series is an excellent place |
4 | to start learning about RCU: | |
5 | ||
6 | 1. What is RCU, Fundamentally? http://lwn.net/Articles/262464/ | |
7 | 2. What is RCU? Part 2: Usage http://lwn.net/Articles/263130/ | |
8 | 3. RCU part 3: the RCU API http://lwn.net/Articles/264090/ | |
d493011a | 9 | 4. The RCU API, 2010 Edition http://lwn.net/Articles/418853/ |
db4855b5 | 10 | 2010 Big API Table http://lwn.net/Articles/419086/ |
2921b123 | 11 | 5. The RCU API, 2014 Edition http://lwn.net/Articles/609904/ |
db4855b5 | 12 | 2014 Big API Table http://lwn.net/Articles/609973/ |
32300751 PM |
13 | |
14 | ||
dd81eca8 PM |
15 | What is RCU? |
16 | ||
17 | RCU is a synchronization mechanism that was added to the Linux kernel | |
18 | during the 2.5 development effort that is optimized for read-mostly | |
19 | situations. Although RCU is actually quite simple once you understand it, | |
20 | getting there can sometimes be a challenge. Part of the problem is that | |
21 | most of the past descriptions of RCU have been written with the mistaken | |
22 | assumption that there is "one true way" to describe RCU. Instead, | |
23 | the experience has been that different people must take different paths | |
24 | to arrive at an understanding of RCU. This document provides several | |
25 | different paths, as follows: | |
26 | ||
27 | 1. RCU OVERVIEW | |
28 | 2. WHAT IS RCU'S CORE API? | |
29 | 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API? | |
30 | 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? | |
31 | 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU? | |
32 | 6. ANALOGY WITH READER-WRITER LOCKING | |
33 | 7. FULL LIST OF RCU APIs | |
34 | 8. ANSWERS TO QUICK QUIZZES | |
35 | ||
36 | People who prefer starting with a conceptual overview should focus on | |
37 | Section 1, though most readers will profit by reading this section at | |
38 | some point. People who prefer to start with an API that they can then | |
39 | experiment with should focus on Section 2. People who prefer to start | |
40 | with example uses should focus on Sections 3 and 4. People who need to | |
41 | understand the RCU implementation should focus on Section 5, then dive | |
42 | into the kernel source code. People who reason best by analogy should | |
43 | focus on Section 6. Section 7 serves as an index to the docbook API | |
44 | documentation, and Section 8 is the traditional answer key. | |
45 | ||
46 | So, start with the section that makes the most sense to you and your | |
47 | preferred method of learning. If you need to know everything about | |
48 | everything, feel free to read the whole thing -- but if you are really | |
49 | that type of person, you have perused the source code and will therefore | |
50 | never need this document anyway. ;-) | |
51 | ||
52 | ||
53 | 1. RCU OVERVIEW | |
54 | ||
55 | The basic idea behind RCU is to split updates into "removal" and | |
56 | "reclamation" phases. The removal phase removes references to data items | |
57 | within a data structure (possibly by replacing them with references to | |
58 | new versions of these data items), and can run concurrently with readers. | |
59 | The reason that it is safe to run the removal phase concurrently with | |
60 | readers is the semantics of modern CPUs guarantee that readers will see | |
61 | either the old or the new version of the data structure rather than a | |
62 | partially updated reference. The reclamation phase does the work of reclaiming | |
63 | (e.g., freeing) the data items removed from the data structure during the | |
64 | removal phase. Because reclaiming data items can disrupt any readers | |
65 | concurrently referencing those data items, the reclamation phase must | |
66 | not start until readers no longer hold references to those data items. | |
67 | ||
68 | Splitting the update into removal and reclamation phases permits the | |
69 | updater to perform the removal phase immediately, and to defer the | |
70 | reclamation phase until all readers active during the removal phase have | |
71 | completed, either by blocking until they finish or by registering a | |
72 | callback that is invoked after they finish. Only readers that are active | |
73 | during the removal phase need be considered, because any reader starting | |
74 | after the removal phase will be unable to gain a reference to the removed | |
75 | data items, and therefore cannot be disrupted by the reclamation phase. | |
76 | ||
77 | So the typical RCU update sequence goes something like the following: | |
78 | ||
79 | a. Remove pointers to a data structure, so that subsequent | |
80 | readers cannot gain a reference to it. | |
81 | ||
82 | b. Wait for all previous readers to complete their RCU read-side | |
83 | critical sections. | |
84 | ||
85 | c. At this point, there cannot be any readers who hold references | |
86 | to the data structure, so it now may safely be reclaimed | |
87 | (e.g., kfree()d). | |
88 | ||
89 | Step (b) above is the key idea underlying RCU's deferred destruction. | |
90 | The ability to wait until all readers are done allows RCU readers to | |
91 | use much lighter-weight synchronization, in some cases, absolutely no | |
92 | synchronization at all. In contrast, in more conventional lock-based | |
93 | schemes, readers must use heavy-weight synchronization in order to | |
94 | prevent an updater from deleting the data structure out from under them. | |
95 | This is because lock-based updaters typically update data items in place, | |
96 | and must therefore exclude readers. In contrast, RCU-based updaters | |
97 | typically take advantage of the fact that writes to single aligned | |
98 | pointers are atomic on modern CPUs, allowing atomic insertion, removal, | |
99 | and replacement of data items in a linked structure without disrupting | |
100 | readers. Concurrent RCU readers can then continue accessing the old | |
101 | versions, and can dispense with the atomic operations, memory barriers, | |
102 | and communications cache misses that are so expensive on present-day | |
103 | SMP computer systems, even in absence of lock contention. | |
104 | ||
