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? https://lwn.net/Articles/262464/
10 | 2. What is RCU? Part 2: Usage https://l 10 | 2. What is RCU? Part 2: Usage https://lwn.net/Articles/263130/
11 | 3. RCU part 3: the RCU API https://l 11 | 3. RCU part 3: the RCU API https://lwn.net/Articles/264090/
12 | 4. The RCU API, 2010 Edition https://l 12 | 4. The RCU API, 2010 Edition https://lwn.net/Articles/418853/
13 | 2010 Big API Table https://l 13 | 2010 Big API Table https://lwn.net/Articles/419086/
14 | 5. The RCU API, 2014 Edition https://l 14 | 5. The RCU API, 2014 Edition https://lwn.net/Articles/609904/
15 | 2014 Big API Table https://l 15 | 2014 Big API Table https://lwn.net/Articles/609973/
16 | 6. The RCU API, 2019 Edition https://l 16 | 6. The RCU API, 2019 Edition https://lwn.net/Articles/777036/
17 | 2019 Big API Table https://l 17 | 2019 Big API Table https://lwn.net/Articles/777165/
18 18
19 For those preferring video: 19 For those preferring video:
20 20
21 | 1. Unraveling RCU Mysteries: Fundamentals 21 | 1. Unraveling RCU Mysteries: Fundamentals https://www.linuxfoundation.org/webinars/unraveling-rcu-usage-mysteries
22 | 2. Unraveling RCU Mysteries: Additional U 22 | 2. Unraveling RCU Mysteries: Additional Use Cases https://www.linuxfoundation.org/webinars/unraveling-rcu-usage-mysteries-additional-use-cases
23 23
24 24
25 What is RCU? 25 What is RCU?
26 26
27 RCU is a synchronization mechanism that was ad 27 RCU is a synchronization mechanism that was added to the Linux kernel
28 during the 2.5 development effort that is opti 28 during the 2.5 development effort that is optimized for read-mostly
29 situations. Although RCU is actually quite si 29 situations. Although RCU is actually quite simple, making effective use
30 of it requires you to think differently about 30 of it requires you to think differently about your code. Another part
31 of the problem is the mistaken assumption that 31 of the problem is the mistaken assumption that there is "one true way" to
32 describe and to use RCU. Instead, the experie 32 describe and to use RCU. Instead, the experience has been that different
33 people must take different paths to arrive at 33 people must take different paths to arrive at an understanding of RCU,
34 depending on their experiences and use cases. 34 depending on their experiences and use cases. This document provides
35 several different paths, as follows: 35 several different paths, as follows:
36 36
37 :ref:`1. RCU OVERVIEW <1_whatisRCU>` 37 :ref:`1. RCU OVERVIEW <1_whatisRCU>`
38 38
39 :ref:`2. WHAT IS RCU'S CORE API? <2_wha 39 :ref:`2. WHAT IS RCU'S CORE API? <2_whatisRCU>`
40 40
41 :ref:`3. WHAT ARE SOME EXAMPLE USES OF 41 :ref:`3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API? <3_whatisRCU>`
42 42
43 :ref:`4. WHAT IF MY UPDATING THREAD CAN 43 :ref:`4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? <4_whatisRCU>`
44 44
45 :ref:`5. WHAT ARE SOME SIMPLE IMPLEMENT 45 :ref:`5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU? <5_whatisRCU>`
46 46
47 :ref:`6. ANALOGY WITH READER-WRITER LOC 47 :ref:`6. ANALOGY WITH READER-WRITER LOCKING <6_whatisRCU>`
48 48
49 :ref:`7. ANALOGY WITH REFERENCE COUNTIN 49 :ref:`7. ANALOGY WITH REFERENCE COUNTING <7_whatisRCU>`
50 50
51 :ref:`8. FULL LIST OF RCU APIs <8_whati 51 :ref:`8. FULL LIST OF RCU APIs <8_whatisRCU>`
52 52
53 :ref:`9. ANSWERS TO QUICK QUIZZES <9_wh 53 :ref:`9. ANSWERS TO QUICK QUIZZES <9_whatisRCU>`
54 54
55 People who prefer starting with a conceptual o 55 People who prefer starting with a conceptual overview should focus on
56 Section 1, though most readers will profit by 56 Section 1, though most readers will profit by reading this section at
57 some point. People who prefer to start with a 57 some point. People who prefer to start with an API that they can then
58 experiment with should focus on Section 2. Pe 58 experiment with should focus on Section 2. People who prefer to start
59 with example uses should focus on Sections 3 a 59 with example uses should focus on Sections 3 and 4. People who need to
60 understand the RCU implementation should focus 60 understand the RCU implementation should focus on Section 5, then dive
61 into the kernel source code. People who reaso 61 into the kernel source code. People who reason best by analogy should
62 focus on Section 6 and 7. Section 8 serves as 62 focus on Section 6 and 7. Section 8 serves as an index to the docbook
63 API documentation, and Section 9 is the tradit 63 API documentation, and Section 9 is the traditional answer key.
64 64
65 So, start with the section that makes the most 65 So, start with the section that makes the most sense to you and your
66 preferred method of learning. If you need to 66 preferred method of learning. If you need to know everything about
67 everything, feel free to read the whole thing 67 everything, feel free to read the whole thing -- but if you are really
68 that type of person, you have perused the sour 68 that type of person, you have perused the source code and will therefore
69 never need this document anyway. ;-) 69 never need this document anyway. ;-)
70 70
71 .. _1_whatisRCU: 71 .. _1_whatisRCU:
72 72
73 1. RCU OVERVIEW 73 1. RCU OVERVIEW
74 ---------------- 74 ----------------
75 75
76 The basic idea behind RCU is to split updates 76 The basic idea behind RCU is to split updates into "removal" and
77 "reclamation" phases. The removal phase remov 77 "reclamation" phases. The removal phase removes references to data items
78 within a data structure (possibly by replacing 78 within a data structure (possibly by replacing them with references to
79 new versions of these data items), and can run 79 new versions of these data items), and can run concurrently with readers.
80 The reason that it is safe to run the removal 80 The reason that it is safe to run the removal phase concurrently with
81 readers is the semantics of modern CPUs guaran 81 readers is the semantics of modern CPUs guarantee that readers will see
82 either the old or the new version of the data 82 either the old or the new version of the data structure rather than a
83 partially updated reference. The reclamation 83 partially updated reference. The reclamation phase does the work of reclaiming
84 (e.g., freeing) the data items removed from th 84 (e.g., freeing) the data items removed from the data structure during the
85 removal phase. Because reclaiming data items 85 removal phase. Because reclaiming data items can disrupt any readers
86 concurrently referencing those data items, the 86 concurrently referencing those data items, the reclamation phase must
87 not start until readers no longer hold referen 87 not start until readers no longer hold references to those data items.
