1 .. _whatisrcu_doc: 1 .. _whatisrcu_doc:
2 2
3 What is RCU? -- "Read, Copy, Update" 3 What is RCU? -- "Read, Copy, Update"
4 ====================================== 4 ======================================
5 5
6 Please note that the "What is RCU?" LWN series 6 Please note that the "What is RCU?" LWN series is an excellent place
7 to start learning about RCU: 7 to start learning about RCU:
8 8
9 | 1. What is RCU, Fundamentally? https://l !! 9 | 1. What is RCU, Fundamentally? http://lwn.net/Articles/262464/
10 | 2. What is RCU? Part 2: Usage https://l !! 10 | 2. What is RCU? Part 2: Usage http://lwn.net/Articles/263130/
11 | 3. RCU part 3: the RCU API https://l !! 11 | 3. RCU part 3: the RCU API http://lwn.net/Articles/264090/
12 | 4. The RCU API, 2010 Edition https://l !! 12 | 4. The RCU API, 2010 Edition http://lwn.net/Articles/418853/
13 | 2010 Big API Table https://l !! 13 | 2010 Big API Table http://lwn.net/Articles/419086/
14 | 5. The RCU API, 2014 Edition https://l !! 14 | 5. The RCU API, 2014 Edition http://lwn.net/Articles/609904/
15 | 2014 Big API Table https://l !! 15 | 2014 Big API Table http://lwn.net/Articles/609973/
16 | 6. The RCU API, 2019 Edition https://l <<
17 | 2019 Big API Table https://l <<
18 <<
19 For those preferring video: <<
20 <<
21 | 1. Unraveling RCU Mysteries: Fundamentals <<
22 | 2. Unraveling RCU Mysteries: Additional U <<
23 16
24 17
25 What is RCU? 18 What is RCU?
26 19
27 RCU is a synchronization mechanism that was ad 20 RCU is a synchronization mechanism that was added to the Linux kernel
28 during the 2.5 development effort that is opti 21 during the 2.5 development effort that is optimized for read-mostly
29 situations. Although RCU is actually quite si !! 22 situations. Although RCU is actually quite simple once you understand it,
30 of it requires you to think differently about !! 23 getting there can sometimes be a challenge. Part of the problem is that
31 of the problem is the mistaken assumption that !! 24 most of the past descriptions of RCU have been written with the mistaken
32 describe and to use RCU. Instead, the experie !! 25 assumption that there is "one true way" to describe RCU. Instead,
33 people must take different paths to arrive at !! 26 the experience has been that different people must take different paths
34 depending on their experiences and use cases. !! 27 to arrive at an understanding of RCU. This document provides several
35 several different paths, as follows: !! 28 different paths, as follows:
36 29
37 :ref:`1. RCU OVERVIEW <1_whatisRCU>` 30 :ref:`1. RCU OVERVIEW <1_whatisRCU>`
38 31
39 :ref:`2. WHAT IS RCU'S CORE API? <2_wha 32 :ref:`2. WHAT IS RCU'S CORE API? <2_whatisRCU>`
40 33
41 :ref:`3. WHAT ARE SOME EXAMPLE USES OF 34 :ref:`3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API? <3_whatisRCU>`
42 35
43 :ref:`4. WHAT IF MY UPDATING THREAD CAN 36 :ref:`4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? <4_whatisRCU>`
44 37
45 :ref:`5. WHAT ARE SOME SIMPLE IMPLEMENT 38 :ref:`5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU? <5_whatisRCU>`
46 39
47 :ref:`6. ANALOGY WITH READER-WRITER LOC 40 :ref:`6. ANALOGY WITH READER-WRITER LOCKING <6_whatisRCU>`
48 41
49 :ref:`7. ANALOGY WITH REFERENCE COUNTIN !! 42 :ref:`7. FULL LIST OF RCU APIs <7_whatisRCU>`
50 <<
51 :ref:`8. FULL LIST OF RCU APIs <8_whati <<
52 43
53 :ref:`9. ANSWERS TO QUICK QUIZZES <9_wh !! 44 :ref:`8. ANSWERS TO QUICK QUIZZES <8_whatisRCU>`
54 45
55 People who prefer starting with a conceptual o 46 People who prefer starting with a conceptual overview should focus on
56 Section 1, though most readers will profit by 47 Section 1, though most readers will profit by reading this section at
57 some point. People who prefer to start with a 48 some point. People who prefer to start with an API that they can then
58 experiment with should focus on Section 2. Pe 49 experiment with should focus on Section 2. People who prefer to start
59 with example uses should focus on Sections 3 a 50 with example uses should focus on Sections 3 and 4. People who need to
60 understand the RCU implementation should focus 51 understand the RCU implementation should focus on Section 5, then dive
61 into the kernel source code. People who reaso 52 into the kernel source code. People who reason best by analogy should
62 focus on Section 6 and 7. Section 8 serves as !! 53 focus on Section 6. Section 7 serves as an index to the docbook API
63 API documentation, and Section 9 is the tradit !! 54 documentation, and Section 8 is the traditional answer key.
64 55
65 So, start with the section that makes the most 56 So, start with the section that makes the most sense to you and your
66 preferred method of learning. If you need to 57 preferred method of learning. If you need to know everything about
67 everything, feel free to read the whole thing 58 everything, feel free to read the whole thing -- but if you are really
68 that type of person, you have perused the sour 59 that type of person, you have perused the source code and will therefore
69 never need this document anyway. ;-) 60 never need this document anyway. ;-)
70 61
71 .. _1_whatisRCU: 62 .. _1_whatisRCU:
72 63
73 1. RCU OVERVIEW 64 1. RCU OVERVIEW
74 ---------------- 65 ----------------
75 66
76 The basic idea behind RCU is to split updates 67 The basic idea behind RCU is to split updates into "removal" and
77 "reclamation" phases. The removal phase remov 68 "reclamation" phases. The removal phase removes references to data items
78 within a data structure (possibly by replacing 69 within a data structure (possibly by replacing them with references to
79 new versions of these data items), and can run 70 new versions of these data items), and can run concurrently with readers.
80 The reason that it is safe to run the removal 71 The reason that it is safe to run the removal phase concurrently with
81 readers is the semantics of modern CPUs guaran 72 readers is the semantics of modern CPUs guarantee that readers will see
82 either the old or the new version of the data 73 either the old or the new version of the data structure rather than a
83 partially updated reference. The reclamation 74 partially updated reference. The reclamation phase does the work of reclaiming
84 (e.g., freeing) the data items removed from th 75 (e.g., freeing) the data items removed from the data structure during the
85 removal phase. Because reclaiming data items 76 removal phase. Because reclaiming data items can disrupt any readers
86 concurrently referencing those data items, the 77 concurrently referencing those data items, the reclamation phase must
87 not start until readers no longer hold referen 78 not start until readers no longer hold references to those data items.
88 79
89 Splitting the update into removal and reclamat 80 Splitting the update into removal and reclamation phases permits the
90 updater to perform the removal phase immediate 81 updater to perform the removal phase immediately, and to defer the
91 reclamation phase until all readers active dur 82 reclamation phase until all readers active during the removal phase have
92 completed, either by blocking until they finis 83 completed, either by blocking until they finish or by registering a
93 callback that is invoked after they finish. O 84 callback that is invoked after they finish. Only readers that are active
94 during the removal phase need be considered, b 85 during the removal phase need be considered, because any reader starting
95 after the removal phase will be unable to gain 86 after the removal phase will be unable to gain a reference to the removed
96 data items, and therefore cannot be disrupted 87 data items, and therefore cannot be disrupted by the reclamation phase.
97 88
98 So the typical RCU update sequence goes someth 89 So the typical RCU update sequence goes something like the following:
99 90
100 a. Remove pointers to a data structure, s 91 a. Remove pointers to a data structure, so that subsequent
101 readers cannot gain a reference to it. 92 readers cannot gain a reference to it.
102 93
103 b. Wait for all previous readers to compl 94 b. Wait for all previous readers to complete their RCU read-side
104 critical sections. 95 critical sections.
105 96
106 c. At this point, there cannot be any rea 97 c. At this point, there cannot be any readers who hold references
107 to the data structure, so it now may s 98 to the data structure, so it now may safely be reclaimed
108 (e.g., kfree()d). 99 (e.g., kfree()d).
109 100
110 Step (b) above is the key idea underlying RCU' 101 Step (b) above is the key idea underlying RCU's deferred destruction.
111 The ability to wait until all readers are done 102 The ability to wait until all readers are done allows RCU readers to
112 use much lighter-weight synchronization, in so 103 use much lighter-weight synchronization, in some cases, absolutely no
113 synchronization at all. In contrast, in more 104 synchronization at all. In contrast, in more conventional lock-based
114 schemes, readers must use heavy-weight synchro 105 schemes, readers must use heavy-weight synchronization in order to
115 prevent an updater from deleting the data stru 106 prevent an updater from deleting the data structure out from under them.
116 This is because lock-based updaters typically 107 This is because lock-based updaters typically update data items in place,
117 and must therefore exclude readers. In contra 108 and must therefore exclude readers. In contrast, RCU-based updaters
118 typically take advantage of the fact that writ 109 typically take advantage of the fact that writes to single aligned
119 pointers are atomic on modern CPUs, allowing a 110 pointers are atomic on modern CPUs, allowing atomic insertion, removal,
120 and replacement of data items in a linked stru 111 and replacement of data items in a linked structure without disrupting
121 readers. Concurrent RCU readers can then cont 112 readers. Concurrent RCU readers can then continue accessing the old
122 versions, and can dispense with the atomic ope 113 versions, and can dispense with the atomic operations, memory barriers,
123 and communications cache misses that are so ex 114 and communications cache misses that are so expensive on present-day
124 SMP computer systems, even in absence of lock 115 SMP computer systems, even in absence of lock contention.
125 116
126 In the three-step procedure shown above, the u 117 In the three-step procedure shown above, the updater is performing both
127 the removal and the reclamation step, but it i 118 the removal and the reclamation step, but it is often helpful for an
128 entirely different thread to do the reclamatio 119 entirely different thread to do the reclamation, as is in fact the case
129 in the Linux kernel's directory-entry cache (d 120 in the Linux kernel's directory-entry cache (dcache). Even if the same
130 thread performs both the update step (step (a) 121 thread performs both the update step (step (a) above) and the reclamation
131 step (step (c) above), it is often helpful to 122 step (step (c) above), it is often helpful to think of them separately.
132 For example, RCU readers and updaters need not 123 For example, RCU readers and updaters need not communicate at all,
133 but RCU provides implicit low-overhead communi 124 but RCU provides implicit low-overhead communication between readers
134 and reclaimers, namely, in step (b) above. 125 and reclaimers, namely, in step (b) above.
135 126
136 So how the heck can a reclaimer tell when a re 127 So how the heck can a reclaimer tell when a reader is done, given
137 that readers are not doing any sort of synchro 128 that readers are not doing any sort of synchronization operations???