105 | In the three-step procedure shown above, the updater is performing both | |
106 | the removal and the reclamation step, but it is often helpful for an | |
107 | entirely different thread to do the reclamation, as is in fact the case | |
108 | in the Linux kernel's directory-entry cache (dcache). Even if the same | |
109 | thread performs both the update step (step (a) above) and the reclamation | |
110 | step (step (c) above), it is often helpful to think of them separately. | |
111 | For example, RCU readers and updaters need not communicate at all, | |
112 | but RCU provides implicit low-overhead communication between readers | |
113 | and reclaimers, namely, in step (b) above. | |
114 | ||
115 | So how the heck can a reclaimer tell when a reader is done, given | |
116 | that readers are not doing any sort of synchronization operations??? | |
117 | Read on to learn about how RCU's API makes this easy. | |
118 | ||
119 | ||
120 | 2. WHAT IS RCU'S CORE API? | |
121 | ||
122 | The core RCU API is quite small: | |
123 | ||
124 | a. rcu_read_lock() | |
125 | b. rcu_read_unlock() | |
126 | c. synchronize_rcu() / call_rcu() | |
127 | d. rcu_assign_pointer() | |
128 | e. rcu_dereference() | |
129 | ||
130 | There are many other members of the RCU API, but the rest can be | |
131 | expressed in terms of these five, though most implementations instead | |
132 | express synchronize_rcu() in terms of the call_rcu() callback API. | |
133 | ||
134 | The five core RCU APIs are described below, the other 18 will be enumerated | |
135 | later. See the kernel docbook documentation for more info, or look directly | |
136 | at the function header comments. | |
137 | ||
138 | rcu_read_lock() | |
139 | ||
140 | void rcu_read_lock(void); | |
141 | ||
142 | Used by a reader to inform the reclaimer that the reader is | |
143 | entering an RCU read-side critical section. It is illegal | |
144 | to block while in an RCU read-side critical section, though | |
28f6569a | 145 | kernels built with CONFIG_PREEMPT_RCU can preempt RCU |
6b3ef48a PM |
146 | read-side critical sections. Any RCU-protected data structure |
147 | accessed during an RCU read-side critical section is guaranteed to | |
148 | remain unreclaimed for the full duration of that critical section. | |
dd81eca8 PM |
149 | Reference counts may be used in conjunction with RCU to maintain |
150 | longer-term references to data structures. | |
151 | ||
152 | rcu_read_unlock() | |
153 | ||
154 | void rcu_read_unlock(void); | |
155 | ||
156 | Used by a reader to inform the reclaimer that the reader is | |
157 | exiting an RCU read-side critical section. Note that RCU | |
158 | read-side critical sections may be nested and/or overlapping. | |
159 | ||
160 | synchronize_rcu() | |
161 | ||
162 | void synchronize_rcu(void); | |
163 | ||
164 | Marks the end of updater code and the beginning of reclaimer | |
165 | code. It does this by blocking until all pre-existing RCU | |
166 | read-side critical sections on all CPUs have completed. | |
167 | Note that synchronize_rcu() will -not- necessarily wait for | |
168 | any subsequent RCU read-side critical sections to complete. | |
169 | For example, consider the following sequence of events: | |
170 | ||
171 | CPU 0 CPU 1 CPU 2 | |
172 | ----------------- ------------------------- --------------- | |
173 | 1. rcu_read_lock() | |
174 | 2. enters synchronize_rcu() | |
175 | 3. rcu_read_lock() | |
176 | 4. rcu_read_unlock() | |
177 | 5. exits synchronize_rcu() | |
178 | 6. rcu_read_unlock() | |
179 | ||
180 | To reiterate, synchronize_rcu() waits only for ongoing RCU | |
181 | read-side critical sections to complete, not necessarily for | |
182 | any that begin after synchronize_rcu() is invoked. | |
183 | ||
184 | Of course, synchronize_rcu() does not necessarily return | |
185 | -immediately- after the last pre-existing RCU read-side critical | |
186 | section completes. For one thing, there might well be scheduling | |
187 | delays. For another thing, many RCU implementations process | |
188 | requests in batches in order to improve efficiencies, which can | |
189 | further delay synchronize_rcu(). | |
190 | ||
191 | Since synchronize_rcu() is the API that must figure out when | |
192 | readers are done, its implementation is key to RCU. For RCU | |
193 | to be useful in all but the most read-intensive situations, | |
194 | synchronize_rcu()'s overhead must also be quite small. | |
195 | ||
196 | The call_rcu() API is a callback form of synchronize_rcu(), | |
197 | and is described in more detail in a later section. Instead of | |
198 | blocking, it registers a function and argument which are invoked | |
199 | after all ongoing RCU read-side critical sections have completed. | |
200 | This callback variant is particularly useful in situations where | |
165d6c78 PM |
201 | it is illegal to block or where update-side performance is |
202 | critically important. | |
203 | ||
204 | However, the call_rcu() API should not be used lightly, as use | |
205 | of the synchronize_rcu() API generally results in simpler code. | |
206 | In addition, the synchronize_rcu() API has the nice property | |
207 | of automatically limiting update rate should grace periods | |
208 | be delayed. This property results in system resilience in face | |
209 | of denial-of-service attacks. Code using call_rcu() should limit | |
210 | update rate in order to gain this same sort of resilience. See | |
211 | checklist.txt for some approaches to limiting the update rate. | |
dd81eca8 PM |
212 | |
213 | rcu_assign_pointer() | |
214 | ||
215 | typeof(p) rcu_assign_pointer(p, typeof(p) v); | |
216 | ||
217 | Yes, rcu_assign_pointer() -is- implemented as a macro, though it | |
218 | would be cool to be able to declare a function in this manner. | |
219 | (Compiler experts will no doubt disagree.) | |
220 | ||
221 | The updater uses this function to assign a new value to an | |
222 | RCU-protected pointer, in order to safely communicate the change | |
223 | in value from the updater to the reader. This function returns | |
224 | the new value, and also executes any memory-barrier instructions | |
225 | required for a given CPU architecture. | |
226 | ||