88 88
89 Splitting the update into removal and reclamat 89 Splitting the update into removal and reclamation phases permits the
90 updater to perform the removal phase immediate 90 updater to perform the removal phase immediately, and to defer the
91 reclamation phase until all readers active dur 91 reclamation phase until all readers active during the removal phase have
92 completed, either by blocking until they finis 92 completed, either by blocking until they finish or by registering a
93 callback that is invoked after they finish. O 93 callback that is invoked after they finish. Only readers that are active
94 during the removal phase need be considered, b 94 during the removal phase need be considered, because any reader starting
95 after the removal phase will be unable to gain 95 after the removal phase will be unable to gain a reference to the removed
96 data items, and therefore cannot be disrupted 96 data items, and therefore cannot be disrupted by the reclamation phase.
97 97
98 So the typical RCU update sequence goes someth 98 So the typical RCU update sequence goes something like the following:
99 99
100 a. Remove pointers to a data structure, s 100 a. Remove pointers to a data structure, so that subsequent
101 readers cannot gain a reference to it. 101 readers cannot gain a reference to it.
102 102
103 b. Wait for all previous readers to compl 103 b. Wait for all previous readers to complete their RCU read-side
104 critical sections. 104 critical sections.
105 105
106 c. At this point, there cannot be any rea 106 c. At this point, there cannot be any readers who hold references
107 to the data structure, so it now may s 107 to the data structure, so it now may safely be reclaimed
108 (e.g., kfree()d). 108 (e.g., kfree()d).
109 109
110 Step (b) above is the key idea underlying RCU' 110 Step (b) above is the key idea underlying RCU's deferred destruction.
111 The ability to wait until all readers are done 111 The ability to wait until all readers are done allows RCU readers to
112 use much lighter-weight synchronization, in so 112 use much lighter-weight synchronization, in some cases, absolutely no
113 synchronization at all. In contrast, in more 113 synchronization at all. In contrast, in more conventional lock-based
114 schemes, readers must use heavy-weight synchro 114 schemes, readers must use heavy-weight synchronization in order to
115 prevent an updater from deleting the data stru 115 prevent an updater from deleting the data structure out from under them.
116 This is because lock-based updaters typically 116 This is because lock-based updaters typically update data items in place,
117 and must therefore exclude readers. In contra 117 and must therefore exclude readers. In contrast, RCU-based updaters
118 typically take advantage of the fact that writ 118 typically take advantage of the fact that writes to single aligned
119 pointers are atomic on modern CPUs, allowing a 119 pointers are atomic on modern CPUs, allowing atomic insertion, removal,
120 and replacement of data items in a linked stru 120 and replacement of data items in a linked structure without disrupting
121 readers. Concurrent RCU readers can then cont 121 readers. Concurrent RCU readers can then continue accessing the old
122 versions, and can dispense with the atomic ope 122 versions, and can dispense with the atomic operations, memory barriers,
123 and communications cache misses that are so ex 123 and communications cache misses that are so expensive on present-day
124 SMP computer systems, even in absence of lock 124 SMP computer systems, even in absence of lock contention.
125 125
126 In the three-step procedure shown above, the u 126 In the three-step procedure shown above, the updater is performing both
127 the removal and the reclamation step, but it i 127 the removal and the reclamation step, but it is often helpful for an
128 entirely different thread to do the reclamatio 128 entirely different thread to do the reclamation, as is in fact the case
129 in the Linux kernel's directory-entry cache (d 129 in the Linux kernel's directory-entry cache (dcache). Even if the same
130 thread performs both the update step (step (a) 130 thread performs both the update step (step (a) above) and the reclamation
131 step (step (c) above), it is often helpful to 131 step (step (c) above), it is often helpful to think of them separately.
132 For example, RCU readers and updaters need not 132 For example, RCU readers and updaters need not communicate at all,
133 but RCU provides implicit low-overhead communi 133 but RCU provides implicit low-overhead communication between readers
134 and reclaimers, namely, in step (b) above. 134 and reclaimers, namely, in step (b) above.
135 135
136 So how the heck can a reclaimer tell when a re 136 So how the heck can a reclaimer tell when a reader is done, given
137 that readers are not doing any sort of synchro 137 that readers are not doing any sort of synchronization operations???
138 Read on to learn about how RCU's API makes thi 138 Read on to learn about how RCU's API makes this easy.
139 139
140 .. _2_whatisRCU: 140 .. _2_whatisRCU:
141 141
142 2. WHAT IS RCU'S CORE API? 142 2. WHAT IS RCU'S CORE API?
143 --------------------------- 143 ---------------------------
144 144
145 The core RCU API is quite small: 145 The core RCU API is quite small:
146 146
147 a. rcu_read_lock() 147 a. rcu_read_lock()
148 b. rcu_read_unlock() 148 b. rcu_read_unlock()
149 c. synchronize_rcu() / call_rcu() 149 c. synchronize_rcu() / call_rcu()
150 d. rcu_assign_pointer() 150 d. rcu_assign_pointer()
151 e. rcu_dereference() 151 e. rcu_dereference()
152 152
153 There are many other members of the RCU API, b 153 There are many other members of the RCU API, but the rest can be
154 expressed in terms of these five, though most 154 expressed in terms of these five, though most implementations instead
155 express synchronize_rcu() in terms of the call 155 express synchronize_rcu() in terms of the call_rcu() callback API.
156 156
157 The five core RCU APIs are described below, th 157 The five core RCU APIs are described below, the other 18 will be enumerated
158 later. See the kernel docbook documentation f 158 later. See the kernel docbook documentation for more info, or look directly
159 at the function header comments. 159 at the function header comments.
160 160
161 rcu_read_lock() 161 rcu_read_lock()
162 ^^^^^^^^^^^^^^^ 162 ^^^^^^^^^^^^^^^
163 void rcu_read_lock(void); 163 void rcu_read_lock(void);
164 164
165 This temporal primitive is used by a r 165 This temporal primitive is used by a reader to inform the
166 reclaimer that the reader is entering 166 reclaimer that the reader is entering an RCU read-side critical
167 section. It is illegal to block while 167 section. It is illegal to block while in an RCU read-side
168 critical section, though kernels built 168 critical section, though kernels built with CONFIG_PREEMPT_RCU
169 can preempt RCU read-side critical sec 169 can preempt RCU read-side critical sections. Any RCU-protected
170 data structure accessed during an RCU 170 data structure accessed during an RCU read-side critical section
171 is guaranteed to remain unreclaimed fo 171 is guaranteed to remain unreclaimed for the full duration of that
172 critical section. Reference counts ma 172 critical section. Reference counts may be used in conjunction
173 with RCU to maintain longer-term refer 173 with RCU to maintain longer-term references to data structures.