138 Read on to learn about how RCU's API makes thi 129 Read on to learn about how RCU's API makes this easy.
139 130
140 .. _2_whatisRCU: 131 .. _2_whatisRCU:
141 132
142 2. WHAT IS RCU'S CORE API? 133 2. WHAT IS RCU'S CORE API?
143 --------------------------- 134 ---------------------------
144 135
145 The core RCU API is quite small: 136 The core RCU API is quite small:
146 137
147 a. rcu_read_lock() 138 a. rcu_read_lock()
148 b. rcu_read_unlock() 139 b. rcu_read_unlock()
149 c. synchronize_rcu() / call_rcu() 140 c. synchronize_rcu() / call_rcu()
150 d. rcu_assign_pointer() 141 d. rcu_assign_pointer()
151 e. rcu_dereference() 142 e. rcu_dereference()
152 143
153 There are many other members of the RCU API, b 144 There are many other members of the RCU API, but the rest can be
154 expressed in terms of these five, though most 145 expressed in terms of these five, though most implementations instead
155 express synchronize_rcu() in terms of the call 146 express synchronize_rcu() in terms of the call_rcu() callback API.
156 147
157 The five core RCU APIs are described below, th 148 The five core RCU APIs are described below, the other 18 will be enumerated
158 later. See the kernel docbook documentation f 149 later. See the kernel docbook documentation for more info, or look directly
159 at the function header comments. 150 at the function header comments.
160 151
161 rcu_read_lock() 152 rcu_read_lock()
162 ^^^^^^^^^^^^^^^ 153 ^^^^^^^^^^^^^^^
163 void rcu_read_lock(void); 154 void rcu_read_lock(void);
164 155
165 This temporal primitive is used by a r !! 156 Used by a reader to inform the reclaimer that the reader is
166 reclaimer that the reader is entering !! 157 entering an RCU read-side critical section. It is illegal
167 section. It is illegal to block while !! 158 to block while in an RCU read-side critical section, though
168 critical section, though kernels built !! 159 kernels built with CONFIG_PREEMPT_RCU can preempt RCU
169 can preempt RCU read-side critical sec !! 160 read-side critical sections. Any RCU-protected data structure
170 data structure accessed during an RCU !! 161 accessed during an RCU read-side critical section is guaranteed to
171 is guaranteed to remain unreclaimed fo !! 162 remain unreclaimed for the full duration of that critical section.
172 critical section. Reference counts ma !! 163 Reference counts may be used in conjunction with RCU to maintain
173 with RCU to maintain longer-term refer !! 164 longer-term references to data structures.
174 <<
175 Note that anything that disables botto <<
176 or interrupts also enters an RCU read- <<
177 Acquiring a spinlock also enters an RC <<
178 sections, even for spinlocks that do n <<
179 as is the case in kernels built with C <<
180 Sleeplocks do *not* enter RCU read-sid <<
181 165
182 rcu_read_unlock() 166 rcu_read_unlock()
183 ^^^^^^^^^^^^^^^^^ 167 ^^^^^^^^^^^^^^^^^
184 void rcu_read_unlock(void); 168 void rcu_read_unlock(void);
185 169
186 This temporal primitives is used by a !! 170 Used by a reader to inform the reclaimer that the reader is
187 reclaimer that the reader is exiting a !! 171 exiting an RCU read-side critical section. Note that RCU
188 section. Anything that enables bottom !! 172 read-side critical sections may be nested and/or overlapping.
189 or interrupts also exits an RCU read-s <<
190 Releasing a spinlock also exits an RCU <<
191 <<
192 Note that RCU read-side critical secti <<
193 overlapping. <<
194 173
195 synchronize_rcu() 174 synchronize_rcu()
196 ^^^^^^^^^^^^^^^^^ 175 ^^^^^^^^^^^^^^^^^
197 void synchronize_rcu(void); 176 void synchronize_rcu(void);
198 177
199 This temporal primitive marks the end !! 178 Marks the end of updater code and the beginning of reclaimer
200 beginning of reclaimer code. It does !! 179 code. It does this by blocking until all pre-existing RCU
201 all pre-existing RCU read-side critica !! 180 read-side critical sections on all CPUs have completed.
202 have completed. Note that synchronize !! 181 Note that synchronize_rcu() will **not** necessarily wait for
203 necessarily wait for any subsequent RC !! 182 any subsequent RCU read-side critical sections to complete.
204 sections to complete. For example, co !! 183 For example, consider the following sequence of events::
205 sequence of events:: <<
206 184
207 CPU 0 CPU 1 185 CPU 0 CPU 1 CPU 2
208 ----------------- --------------- 186 ----------------- ------------------------- ---------------
209 1. rcu_read_lock() 187 1. rcu_read_lock()
210 2. enters synchron 188 2. enters synchronize_rcu()
211 3. 189 3. rcu_read_lock()
212 4. rcu_read_unlock() 190 4. rcu_read_unlock()
213 5. exits synchron 191 5. exits synchronize_rcu()
214 6. 192 6. rcu_read_unlock()
215 193
216 To reiterate, synchronize_rcu() waits 194 To reiterate, synchronize_rcu() waits only for ongoing RCU
217 read-side critical sections to complet 195 read-side critical sections to complete, not necessarily for
218 any that begin after synchronize_rcu() 196 any that begin after synchronize_rcu() is invoked.
219 197
220 Of course, synchronize_rcu() does not 198 Of course, synchronize_rcu() does not necessarily return
221 **immediately** after the last pre-exi 199 **immediately** after the last pre-existing RCU read-side critical
222 section completes. For one thing, the 200 section completes. For one thing, there might well be scheduling
223 delays. For another thing, many RCU i 201 delays. For another thing, many RCU implementations process
224 requests in batches in order to improv 202 requests in batches in order to improve efficiencies, which can
225 further delay synchronize_rcu(). 203 further delay synchronize_rcu().
226 204
227 Since synchronize_rcu() is the API tha 205 Since synchronize_rcu() is the API that must figure out when
228 readers are done, its implementation i 206 readers are done, its implementation is key to RCU. For RCU
229 to be useful in all but the most read- 207 to be useful in all but the most read-intensive situations,
230 synchronize_rcu()'s overhead must also 208 synchronize_rcu()'s overhead must also be quite small.
231 209
232 The call_rcu() API is an asynchronous !! 210 The call_rcu() API is a callback form of synchronize_rcu(),
233 synchronize_rcu(), and is described in !! 211 and is described in more detail in a later section. Instead of
234 section. Instead of blocking, it regi !! 212 blocking, it registers a function and argument which are invoked
235 argument which are invoked after all o !! 213 after all ongoing RCU read-side critical sections have completed.
236 critical sections have completed. Thi !! 214 This callback variant is particularly useful in situations where
237 particularly useful in situations wher !! 215 it is illegal to block or where update-side performance is
238 or where update-side performance is cr !! 216 critically important.
239 217
240 However, the call_rcu() API should not 218 However, the call_rcu() API should not be used lightly, as use
241 of the synchronize_rcu() API generally 219 of the synchronize_rcu() API generally results in simpler code.
242 In addition, the synchronize_rcu() API 220 In addition, the synchronize_rcu() API has the nice property
243 of automatically limiting update rate 221 of automatically limiting update rate should grace periods
244 be delayed. This property results in 222 be delayed. This property results in system resilience in face
245 of denial-of-service attacks. Code us 223 of denial-of-service attacks. Code using call_rcu() should limit
246 update rate in order to gain this same 224 update rate in order to gain this same sort of resilience. See
247 checklist.rst for some approaches to l !! 225 checklist.txt for some approaches to limiting the update rate.
248 226
249 rcu_assign_pointer() 227 rcu_assign_pointer()
250 ^^^^^^^^^^^^^^^^^^^^ 228 ^^^^^^^^^^^^^^^^^^^^
251 void rcu_assign_pointer(p, typeof(p) v 229 void rcu_assign_pointer(p, typeof(p) v);
252 230
253 Yes, rcu_assign_pointer() **is** imple !! 231 Yes, rcu_assign_pointer() **is** implemented as a macro, though it
254 it would be cool to be able to declare !! 232 would be cool to be able to declare a function in this manner.
255 (And there has been some discussion of !! 233 (Compiler experts will no doubt disagree.)
256 to the C language, so who knows?) <<
257 234
258 The updater uses this spatial macro to !! 235 The updater uses this function to assign a new value to an
259 RCU-protected pointer, in order to saf 236 RCU-protected pointer, in order to safely communicate the change
260 in value from the updater to the reade !! 237 in value from the updater to the reader. This macro does not
261 opposed to temporal) macro. It does n !! 238 evaluate to an rvalue, but it does execute any memory-barrier
262 but it does provide any compiler direc !! 239 instructions required for a given CPU architecture.
263 instructions required for a given comp !! 240
264 Its ordering properties are that of a !! 241 Perhaps just as important, it serves to document (1) which
265 that is, any prior loads and stores re !! 242 pointers are protected by RCU and (2) the point at which a
266 structure are ordered before the store !! 243 given structure becomes accessible to other CPUs. That said,
267 to that structure. <<
268 <<
269 Perhaps just as important, rcu_assign_ <<
270 (1) which pointers are protected by RC <<
271 a given structure becomes accessible t <<
272 rcu_assign_pointer() is most frequentl 244 rcu_assign_pointer() is most frequently used indirectly, via
273 the _rcu list-manipulation primitives 245 the _rcu list-manipulation primitives such as list_add_rcu().
274 246
275 rcu_dereference() 247 rcu_dereference()
276 ^^^^^^^^^^^^^^^^^ 248 ^^^^^^^^^^^^^^^^^
277 typeof(p) rcu_dereference(p); 249 typeof(p) rcu_dereference(p);
278 250
279 Like rcu_assign_pointer(), rcu_derefer 251 Like rcu_assign_pointer(), rcu_dereference() must be implemented
280 as a macro. 252 as a macro.
281 253
282 The reader uses the spatial rcu_derefe !! 254 The reader uses rcu_dereference() to fetch an RCU-protected
283 an RCU-protected pointer, which return !! 255 pointer, which returns a value that may then be safely
284 then be safely dereferenced. Note tha !! 256 dereferenced. Note that rcu_dereference() does not actually
285 does not actually dereference the poin !! 257 dereference the pointer, instead, it protects the pointer for
286 protects the pointer for later derefer !! 258 later dereferencing. It also executes any needed memory-barrier
287 executes any needed memory-barrier ins !! 259 instructions for a given CPU architecture. Currently, only Alpha
288 CPU architecture. Currently, only Alp !! 260 needs memory barriers within rcu_dereference() -- on other CPUs,
289 within rcu_dereference() -- on other C !! 261 it compiles to nothing, not even a compiler directive.