d19720a9 PM |
227 | Perhaps just as important, it serves to document (1) which |
228 | pointers are protected by RCU and (2) the point at which a | |
229 | given structure becomes accessible to other CPUs. That said, | |
230 | rcu_assign_pointer() is most frequently used indirectly, via | |
231 | the _rcu list-manipulation primitives such as list_add_rcu(). | |
dd81eca8 PM |
232 | |
233 | rcu_dereference() | |
234 | ||
235 | typeof(p) rcu_dereference(p); | |
236 | ||
237 | Like rcu_assign_pointer(), rcu_dereference() must be implemented | |
238 | as a macro. | |
239 | ||
240 | The reader uses rcu_dereference() to fetch an RCU-protected | |
241 | pointer, which returns a value that may then be safely | |
8cf503d3 | 242 | dereferenced. Note that rcu_dereference() does not actually |
dd81eca8 PM |
243 | dereference the pointer, instead, it protects the pointer for |
244 | later dereferencing. It also executes any needed memory-barrier | |
245 | instructions for a given CPU architecture. Currently, only Alpha | |
246 | needs memory barriers within rcu_dereference() -- on other CPUs, | |
247 | it compiles to nothing, not even a compiler directive. | |
248 | ||
249 | Common coding practice uses rcu_dereference() to copy an | |
250 | RCU-protected pointer to a local variable, then dereferences | |
251 | this local variable, for example as follows: | |
252 | ||
253 | p = rcu_dereference(head.next); | |
254 | return p->data; | |
255 | ||
256 | However, in this case, one could just as easily combine these | |
257 | into one statement: | |
258 | ||
259 | return rcu_dereference(head.next)->data; | |
260 | ||
261 | If you are going to be fetching multiple fields from the | |
262 | RCU-protected structure, using the local variable is of | |
263 | course preferred. Repeated rcu_dereference() calls look | |
ed384464 MV |
264 | ugly, do not guarantee that the same pointer will be returned |
265 | if an update happened while in the critical section, and incur | |
266 | unnecessary overhead on Alpha CPUs. | |
dd81eca8 PM |
267 | |
268 | Note that the value returned by rcu_dereference() is valid | |
269 | only within the enclosing RCU read-side critical section. | |
270 | For example, the following is -not- legal: | |
271 | ||
272 | rcu_read_lock(); | |
273 | p = rcu_dereference(head.next); | |
274 | rcu_read_unlock(); | |
4357fb57 | 275 | x = p->address; /* BUG!!! */ |
dd81eca8 | 276 | rcu_read_lock(); |
4357fb57 | 277 | y = p->data; /* BUG!!! */ |
dd81eca8 PM |
278 | rcu_read_unlock(); |
279 | ||
280 | Holding a reference from one RCU read-side critical section | |
281 | to another is just as illegal as holding a reference from | |
282 | one lock-based critical section to another! Similarly, | |
283 | using a reference outside of the critical section in which | |
284 | it was acquired is just as illegal as doing so with normal | |
285 | locking. | |
286 | ||
287 | As with rcu_assign_pointer(), an important function of | |
d19720a9 PM |
288 | rcu_dereference() is to document which pointers are protected by |
289 | RCU, in particular, flagging a pointer that is subject to changing | |
290 | at any time, including immediately after the rcu_dereference(). | |
291 | And, again like rcu_assign_pointer(), rcu_dereference() is | |
292 | typically used indirectly, via the _rcu list-manipulation | |
dd81eca8 PM |
293 | primitives, such as list_for_each_entry_rcu(). |
294 | ||
295 | The following diagram shows how each API communicates among the | |
296 | reader, updater, and reclaimer. | |
297 | ||
298 | ||
299 | rcu_assign_pointer() | |
300 | +--------+ | |
301 | +---------------------->| reader |---------+ | |
302 | | +--------+ | | |
303 | | | | | |
304 | | | | Protect: | |
305 | | | | rcu_read_lock() | |
306 | | | | rcu_read_unlock() | |
307 | | rcu_dereference() | | | |
308 | +---------+ | | | |
309 | | updater |<---------------------+ | | |
310 | +---------+ V | |
311 | | +-----------+ | |
312 | +----------------------------------->| reclaimer | | |
313 | +-----------+ | |
314 | Defer: | |
315 | synchronize_rcu() & call_rcu() | |
316 | ||
317 | ||
318 | The RCU infrastructure observes the time sequence of rcu_read_lock(), | |
319 | rcu_read_unlock(), synchronize_rcu(), and call_rcu() invocations in | |
320 | order to determine when (1) synchronize_rcu() invocations may return | |
321 | to their callers and (2) call_rcu() callbacks may be invoked. Efficient | |
322 | implementations of the RCU infrastructure make heavy use of batching in | |
323 | order to amortize their overhead over many uses of the corresponding APIs. | |
324 | ||
325 | There are no fewer than three RCU mechanisms in the Linux kernel; the | |
326 | diagram above shows the first one, which is by far the most commonly used. | |
327 | The rcu_dereference() and rcu_assign_pointer() primitives are used for | |
328 | all three mechanisms, but different defer and protect primitives are | |
329 | used as follows: | |
330 | ||
331 | Defer Protect | |
332 | ||
333 | a. synchronize_rcu() rcu_read_lock() / rcu_read_unlock() | |
c598a070 | 334 | call_rcu() rcu_dereference() |
dd81eca8 | 335 | |
d07e6d08 PM |
336 | b. synchronize_rcu_bh() rcu_read_lock_bh() / rcu_read_unlock_bh() |
337 | call_rcu_bh() rcu_dereference_bh() | |
dd81eca8 | 338 | |
4c54005c | 339 | c. synchronize_sched() rcu_read_lock_sched() / rcu_read_unlock_sched() |
d07e6d08 | 340 | call_rcu_sched() preempt_disable() / preempt_enable() |
dd81eca8 PM |
341 | local_irq_save() / local_irq_restore() |
342 | hardirq enter / hardirq exit | |
343 | NMI enter / NMI exit | |
c598a070 | 344 | rcu_dereference_sched() |
dd81eca8 PM |
345 | |
346 | These three mechanisms are used as follows: | |
347 | ||
348 | a. RCU applied to normal data structures. | |
349 | ||
350 | b. RCU applied to networking data structures that may be subjected | |
351 | to remote denial-of-service attacks. | |
352 | ||
353 | c. RCU applied to scheduler and interrupt/NMI-handler tasks. | |
354 | ||
355 | Again, most uses will be of (a). The (b) and (c) cases are important | |
356 | for specialized uses, but are relatively uncommon. | |
357 | ||
358 | ||
359 | 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API? | |