174 174
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 <<
182 rcu_read_unlock() 175 rcu_read_unlock()
183 ^^^^^^^^^^^^^^^^^ 176 ^^^^^^^^^^^^^^^^^
184 void rcu_read_unlock(void); 177 void rcu_read_unlock(void);
185 178
186 This temporal primitives is used by a 179 This temporal primitives is used by a reader to inform the
187 reclaimer that the reader is exiting a 180 reclaimer that the reader is exiting an RCU read-side critical
188 section. Anything that enables bottom !! 181 section. Note that RCU read-side critical sections may be nested
189 or interrupts also exits an RCU read-s !! 182 and/or overlapping.
190 Releasing a spinlock also exits an RCU <<
191 <<
192 Note that RCU read-side critical secti <<
193 overlapping. <<
194 183
195 synchronize_rcu() 184 synchronize_rcu()
196 ^^^^^^^^^^^^^^^^^ 185 ^^^^^^^^^^^^^^^^^
197 void synchronize_rcu(void); 186 void synchronize_rcu(void);
198 187
199 This temporal primitive marks the end 188 This temporal primitive marks the end of updater code and the
200 beginning of reclaimer code. It does 189 beginning of reclaimer code. It does this by blocking until
201 all pre-existing RCU read-side critica 190 all pre-existing RCU read-side critical sections on all CPUs
202 have completed. Note that synchronize 191 have completed. Note that synchronize_rcu() will **not**
203 necessarily wait for any subsequent RC 192 necessarily wait for any subsequent RCU read-side critical
204 sections to complete. For example, co 193 sections to complete. For example, consider the following
205 sequence of events:: 194 sequence of events::
206 195
207 CPU 0 CPU 1 196 CPU 0 CPU 1 CPU 2
208 ----------------- --------------- 197 ----------------- ------------------------- ---------------
209 1. rcu_read_lock() 198 1. rcu_read_lock()
210 2. enters synchron 199 2. enters synchronize_rcu()
211 3. 200 3. rcu_read_lock()
212 4. rcu_read_unlock() 201 4. rcu_read_unlock()
213 5. exits synchron 202 5. exits synchronize_rcu()
214 6. 203 6. rcu_read_unlock()
215 204
216 To reiterate, synchronize_rcu() waits 205 To reiterate, synchronize_rcu() waits only for ongoing RCU
217 read-side critical sections to complet 206 read-side critical sections to complete, not necessarily for
218 any that begin after synchronize_rcu() 207 any that begin after synchronize_rcu() is invoked.
219 208
220 Of course, synchronize_rcu() does not 209 Of course, synchronize_rcu() does not necessarily return
221 **immediately** after the last pre-exi 210 **immediately** after the last pre-existing RCU read-side critical
222 section completes. For one thing, the 211 section completes. For one thing, there might well be scheduling
223 delays. For another thing, many RCU i 212 delays. For another thing, many RCU implementations process
224 requests in batches in order to improv 213 requests in batches in order to improve efficiencies, which can
225 further delay synchronize_rcu(). 214 further delay synchronize_rcu().
226 215
227 Since synchronize_rcu() is the API tha 216 Since synchronize_rcu() is the API that must figure out when
228 readers are done, its implementation i 217 readers are done, its implementation is key to RCU. For RCU
229 to be useful in all but the most read- 218 to be useful in all but the most read-intensive situations,
230 synchronize_rcu()'s overhead must also 219 synchronize_rcu()'s overhead must also be quite small.
231 220
232 The call_rcu() API is an asynchronous 221 The call_rcu() API is an asynchronous callback form of
233 synchronize_rcu(), and is described in 222 synchronize_rcu(), and is described in more detail in a later
234 section. Instead of blocking, it regi 223 section. Instead of blocking, it registers a function and
235 argument which are invoked after all o 224 argument which are invoked after all ongoing RCU read-side
236 critical sections have completed. Thi 225 critical sections have completed. This callback variant is
237 particularly useful in situations wher 226 particularly useful in situations where it is illegal to block
238 or where update-side performance is cr 227 or where update-side performance is critically important.
239 228
240 However, the call_rcu() API should not 229 However, the call_rcu() API should not be used lightly, as use
241 of the synchronize_rcu() API generally 230 of the synchronize_rcu() API generally results in simpler code.
242 In addition, the synchronize_rcu() API 231 In addition, the synchronize_rcu() API has the nice property
243 of automatically limiting update rate 232 of automatically limiting update rate should grace periods
244 be delayed. This property results in 233 be delayed. This property results in system resilience in face
245 of denial-of-service attacks. Code us 234 of denial-of-service attacks. Code using call_rcu() should limit
246 update rate in order to gain this same 235 update rate in order to gain this same sort of resilience. See
247 checklist.rst for some approaches to l 236 checklist.rst for some approaches to limiting the update rate.
248 237
249 rcu_assign_pointer() 238 rcu_assign_pointer()
250 ^^^^^^^^^^^^^^^^^^^^ 239 ^^^^^^^^^^^^^^^^^^^^
251 void rcu_assign_pointer(p, typeof(p) v 240 void rcu_assign_pointer(p, typeof(p) v);
252 241
253 Yes, rcu_assign_pointer() **is** imple !! 242 Yes, rcu_assign_pointer() **is** implemented as a macro, though it
254 it would be cool to be able to declare !! 243 would be cool to be able to declare a function in this manner.
255 (And there has been some discussion of !! 244 (Compiler experts will no doubt disagree.)
256 to the C language, so who knows?) <<
257 245
258 The updater uses this spatial macro to 246 The updater uses this spatial macro to assign a new value to an
259 RCU-protected pointer, in order to saf 247 RCU-protected pointer, in order to safely communicate the change
260 in value from the updater to the reade 248 in value from the updater to the reader. This is a spatial (as
261 opposed to temporal) macro. It does n 249 opposed to temporal) macro. It does not evaluate to an rvalue,
262 but it does provide any compiler direc !! 250 but it does execute any memory-barrier instructions required
263 instructions required for a given comp !! 251 for a given CPU architecture. Its ordering properties are that
264 Its ordering properties are that of a !! 252 of a store-release operation.
265 that is, any prior loads and stores re !! 253
266 structure are ordered before the store !! 254 Perhaps just as important, it serves to document (1) which
267 to that structure. !! 255 pointers are protected by RCU and (2) the point at which a
268 !! 256 given structure becomes accessible to other CPUs. That said,
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 257 rcu_assign_pointer() is most frequently used indirectly, via
273 the _rcu list-manipulation primitives 258 the _rcu list-manipulation primitives such as list_add_rcu().
274 259
275 rcu_dereference() 260 rcu_dereference()
276 ^^^^^^^^^^^^^^^^^ 261 ^^^^^^^^^^^^^^^^^
277 typeof(p) rcu_dereference(p); 262 typeof(p) rcu_dereference(p);
278 263
279 Like rcu_assign_pointer(), rcu_derefer 264 Like rcu_assign_pointer(), rcu_dereference() must be implemented
280 as a macro. 265 as a macro.