290 volatile load. However, no mainstream <<
291 address dependencies, so rcu_dereferen <<
292 which, in combination with the coding <<
293 rcu_dereference.rst, prevent current c <<
294 these dependencies. <<
295 262
296 Common coding practice uses rcu_derefe 263 Common coding practice uses rcu_dereference() to copy an
297 RCU-protected pointer to a local varia 264 RCU-protected pointer to a local variable, then dereferences
298 this local variable, for example as fo 265 this local variable, for example as follows::
299 266
300 p = rcu_dereference(head.next) 267 p = rcu_dereference(head.next);
301 return p->data; 268 return p->data;
302 269
303 However, in this case, one could just 270 However, in this case, one could just as easily combine these
304 into one statement:: 271 into one statement::
305 272
306 return rcu_dereference(head.ne 273 return rcu_dereference(head.next)->data;
307 274
308 If you are going to be fetching multip 275 If you are going to be fetching multiple fields from the
309 RCU-protected structure, using the loc 276 RCU-protected structure, using the local variable is of
310 course preferred. Repeated rcu_derefe 277 course preferred. Repeated rcu_dereference() calls look
311 ugly, do not guarantee that the same p 278 ugly, do not guarantee that the same pointer will be returned
312 if an update happened while in the cri 279 if an update happened while in the critical section, and incur
313 unnecessary overhead on Alpha CPUs. 280 unnecessary overhead on Alpha CPUs.
314 281
315 Note that the value returned by rcu_de 282 Note that the value returned by rcu_dereference() is valid
316 only within the enclosing RCU read-sid 283 only within the enclosing RCU read-side critical section [1]_.
317 For example, the following is **not** 284 For example, the following is **not** legal::
318 285
319 rcu_read_lock(); 286 rcu_read_lock();
320 p = rcu_dereference(head.next) 287 p = rcu_dereference(head.next);
321 rcu_read_unlock(); 288 rcu_read_unlock();
322 x = p->address; /* BUG!!! */ 289 x = p->address; /* BUG!!! */
323 rcu_read_lock(); 290 rcu_read_lock();
324 y = p->data; /* BUG!!! */ 291 y = p->data; /* BUG!!! */
325 rcu_read_unlock(); 292 rcu_read_unlock();
326 293
327 Holding a reference from one RCU read- 294 Holding a reference from one RCU read-side critical section
328 to another is just as illegal as holdi 295 to another is just as illegal as holding a reference from
329 one lock-based critical section to ano 296 one lock-based critical section to another! Similarly,
330 using a reference outside of the criti 297 using a reference outside of the critical section in which
331 it was acquired is just as illegal as 298 it was acquired is just as illegal as doing so with normal
332 locking. 299 locking.
333 300
334 As with rcu_assign_pointer(), an impor 301 As with rcu_assign_pointer(), an important function of
335 rcu_dereference() is to document which 302 rcu_dereference() is to document which pointers are protected by
336 RCU, in particular, flagging a pointer 303 RCU, in particular, flagging a pointer that is subject to changing
337 at any time, including immediately aft 304 at any time, including immediately after the rcu_dereference().
338 And, again like rcu_assign_pointer(), 305 And, again like rcu_assign_pointer(), rcu_dereference() is
339 typically used indirectly, via the _rc 306 typically used indirectly, via the _rcu list-manipulation
340 primitives, such as list_for_each_entr 307 primitives, such as list_for_each_entry_rcu() [2]_.
341 308
342 .. [1] The variant rcu_dereference_protec 309 .. [1] The variant rcu_dereference_protected() can be used outside
343 of an RCU read-side critical section a 310 of an RCU read-side critical section as long as the usage is
344 protected by locks acquired by the upd 311 protected by locks acquired by the update-side code. This variant
345 avoids the lockdep warning that would 312 avoids the lockdep warning that would happen when using (for
346 example) rcu_dereference() without rcu 313 example) rcu_dereference() without rcu_read_lock() protection.
347 Using rcu_dereference_protected() also 314 Using rcu_dereference_protected() also has the advantage
348 of permitting compiler optimizations t 315 of permitting compiler optimizations that rcu_dereference()
349 must prohibit. The rcu_dereference_pr 316 must prohibit. The rcu_dereference_protected() variant takes
350 a lockdep expression to indicate which 317 a lockdep expression to indicate which locks must be acquired
351 by the caller. If the indicated protec 318 by the caller. If the indicated protection is not provided,
352 a lockdep splat is emitted. See Desig !! 319 a lockdep splat is emitted. See Documentation/RCU/Design/Requirements/Requirements.rst
353 and the API's code comments for more d 320 and the API's code comments for more details and example usage.
354 321
355 .. [2] If the list_for_each_entry_rcu() i 322 .. [2] If the list_for_each_entry_rcu() instance might be used by
356 update-side code as well as by RCU rea 323 update-side code as well as by RCU readers, then an additional
357 lockdep expression can be added to its 324 lockdep expression can be added to its list of arguments.
358 For example, given an additional "lock 325 For example, given an additional "lock_is_held(&mylock)" argument,
359 the RCU lockdep code would complain on 326 the RCU lockdep code would complain only if this instance was
360 invoked outside of an RCU read-side cr 327 invoked outside of an RCU read-side critical section and without
361 the protection of mylock. 328 the protection of mylock.
362 329
363 The following diagram shows how each API commu 330 The following diagram shows how each API communicates among the
364 reader, updater, and reclaimer. 331 reader, updater, and reclaimer.
365 :: 332 ::
366 333
367 334
368 rcu_assign_pointer() 335 rcu_assign_pointer()
369 +--------+ 336 +--------+
370 +---------------------->| reader | 337 +---------------------->| reader |---------+
371 | +--------+ 338 | +--------+ |
372 | | 339 | | |
373 | | 340 | | | Protect:
374 | | 341 | | | rcu_read_lock()
375 | | 342 | | | rcu_read_unlock()
376 | rcu_dereference() | 343 | rcu_dereference() | |
377 +---------+ | 344 +---------+ | |
378 | updater |<----------------+ 345 | updater |<----------------+ |
379 +---------+ 346 +---------+ V
380 | 347 | +-----------+
381 +--------------------------------- 348 +----------------------------------->| reclaimer |
382 349 +-----------+
383 Defer: 350 Defer:
384 synchronize_rcu() & call_rcu() 351 synchronize_rcu() & call_rcu()
385 352
386 353
387 The RCU infrastructure observes the temporal s !! 354 The RCU infrastructure observes the time sequence of rcu_read_lock(),
388 rcu_read_unlock(), synchronize_rcu(), and call 355 rcu_read_unlock(), synchronize_rcu(), and call_rcu() invocations in
389 order to determine when (1) synchronize_rcu() 356 order to determine when (1) synchronize_rcu() invocations may return
390 to their callers and (2) call_rcu() callbacks 357 to their callers and (2) call_rcu() callbacks may be invoked. Efficient
391 implementations of the RCU infrastructure make 358 implementations of the RCU infrastructure make heavy use of batching in
392 order to amortize their overhead over many use 359 order to amortize their overhead over many uses of the corresponding APIs.
393 The rcu_assign_pointer() and rcu_dereference() <<
394 spatial changes via stores to and loads from t <<
395 question. <<
396 360
397 There are at least three flavors of RCU usage 361 There are at least three flavors of RCU usage in the Linux kernel. The diagram
398 above shows the most common one. On the update 362 above shows the most common one. On the updater side, the rcu_assign_pointer(),
399 synchronize_rcu() and call_rcu() primitives us !! 363 sychronize_rcu() and call_rcu() primitives used are the same for all three
400 flavors. However for protection (on the reader 364 flavors. However for protection (on the reader side), the primitives used vary
401 depending on the flavor: 365 depending on the flavor:
402 366
403 a. rcu_read_lock() / rcu_read_unlock() 367 a. rcu_read_lock() / rcu_read_unlock()
404 rcu_dereference() 368 rcu_dereference()
405 369
406 b. rcu_read_lock_bh() / rcu_read_unlock_b 370 b. rcu_read_lock_bh() / rcu_read_unlock_bh()
407 local_bh_disable() / local_bh_enable() 371 local_bh_disable() / local_bh_enable()
408 rcu_dereference_bh() 372 rcu_dereference_bh()
409 373
410 c. rcu_read_lock_sched() / rcu_read_unloc 374 c. rcu_read_lock_sched() / rcu_read_unlock_sched()
411 preempt_disable() / preempt_enable() 375 preempt_disable() / preempt_enable()
412 local_irq_save() / local_irq_restore() 376 local_irq_save() / local_irq_restore()
413 hardirq enter / hardirq exit 377 hardirq enter / hardirq exit
414 NMI enter / NMI exit 378 NMI enter / NMI exit
415 rcu_dereference_sched() 379 rcu_dereference_sched()
416 380
417 These three flavors are used as follows: 381 These three flavors are used as follows:
418 382
419 a. RCU applied to normal data structures. 383 a. RCU applied to normal data structures.
420 384
421 b. RCU applied to networking data structu 385 b. RCU applied to networking data structures that may be subjected
422 to remote denial-of-service attacks. 386 to remote denial-of-service attacks.
423 387
424 c. RCU applied to scheduler and interrupt 388 c. RCU applied to scheduler and interrupt/NMI-handler tasks.
425 389
426 Again, most uses will be of (a). The (b) and 390 Again, most uses will be of (a). The (b) and (c) cases are important
427 for specialized uses, but are relatively uncom !! 391 for specialized uses, but are relatively uncommon.
428 RCU-Tasks-Rude, and RCU-Tasks-Trace have simil <<
429 their assorted primitives. <<
430 392
431 .. _3_whatisRCU: 393 .. _3_whatisRCU:
432 394
433 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API 395 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API?
434 ---------------------------------------------- 396 -----------------------------------------------
435 397
436 This section shows a simple use of the core RC 398 This section shows a simple use of the core RCU API to protect a
437 global pointer to a dynamically allocated stru 399 global pointer to a dynamically allocated structure. More-typical
438 uses of RCU may be found in listRCU.rst and NM !! 400 uses of RCU may be found in :ref:`listRCU.rst <list_rcu_doc>`,
>> 401 :ref:`arrayRCU.rst <array_rcu_doc>`, and :ref:`NMI-RCU.rst <NMI_rcu_doc>`.
439 :: 402 ::
440 403
441 struct foo { 404 struct foo {
442 int a; 405 int a;
443 char b; 406 char b;
444 long c; 407 long c;
445 }; 408 };
446 DEFINE_SPINLOCK(foo_mutex); 409 DEFINE_SPINLOCK(foo_mutex);
447 410
448 struct foo __rcu *gbl_foo; 411 struct foo __rcu *gbl_foo;
449 412
450 /* 413 /*
451 * Create a new struct foo that is the 414 * Create a new struct foo that is the same as the one currently
452 * pointed to by gbl_foo, except that 415 * pointed to by gbl_foo, except that field "a" is replaced
453 * with "new_a". Points gbl_foo to th 416 * with "new_a". Points gbl_foo to the new structure, and
454 * frees up the old structure after a 417 * frees up the old structure after a grace period.
455 * 418 *
456 * Uses rcu_assign_pointer() to ensure 419 * Uses rcu_assign_pointer() to ensure that concurrent readers
457 * see the initialized version of the 420 * see the initialized version of the new structure.