360 | ||
361 | This section shows a simple use of the core RCU API to protect a | |
d19720a9 | 362 | global pointer to a dynamically allocated structure. More-typical |
dd81eca8 PM |
363 | uses of RCU may be found in listRCU.txt, arrayRCU.txt, and NMI-RCU.txt. |
364 | ||
365 | struct foo { | |
366 | int a; | |
367 | char b; | |
368 | long c; | |
369 | }; | |
370 | DEFINE_SPINLOCK(foo_mutex); | |
371 | ||
2c4ac34b | 372 | struct foo __rcu *gbl_foo; |
dd81eca8 PM |
373 | |
374 | /* | |
375 | * Create a new struct foo that is the same as the one currently | |
376 | * pointed to by gbl_foo, except that field "a" is replaced | |
377 | * with "new_a". Points gbl_foo to the new structure, and | |
378 | * frees up the old structure after a grace period. | |
379 | * | |
380 | * Uses rcu_assign_pointer() to ensure that concurrent readers | |
381 | * see the initialized version of the new structure. | |
382 | * | |
383 | * Uses synchronize_rcu() to ensure that any readers that might | |
384 | * have references to the old structure complete before freeing | |
385 | * the old structure. | |
386 | */ | |
387 | void foo_update_a(int new_a) | |
388 | { | |
389 | struct foo *new_fp; | |
390 | struct foo *old_fp; | |
391 | ||
de0dfcdf | 392 | new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL); |
dd81eca8 | 393 | spin_lock(&foo_mutex); |
2c4ac34b | 394 | old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex)); |
dd81eca8 PM |
395 | *new_fp = *old_fp; |
396 | new_fp->a = new_a; | |
397 | rcu_assign_pointer(gbl_foo, new_fp); | |
398 | spin_unlock(&foo_mutex); | |
399 | synchronize_rcu(); | |
400 | kfree(old_fp); | |
401 | } | |
402 | ||
403 | /* | |
404 | * Return the value of field "a" of the current gbl_foo | |
405 | * structure. Use rcu_read_lock() and rcu_read_unlock() | |
406 | * to ensure that the structure does not get deleted out | |
407 | * from under us, and use rcu_dereference() to ensure that | |
408 | * we see the initialized version of the structure (important | |
409 | * for DEC Alpha and for people reading the code). | |
410 | */ | |
411 | int foo_get_a(void) | |
412 | { | |
413 | int retval; | |
414 | ||
415 | rcu_read_lock(); | |
416 | retval = rcu_dereference(gbl_foo)->a; | |
417 | rcu_read_unlock(); | |
418 | return retval; | |
419 | } | |
420 | ||
421 | So, to sum up: | |
422 | ||
423 | o Use rcu_read_lock() and rcu_read_unlock() to guard RCU | |
424 | read-side critical sections. | |
425 | ||
426 | o Within an RCU read-side critical section, use rcu_dereference() | |
427 | to dereference RCU-protected pointers. | |
428 | ||
429 | o Use some solid scheme (such as locks or semaphores) to | |
430 | keep concurrent updates from interfering with each other. | |
431 | ||
432 | o Use rcu_assign_pointer() to update an RCU-protected pointer. | |
433 | This primitive protects concurrent readers from the updater, | |
434 | -not- concurrent updates from each other! You therefore still | |
435 | need to use locking (or something similar) to keep concurrent | |
436 | rcu_assign_pointer() primitives from interfering with each other. | |
437 | ||
438 | o Use synchronize_rcu() -after- removing a data element from an | |
439 | RCU-protected data structure, but -before- reclaiming/freeing | |
440 | the data element, in order to wait for the completion of all | |
441 | RCU read-side critical sections that might be referencing that | |
442 | data item. | |
443 | ||
444 | See checklist.txt for additional rules to follow when using RCU. | |
d19720a9 PM |
445 | And again, more-typical uses of RCU may be found in listRCU.txt, |
446 | arrayRCU.txt, and NMI-RCU.txt. | |
dd81eca8 PM |
447 | |
448 | ||
449 | 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? | |
450 | ||
451 | In the example above, foo_update_a() blocks until a grace period elapses. | |
452 | This is quite simple, but in some cases one cannot afford to wait so | |
453 | long -- there might be other high-priority work to be done. | |
454 | ||
455 | In such cases, one uses call_rcu() rather than synchronize_rcu(). | |
456 | The call_rcu() API is as follows: | |
457 | ||
458 | void call_rcu(struct rcu_head * head, | |
459 | void (*func)(struct rcu_head *head)); | |
460 | ||
461 | This function invokes func(head) after a grace period has elapsed. | |
462 | This invocation might happen from either softirq or process context, | |
463 | so the function is not permitted to block. The foo struct needs to | |
464 | have an rcu_head structure added, perhaps as follows: | |
465 | ||
466 | struct foo { | |
467 | int a; | |
468 | char b; | |
469 | long c; | |
470 | struct rcu_head rcu; | |
471 | }; | |
472 | ||
473 | The foo_update_a() function might then be written as follows: | |
474 | ||
475 | /* | |
476 | * Create a new struct foo that is the same as the one currently | |
477 | * pointed to by gbl_foo, except that field "a" is replaced | |
478 | * with "new_a". Points gbl_foo to the new structure, and | |
479 | * frees up the old structure after a grace period. | |
480 | * | |
481 | * Uses rcu_assign_pointer() to ensure that concurrent readers | |
482 | * see the initialized version of the new structure. | |
483 | * | |
484 | * Uses call_rcu() to ensure that any readers that might have | |
485 | * references to the old structure complete before freeing the | |
486 | * old structure. | |
487 | */ | |
488 | void foo_update_a(int new_a) | |
489 | { | |
490 | struct foo *new_fp; | |
491 | struct foo *old_fp; | |
492 | ||
de0dfcdf | 493 | new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL); |
dd81eca8 | 494 | spin_lock(&foo_mutex); |
2c4ac34b | 495 | old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex)); |
dd81eca8 PM |
496 | *new_fp = *old_fp; |
497 | new_fp->a = new_a; | |
498 | rcu_assign_pointer(gbl_foo, new_fp); | |
499 | spin_unlock(&foo_mutex); | |
500 | call_rcu(&old_fp->rcu, foo_reclaim); | |
501 | } | |
502 | ||
503 | The foo_reclaim() function might appear as follows: | |
504 | ||
505 | void foo_reclaim(struct rcu_head *rp) | |
506 | { | |
507 | struct foo *fp = container_of(rp, struct foo, rcu); | |
508 | ||
57d34a6c KC |
509 | foo_cleanup(fp->a); |
510 | ||
dd81eca8 PM |