281 266
282 The reader uses the spatial rcu_derefe 267 The reader uses the spatial rcu_dereference() macro to fetch
283 an RCU-protected pointer, which return 268 an RCU-protected pointer, which returns a value that may
284 then be safely dereferenced. Note tha 269 then be safely dereferenced. Note that rcu_dereference()
285 does not actually dereference the poin 270 does not actually dereference the pointer, instead, it
286 protects the pointer for later derefer 271 protects the pointer for later dereferencing. It also
287 executes any needed memory-barrier ins 272 executes any needed memory-barrier instructions for a given
288 CPU architecture. Currently, only Alp 273 CPU architecture. Currently, only Alpha needs memory barriers
289 within rcu_dereference() -- on other C 274 within rcu_dereference() -- on other CPUs, it compiles to a
290 volatile load. However, no mainstream !! 275 volatile load.
291 address dependencies, so rcu_dereferen <<
292 which, in combination with the coding <<
293 rcu_dereference.rst, prevent current c <<
294 these dependencies. <<
295 276
296 Common coding practice uses rcu_derefe 277 Common coding practice uses rcu_dereference() to copy an
297 RCU-protected pointer to a local varia 278 RCU-protected pointer to a local variable, then dereferences
298 this local variable, for example as fo 279 this local variable, for example as follows::
299 280
300 p = rcu_dereference(head.next) 281 p = rcu_dereference(head.next);
301 return p->data; 282 return p->data;
302 283
303 However, in this case, one could just 284 However, in this case, one could just as easily combine these
304 into one statement:: 285 into one statement::
305 286
306 return rcu_dereference(head.ne 287 return rcu_dereference(head.next)->data;
307 288
308 If you are going to be fetching multip 289 If you are going to be fetching multiple fields from the
309 RCU-protected structure, using the loc 290 RCU-protected structure, using the local variable is of
310 course preferred. Repeated rcu_derefe 291 course preferred. Repeated rcu_dereference() calls look
311 ugly, do not guarantee that the same p 292 ugly, do not guarantee that the same pointer will be returned
312 if an update happened while in the cri 293 if an update happened while in the critical section, and incur
313 unnecessary overhead on Alpha CPUs. 294 unnecessary overhead on Alpha CPUs.
314 295
315 Note that the value returned by rcu_de 296 Note that the value returned by rcu_dereference() is valid
316 only within the enclosing RCU read-sid 297 only within the enclosing RCU read-side critical section [1]_.
317 For example, the following is **not** 298 For example, the following is **not** legal::
318 299
319 rcu_read_lock(); 300 rcu_read_lock();
320 p = rcu_dereference(head.next) 301 p = rcu_dereference(head.next);
321 rcu_read_unlock(); 302 rcu_read_unlock();
322 x = p->address; /* BUG!!! */ 303 x = p->address; /* BUG!!! */
323 rcu_read_lock(); 304 rcu_read_lock();
324 y = p->data; /* BUG!!! */ 305 y = p->data; /* BUG!!! */
325 rcu_read_unlock(); 306 rcu_read_unlock();
326 307
327 Holding a reference from one RCU read- 308 Holding a reference from one RCU read-side critical section
328 to another is just as illegal as holdi 309 to another is just as illegal as holding a reference from
329 one lock-based critical section to ano 310 one lock-based critical section to another! Similarly,
330 using a reference outside of the criti 311 using a reference outside of the critical section in which
331 it was acquired is just as illegal as 312 it was acquired is just as illegal as doing so with normal
332 locking. 313 locking.
333 314
334 As with rcu_assign_pointer(), an impor 315 As with rcu_assign_pointer(), an important function of
335 rcu_dereference() is to document which 316 rcu_dereference() is to document which pointers are protected by
336 RCU, in particular, flagging a pointer 317 RCU, in particular, flagging a pointer that is subject to changing
337 at any time, including immediately aft 318 at any time, including immediately after the rcu_dereference().
338 And, again like rcu_assign_pointer(), 319 And, again like rcu_assign_pointer(), rcu_dereference() is
339 typically used indirectly, via the _rc 320 typically used indirectly, via the _rcu list-manipulation
340 primitives, such as list_for_each_entr 321 primitives, such as list_for_each_entry_rcu() [2]_.
341 322
342 .. [1] The variant rcu_dereference_protec 323 .. [1] The variant rcu_dereference_protected() can be used outside
343 of an RCU read-side critical section a 324 of an RCU read-side critical section as long as the usage is
344 protected by locks acquired by the upd 325 protected by locks acquired by the update-side code. This variant
345 avoids the lockdep warning that would 326 avoids the lockdep warning that would happen when using (for
346 example) rcu_dereference() without rcu 327 example) rcu_dereference() without rcu_read_lock() protection.
347 Using rcu_dereference_protected() also 328 Using rcu_dereference_protected() also has the advantage
348 of permitting compiler optimizations t 329 of permitting compiler optimizations that rcu_dereference()
349 must prohibit. The rcu_dereference_pr 330 must prohibit. The rcu_dereference_protected() variant takes
350 a lockdep expression to indicate which 331 a lockdep expression to indicate which locks must be acquired
351 by the caller. If the indicated protec 332 by the caller. If the indicated protection is not provided,
352 a lockdep splat is emitted. See Desig 333 a lockdep splat is emitted. See Design/Requirements/Requirements.rst
353 and the API's code comments for more d 334 and the API's code comments for more details and example usage.
354 335
355 .. [2] If the list_for_each_entry_rcu() i 336 .. [2] If the list_for_each_entry_rcu() instance might be used by
356 update-side code as well as by RCU rea 337 update-side code as well as by RCU readers, then an additional
357 lockdep expression can be added to its 338 lockdep expression can be added to its list of arguments.
358 For example, given an additional "lock 339 For example, given an additional "lock_is_held(&mylock)" argument,
359 the RCU lockdep code would complain on 340 the RCU lockdep code would complain only if this instance was
360 invoked outside of an RCU read-side cr 341 invoked outside of an RCU read-side critical section and without
361 the protection of mylock. 342 the protection of mylock.
362 343
363 The following diagram shows how each API commu 344 The following diagram shows how each API communicates among the
364 reader, updater, and reclaimer. 345 reader, updater, and reclaimer.