458 * 421 *
459 * Uses synchronize_rcu() to ensure th 422 * Uses synchronize_rcu() to ensure that any readers that might
460 * have references to the old structur 423 * have references to the old structure complete before freeing
461 * the old structure. 424 * the old structure.
462 */ 425 */
463 void foo_update_a(int new_a) 426 void foo_update_a(int new_a)
464 { 427 {
465 struct foo *new_fp; 428 struct foo *new_fp;
466 struct foo *old_fp; 429 struct foo *old_fp;
467 430
468 new_fp = kmalloc(sizeof(*new_f 431 new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL);
469 spin_lock(&foo_mutex); 432 spin_lock(&foo_mutex);
470 old_fp = rcu_dereference_prote 433 old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex));
471 *new_fp = *old_fp; 434 *new_fp = *old_fp;
472 new_fp->a = new_a; 435 new_fp->a = new_a;
473 rcu_assign_pointer(gbl_foo, ne 436 rcu_assign_pointer(gbl_foo, new_fp);
474 spin_unlock(&foo_mutex); 437 spin_unlock(&foo_mutex);
475 synchronize_rcu(); 438 synchronize_rcu();
476 kfree(old_fp); 439 kfree(old_fp);
477 } 440 }
478 441
479 /* 442 /*
480 * Return the value of field "a" of th 443 * Return the value of field "a" of the current gbl_foo
481 * structure. Use rcu_read_lock() and 444 * structure. Use rcu_read_lock() and rcu_read_unlock()
482 * to ensure that the structure does n 445 * to ensure that the structure does not get deleted out
483 * from under us, and use rcu_derefere 446 * from under us, and use rcu_dereference() to ensure that
484 * we see the initialized version of t 447 * we see the initialized version of the structure (important
485 * for DEC Alpha and for people readin 448 * for DEC Alpha and for people reading the code).
486 */ 449 */
487 int foo_get_a(void) 450 int foo_get_a(void)
488 { 451 {
489 int retval; 452 int retval;
490 453
491 rcu_read_lock(); 454 rcu_read_lock();
492 retval = rcu_dereference(gbl_f 455 retval = rcu_dereference(gbl_foo)->a;
493 rcu_read_unlock(); 456 rcu_read_unlock();
494 return retval; 457 return retval;
495 } 458 }
496 459
497 So, to sum up: 460 So, to sum up:
498 461
499 - Use rcu_read_lock() and rcu_read_unloc 462 - Use rcu_read_lock() and rcu_read_unlock() to guard RCU
500 read-side critical sections. 463 read-side critical sections.
501 464
502 - Within an RCU read-side critical secti 465 - Within an RCU read-side critical section, use rcu_dereference()
503 to dereference RCU-protected pointers. 466 to dereference RCU-protected pointers.
504 467
505 - Use some solid design (such as locks o !! 468 - Use some solid scheme (such as locks or semaphores) to
506 keep concurrent updates from interferi 469 keep concurrent updates from interfering with each other.
507 470
508 - Use rcu_assign_pointer() to update an 471 - Use rcu_assign_pointer() to update an RCU-protected pointer.
509 This primitive protects concurrent rea 472 This primitive protects concurrent readers from the updater,
510 **not** concurrent updates from each o 473 **not** concurrent updates from each other! You therefore still
511 need to use locking (or something simi 474 need to use locking (or something similar) to keep concurrent
512 rcu_assign_pointer() primitives from i 475 rcu_assign_pointer() primitives from interfering with each other.
513 476
514 - Use synchronize_rcu() **after** removi 477 - Use synchronize_rcu() **after** removing a data element from an
515 RCU-protected data structure, but **be 478 RCU-protected data structure, but **before** reclaiming/freeing
516 the data element, in order to wait for 479 the data element, in order to wait for the completion of all
517 RCU read-side critical sections that m 480 RCU read-side critical sections that might be referencing that
518 data item. 481 data item.
519 482
520 See checklist.rst for additional rules to foll !! 483 See checklist.txt for additional rules to follow when using RCU.
521 And again, more-typical uses of RCU may be fou !! 484 And again, more-typical uses of RCU may be found in :ref:`listRCU.rst
522 and NMI-RCU.rst. !! 485 <list_rcu_doc>`, :ref:`arrayRCU.rst <array_rcu_doc>`, and :ref:`NMI-RCU.rst
>> 486 <NMI_rcu_doc>`.
523 487
524 .. _4_whatisRCU: 488 .. _4_whatisRCU:
525 489
526 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? 490 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK?
527 -------------------------------------------- 491 --------------------------------------------
528 492
529 In the example above, foo_update_a() blocks un 493 In the example above, foo_update_a() blocks until a grace period elapses.
530 This is quite simple, but in some cases one ca 494 This is quite simple, but in some cases one cannot afford to wait so
531 long -- there might be other high-priority wor 495 long -- there might be other high-priority work to be done.
532 496
533 In such cases, one uses call_rcu() rather than 497 In such cases, one uses call_rcu() rather than synchronize_rcu().
534 The call_rcu() API is as follows:: 498 The call_rcu() API is as follows::
535 499
536 void call_rcu(struct rcu_head *head, r !! 500 void call_rcu(struct rcu_head * head,
>> 501 void (*func)(struct rcu_head *head));
537 502
538 This function invokes func(head) after a grace 503 This function invokes func(head) after a grace period has elapsed.
539 This invocation might happen from either softi 504 This invocation might happen from either softirq or process context,
540 so the function is not permitted to block. Th 505 so the function is not permitted to block. The foo struct needs to
541 have an rcu_head structure added, perhaps as f 506 have an rcu_head structure added, perhaps as follows::
542 507
543 struct foo { 508 struct foo {
544 int a; 509 int a;
545 char b; 510 char b;
546 long c; 511 long c;
547 struct rcu_head rcu; 512 struct rcu_head rcu;
548 }; 513 };
549 514
550 The foo_update_a() function might then be writ 515 The foo_update_a() function might then be written as follows::
551 516
552 /* 517 /*
553 * Create a new struct foo that is the 518 * Create a new struct foo that is the same as the one currently
554 * pointed to by gbl_foo, except that 519 * pointed to by gbl_foo, except that field "a" is replaced
555 * with "new_a". Points gbl_foo to th 520 * with "new_a". Points gbl_foo to the new structure, and
556 * frees up the old structure after a 521 * frees up the old structure after a grace period.
557 * 522 *
558 * Uses rcu_assign_pointer() to ensure 523 * Uses rcu_assign_pointer() to ensure that concurrent readers
559 * see the initialized version of the 524 * see the initialized version of the new structure.
560 * 525 *
561 * Uses call_rcu() to ensure that any 526 * Uses call_rcu() to ensure that any readers that might have
562 * references to the old structure com 527 * references to the old structure complete before freeing the
563 * old structure. 528 * old structure.
564 */ 529 */
565 void foo_update_a(int new_a) 530 void foo_update_a(int new_a)
566 { 531 {
567 struct foo *new_fp; 532 struct foo *new_fp;
568 struct foo *old_fp; 533 struct foo *old_fp;
569 534
570 new_fp = kmalloc(sizeof(*new_f 535 new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL);
571 spin_lock(&foo_mutex); 536 spin_lock(&foo_mutex);
572 old_fp = rcu_dereference_prote 537 old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex));
573 *new_fp = *old_fp; 538 *new_fp = *old_fp;
574 new_fp->a = new_a; 539 new_fp->a = new_a;
575 rcu_assign_pointer(gbl_foo, ne 540 rcu_assign_pointer(gbl_foo, new_fp);
576 spin_unlock(&foo_mutex); 541 spin_unlock(&foo_mutex);
577 call_rcu(&old_fp->rcu, foo_rec 542 call_rcu(&old_fp->rcu, foo_reclaim);
578 } 543 }
579 544
580 The foo_reclaim() function might appear as fol 545 The foo_reclaim() function might appear as follows::
581 546
582 void foo_reclaim(struct rcu_head *rp) 547 void foo_reclaim(struct rcu_head *rp)
583 { 548 {
584 struct foo *fp = container_of( 549 struct foo *fp = container_of(rp, struct foo, rcu);
585 550
586 foo_cleanup(fp->a); 551 foo_cleanup(fp->a);
587 552
588 kfree(fp); 553 kfree(fp);
589 } 554 }
590 555
591 The container_of() primitive is a macro that, 556 The container_of() primitive is a macro that, given a pointer into a
592 struct, the type of the struct, and the pointe 557 struct, the type of the struct, and the pointed-to field within the
593 struct, returns a pointer to the beginning of 558 struct, returns a pointer to the beginning of the struct.
594 559
595 The use of call_rcu() permits the caller of fo 560 The use of call_rcu() permits the caller of foo_update_a() to
596 immediately regain control, without needing to 561 immediately regain control, without needing to worry further about the
597 old version of the newly updated element. It 562 old version of the newly updated element. It also clearly shows the
598 RCU distinction between updater, namely foo_up 563 RCU distinction between updater, namely foo_update_a(), and reclaimer,
599 namely foo_reclaim(). 564 namely foo_reclaim().
600 565
601 The summary of advice is the same as for the p 566 The summary of advice is the same as for the previous section, except
602 that we are now using call_rcu() rather than s 567 that we are now using call_rcu() rather than synchronize_rcu():
603 568
604 - Use call_rcu() **after** removing a da 569 - Use call_rcu() **after** removing a data element from an
605 RCU-protected data structure in order 570 RCU-protected data structure in order to register a callback
606 function that will be invoked after th 571 function that will be invoked after the completion of all RCU
607 read-side critical sections that might 572 read-side critical sections that might be referencing that
608 data item. 573 data item.
609 574
610 If the callback for call_rcu() is not doing an 575 If the callback for call_rcu() is not doing anything more than calling
611 kfree() on the structure, you can use kfree_rc 576 kfree() on the structure, you can use kfree_rcu() instead of call_rcu()
612 to avoid having to write your own callback:: 577 to avoid having to write your own callback::
613 578
614 kfree_rcu(old_fp, rcu); 579 kfree_rcu(old_fp, rcu);
615 580
616 If the occasional sleep is permitted, the sing !! 581 Again, see checklist.txt for additional rules governing the use of RCU.
617 be used, omitting the rcu_head structure from <<
618 <<
619 kfree_rcu_mightsleep(old_fp); <<
620 <<
621 This variant almost never blocks, but might do <<
622 synchronize_rcu() in response to memory-alloca <<
623 <<
624 Again, see checklist.rst for additional rules <<
625 582
626 .. _5_whatisRCU: 583 .. _5_whatisRCU:
627 584
628 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RC 585 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU?