511 | kfree(fp); |
512 | } | |
513 | ||
514 | The container_of() primitive is a macro that, given a pointer into a | |
515 | struct, the type of the struct, and the pointed-to field within the | |
516 | struct, returns a pointer to the beginning of the struct. | |
517 | ||
518 | The use of call_rcu() permits the caller of foo_update_a() to | |
519 | immediately regain control, without needing to worry further about the | |
520 | old version of the newly updated element. It also clearly shows the | |
521 | RCU distinction between updater, namely foo_update_a(), and reclaimer, | |
522 | namely foo_reclaim(). | |
523 | ||
524 | The summary of advice is the same as for the previous section, except | |
525 | that we are now using call_rcu() rather than synchronize_rcu(): | |
526 | ||
527 | o Use call_rcu() -after- removing a data element from an | |
528 | RCU-protected data structure in order to register a callback | |
529 | function that will be invoked after the completion of all RCU | |
530 | read-side critical sections that might be referencing that | |
531 | data item. | |
532 | ||
57d34a6c KC |
533 | If the callback for call_rcu() is not doing anything more than calling |
534 | kfree() on the structure, you can use kfree_rcu() instead of call_rcu() | |
535 | to avoid having to write your own callback: | |
536 | ||
537 | kfree_rcu(old_fp, rcu); | |
538 | ||
dd81eca8 PM |
539 | Again, see checklist.txt for additional rules governing the use of RCU. |
540 | ||
541 | ||
542 | 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU? | |
543 | ||
544 | One of the nice things about RCU is that it has extremely simple "toy" | |
545 | implementations that are a good first step towards understanding the | |
546 | production-quality implementations in the Linux kernel. This section | |
547 | presents two such "toy" implementations of RCU, one that is implemented | |
548 | in terms of familiar locking primitives, and another that more closely | |
549 | resembles "classic" RCU. Both are way too simple for real-world use, | |
550 | lacking both functionality and performance. However, they are useful | |
551 | in getting a feel for how RCU works. See kernel/rcupdate.c for a | |
552 | production-quality implementation, and see: | |
553 | ||
554 | http://www.rdrop.com/users/paulmck/RCU | |
555 | ||
556 | for papers describing the Linux kernel RCU implementation. The OLS'01 | |
557 | and OLS'02 papers are a good introduction, and the dissertation provides | |
d19720a9 | 558 | more details on the current implementation as of early 2004. |
dd81eca8 PM |
559 | |
560 | ||
561 | 5A. "TOY" IMPLEMENTATION #1: LOCKING | |
562 | ||
563 | This section presents a "toy" RCU implementation that is based on | |
564 | familiar locking primitives. Its overhead makes it a non-starter for | |
565 | real-life use, as does its lack of scalability. It is also unsuitable | |
566 | for realtime use, since it allows scheduling latency to "bleed" from | |
d3d3a3cc PM |
567 | one read-side critical section to another. It also assumes recursive |
568 | reader-writer locks: If you try this with non-recursive locks, and | |
569 | you allow nested rcu_read_lock() calls, you can deadlock. | |
dd81eca8 PM |
570 | |
571 | However, it is probably the easiest implementation to relate to, so is | |
572 | a good starting point. | |
573 | ||
574 | It is extremely simple: | |
575 | ||
576 | static DEFINE_RWLOCK(rcu_gp_mutex); | |
577 | ||
578 | void rcu_read_lock(void) | |
579 | { | |
580 | read_lock(&rcu_gp_mutex); | |
581 | } | |
582 | ||
583 | void rcu_read_unlock(void) | |
584 | { | |
585 | read_unlock(&rcu_gp_mutex); | |
586 | } | |
587 | ||
588 | void synchronize_rcu(void) | |
589 | { | |
590 | write_lock(&rcu_gp_mutex); | |
264d4f88 | 591 | smp_mb__after_spinlock(); |
dd81eca8 PM |
592 | write_unlock(&rcu_gp_mutex); |
593 | } | |
594 | ||
066bb1c8 PM |
595 | [You can ignore rcu_assign_pointer() and rcu_dereference() without missing |
596 | much. But here are simplified versions anyway. And whatever you do, | |
597 | don't forget about them when submitting patches making use of RCU!] | |
598 | ||
599 | #define rcu_assign_pointer(p, v) \ | |
600 | ({ \ | |
601 | smp_store_release(&(p), (v)); \ | |
602 | }) | |
603 | ||
604 | #define rcu_dereference(p) \ | |
605 | ({ \ | |
9ad3c143 | 606 | typeof(p) _________p1 = READ_ONCE(p); \ |
066bb1c8 PM |
607 | (_________p1); \ |
608 | }) | |
dd81eca8 PM |
609 | |
610 | ||
611 | The rcu_read_lock() and rcu_read_unlock() primitive read-acquire | |
612 | and release a global reader-writer lock. The synchronize_rcu() | |
264d4f88 AP |
613 | primitive write-acquires this same lock, then releases it. This means |
614 | that once synchronize_rcu() exits, all RCU read-side critical sections | |
615 | that were in progress before synchronize_rcu() was called are guaranteed | |
616 | to have completed -- there is no way that synchronize_rcu() would have | |
617 | been able to write-acquire the lock otherwise. The smp_mb__after_spinlock() | |
618 | promotes synchronize_rcu() to a full memory barrier in compliance with | |
619 | the "Memory-Barrier Guarantees" listed in: | |
620 | ||
621 | Documentation/RCU/Design/Requirements/Requirements.html. | |
dd81eca8 PM |
622 | |
623 | It is possible to nest rcu_read_lock(), since reader-writer locks may | |
624 | be recursively acquired. Note also that rcu_read_lock() is immune | |
625 | from deadlock (an important property of RCU). The reason for this is | |
626 | that the only thing that can block rcu_read_lock() is a synchronize_rcu(). | |
627 | But synchronize_rcu() does not acquire any locks while holding rcu_gp_mutex, | |
628 | so there can be no deadlock cycle. | |
629 | ||
630 | Quick Quiz #1: Why is this argument naive? How could a deadlock | |
631 | occur when using this algorithm in a real-world Linux | |
632 | kernel? How could this deadlock be avoided? | |
633 | ||
634 | ||
635 | 5B. "TOY" EXAMPLE #2: CLASSIC RCU | |
636 | ||
637 | This section presents a "toy" RCU implementation that is based on | |
638 | "classic RCU". It is also short on performance (but only for updates) and | |