365 :: 346 ::
366 347
367 348
368 rcu_assign_pointer() 349 rcu_assign_pointer()
369 +--------+ 350 +--------+
370 +---------------------->| reader | 351 +---------------------->| reader |---------+
371 | +--------+ 352 | +--------+ |
372 | | 353 | | |
373 | | 354 | | | Protect:
374 | | 355 | | | rcu_read_lock()
375 | | 356 | | | rcu_read_unlock()
376 | rcu_dereference() | 357 | rcu_dereference() | |
377 +---------+ | 358 +---------+ | |
378 | updater |<----------------+ 359 | updater |<----------------+ |
379 +---------+ 360 +---------+ V
380 | 361 | +-----------+
381 +--------------------------------- 362 +----------------------------------->| reclaimer |
382 363 +-----------+
383 Defer: 364 Defer:
384 synchronize_rcu() & call_rcu() 365 synchronize_rcu() & call_rcu()
385 366
386 367
387 The RCU infrastructure observes the temporal s 368 The RCU infrastructure observes the temporal sequence of rcu_read_lock(),
388 rcu_read_unlock(), synchronize_rcu(), and call 369 rcu_read_unlock(), synchronize_rcu(), and call_rcu() invocations in
389 order to determine when (1) synchronize_rcu() 370 order to determine when (1) synchronize_rcu() invocations may return
390 to their callers and (2) call_rcu() callbacks 371 to their callers and (2) call_rcu() callbacks may be invoked. Efficient
391 implementations of the RCU infrastructure make 372 implementations of the RCU infrastructure make heavy use of batching in
392 order to amortize their overhead over many use 373 order to amortize their overhead over many uses of the corresponding APIs.
393 The rcu_assign_pointer() and rcu_dereference() 374 The rcu_assign_pointer() and rcu_dereference() invocations communicate
394 spatial changes via stores to and loads from t 375 spatial changes via stores to and loads from the RCU-protected pointer in
395 question. 376 question.
396 377
397 There are at least three flavors of RCU usage 378 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 379 above shows the most common one. On the updater side, the rcu_assign_pointer(),
399 synchronize_rcu() and call_rcu() primitives us 380 synchronize_rcu() and call_rcu() primitives used are the same for all three
400 flavors. However for protection (on the reader 381 flavors. However for protection (on the reader side), the primitives used vary
401 depending on the flavor: 382 depending on the flavor:
402 383
403 a. rcu_read_lock() / rcu_read_unlock() 384 a. rcu_read_lock() / rcu_read_unlock()
404 rcu_dereference() 385 rcu_dereference()
405 386
406 b. rcu_read_lock_bh() / rcu_read_unlock_b 387 b. rcu_read_lock_bh() / rcu_read_unlock_bh()
407 local_bh_disable() / local_bh_enable() 388 local_bh_disable() / local_bh_enable()
408 rcu_dereference_bh() 389 rcu_dereference_bh()
409 390
410 c. rcu_read_lock_sched() / rcu_read_unloc 391 c. rcu_read_lock_sched() / rcu_read_unlock_sched()
411 preempt_disable() / preempt_enable() 392 preempt_disable() / preempt_enable()
412 local_irq_save() / local_irq_restore() 393 local_irq_save() / local_irq_restore()
413 hardirq enter / hardirq exit 394 hardirq enter / hardirq exit
414 NMI enter / NMI exit 395 NMI enter / NMI exit
415 rcu_dereference_sched() 396 rcu_dereference_sched()
416 397
417 These three flavors are used as follows: 398 These three flavors are used as follows:
418 399
419 a. RCU applied to normal data structures. 400 a. RCU applied to normal data structures.
420 401
421 b. RCU applied to networking data structu 402 b. RCU applied to networking data structures that may be subjected
422 to remote denial-of-service attacks. 403 to remote denial-of-service attacks.
423 404
424 c. RCU applied to scheduler and interrupt 405 c. RCU applied to scheduler and interrupt/NMI-handler tasks.
425 406
426 Again, most uses will be of (a). The (b) and 407 Again, most uses will be of (a). The (b) and (c) cases are important
427 for specialized uses, but are relatively uncom 408 for specialized uses, but are relatively uncommon. The SRCU, RCU-Tasks,
428 RCU-Tasks-Rude, and RCU-Tasks-Trace have simil 409 RCU-Tasks-Rude, and RCU-Tasks-Trace have similar relationships among
429 their assorted primitives. 410 their assorted primitives.
430 411
431 .. _3_whatisRCU: 412 .. _3_whatisRCU:
432 413
433 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API 414 3. WHAT ARE SOME EXAMPLE USES OF CORE RCU API?
434 ---------------------------------------------- 415 -----------------------------------------------
435 416
436 This section shows a simple use of the core RC 417 This section shows a simple use of the core RCU API to protect a
437 global pointer to a dynamically allocated stru 418 global pointer to a dynamically allocated structure. More-typical
438 uses of RCU may be found in listRCU.rst and NM !! 419 uses of RCU may be found in listRCU.rst, arrayRCU.rst, and NMI-RCU.rst.
439 :: 420 ::
440 421
441 struct foo { 422 struct foo {
442 int a; 423 int a;
443 char b; 424 char b;
444 long c; 425 long c;
445 }; 426 };
446 DEFINE_SPINLOCK(foo_mutex); 427 DEFINE_SPINLOCK(foo_mutex);
447 428
448 struct foo __rcu *gbl_foo; 429 struct foo __rcu *gbl_foo;
449 430
450 /* 431 /*
451 * Create a new struct foo that is the 432 * Create a new struct foo that is the same as the one currently
452 * pointed to by gbl_foo, except that 433 * pointed to by gbl_foo, except that field "a" is replaced
453 * with "new_a". Points gbl_foo to th 434 * with "new_a". Points gbl_foo to the new structure, and
454 * frees up the old structure after a 435 * frees up the old structure after a grace period.
455 * 436 *
456 * Uses rcu_assign_pointer() to ensure 437 * Uses rcu_assign_pointer() to ensure that concurrent readers
457 * see the initialized version of the 438 * see the initialized version of the new structure.
458 * 439 *
459 * Uses synchronize_rcu() to ensure th 440 * Uses synchronize_rcu() to ensure that any readers that might
460 * have references to the old structur 441 * have references to the old structure complete before freeing
461 * the old structure. 442 * the old structure.
462 */ 443 */
463 void foo_update_a(int new_a) 444 void foo_update_a(int new_a)
464 { 445 {
465 struct foo *new_fp; 446 struct foo *new_fp;
466 struct foo *old_fp; 447 struct foo *old_fp;
467 448
468 new_fp = kmalloc(sizeof(*new_f 449 new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL);
469 spin_lock(&foo_mutex); 450 spin_lock(&foo_mutex);
470 old_fp = rcu_dereference_prote 451 old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex));
471 *new_fp = *old_fp; 452 *new_fp = *old_fp;
472 new_fp->a = new_a; 453 new_fp->a = new_a;
473 rcu_assign_pointer(gbl_foo, ne 454 rcu_assign_pointer(gbl_foo, new_fp);
474 spin_unlock(&foo_mutex); 455 spin_unlock(&foo_mutex);
475 synchronize_rcu(); 456 synchronize_rcu();
476 kfree(old_fp); 457 kfree(old_fp);
477 } 458 }
478 459
479 /* 460 /*
480 * Return the value of field "a" of th 461 * Return the value of field "a" of the current gbl_foo
481 * structure. Use rcu_read_lock() and 462 * structure. Use rcu_read_lock() and rcu_read_unlock()
482 * to ensure that the structure does n 463 * to ensure that the structure does not get deleted out
483 * from under us, and use rcu_derefere 464 * from under us, and use rcu_dereference() to ensure that
484 * we see the initialized version of t 465 * we see the initialized version of the structure (important
485 * for DEC Alpha and for people readin 466 * for DEC Alpha and for people reading the code).