629 ---------------------------------------------- 586 ------------------------------------------------
630 587
631 One of the nice things about RCU is that it ha 588 One of the nice things about RCU is that it has extremely simple "toy"
632 implementations that are a good first step tow 589 implementations that are a good first step towards understanding the
633 production-quality implementations in the Linu 590 production-quality implementations in the Linux kernel. This section
634 presents two such "toy" implementations of RCU 591 presents two such "toy" implementations of RCU, one that is implemented
635 in terms of familiar locking primitives, and a 592 in terms of familiar locking primitives, and another that more closely
636 resembles "classic" RCU. Both are way too sim 593 resembles "classic" RCU. Both are way too simple for real-world use,
637 lacking both functionality and performance. H 594 lacking both functionality and performance. However, they are useful
638 in getting a feel for how RCU works. See kern 595 in getting a feel for how RCU works. See kernel/rcu/update.c for a
639 production-quality implementation, and see: 596 production-quality implementation, and see:
640 597
641 https://docs.google.com/document/d/1X0 !! 598 http://www.rdrop.com/users/paulmck/RCU
642 599
643 for papers describing the Linux kernel RCU imp 600 for papers describing the Linux kernel RCU implementation. The OLS'01
644 and OLS'02 papers are a good introduction, and 601 and OLS'02 papers are a good introduction, and the dissertation provides
645 more details on the current implementation as 602 more details on the current implementation as of early 2004.
646 603
647 604
648 5A. "TOY" IMPLEMENTATION #1: LOCKING 605 5A. "TOY" IMPLEMENTATION #1: LOCKING
649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
650 This section presents a "toy" RCU implementati 607 This section presents a "toy" RCU implementation that is based on
651 familiar locking primitives. Its overhead mak 608 familiar locking primitives. Its overhead makes it a non-starter for
652 real-life use, as does its lack of scalability 609 real-life use, as does its lack of scalability. It is also unsuitable
653 for realtime use, since it allows scheduling l 610 for realtime use, since it allows scheduling latency to "bleed" from
654 one read-side critical section to another. It 611 one read-side critical section to another. It also assumes recursive
655 reader-writer locks: If you try this with non 612 reader-writer locks: If you try this with non-recursive locks, and
656 you allow nested rcu_read_lock() calls, you ca 613 you allow nested rcu_read_lock() calls, you can deadlock.
657 614
658 However, it is probably the easiest implementa 615 However, it is probably the easiest implementation to relate to, so is
659 a good starting point. 616 a good starting point.
660 617
661 It is extremely simple:: 618 It is extremely simple::
662 619
663 static DEFINE_RWLOCK(rcu_gp_mutex); 620 static DEFINE_RWLOCK(rcu_gp_mutex);
664 621
665 void rcu_read_lock(void) 622 void rcu_read_lock(void)
666 { 623 {
667 read_lock(&rcu_gp_mutex); 624 read_lock(&rcu_gp_mutex);
668 } 625 }
669 626
670 void rcu_read_unlock(void) 627 void rcu_read_unlock(void)
671 { 628 {
672 read_unlock(&rcu_gp_mutex); 629 read_unlock(&rcu_gp_mutex);
673 } 630 }
674 631
675 void synchronize_rcu(void) 632 void synchronize_rcu(void)
676 { 633 {
677 write_lock(&rcu_gp_mutex); 634 write_lock(&rcu_gp_mutex);
678 smp_mb__after_spinlock(); 635 smp_mb__after_spinlock();
679 write_unlock(&rcu_gp_mutex); 636 write_unlock(&rcu_gp_mutex);
680 } 637 }
681 638
682 [You can ignore rcu_assign_pointer() and rcu_d 639 [You can ignore rcu_assign_pointer() and rcu_dereference() without missing
683 much. But here are simplified versions anyway 640 much. But here are simplified versions anyway. And whatever you do,
684 don't forget about them when submitting patche 641 don't forget about them when submitting patches making use of RCU!]::
685 642
686 #define rcu_assign_pointer(p, v) \ 643 #define rcu_assign_pointer(p, v) \
687 ({ \ 644 ({ \
688 smp_store_release(&(p), (v)); 645 smp_store_release(&(p), (v)); \
689 }) 646 })
690 647
691 #define rcu_dereference(p) \ 648 #define rcu_dereference(p) \
692 ({ \ 649 ({ \
693 typeof(p) _________p1 = READ_O 650 typeof(p) _________p1 = READ_ONCE(p); \
694 (_________p1); \ 651 (_________p1); \
695 }) 652 })
696 653
697 654
698 The rcu_read_lock() and rcu_read_unlock() prim 655 The rcu_read_lock() and rcu_read_unlock() primitive read-acquire
699 and release a global reader-writer lock. The 656 and release a global reader-writer lock. The synchronize_rcu()
700 primitive write-acquires this same lock, then 657 primitive write-acquires this same lock, then releases it. This means
701 that once synchronize_rcu() exits, all RCU rea 658 that once synchronize_rcu() exits, all RCU read-side critical sections
702 that were in progress before synchronize_rcu() 659 that were in progress before synchronize_rcu() was called are guaranteed
703 to have completed -- there is no way that sync 660 to have completed -- there is no way that synchronize_rcu() would have
704 been able to write-acquire the lock otherwise. 661 been able to write-acquire the lock otherwise. The smp_mb__after_spinlock()
705 promotes synchronize_rcu() to a full memory ba 662 promotes synchronize_rcu() to a full memory barrier in compliance with
706 the "Memory-Barrier Guarantees" listed in: 663 the "Memory-Barrier Guarantees" listed in:
707 664
708 Design/Requirements/Requirements.rst !! 665 Documentation/RCU/Design/Requirements/Requirements.rst
709 666
710 It is possible to nest rcu_read_lock(), since 667 It is possible to nest rcu_read_lock(), since reader-writer locks may
711 be recursively acquired. Note also that rcu_r 668 be recursively acquired. Note also that rcu_read_lock() is immune
712 from deadlock (an important property of RCU). 669 from deadlock (an important property of RCU). The reason for this is
713 that the only thing that can block rcu_read_lo 670 that the only thing that can block rcu_read_lock() is a synchronize_rcu().
714 But synchronize_rcu() does not acquire any loc 671 But synchronize_rcu() does not acquire any locks while holding rcu_gp_mutex,
715 so there can be no deadlock cycle. 672 so there can be no deadlock cycle.
716 673
717 .. _quiz_1: 674 .. _quiz_1:
718 675
719 Quick Quiz #1: 676 Quick Quiz #1:
720 Why is this argument naive? H 677 Why is this argument naive? How could a deadlock
721 occur when using this algorith 678 occur when using this algorithm in a real-world Linux
722 kernel? How could this deadlo 679 kernel? How could this deadlock be avoided?
723 680
724 :ref:`Answers to Quick Quiz <9_whatisRCU>` !! 681 :ref:`Answers to Quick Quiz <8_whatisRCU>`
725 682
726 5B. "TOY" EXAMPLE #2: CLASSIC RCU 683 5B. "TOY" EXAMPLE #2: CLASSIC RCU
727 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 684 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
728 This section presents a "toy" RCU implementati 685 This section presents a "toy" RCU implementation that is based on
729 "classic RCU". It is also short on performanc 686 "classic RCU". It is also short on performance (but only for updates) and
730 on features such as hotplug CPU and the abilit !! 687 on features such as hotplug CPU and the ability to run in CONFIG_PREEMPT
731 kernels. The definitions of rcu_dereference() 688 kernels. The definitions of rcu_dereference() and rcu_assign_pointer()
732 are the same as those shown in the preceding s 689 are the same as those shown in the preceding section, so they are omitted.
733 :: 690 ::
734 691
735 void rcu_read_lock(void) { } 692 void rcu_read_lock(void) { }
736 693
737 void rcu_read_unlock(void) { } 694 void rcu_read_unlock(void) { }
738 695
739 void synchronize_rcu(void) 696 void synchronize_rcu(void)
740 { 697 {
741 int cpu; 698 int cpu;
742 699
743 for_each_possible_cpu(cpu) 700 for_each_possible_cpu(cpu)
744 run_on(cpu); 701 run_on(cpu);
745 } 702 }
746 703
747 Note that rcu_read_lock() and rcu_read_unlock( 704 Note that rcu_read_lock() and rcu_read_unlock() do absolutely nothing.
748 This is the great strength of classic RCU in a 705 This is the great strength of classic RCU in a non-preemptive kernel:
749 read-side overhead is precisely zero, at least 706 read-side overhead is precisely zero, at least on non-Alpha CPUs.
750 And there is absolutely no way that rcu_read_l 707 And there is absolutely no way that rcu_read_lock() can possibly
751 participate in a deadlock cycle! 708 participate in a deadlock cycle!
752 709
753 The implementation of synchronize_rcu() simply 710 The implementation of synchronize_rcu() simply schedules itself on each
754 CPU in turn. The run_on() primitive can be im 711 CPU in turn. The run_on() primitive can be implemented straightforwardly
755 in terms of the sched_setaffinity() primitive. 712 in terms of the sched_setaffinity() primitive. Of course, a somewhat less
756 "toy" implementation would restore the affinit 713 "toy" implementation would restore the affinity upon completion rather
757 than just leaving all tasks running on the las 714 than just leaving all tasks running on the last CPU, but when I said
758 "toy", I meant **toy**! 715 "toy", I meant **toy**!
759 716
760 So how the heck is this supposed to work??? 717 So how the heck is this supposed to work???
761 718
762 Remember that it is illegal to block while in 719 Remember that it is illegal to block while in an RCU read-side critical
763 section. Therefore, if a given CPU executes a 720 section. Therefore, if a given CPU executes a context switch, we know
764 that it must have completed all preceding RCU 721 that it must have completed all preceding RCU read-side critical sections.
765 Once **all** CPUs have executed a context swit 722 Once **all** CPUs have executed a context switch, then **all** preceding
766 RCU read-side critical sections will have comp 723 RCU read-side critical sections will have completed.
767 724
768 So, suppose that we remove a data item from it 725 So, suppose that we remove a data item from its structure and then invoke
769 synchronize_rcu(). Once synchronize_rcu() ret 726 synchronize_rcu(). Once synchronize_rcu() returns, we are guaranteed
770 that there are no RCU read-side critical secti 727 that there are no RCU read-side critical sections holding a reference
771 to that data item, so we can safely reclaim it 728 to that data item, so we can safely reclaim it.
772 729
773 .. _quiz_2: 730 .. _quiz_2:
774 731
775 Quick Quiz #2: 732 Quick Quiz #2:
776 Give an example where Classic 733 Give an example where Classic RCU's read-side
777 overhead is **negative**. 734 overhead is **negative**.
778 735
779 :ref:`Answers to Quick Quiz <9_whatisRCU>` !! 736 :ref:`Answers to Quick Quiz <8_whatisRCU>`
780 737
781 .. _quiz_3: 738 .. _quiz_3:
782 739
783 Quick Quiz #3: 740 Quick Quiz #3:
784 If it is illegal to block in a 741 If it is illegal to block in an RCU read-side
785 critical section, what the hec 742 critical section, what the heck do you do in
786 CONFIG_PREEMPT_RT, where norma !! 743 PREEMPT_RT, where normal spinlocks can block???
787 744
788 :ref:`Answers to Quick Quiz <9_whatisRCU>` !! 745 :ref:`Answers to Quick Quiz <8_whatisRCU>`
789 746
790 .. _6_whatisRCU: 747 .. _6_whatisRCU:
791 748
792 6. ANALOGY WITH READER-WRITER LOCKING 749 6. ANALOGY WITH READER-WRITER LOCKING
793 -------------------------------------- 750 --------------------------------------
794 751
795 Although RCU can be used in many different way 752 Although RCU can be used in many different ways, a very common use of
796 RCU is analogous to reader-writer locking. Th 753 RCU is analogous to reader-writer locking. The following unified
797 diff shows how closely related RCU and reader- 754 diff shows how closely related RCU and reader-writer locking can be.