639 | on features such as hotplug CPU and the ability to run in CONFIG_PREEMPT | |
640 | kernels. The definitions of rcu_dereference() and rcu_assign_pointer() | |
641 | are the same as those shown in the preceding section, so they are omitted. | |
642 | ||
643 | void rcu_read_lock(void) { } | |
644 | ||
645 | void rcu_read_unlock(void) { } | |
646 | ||
647 | void synchronize_rcu(void) | |
648 | { | |
649 | int cpu; | |
650 | ||
3c30a752 | 651 | for_each_possible_cpu(cpu) |
dd81eca8 PM |
652 | run_on(cpu); |
653 | } | |
654 | ||
655 | Note that rcu_read_lock() and rcu_read_unlock() do absolutely nothing. | |
656 | This is the great strength of classic RCU in a non-preemptive kernel: | |
657 | read-side overhead is precisely zero, at least on non-Alpha CPUs. | |
658 | And there is absolutely no way that rcu_read_lock() can possibly | |
659 | participate in a deadlock cycle! | |
660 | ||
661 | The implementation of synchronize_rcu() simply schedules itself on each | |
662 | CPU in turn. The run_on() primitive can be implemented straightforwardly | |
663 | in terms of the sched_setaffinity() primitive. Of course, a somewhat less | |
664 | "toy" implementation would restore the affinity upon completion rather | |
665 | than just leaving all tasks running on the last CPU, but when I said | |
666 | "toy", I meant -toy-! | |
667 | ||
668 | So how the heck is this supposed to work??? | |
669 | ||
670 | Remember that it is illegal to block while in an RCU read-side critical | |
671 | section. Therefore, if a given CPU executes a context switch, we know | |
672 | that it must have completed all preceding RCU read-side critical sections. | |
673 | Once -all- CPUs have executed a context switch, then -all- preceding | |
674 | RCU read-side critical sections will have completed. | |
675 | ||
676 | So, suppose that we remove a data item from its structure and then invoke | |
677 | synchronize_rcu(). Once synchronize_rcu() returns, we are guaranteed | |
678 | that there are no RCU read-side critical sections holding a reference | |
679 | to that data item, so we can safely reclaim it. | |
680 | ||
681 | Quick Quiz #2: Give an example where Classic RCU's read-side | |
682 | overhead is -negative-. | |
683 | ||
684 | Quick Quiz #3: If it is illegal to block in an RCU read-side | |
685 | critical section, what the heck do you do in | |
686 | PREEMPT_RT, where normal spinlocks can block??? | |
687 | ||
688 | ||
689 | 6. ANALOGY WITH READER-WRITER LOCKING | |
690 | ||
691 | Although RCU can be used in many different ways, a very common use of | |
692 | RCU is analogous to reader-writer locking. The following unified | |
693 | diff shows how closely related RCU and reader-writer locking can be. | |
694 | ||
70946a44 YD |
695 | @@ -5,5 +5,5 @@ struct el { |
696 | int data; | |
697 | /* Other data fields */ | |
698 | }; | |
699 | -rwlock_t listmutex; | |
700 | +spinlock_t listmutex; | |
701 | struct el head; | |
702 | ||
dd81eca8 PM |
703 | @@ -13,15 +14,15 @@ |
704 | struct list_head *lp; | |
705 | struct el *p; | |
706 | ||
70946a44 | 707 | - read_lock(&listmutex); |
dd81eca8 PM |
708 | - list_for_each_entry(p, head, lp) { |
709 | + rcu_read_lock(); | |
710 | + list_for_each_entry_rcu(p, head, lp) { | |
711 | if (p->key == key) { | |
712 | *result = p->data; | |
70946a44 | 713 | - read_unlock(&listmutex); |
dd81eca8 PM |
714 | + rcu_read_unlock(); |
715 | return 1; | |
716 | } | |
717 | } | |
70946a44 | 718 | - read_unlock(&listmutex); |
dd81eca8 PM |
719 | + rcu_read_unlock(); |
720 | return 0; | |
721 | } | |
722 | ||
723 | @@ -29,15 +30,16 @@ | |
724 | { | |
725 | struct el *p; | |
726 | ||
727 | - write_lock(&listmutex); | |
728 | + spin_lock(&listmutex); | |
729 | list_for_each_entry(p, head, lp) { | |
730 | if (p->key == key) { | |
82a854ec | 731 | - list_del(&p->list); |
dd81eca8 | 732 | - write_unlock(&listmutex); |
82a854ec | 733 | + list_del_rcu(&p->list); |
dd81eca8 PM |
734 | + spin_unlock(&listmutex); |
735 | + synchronize_rcu(); | |
736 | kfree(p); | |
737 | return 1; | |
738 | } | |
739 | } | |
740 | - write_unlock(&listmutex); | |
741 | + spin_unlock(&listmutex); | |
742 | return 0; | |
743 | } | |
744 | ||
745 | Or, for those who prefer a side-by-side listing: | |
746 | ||
747 | 1 struct el { 1 struct el { | |
748 | 2 struct list_head list; 2 struct list_head list; | |
749 | 3 long key; 3 long key; | |
750 | 4 spinlock_t mutex; 4 spinlock_t mutex; | |
751 | 5 int data; 5 int data; | |
752 | 6 /* Other data fields */ 6 /* Other data fields */ | |
753 | 7 }; 7 }; | |
70946a44 | 754 | 8 rwlock_t listmutex; 8 spinlock_t listmutex; |
dd81eca8 PM |
755 | 9 struct el head; 9 struct el head; |
756 | ||
757 | 1 int search(long key, int *result) 1 int search(long key, int *result) | |
758 | 2 { 2 { | |
759 | 3 struct list_head *lp; 3 struct list_head *lp; | |
760 | 4 struct el *p; 4 struct el *p; | |
761 | 5 5 | |
70946a44 | 762 | 6 read_lock(&listmutex); 6 rcu_read_lock(); |
dd81eca8 PM |
763 | 7 list_for_each_entry(p, head, lp) { 7 list_for_each_entry_rcu(p, head, lp) { |
764 | 8 if (p->key == key) { 8 if (p->key == key) { | |
765 | 9 *result = p->data; 9 *result = p->data; | |
70946a44 | 766 | 10 read_unlock(&listmutex); 10 rcu_read_unlock(); |
dd81eca8 PM |
767 | 11 return 1; 11 return 1; |
768 | 12 } 12 } | |
769 | 13 } 13 } | |
70946a44 | 770 | 14 read_unlock(&listmutex); 14 rcu_read_unlock(); |
dd81eca8 PM |
771 | 15 return 0; 15 return 0; |
772 | 16 } 16 } | |
773 | ||
774 | 1 int delete(long key) 1 int delete(long key) | |
775 | 2 { 2 { | |
776 | 3 struct el *p; 3 struct el *p; | |
777 | 4 4 | |
778 | 5 write_lock(&listmutex); 5 spin_lock(&listmutex); | |
779 | 6 list_for_each_entry(p, head, lp) { 6 list_for_each_entry(p, head, lp) { | |
780 | 7 if (p->key == key) { 7 if (p->key == key) { | |
82a854ec | 781 | 8 list_del(&p->list); 8 list_del_rcu(&p->list); |
dd81eca8 PM |
782 | 9 write_unlock(&listmutex); 9 spin_unlock(&listmutex); |
783 | 10 synchronize_rcu(); | |
784 | 10 kfree(p); 11 kfree(p); | |
785 | 11 return 1; 12 return 1; | |
786 | 12 } 13 } | |
787 | 13 } 14 } | |
788 | 14 write_unlock(&listmutex); 15 spin_unlock(&listmutex); | |
789 | 15 return 0; 16 return 0; | |
790 | 16 } 17 } | |