486 */ 467 */
487 int foo_get_a(void) 468 int foo_get_a(void)
488 { 469 {
489 int retval; 470 int retval;
490 471
491 rcu_read_lock(); 472 rcu_read_lock();
492 retval = rcu_dereference(gbl_f 473 retval = rcu_dereference(gbl_foo)->a;
493 rcu_read_unlock(); 474 rcu_read_unlock();
494 return retval; 475 return retval;
495 } 476 }
496 477
497 So, to sum up: 478 So, to sum up:
498 479
499 - Use rcu_read_lock() and rcu_read_unloc 480 - Use rcu_read_lock() and rcu_read_unlock() to guard RCU
500 read-side critical sections. 481 read-side critical sections.
501 482
502 - Within an RCU read-side critical secti 483 - Within an RCU read-side critical section, use rcu_dereference()
503 to dereference RCU-protected pointers. 484 to dereference RCU-protected pointers.
504 485
505 - Use some solid design (such as locks o 486 - Use some solid design (such as locks or semaphores) to
506 keep concurrent updates from interferi 487 keep concurrent updates from interfering with each other.
507 488
508 - Use rcu_assign_pointer() to update an 489 - Use rcu_assign_pointer() to update an RCU-protected pointer.
509 This primitive protects concurrent rea 490 This primitive protects concurrent readers from the updater,
510 **not** concurrent updates from each o 491 **not** concurrent updates from each other! You therefore still
511 need to use locking (or something simi 492 need to use locking (or something similar) to keep concurrent
512 rcu_assign_pointer() primitives from i 493 rcu_assign_pointer() primitives from interfering with each other.
513 494
514 - Use synchronize_rcu() **after** removi 495 - Use synchronize_rcu() **after** removing a data element from an
515 RCU-protected data structure, but **be 496 RCU-protected data structure, but **before** reclaiming/freeing
516 the data element, in order to wait for 497 the data element, in order to wait for the completion of all
517 RCU read-side critical sections that m 498 RCU read-side critical sections that might be referencing that
518 data item. 499 data item.
519 500
520 See checklist.rst for additional rules to foll 501 See checklist.rst for additional rules to follow when using RCU.
521 And again, more-typical uses of RCU may be fou !! 502 And again, more-typical uses of RCU may be found in listRCU.rst,
522 and NMI-RCU.rst. !! 503 arrayRCU.rst, and NMI-RCU.rst.
523 504
524 .. _4_whatisRCU: 505 .. _4_whatisRCU:
525 506
526 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK? 507 4. WHAT IF MY UPDATING THREAD CANNOT BLOCK?
527 -------------------------------------------- 508 --------------------------------------------
528 509
529 In the example above, foo_update_a() blocks un 510 In the example above, foo_update_a() blocks until a grace period elapses.
530 This is quite simple, but in some cases one ca 511 This is quite simple, but in some cases one cannot afford to wait so
531 long -- there might be other high-priority wor 512 long -- there might be other high-priority work to be done.
532 513
533 In such cases, one uses call_rcu() rather than 514 In such cases, one uses call_rcu() rather than synchronize_rcu().
534 The call_rcu() API is as follows:: 515 The call_rcu() API is as follows::
535 516
536 void call_rcu(struct rcu_head *head, r 517 void call_rcu(struct rcu_head *head, rcu_callback_t func);
537 518
538 This function invokes func(head) after a grace 519 This function invokes func(head) after a grace period has elapsed.
539 This invocation might happen from either softi 520 This invocation might happen from either softirq or process context,
540 so the function is not permitted to block. Th 521 so the function is not permitted to block. The foo struct needs to
541 have an rcu_head structure added, perhaps as f 522 have an rcu_head structure added, perhaps as follows::
542 523
543 struct foo { 524 struct foo {
544 int a; 525 int a;
545 char b; 526 char b;
546 long c; 527 long c;
547 struct rcu_head rcu; 528 struct rcu_head rcu;
548 }; 529 };
549 530
550 The foo_update_a() function might then be writ 531 The foo_update_a() function might then be written as follows::
551 532
552 /* 533 /*
553 * Create a new struct foo that is the 534 * Create a new struct foo that is the same as the one currently
554 * pointed to by gbl_foo, except that 535 * pointed to by gbl_foo, except that field "a" is replaced
555 * with "new_a". Points gbl_foo to th 536 * with "new_a". Points gbl_foo to the new structure, and
556 * frees up the old structure after a 537 * frees up the old structure after a grace period.
557 * 538 *
558 * Uses rcu_assign_pointer() to ensure 539 * Uses rcu_assign_pointer() to ensure that concurrent readers
559 * see the initialized version of the 540 * see the initialized version of the new structure.
560 * 541 *
561 * Uses call_rcu() to ensure that any 542 * Uses call_rcu() to ensure that any readers that might have
562 * references to the old structure com 543 * references to the old structure complete before freeing the
563 * old structure. 544 * old structure.
564 */ 545 */
565 void foo_update_a(int new_a) 546 void foo_update_a(int new_a)
566 { 547 {
567 struct foo *new_fp; 548 struct foo *new_fp;
568 struct foo *old_fp; 549 struct foo *old_fp;
569 550
570 new_fp = kmalloc(sizeof(*new_f 551 new_fp = kmalloc(sizeof(*new_fp), GFP_KERNEL);
571 spin_lock(&foo_mutex); 552 spin_lock(&foo_mutex);
572 old_fp = rcu_dereference_prote 553 old_fp = rcu_dereference_protected(gbl_foo, lockdep_is_held(&foo_mutex));
573 *new_fp = *old_fp; 554 *new_fp = *old_fp;
574 new_fp->a = new_a; 555 new_fp->a = new_a;
575 rcu_assign_pointer(gbl_foo, ne 556 rcu_assign_pointer(gbl_foo, new_fp);
576 spin_unlock(&foo_mutex); 557 spin_unlock(&foo_mutex);
577 call_rcu(&old_fp->rcu, foo_rec 558 call_rcu(&old_fp->rcu, foo_reclaim);
578 } 559 }
579 560
580 The foo_reclaim() function might appear as fol 561 The foo_reclaim() function might appear as follows::
581 562
582 void foo_reclaim(struct rcu_head *rp) 563 void foo_reclaim(struct rcu_head *rp)
583 { 564 {
584 struct foo *fp = container_of( 565 struct foo *fp = container_of(rp, struct foo, rcu);
585 566
586 foo_cleanup(fp->a); 567 foo_cleanup(fp->a);
587 568
588 kfree(fp); 569 kfree(fp);
589 } 570 }
590 571
591 The container_of() primitive is a macro that, 572 The container_of() primitive is a macro that, given a pointer into a
592 struct, the type of the struct, and the pointe 573 struct, the type of the struct, and the pointed-to field within the
593 struct, returns a pointer to the beginning of 574 struct, returns a pointer to the beginning of the struct.