798 :: 755 ::
799 756
800 @@ -5,5 +5,5 @@ struct el { 757 @@ -5,5 +5,5 @@ struct el {
801 int data; 758 int data;
802 /* Other data fields */ 759 /* Other data fields */
803 }; 760 };
804 -rwlock_t listmutex; 761 -rwlock_t listmutex;
805 +spinlock_t listmutex; 762 +spinlock_t listmutex;
806 struct el head; 763 struct el head;
807 764
808 @@ -13,15 +14,15 @@ 765 @@ -13,15 +14,15 @@
809 struct list_head *lp; 766 struct list_head *lp;
810 struct el *p; 767 struct el *p;
811 768
812 - read_lock(&listmutex); 769 - read_lock(&listmutex);
813 - list_for_each_entry(p, head, l 770 - list_for_each_entry(p, head, lp) {
814 + rcu_read_lock(); 771 + rcu_read_lock();
815 + list_for_each_entry_rcu(p, hea 772 + list_for_each_entry_rcu(p, head, lp) {
816 if (p->key == key) { 773 if (p->key == key) {
817 *result = p->d 774 *result = p->data;
818 - read_unlock(&l 775 - read_unlock(&listmutex);
819 + rcu_read_unloc 776 + rcu_read_unlock();
820 return 1; 777 return 1;
821 } 778 }
822 } 779 }
823 - read_unlock(&listmutex); 780 - read_unlock(&listmutex);
824 + rcu_read_unlock(); 781 + rcu_read_unlock();
825 return 0; 782 return 0;
826 } 783 }
827 784
828 @@ -29,15 +30,16 @@ 785 @@ -29,15 +30,16 @@
829 { 786 {
830 struct el *p; 787 struct el *p;
831 788
832 - write_lock(&listmutex); 789 - write_lock(&listmutex);
833 + spin_lock(&listmutex); 790 + spin_lock(&listmutex);
834 list_for_each_entry(p, head, l 791 list_for_each_entry(p, head, lp) {
835 if (p->key == key) { 792 if (p->key == key) {
836 - list_del(&p->l 793 - list_del(&p->list);
837 - write_unlock(& 794 - write_unlock(&listmutex);
838 + list_del_rcu(& 795 + list_del_rcu(&p->list);
839 + spin_unlock(&l 796 + spin_unlock(&listmutex);
840 + synchronize_rc 797 + synchronize_rcu();
841 kfree(p); 798 kfree(p);
842 return 1; 799 return 1;
843 } 800 }
844 } 801 }
845 - write_unlock(&listmutex); 802 - write_unlock(&listmutex);
846 + spin_unlock(&listmutex); 803 + spin_unlock(&listmutex);
847 return 0; 804 return 0;
848 } 805 }
849 806
850 Or, for those who prefer a side-by-side listin 807 Or, for those who prefer a side-by-side listing::
851 808
852 1 struct el { 1 stru 809 1 struct el { 1 struct el {
853 2 struct list_head list; 2 st 810 2 struct list_head list; 2 struct list_head list;
854 3 long key; 3 lo 811 3 long key; 3 long key;
855 4 spinlock_t mutex; 4 sp 812 4 spinlock_t mutex; 4 spinlock_t mutex;
856 5 int data; 5 in 813 5 int data; 5 int data;
857 6 /* Other data fields */ 6 /* 814 6 /* Other data fields */ 6 /* Other data fields */
858 7 }; 7 }; 815 7 }; 7 };
859 8 rwlock_t listmutex; 8 spin 816 8 rwlock_t listmutex; 8 spinlock_t listmutex;
860 9 struct el head; 9 stru 817 9 struct el head; 9 struct el head;
861 818
862 :: 819 ::
863 820
864 1 int search(long key, int *result) 1 int 821 1 int search(long key, int *result) 1 int search(long key, int *result)
865 2 { 2 { 822 2 { 2 {
866 3 struct list_head *lp; 3 s 823 3 struct list_head *lp; 3 struct list_head *lp;
867 4 struct el *p; 4 s 824 4 struct el *p; 4 struct el *p;
868 5 5 825 5 5
869 6 read_lock(&listmutex); 6 r 826 6 read_lock(&listmutex); 6 rcu_read_lock();
870 7 list_for_each_entry(p, head, lp) { 7 l 827 7 list_for_each_entry(p, head, lp) { 7 list_for_each_entry_rcu(p, head, lp) {
871 8 if (p->key == key) { 8 828 8 if (p->key == key) { 8 if (p->key == key) {
872 9 *result = p->data; 9 829 9 *result = p->data; 9 *result = p->data;
873 10 read_unlock(&listmutex); 10 830 10 read_unlock(&listmutex); 10 rcu_read_unlock();
874 11 return 1; 11 831 11 return 1; 11 return 1;
875 12 } 12 832 12 } 12 }
876 13 } 13 } 833 13 } 13 }
877 14 read_unlock(&listmutex); 14 r 834 14 read_unlock(&listmutex); 14 rcu_read_unlock();
878 15 return 0; 15 r 835 15 return 0; 15 return 0;
879 16 } 16 } 836 16 } 16 }
880 837
881 :: 838 ::
882 839
883 1 int delete(long key) 1 int 840 1 int delete(long key) 1 int delete(long key)
884 2 { 2 { 841 2 { 2 {
885 3 struct el *p; 3 s 842 3 struct el *p; 3 struct el *p;
886 4 4 843 4 4
887 5 write_lock(&listmutex); 5 s 844 5 write_lock(&listmutex); 5 spin_lock(&listmutex);
888 6 list_for_each_entry(p, head, lp) { 6 l 845 6 list_for_each_entry(p, head, lp) { 6 list_for_each_entry(p, head, lp) {
889 7 if (p->key == key) { 7 846 7 if (p->key == key) { 7 if (p->key == key) {
890 8 list_del(&p->list); 8 847 8 list_del(&p->list); 8 list_del_rcu(&p->list);
891 9 write_unlock(&listmutex); 9 848 9 write_unlock(&listmutex); 9 spin_unlock(&listmutex);
892 10 849 10 synchronize_rcu();
893 10 kfree(p); 11 850 10 kfree(p); 11 kfree(p);
894 11 return 1; 12 851 11 return 1; 12 return 1;
895 12 } 13 852 12 } 13 }
896 13 } 14 } 853 13 } 14 }
897 14 write_unlock(&listmutex); 15 s 854 14 write_unlock(&listmutex); 15 spin_unlock(&listmutex);
898 15 return 0; 16 r 855 15 return 0; 16 return 0;
899 16 } 17 } 856 16 } 17 }
900 857
901 Either way, the differences are quite small. 858 Either way, the differences are quite small. Read-side locking moves
902 to rcu_read_lock() and rcu_read_unlock, update 859 to rcu_read_lock() and rcu_read_unlock, update-side locking moves from
903 a reader-writer lock to a simple spinlock, and 860 a reader-writer lock to a simple spinlock, and a synchronize_rcu()
904 precedes the kfree(). 861 precedes the kfree().
905 862
906 However, there is one potential catch: the rea 863 However, there is one potential catch: the read-side and update-side
907 critical sections can now run concurrently. I 864 critical sections can now run concurrently. In many cases, this will
908 not be a problem, but it is necessary to check 865 not be a problem, but it is necessary to check carefully regardless.
909 For example, if multiple independent list upda 866 For example, if multiple independent list updates must be seen as
910 a single atomic update, converting to RCU will 867 a single atomic update, converting to RCU will require special care.
911 868
912 Also, the presence of synchronize_rcu() means 869 Also, the presence of synchronize_rcu() means that the RCU version of
913 delete() can now block. If this is a problem, 870 delete() can now block. If this is a problem, there is a callback-based
914 mechanism that never blocks, namely call_rcu() 871 mechanism that never blocks, namely call_rcu() or kfree_rcu(), that can
915 be used in place of synchronize_rcu(). 872 be used in place of synchronize_rcu().
916 873
917 .. _7_whatisRCU: 874 .. _7_whatisRCU:
918 875
919 7. ANALOGY WITH REFERENCE COUNTING !! 876 7. FULL LIST OF RCU APIs
920 ----------------------------------- <<
921 <<
922 The reader-writer analogy (illustrated by the <<
923 always the best way to think about using RCU. <<
924 considers RCU an effective reference count on <<
925 protected by RCU. <<
926 <<
927 A reference count typically does not prevent t <<
928 values from changing, but does prevent changes <<
929 gross change of type that happens when that ob <<
930 re-allocated for some other purpose. Once a t <<
931 object is obtained, some other mechanism is ne <<
932 access to the data in the object. This could <<
933 but with RCU the typical approach is to perfor <<
934 operations such as smp_load_acquire(), to perf <<
935 read-modify-write operations, and to provide t <<
936 RCU provides a number of support functions tha <<
937 operations and ordering, such as the list_for_ <<
938 used in the previous section. <<
939 <<
940 A more focused view of the reference counting <<
941 between rcu_read_lock() and rcu_read_unlock(), <<
942 rcu_dereference() on a pointer marked as ``__r <<
943 though a reference-count on that object has be <<
944 This prevents the object from changing type. <<
945 will depend on normal expectations of objects <<
946 typically includes that spinlocks can still be <<
947 reference counters can be safely manipulated, <<
948 can be safely dereferenced. <<
949 <<
950 Some operations that one might expect to see o <<
951 which an RCU reference is held include: <<
952 <<
953 - Copying out data that is guaranteed to be s <<
954 - Using kref_get_unless_zero() or similar to <<
955 reference. This may fail of course. <<
956 - Acquiring a spinlock in the object, and che <<
957 is the expected object and if so, manipulat <<
958 <<
959 The understanding that RCU provides a referenc <<
960 change of type is particularly visible with ob <<
961 slab cache marked ``SLAB_TYPESAFE_BY_RCU``. R <<
962 reference to an object from such a cache that <<
963 and the memory reallocated to a completely dif <<
964 the same type. In this case RCU doesn't even <<
965 object from changing, only its type. So the o <<
966 one expected, but it will be one where it is s <<
967 (and then potentially acquiring a spinlock), a <<
968 to check whether the identity matches expectat <<
969 to simply acquire the spinlock without first t <<
970 unfortunately any spinlock in a ``SLAB_TYPESAF <<
971 initialized after each and every call to kmem_ <<
972 reference-free spinlock acquisition completely <<
973 using ``SLAB_TYPESAFE_BY_RCU``, make proper us <<
974 (Those willing to initialize their locks in a <<
975 may also use locking, including cache-friendly <<
976 <<
977 With traditional reference counting -- such as <<
978 kref library in Linux -- there is typically co <<
979 reference to an object is dropped. With kref, <<
980 passed to kref_put(). When RCU is being used, <<
981 must not be run until all ``__rcu`` pointers r <<
982 been updated, and then a grace period has pass <<
983 globally visible pointer to the object must be <<
984 potential counted reference, and the finalizat <<
985 using call_rcu() only after all those pointers <<
986 <<
987 To see how to choose between these two analogi <<
988 reader-writer lock and RCU as a reference coun <<
989 to reflect on the scale of the thing being pro <<
990 lock analogy looks at larger multi-part object <<
991 and shows how RCU can facilitate concurrency w <<
992 to, and removed from, the list. The reference <<
993 the individual objects and looks at how they c <<
994 within whatever whole they are a part of. <<
995 <<
996 .. _8_whatisRCU: <<
997 <<
998 8. FULL LIST OF RCU APIs <<
999 ------------------------- 877 -------------------------
1000 878
1001 The RCU APIs are documented in docbook-format 879 The RCU APIs are documented in docbook-format header comments in the
1002 Linux-kernel source code, but it helps to hav 880 Linux-kernel source code, but it helps to have a full list of the
1003 APIs, since there does not appear to be a way 881 APIs, since there does not appear to be a way to categorize them
1004 in docbook. Here is the list, by category. 882 in docbook. Here is the list, by category.