791 | ||
792 | Either way, the differences are quite small. Read-side locking moves | |
793 | to rcu_read_lock() and rcu_read_unlock, update-side locking moves from | |
670e9f34 | 794 | a reader-writer lock to a simple spinlock, and a synchronize_rcu() |
dd81eca8 PM |
795 | precedes the kfree(). |
796 | ||
797 | However, there is one potential catch: the read-side and update-side | |
798 | critical sections can now run concurrently. In many cases, this will | |
799 | not be a problem, but it is necessary to check carefully regardless. | |
800 | For example, if multiple independent list updates must be seen as | |
801 | a single atomic update, converting to RCU will require special care. | |
802 | ||
803 | Also, the presence of synchronize_rcu() means that the RCU version of | |
804 | delete() can now block. If this is a problem, there is a callback-based | |
57d34a6c KC |
805 | mechanism that never blocks, namely call_rcu() or kfree_rcu(), that can |
806 | be used in place of synchronize_rcu(). | |
dd81eca8 PM |
807 | |
808 | ||
809 | 7. FULL LIST OF RCU APIs | |
810 | ||
811 | The RCU APIs are documented in docbook-format header comments in the | |
812 | Linux-kernel source code, but it helps to have a full list of the | |
813 | APIs, since there does not appear to be a way to categorize them | |
814 | in docbook. Here is the list, by category. | |
815 | ||
c598a070 | 816 | RCU list traversal: |
dd81eca8 | 817 | |
d07e6d08 PM |
818 | list_entry_rcu |
819 | list_first_entry_rcu | |
820 | list_next_rcu | |
32300751 | 821 | list_for_each_entry_rcu |
d07e6d08 | 822 | list_for_each_entry_continue_rcu |
b7b6f94c | 823 | list_for_each_entry_from_rcu |
d07e6d08 PM |
824 | hlist_first_rcu |
825 | hlist_next_rcu | |
826 | hlist_pprev_rcu | |
32300751 | 827 | hlist_for_each_entry_rcu |
d07e6d08 | 828 | hlist_for_each_entry_rcu_bh |
b7b6f94c | 829 | hlist_for_each_entry_from_rcu |
d07e6d08 PM |
830 | hlist_for_each_entry_continue_rcu |
831 | hlist_for_each_entry_continue_rcu_bh | |
832 | hlist_nulls_first_rcu | |
240ebbf8 | 833 | hlist_nulls_for_each_entry_rcu |
d07e6d08 PM |
834 | hlist_bl_first_rcu |
835 | hlist_bl_for_each_entry_rcu | |
dd81eca8 | 836 | |
32300751 | 837 | RCU pointer/list update: |
dd81eca8 PM |
838 | |
839 | rcu_assign_pointer | |
840 | list_add_rcu | |
841 | list_add_tail_rcu | |
842 | list_del_rcu | |
843 | list_replace_rcu | |
1d023284 | 844 | hlist_add_behind_rcu |
32300751 | 845 | hlist_add_before_rcu |
dd81eca8 | 846 | hlist_add_head_rcu |
d07e6d08 PM |
847 | hlist_del_rcu |
848 | hlist_del_init_rcu | |
32300751 PM |
849 | hlist_replace_rcu |
850 | list_splice_init_rcu() | |
d07e6d08 PM |
851 | hlist_nulls_del_init_rcu |
852 | hlist_nulls_del_rcu | |
853 | hlist_nulls_add_head_rcu | |
854 | hlist_bl_add_head_rcu | |
855 | hlist_bl_del_init_rcu | |
856 | hlist_bl_del_rcu | |
857 | hlist_bl_set_first_rcu | |
dd81eca8 | 858 | |
32300751 PM |
859 | RCU: Critical sections Grace period Barrier |
860 | ||
861 | rcu_read_lock synchronize_net rcu_barrier | |
862 | rcu_read_unlock synchronize_rcu | |
c598a070 | 863 | rcu_dereference synchronize_rcu_expedited |
d07e6d08 PM |
864 | rcu_read_lock_held call_rcu |
865 | rcu_dereference_check kfree_rcu | |
866 | rcu_dereference_protected | |
32300751 PM |
867 | |
868 | bh: Critical sections Grace period Barrier | |
869 | ||
870 | rcu_read_lock_bh call_rcu_bh rcu_barrier_bh | |
240ebbf8 | 871 | rcu_read_unlock_bh synchronize_rcu_bh |
c598a070 | 872 | rcu_dereference_bh synchronize_rcu_bh_expedited |
d07e6d08 PM |
873 | rcu_dereference_bh_check |
874 | rcu_dereference_bh_protected | |
875 | rcu_read_lock_bh_held | |
32300751 PM |
876 | |
877 | sched: Critical sections Grace period Barrier | |
878 | ||
240ebbf8 PM |
879 | rcu_read_lock_sched synchronize_sched rcu_barrier_sched |
880 | rcu_read_unlock_sched call_rcu_sched | |
881 | [preempt_disable] synchronize_sched_expedited | |
882 | [and friends] | |
d07e6d08 PM |
883 | rcu_read_lock_sched_notrace |
884 | rcu_read_unlock_sched_notrace | |
c598a070 | 885 | rcu_dereference_sched |
d07e6d08 PM |
886 | rcu_dereference_sched_check |
887 | rcu_dereference_sched_protected | |
888 | rcu_read_lock_sched_held | |
32300751 PM |
889 | |
890 | ||
891 | SRCU: Critical sections Grace period Barrier | |
892 | ||
74d874e7 PM |
893 | srcu_read_lock synchronize_srcu srcu_barrier |
894 | srcu_read_unlock call_srcu | |
99f88919 | 895 | srcu_dereference synchronize_srcu_expedited |
d07e6d08 PM |
896 | srcu_dereference_check |
897 | srcu_read_lock_held | |
dd81eca8 | 898 | |
240ebbf8 | 899 | SRCU: Initialization/cleanup |
4de5f89e PM |
900 | DEFINE_SRCU |
901 | DEFINE_STATIC_SRCU | |
240ebbf8 PM |
902 | init_srcu_struct |
903 | cleanup_srcu_struct | |
dd81eca8 | 904 | |
50aec002 PM |
905 | All: lockdep-checked RCU-protected pointer access |
906 | ||
50aec002 | 907 | rcu_access_pointer |
d07e6d08 | 908 | rcu_dereference_raw |
f78f5b90 | 909 | RCU_LOCKDEP_WARN |
d07e6d08 PM |
910 | rcu_sleep_check |
911 | RCU_NONIDLE | |
50aec002 | 912 | |
dd81eca8 PM |
913 | See the comment headers in the source code (or the docbook generated |
914 | from them) for more information. | |
915 | ||
fea65126 PM |
916 | However, given that there are no fewer than four families of RCU APIs |
917 | in the Linux kernel, how do you choose which one to use? The following | |
918 | list can be helpful: | |
919 | ||
920 | a. Will readers need to block? If so, you need SRCU. | |
921 | ||
99f88919 | 922 | b. What about the -rt patchset? If readers would need to block |
fea65126 PM |
923 | in an non-rt kernel, you need SRCU. If readers would block |
924 | in a -rt kernel, but not in a non-rt kernel, SRCU is not | |
4de5f89e PM |
925 | necessary. (The -rt patchset turns spinlocks into sleeplocks, |
926 | hence this distinction.) | |
fea65126 | 927 | |
99f88919 | 928 | c. Do you need to treat NMI handlers, hardirq handlers, |
fea65126 PM |
929 | and code segments with preemption disabled (whether |
930 | via preempt_disable(), local_irq_save(), local_bh_disable(), | |
931 | or some other mechanism) as if they were explicit RCU readers? | |
2aef619c | 932 | If so, RCU-sched is the only choice that will work for you. |