594 575
595 The use of call_rcu() permits the caller of fo 576 The use of call_rcu() permits the caller of foo_update_a() to
596 immediately regain control, without needing to 577 immediately regain control, without needing to worry further about the
597 old version of the newly updated element. It 578 old version of the newly updated element. It also clearly shows the
598 RCU distinction between updater, namely foo_up 579 RCU distinction between updater, namely foo_update_a(), and reclaimer,
599 namely foo_reclaim(). 580 namely foo_reclaim().
600 581
601 The summary of advice is the same as for the p 582 The summary of advice is the same as for the previous section, except
602 that we are now using call_rcu() rather than s 583 that we are now using call_rcu() rather than synchronize_rcu():
603 584
604 - Use call_rcu() **after** removing a da 585 - Use call_rcu() **after** removing a data element from an
605 RCU-protected data structure in order 586 RCU-protected data structure in order to register a callback
606 function that will be invoked after th 587 function that will be invoked after the completion of all RCU
607 read-side critical sections that might 588 read-side critical sections that might be referencing that
608 data item. 589 data item.
609 590
610 If the callback for call_rcu() is not doing an 591 If the callback for call_rcu() is not doing anything more than calling
611 kfree() on the structure, you can use kfree_rc 592 kfree() on the structure, you can use kfree_rcu() instead of call_rcu()
612 to avoid having to write your own callback:: 593 to avoid having to write your own callback::
613 594
614 kfree_rcu(old_fp, rcu); 595 kfree_rcu(old_fp, rcu);
615 596
616 If the occasional sleep is permitted, the sing 597 If the occasional sleep is permitted, the single-argument form may
617 be used, omitting the rcu_head structure from 598 be used, omitting the rcu_head structure from struct foo.
618 599
619 kfree_rcu_mightsleep(old_fp); 600 kfree_rcu_mightsleep(old_fp);
620 601
621 This variant almost never blocks, but might do 602 This variant almost never blocks, but might do so by invoking
622 synchronize_rcu() in response to memory-alloca 603 synchronize_rcu() in response to memory-allocation failure.
623 604
624 Again, see checklist.rst for additional rules 605 Again, see checklist.rst for additional rules governing the use of RCU.
625 606
626 .. _5_whatisRCU: 607 .. _5_whatisRCU:
627 608
628 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RC 609 5. WHAT ARE SOME SIMPLE IMPLEMENTATIONS OF RCU?
629 ---------------------------------------------- 610 ------------------------------------------------
630 611
631 One of the nice things about RCU is that it ha 612 One of the nice things about RCU is that it has extremely simple "toy"
632 implementations that are a good first step tow 613 implementations that are a good first step towards understanding the
633 production-quality implementations in the Linu 614 production-quality implementations in the Linux kernel. This section
634 presents two such "toy" implementations of RCU 615 presents two such "toy" implementations of RCU, one that is implemented
635 in terms of familiar locking primitives, and a 616 in terms of familiar locking primitives, and another that more closely
636 resembles "classic" RCU. Both are way too sim 617 resembles "classic" RCU. Both are way too simple for real-world use,
637 lacking both functionality and performance. H 618 lacking both functionality and performance. However, they are useful
638 in getting a feel for how RCU works. See kern 619 in getting a feel for how RCU works. See kernel/rcu/update.c for a
639 production-quality implementation, and see: 620 production-quality implementation, and see:
640 621
641 https://docs.google.com/document/d/1X0 622 https://docs.google.com/document/d/1X0lThx8OK0ZgLMqVoXiR4ZrGURHrXK6NyLRbeXe3Xac/edit
642 623
643 for papers describing the Linux kernel RCU imp 624 for papers describing the Linux kernel RCU implementation. The OLS'01
644 and OLS'02 papers are a good introduction, and 625 and OLS'02 papers are a good introduction, and the dissertation provides
645 more details on the current implementation as 626 more details on the current implementation as of early 2004.
646 627
647 628
648 5A. "TOY" IMPLEMENTATION #1: LOCKING 629 5A. "TOY" IMPLEMENTATION #1: LOCKING
649 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 630 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
650 This section presents a "toy" RCU implementati 631 This section presents a "toy" RCU implementation that is based on
651 familiar locking primitives. Its overhead mak 632 familiar locking primitives. Its overhead makes it a non-starter for
652 real-life use, as does its lack of scalability 633 real-life use, as does its lack of scalability. It is also unsuitable
653 for realtime use, since it allows scheduling l 634 for realtime use, since it allows scheduling latency to "bleed" from
654 one read-side critical section to another. It 635 one read-side critical section to another. It also assumes recursive
655 reader-writer locks: If you try this with non 636 reader-writer locks: If you try this with non-recursive locks, and
656 you allow nested rcu_read_lock() calls, you ca 637 you allow nested rcu_read_lock() calls, you can deadlock.
657 638
658 However, it is probably the easiest implementa 639 However, it is probably the easiest implementation to relate to, so is
659 a good starting point. 640 a good starting point.