1005 883
1006 RCU list traversal:: 884 RCU list traversal::
1007 885
1008 list_entry_rcu 886 list_entry_rcu
1009 list_entry_lockless 887 list_entry_lockless
1010 list_first_entry_rcu 888 list_first_entry_rcu
1011 list_next_rcu 889 list_next_rcu
1012 list_for_each_entry_rcu 890 list_for_each_entry_rcu
1013 list_for_each_entry_continue_rcu 891 list_for_each_entry_continue_rcu
1014 list_for_each_entry_from_rcu 892 list_for_each_entry_from_rcu
1015 list_first_or_null_rcu 893 list_first_or_null_rcu
1016 list_next_or_null_rcu 894 list_next_or_null_rcu
1017 hlist_first_rcu 895 hlist_first_rcu
1018 hlist_next_rcu 896 hlist_next_rcu
1019 hlist_pprev_rcu 897 hlist_pprev_rcu
1020 hlist_for_each_entry_rcu 898 hlist_for_each_entry_rcu
1021 hlist_for_each_entry_rcu_bh 899 hlist_for_each_entry_rcu_bh
1022 hlist_for_each_entry_from_rcu 900 hlist_for_each_entry_from_rcu
1023 hlist_for_each_entry_continue_rcu 901 hlist_for_each_entry_continue_rcu
1024 hlist_for_each_entry_continue_rcu_bh 902 hlist_for_each_entry_continue_rcu_bh
1025 hlist_nulls_first_rcu 903 hlist_nulls_first_rcu
1026 hlist_nulls_for_each_entry_rcu 904 hlist_nulls_for_each_entry_rcu
1027 hlist_bl_first_rcu 905 hlist_bl_first_rcu
1028 hlist_bl_for_each_entry_rcu 906 hlist_bl_for_each_entry_rcu
1029 907
1030 RCU pointer/list update:: 908 RCU pointer/list update::
1031 909
1032 rcu_assign_pointer 910 rcu_assign_pointer
1033 list_add_rcu 911 list_add_rcu
1034 list_add_tail_rcu 912 list_add_tail_rcu
1035 list_del_rcu 913 list_del_rcu
1036 list_replace_rcu 914 list_replace_rcu
1037 hlist_add_behind_rcu 915 hlist_add_behind_rcu
1038 hlist_add_before_rcu 916 hlist_add_before_rcu
1039 hlist_add_head_rcu 917 hlist_add_head_rcu
1040 hlist_add_tail_rcu 918 hlist_add_tail_rcu
1041 hlist_del_rcu 919 hlist_del_rcu
1042 hlist_del_init_rcu 920 hlist_del_init_rcu
1043 hlist_replace_rcu 921 hlist_replace_rcu
1044 list_splice_init_rcu 922 list_splice_init_rcu
1045 list_splice_tail_init_rcu 923 list_splice_tail_init_rcu
1046 hlist_nulls_del_init_rcu 924 hlist_nulls_del_init_rcu
1047 hlist_nulls_del_rcu 925 hlist_nulls_del_rcu
1048 hlist_nulls_add_head_rcu 926 hlist_nulls_add_head_rcu
1049 hlist_bl_add_head_rcu 927 hlist_bl_add_head_rcu
1050 hlist_bl_del_init_rcu 928 hlist_bl_del_init_rcu
1051 hlist_bl_del_rcu 929 hlist_bl_del_rcu
1052 hlist_bl_set_first_rcu 930 hlist_bl_set_first_rcu
1053 931
1054 RCU:: 932 RCU::
1055 933
1056 Critical sections Grace period 934 Critical sections Grace period Barrier
1057 935
1058 rcu_read_lock synchronize_n 936 rcu_read_lock synchronize_net rcu_barrier
1059 rcu_read_unlock synchronize_r 937 rcu_read_unlock synchronize_rcu
1060 rcu_dereference synchronize_r 938 rcu_dereference synchronize_rcu_expedited
1061 rcu_read_lock_held call_rcu 939 rcu_read_lock_held call_rcu
1062 rcu_dereference_check kfree_rcu 940 rcu_dereference_check kfree_rcu
1063 rcu_dereference_protected 941 rcu_dereference_protected
1064 942
1065 bh:: 943 bh::
1066 944
1067 Critical sections Grace period 945 Critical sections Grace period Barrier
1068 946
1069 rcu_read_lock_bh call_rcu 947 rcu_read_lock_bh call_rcu rcu_barrier
1070 rcu_read_unlock_bh synchronize_r 948 rcu_read_unlock_bh synchronize_rcu
1071 [local_bh_disable] synchronize_r 949 [local_bh_disable] synchronize_rcu_expedited
1072 [and friends] 950 [and friends]
1073 rcu_dereference_bh 951 rcu_dereference_bh
1074 rcu_dereference_bh_check 952 rcu_dereference_bh_check
1075 rcu_dereference_bh_protected 953 rcu_dereference_bh_protected
1076 rcu_read_lock_bh_held 954 rcu_read_lock_bh_held
1077 955
1078 sched:: 956 sched::
1079 957
1080 Critical sections Grace period 958 Critical sections Grace period Barrier
1081 959
1082 rcu_read_lock_sched call_rcu 960 rcu_read_lock_sched call_rcu rcu_barrier
1083 rcu_read_unlock_sched synchronize_r 961 rcu_read_unlock_sched synchronize_rcu
1084 [preempt_disable] synchronize_r 962 [preempt_disable] synchronize_rcu_expedited
1085 [and friends] 963 [and friends]
1086 rcu_read_lock_sched_notrace 964 rcu_read_lock_sched_notrace
1087 rcu_read_unlock_sched_notrace 965 rcu_read_unlock_sched_notrace
1088 rcu_dereference_sched 966 rcu_dereference_sched
1089 rcu_dereference_sched_check 967 rcu_dereference_sched_check
1090 rcu_dereference_sched_protected 968 rcu_dereference_sched_protected
1091 rcu_read_lock_sched_held 969 rcu_read_lock_sched_held
1092 970
1093 971
1094 RCU-Tasks:: <<
1095 <<
1096 Critical sections Grace period <<
1097 <<
1098 N/A call_rcu_task <<
1099 synchronize_r <<
1100 <<
1101 <<
1102 RCU-Tasks-Rude:: <<
1103 <<
1104 Critical sections Grace period <<
1105 <<
1106 N/A call_rcu_task <<
1107 synchronize_r <<
1108 <<
1109 <<
1110 RCU-Tasks-Trace:: <<
1111 <<
1112 Critical sections Grace period <<
1113 <<
1114 rcu_read_lock_trace call_rcu_task <<
1115 rcu_read_unlock_trace synchronize_r <<
1116 <<
1117 <<
1118 SRCU:: 972 SRCU::
1119 973
1120 Critical sections Grace period 974 Critical sections Grace period Barrier
1121 975
1122 srcu_read_lock call_srcu 976 srcu_read_lock call_srcu srcu_barrier
1123 srcu_read_unlock synchronize_s 977 srcu_read_unlock synchronize_srcu
1124 srcu_dereference synchronize_s 978 srcu_dereference synchronize_srcu_expedited
1125 srcu_dereference_check 979 srcu_dereference_check
1126 srcu_read_lock_held 980 srcu_read_lock_held
1127 981
1128 SRCU: Initialization/cleanup:: 982 SRCU: Initialization/cleanup::
1129 983
1130 DEFINE_SRCU 984 DEFINE_SRCU
1131 DEFINE_STATIC_SRCU 985 DEFINE_STATIC_SRCU
1132 init_srcu_struct 986 init_srcu_struct
1133 cleanup_srcu_struct 987 cleanup_srcu_struct
1134 988
1135 All: lockdep-checked RCU utility APIs:: !! 989 All: lockdep-checked RCU-protected pointer access::
1136 990
>> 991 rcu_access_pointer
>> 992 rcu_dereference_raw
1137 RCU_LOCKDEP_WARN 993 RCU_LOCKDEP_WARN
1138 rcu_sleep_check 994 rcu_sleep_check
1139 !! 995 RCU_NONIDLE
1140 All: Unchecked RCU-protected pointer access:: <<
1141 <<
1142 rcu_dereference_raw <<
1143 <<
1144 All: Unchecked RCU-protected pointer access w <<
1145 <<
1146 rcu_access_pointer <<
1147 996
1148 See the comment headers in the source code (o 997 See the comment headers in the source code (or the docbook generated
1149 from them) for more information. 998 from them) for more information.
1150 999
1151 However, given that there are no fewer than f 1000 However, given that there are no fewer than four families of RCU APIs
1152 in the Linux kernel, how do you choose which 1001 in the Linux kernel, how do you choose which one to use? The following
1153 list can be helpful: 1002 list can be helpful:
1154 1003
1155 a. Will readers need to block? If so, y 1004 a. Will readers need to block? If so, you need SRCU.
1156 1005
1157 b. Will readers need to block and are yo !! 1006 b. What about the -rt patchset? If readers would need to block
1158 example, ftrace or BPF? If so, you n !! 1007 in an non-rt kernel, you need SRCU. If readers would block
1159 RCU-tasks-rude, and/or RCU-tasks-trac !! 1008 in a -rt kernel, but not in a non-rt kernel, SRCU is not
1160 !! 1009 necessary. (The -rt patchset turns spinlocks into sleeplocks,
1161 c. What about the -rt patchset? If read !! 1010 hence this distinction.)