fea65126 | 933 | |
99f88919 | 934 | d. Do you need RCU grace periods to complete even in the face |
fea65126 PM |
935 | of softirq monopolization of one or more of the CPUs? For |
936 | example, is your code subject to network-based denial-of-service | |
77095901 PM |
937 | attacks? If so, you should disable softirq across your readers, |
938 | for example, by using rcu_read_lock_bh(). | |
fea65126 | 939 | |
99f88919 | 940 | e. Is your workload too update-intensive for normal use of |
fea65126 | 941 | RCU, but inappropriate for other synchronization mechanisms? |
5f0d5a3a PM |
942 | If so, consider SLAB_TYPESAFE_BY_RCU (which was originally |
943 | named SLAB_DESTROY_BY_RCU). But please be careful! | |
fea65126 | 944 | |
99f88919 | 945 | f. Do you need read-side critical sections that are respected |
2aef619c PM |
946 | even though they are in the middle of the idle loop, during |
947 | user-mode execution, or on an offlined CPU? If so, SRCU is the | |
948 | only choice that will work for you. | |
949 | ||
99f88919 | 950 | g. Otherwise, use RCU. |
fea65126 PM |
951 | |
952 | Of course, this all assumes that you have determined that RCU is in fact | |
953 | the right tool for your job. | |
954 | ||
dd81eca8 PM |
955 | |
956 | 8. ANSWERS TO QUICK QUIZZES | |
957 | ||
958 | Quick Quiz #1: Why is this argument naive? How could a deadlock | |
959 | occur when using this algorithm in a real-world Linux | |
960 | kernel? [Referring to the lock-based "toy" RCU | |
961 | algorithm.] | |
962 | ||
963 | Answer: Consider the following sequence of events: | |
964 | ||
965 | 1. CPU 0 acquires some unrelated lock, call it | |
d19720a9 PM |
966 | "problematic_lock", disabling irq via |
967 | spin_lock_irqsave(). | |
dd81eca8 PM |
968 | |
969 | 2. CPU 1 enters synchronize_rcu(), write-acquiring | |
970 | rcu_gp_mutex. | |
971 | ||
972 | 3. CPU 0 enters rcu_read_lock(), but must wait | |
973 | because CPU 1 holds rcu_gp_mutex. | |
974 | ||
975 | 4. CPU 1 is interrupted, and the irq handler | |
976 | attempts to acquire problematic_lock. | |
977 | ||
978 | The system is now deadlocked. | |
979 | ||
980 | One way to avoid this deadlock is to use an approach like | |
981 | that of CONFIG_PREEMPT_RT, where all normal spinlocks | |
982 | become blocking locks, and all irq handlers execute in | |
983 | the context of special tasks. In this case, in step 4 | |
984 | above, the irq handler would block, allowing CPU 1 to | |
985 | release rcu_gp_mutex, avoiding the deadlock. | |
986 | ||
987 | Even in the absence of deadlock, this RCU implementation | |
988 | allows latency to "bleed" from readers to other | |
989 | readers through synchronize_rcu(). To see this, | |
990 | consider task A in an RCU read-side critical section | |
991 | (thus read-holding rcu_gp_mutex), task B blocked | |
992 | attempting to write-acquire rcu_gp_mutex, and | |
993 | task C blocked in rcu_read_lock() attempting to | |
994 | read_acquire rcu_gp_mutex. Task A's RCU read-side | |
995 | latency is holding up task C, albeit indirectly via | |
996 | task B. | |
997 | ||
998 | Realtime RCU implementations therefore use a counter-based | |
999 | approach where tasks in RCU read-side critical sections | |
1000 | cannot be blocked by tasks executing synchronize_rcu(). | |
1001 | ||
1002 | Quick Quiz #2: Give an example where Classic RCU's read-side | |
1003 | overhead is -negative-. | |
1004 | ||
1005 | Answer: Imagine a single-CPU system with a non-CONFIG_PREEMPT | |
1006 | kernel where a routing table is used by process-context | |
1007 | code, but can be updated by irq-context code (for example, | |
1008 | by an "ICMP REDIRECT" packet). The usual way of handling | |
1009 | this would be to have the process-context code disable | |
1010 | interrupts while searching the routing table. Use of | |
1011 | RCU allows such interrupt-disabling to be dispensed with. | |
1012 | Thus, without RCU, you pay the cost of disabling interrupts, | |
1013 | and with RCU you don't. | |
1014 | ||
1015 | One can argue that the overhead of RCU in this | |
1016 | case is negative with respect to the single-CPU | |
1017 | interrupt-disabling approach. Others might argue that | |
1018 | the overhead of RCU is merely zero, and that replacing | |
1019 | the positive overhead of the interrupt-disabling scheme | |
1020 | with the zero-overhead RCU scheme does not constitute | |
1021 | negative overhead. | |
1022 | ||
1023 | In real life, of course, things are more complex. But | |
1024 | even the theoretical possibility of negative overhead for | |
1025 | a synchronization primitive is a bit unexpected. ;-) | |
1026 | ||
1027 | Quick Quiz #3: If it is illegal to block in an RCU read-side | |
1028 | critical section, what the heck do you do in | |
1029 | PREEMPT_RT, where normal spinlocks can block??? | |
1030 | ||
1031 | Answer: Just as PREEMPT_RT permits preemption of spinlock | |
1032 | critical sections, it permits preemption of RCU | |
1033 | read-side critical sections. It also permits | |
1034 | spinlocks blocking while in RCU read-side critical | |
1035 | sections. | |
1036 | ||
1037 | Why the apparent inconsistency? Because it is it | |
1038 | possible to use priority boosting to keep the RCU | |
1039 | grace periods short if need be (for example, if running | |
1040 | short of memory). In contrast, if blocking waiting | |
1041 | for (say) network reception, there is no way to know | |
1042 | what should be boosted. Especially given that the | |
1043 | process we need to boost might well be a human being | |
1044 | who just went out for a pizza or something. And although | |
1045 | a computer-operated cattle prod might arouse serious | |
1046 | interest, it might also provoke serious objections. | |
1047 | Besides, how does the computer know what pizza parlor | |
1048 | the human being went to??? | |
1049 | ||
1050 | ||
1051 | ACKNOWLEDGEMENTS | |
1052 | ||
1053 | My thanks to the people who helped make this human-readable, including | |
d19720a9 | 1054 | Jon Walpole, Josh Triplett, Serge Hallyn, Suzanne Wood, and Alan Stern. |
dd81eca8 PM |
1055 | |
1056 | ||
1057 | For more information, see http://www.rdrop.com/users/paulmck/RCU. |