1162 an non-rt kernel, you need SRCU. If <<
1163 acquiring spinlocks in a -rt kernel, <<
1164 SRCU is not necessary. (The -rt patc <<
1165 sleeplocks, hence this distinction.) <<
1166 1011
1167 d. Do you need to treat NMI handlers, ha !! 1012 c. Do you need to treat NMI handlers, hardirq handlers,
1168 and code segments with preemption dis 1013 and code segments with preemption disabled (whether
1169 via preempt_disable(), local_irq_save 1014 via preempt_disable(), local_irq_save(), local_bh_disable(),
1170 or some other mechanism) as if they w 1015 or some other mechanism) as if they were explicit RCU readers?
1171 If so, RCU-sched readers are the only !! 1016 If so, RCU-sched is the only choice that will work for you.
1172 for you, but since about v4.20 you us <<
1173 update primitives. <<
1174 <<
1175 e. Do you need RCU grace periods to comp <<
1176 softirq monopolization of one or more <<
1177 is your code subject to network-based <<
1178 If so, you should disable softirq acr <<
1179 example, by using rcu_read_lock_bh(). <<
1180 use can use the vanilla RCU update pr <<
1181 1017
1182 f. Is your workload too update-intensive !! 1018 d. Do you need RCU grace periods to complete even in the face
>> 1019 of softirq monopolization of one or more of the CPUs? For
>> 1020 example, is your code subject to network-based denial-of-service
>> 1021 attacks? If so, you should disable softirq across your readers,
>> 1022 for example, by using rcu_read_lock_bh().
>> 1023
>> 1024 e. Is your workload too update-intensive for normal use of
1183 RCU, but inappropriate for other sync 1025 RCU, but inappropriate for other synchronization mechanisms?
1184 If so, consider SLAB_TYPESAFE_BY_RCU 1026 If so, consider SLAB_TYPESAFE_BY_RCU (which was originally
1185 named SLAB_DESTROY_BY_RCU). But plea 1027 named SLAB_DESTROY_BY_RCU). But please be careful!
1186 1028
1187 g. Do you need read-side critical sectio !! 1029 f. Do you need read-side critical sections that are respected
1188 on CPUs that are deep in the idle loo !! 1030 even though they are in the middle of the idle loop, during
1189 from user-mode execution, or on an of !! 1031 user-mode execution, or on an offlined CPU? If so, SRCU is the
1190 and RCU Tasks Trace are the only choi !! 1032 only choice that will work for you.
1191 with SRCU being strongly preferred in <<
1192 1033
1193 h. Otherwise, use RCU. !! 1034 g. Otherwise, use RCU.
1194 1035
1195 Of course, this all assumes that you have det 1036 Of course, this all assumes that you have determined that RCU is in fact
1196 the right tool for your job. 1037 the right tool for your job.
1197 1038
1198 .. _9_whatisRCU: !! 1039 .. _8_whatisRCU:
1199 1040
1200 9. ANSWERS TO QUICK QUIZZES !! 1041 8. ANSWERS TO QUICK QUIZZES
1201 ---------------------------- 1042 ----------------------------
1202 1043
1203 Quick Quiz #1: 1044 Quick Quiz #1:
1204 Why is this argument naive? 1045 Why is this argument naive? How could a deadlock
1205 occur when using this algorit 1046 occur when using this algorithm in a real-world Linux
1206 kernel? [Referring to the lo 1047 kernel? [Referring to the lock-based "toy" RCU
1207 algorithm.] 1048 algorithm.]
1208 1049
1209 Answer: 1050 Answer:
1210 Consider the following sequen 1051 Consider the following sequence of events:
1211 1052
1212 1. CPU 0 acquires some u 1053 1. CPU 0 acquires some unrelated lock, call it
1213 "problematic_lock", d 1054 "problematic_lock", disabling irq via
1214 spin_lock_irqsave(). 1055 spin_lock_irqsave().
1215 1056
1216 2. CPU 1 enters synchron 1057 2. CPU 1 enters synchronize_rcu(), write-acquiring
1217 rcu_gp_mutex. 1058 rcu_gp_mutex.
1218 1059
1219 3. CPU 0 enters rcu_read 1060 3. CPU 0 enters rcu_read_lock(), but must wait
1220 because CPU 1 holds r 1061 because CPU 1 holds rcu_gp_mutex.
1221 1062
1222 4. CPU 1 is interrupted, 1063 4. CPU 1 is interrupted, and the irq handler
1223 attempts to acquire p 1064 attempts to acquire problematic_lock.
1224 1065
1225 The system is now deadlocked. 1066 The system is now deadlocked.
1226 1067
1227 One way to avoid this deadloc 1068 One way to avoid this deadlock is to use an approach like
1228 that of CONFIG_PREEMPT_RT, wh 1069 that of CONFIG_PREEMPT_RT, where all normal spinlocks
1229 become blocking locks, and al 1070 become blocking locks, and all irq handlers execute in
1230 the context of special tasks. 1071 the context of special tasks. In this case, in step 4
1231 above, the irq handler would 1072 above, the irq handler would block, allowing CPU 1 to
1232 release rcu_gp_mutex, avoidin 1073 release rcu_gp_mutex, avoiding the deadlock.
1233 1074
1234 Even in the absence of deadlo 1075 Even in the absence of deadlock, this RCU implementation
1235 allows latency to "bleed" fro 1076 allows latency to "bleed" from readers to other
1236 readers through synchronize_r 1077 readers through synchronize_rcu(). To see this,
1237 consider task A in an RCU rea 1078 consider task A in an RCU read-side critical section
1238 (thus read-holding rcu_gp_mut 1079 (thus read-holding rcu_gp_mutex), task B blocked
1239 attempting to write-acquire r 1080 attempting to write-acquire rcu_gp_mutex, and
1240 task C blocked in rcu_read_lo 1081 task C blocked in rcu_read_lock() attempting to
1241 read_acquire rcu_gp_mutex. T 1082 read_acquire rcu_gp_mutex. Task A's RCU read-side
1242 latency is holding up task C, 1083 latency is holding up task C, albeit indirectly via
1243 task B. 1084 task B.
1244 1085
1245 Realtime RCU implementations 1086 Realtime RCU implementations therefore use a counter-based
1246 approach where tasks in RCU r 1087 approach where tasks in RCU read-side critical sections
1247 cannot be blocked by tasks ex 1088 cannot be blocked by tasks executing synchronize_rcu().
1248 1089
1249 :ref:`Back to Quick Quiz #1 <quiz_1>` 1090 :ref:`Back to Quick Quiz #1 <quiz_1>`
1250 1091
1251 Quick Quiz #2: 1092 Quick Quiz #2:
1252 Give an example where Classic 1093 Give an example where Classic RCU's read-side
1253 overhead is **negative**. 1094 overhead is **negative**.
1254 1095
1255 Answer: 1096 Answer:
1256 Imagine a single-CPU system w !! 1097 Imagine a single-CPU system with a non-CONFIG_PREEMPT
1257 kernel where a routing table 1098 kernel where a routing table is used by process-context
1258 code, but can be updated by i 1099 code, but can be updated by irq-context code (for example,
1259 by an "ICMP REDIRECT" packet) 1100 by an "ICMP REDIRECT" packet). The usual way of handling
1260 this would be to have the pro 1101 this would be to have the process-context code disable
1261 interrupts while searching th 1102 interrupts while searching the routing table. Use of
1262 RCU allows such interrupt-dis 1103 RCU allows such interrupt-disabling to be dispensed with.
1263 Thus, without RCU, you pay th 1104 Thus, without RCU, you pay the cost of disabling interrupts,
1264 and with RCU you don't. 1105 and with RCU you don't.
1265 1106
1266 One can argue that the overhe 1107 One can argue that the overhead of RCU in this
1267 case is negative with respect 1108 case is negative with respect to the single-CPU
1268 interrupt-disabling approach. 1109 interrupt-disabling approach. Others might argue that
1269 the overhead of RCU is merely 1110 the overhead of RCU is merely zero, and that replacing
1270 the positive overhead of the 1111 the positive overhead of the interrupt-disabling scheme
1271 with the zero-overhead RCU sc 1112 with the zero-overhead RCU scheme does not constitute
1272 negative overhead. 1113 negative overhead.
1273 1114
1274 In real life, of course, thin 1115 In real life, of course, things are more complex. But
1275 even the theoretical possibil 1116 even the theoretical possibility of negative overhead for
1276 a synchronization primitive i 1117 a synchronization primitive is a bit unexpected. ;-)
1277 1118
1278 :ref:`Back to Quick Quiz #2 <quiz_2>` 1119 :ref:`Back to Quick Quiz #2 <quiz_2>`
1279 1120
1280 Quick Quiz #3: 1121 Quick Quiz #3:
1281 If it is illegal to block in 1122 If it is illegal to block in an RCU read-side
1282 critical section, what the he 1123 critical section, what the heck do you do in
1283 CONFIG_PREEMPT_RT, where norm !! 1124 PREEMPT_RT, where normal spinlocks can block???
1284 1125
1285 Answer: 1126 Answer:
1286 Just as CONFIG_PREEMPT_RT per !! 1127 Just as PREEMPT_RT permits preemption of spinlock
1287 critical sections, it permits 1128 critical sections, it permits preemption of RCU
1288 read-side critical sections. 1129 read-side critical sections. It also permits
1289 spinlocks blocking while in R 1130 spinlocks blocking while in RCU read-side critical
1290 sections. 1131 sections.
1291 1132
1292 Why the apparent inconsistenc 1133 Why the apparent inconsistency? Because it is
1293 possible to use priority boos 1134 possible to use priority boosting to keep the RCU
1294 grace periods short if need b 1135 grace periods short if need be (for example, if running
1295 short of memory). In contras 1136 short of memory). In contrast, if blocking waiting
1296 for (say) network reception, 1137 for (say) network reception, there is no way to know
1297 what should be boosted. Espe 1138 what should be boosted. Especially given that the
1298 process we need to boost migh 1139 process we need to boost might well be a human being
1299 who just went out for a pizza 1140 who just went out for a pizza or something. And although
1300 a computer-operated cattle pr 1141 a computer-operated cattle prod might arouse serious
1301 interest, it might also provo 1142 interest, it might also provoke serious objections.
1302 Besides, how does the compute 1143 Besides, how does the computer know what pizza parlor
1303 the human being went to??? 1144 the human being went to???
1304 1145
1305 :ref:`Back to Quick Quiz #3 <quiz_3>` 1146 :ref:`Back to Quick Quiz #3 <quiz_3>`
1306 1147
1307 ACKNOWLEDGEMENTS 1148 ACKNOWLEDGEMENTS
1308 1149
1309 My thanks to the people who helped make this 1150 My thanks to the people who helped make this human-readable, including
1310 Jon Walpole, Josh Triplett, Serge Hallyn, Suz 1151 Jon Walpole, Josh Triplett, Serge Hallyn, Suzanne Wood, and Alan Stern.
1311 1152
1312 1153
1313 For more information, see http://www.rdrop.co 1154 For more information, see http://www.rdrop.com/users/paulmck/RCU.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.