660 641
661 It is extremely simple:: 642 It is extremely simple::
662 643
663 static DEFINE_RWLOCK(rcu_gp_mutex); 644 static DEFINE_RWLOCK(rcu_gp_mutex);
664 645
665 void rcu_read_lock(void) 646 void rcu_read_lock(void)
666 { 647 {
667 read_lock(&rcu_gp_mutex); 648 read_lock(&rcu_gp_mutex);
668 } 649 }
669 650
670 void rcu_read_unlock(void) 651 void rcu_read_unlock(void)
671 { 652 {
672 read_unlock(&rcu_gp_mutex); 653 read_unlock(&rcu_gp_mutex);
673 } 654 }
674 655
675 void synchronize_rcu(void) 656 void synchronize_rcu(void)
676 { 657 {
677 write_lock(&rcu_gp_mutex); 658 write_lock(&rcu_gp_mutex);
678 smp_mb__after_spinlock(); 659 smp_mb__after_spinlock();
679 write_unlock(&rcu_gp_mutex); 660 write_unlock(&rcu_gp_mutex);
680 } 661 }
681 662
682 [You can ignore rcu_assign_pointer() and rcu_d 663 [You can ignore rcu_assign_pointer() and rcu_dereference() without missing
683 much. But here are simplified versions anyway 664 much. But here are simplified versions anyway. And whatever you do,
684 don't forget about them when submitting patche 665 don't forget about them when submitting patches making use of RCU!]::
685 666
686 #define rcu_assign_pointer(p, v) \ 667 #define rcu_assign_pointer(p, v) \
687 ({ \ 668 ({ \
688 smp_store_release(&(p), (v)); 669 smp_store_release(&(p), (v)); \
689 }) 670 })
690 671
691 #define rcu_dereference(p) \ 672 #define rcu_dereference(p) \
692 ({ \ 673 ({ \
693 typeof(p) _________p1 = READ_O 674 typeof(p) _________p1 = READ_ONCE(p); \
694 (_________p1); \ 675 (_________p1); \
695 }) 676 })
696 677
697 678
698 The rcu_read_lock() and rcu_read_unlock() prim 679 The rcu_read_lock() and rcu_read_unlock() primitive read-acquire
699 and release a global reader-writer lock. The 680 and release a global reader-writer lock. The synchronize_rcu()
700 primitive write-acquires this same lock, then 681 primitive write-acquires this same lock, then releases it. This means
701 that once synchronize_rcu() exits, all RCU rea 682 that once synchronize_rcu() exits, all RCU read-side critical sections
702 that were in progress before synchronize_rcu() 683 that were in progress before synchronize_rcu() was called are guaranteed
703 to have completed -- there is no way that sync 684 to have completed -- there is no way that synchronize_rcu() would have
704 been able to write-acquire the lock otherwise. 685 been able to write-acquire the lock otherwise. The smp_mb__after_spinlock()
705 promotes synchronize_rcu() to a full memory ba 686 promotes synchronize_rcu() to a full memory barrier in compliance with
706 the "Memory-Barrier Guarantees" listed in: 687 the "Memory-Barrier Guarantees" listed in:
707 688
708 Design/Requirements/Requirements.rst 689 Design/Requirements/Requirements.rst
709 690
710 It is possible to nest rcu_read_lock(), since 691 It is possible to nest rcu_read_lock(), since reader-writer locks may
711 be recursively acquired. Note also that rcu_r 692 be recursively acquired. Note also that rcu_read_lock() is immune
712 from deadlock (an important property of RCU). 693 from deadlock (an important property of RCU). The reason for this is
713 that the only thing that can block rcu_read_lo 694 that the only thing that can block rcu_read_lock() is a synchronize_rcu().
714 But synchronize_rcu() does not acquire any loc 695 But synchronize_rcu() does not acquire any locks while holding rcu_gp_mutex,
715 so there can be no deadlock cycle. 696 so there can be no deadlock cycle.
716 697
717 .. _quiz_1: 698 .. _quiz_1:
718 699
719 Quick Quiz #1: 700 Quick Quiz #1:
720 Why is this argument naive? H 701 Why is this argument naive? How could a deadlock
721 occur when using this algorith 702 occur when using this algorithm in a real-world Linux
722 kernel? How could this deadlo 703 kernel? How could this deadlock be avoided?
723 704
724 :ref:`Answers to Quick Quiz <9_whatisRCU>` 705 :ref:`Answers to Quick Quiz <9_whatisRCU>`
725 706
726 5B. "TOY" EXAMPLE #2: CLASSIC RCU 707 5B. "TOY" EXAMPLE #2: CLASSIC RCU
727 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
728 This section presents a "toy" RCU implementati 709 This section presents a "toy" RCU implementation that is based on
729 "classic RCU". It is also short on performanc 710 "classic RCU". It is also short on performance (but only for updates) and
730 on features such as hotplug CPU and the abilit 711 on features such as hotplug CPU and the ability to run in CONFIG_PREEMPTION
731 kernels. The definitions of rcu_dereference() 712 kernels. The definitions of rcu_dereference() and rcu_assign_pointer()
732 are the same as those shown in the preceding s 713 are the same as those shown in the preceding section, so they are omitted.
733 :: 714 ::
734 715
735 void rcu_read_lock(void) { } 716 void rcu_read_lock(void) { }
736 717
737 void rcu_read_unlock(void) { } 718 void rcu_read_unlock(void) { }
738 719
739 void synchronize_rcu(void) 720 void synchronize_rcu(void)
740 { 721 {
741 int cpu; 722 int cpu;
742 723
743 for_each_possible_cpu(cpu) 724 for_each_possible_cpu(cpu)
744 run_on(cpu); 725 run_on(cpu);
745 } 726 }
746 727
747 Note that rcu_read_lock() and rcu_read_unlock( 728 Note that rcu_read_lock() and rcu_read_unlock() do absolutely nothing.
748 This is the great strength of classic RCU in a 729 This is the great strength of classic RCU in a non-preemptive kernel:
749 read-side overhead is precisely zero, at least 730 read-side overhead is precisely zero, at least on non-Alpha CPUs.
750 And there is absolutely no way that rcu_read_l 731 And there is absolutely no way that rcu_read_lock() can possibly
751 participate in a deadlock cycle! 732 participate in a deadlock cycle!
752 733
753 The implementation of synchronize_rcu() simply 734 The implementation of synchronize_rcu() simply schedules itself on each
754 CPU in turn. The run_on() primitive can be im 735 CPU in turn. The run_on() primitive can be implemented straightforwardly
755 in terms of the sched_setaffinity() primitive. 736 in terms of the sched_setaffinity() primitive. Of course, a somewhat less
756 "toy" implementation would restore the affinit 737 "toy" implementation would restore the affinity upon completion rather
757 than just leaving all tasks running on the las 738 than just leaving all tasks running on the last CPU, but when I said
758 "toy", I meant **toy**! 739 "toy", I meant **toy**!
759 740
760 So how the heck is this supposed to work??? 741 So how the heck is this supposed to work???
761 742
762 Remember that it is illegal to block while in 743 Remember that it is illegal to block while in an RCU read-side critical
763 section. Therefore, if a given CPU executes a 744 section. Therefore, if a given CPU executes a context switch, we know
764 that it must have completed all preceding RCU 745 that it must have completed all preceding RCU read-side critical sections.
765 Once **all** CPUs have executed a context swit 746 Once **all** CPUs have executed a context switch, then **all** preceding
766 RCU read-side critical sections will have comp 747 RCU read-side critical sections will have completed.
767 748
768 So, suppose that we remove a data item from it 749 So, suppose that we remove a data item from its structure and then invoke
769 synchronize_rcu(). Once synchronize_rcu() ret 750 synchronize_rcu(). Once synchronize_rcu() returns, we are guaranteed
770 that there are no RCU read-side critical secti 751 that there are no RCU read-side critical sections holding a reference
771 to that data item, so we can safely reclaim it 752 to that data item, so we can safely reclaim it.
772 753
773 .. _quiz_2: 754 .. _quiz_2:
774 755
775 Quick Quiz #2: 756 Quick Quiz #2:
776 Give an example where Classic 757 Give an example where Classic RCU's read-side
777 overhead is **negative**. 758 overhead is **negative**.
778 759
779 :ref:`Answers to Quick Quiz <9_whatisRCU>` 760 :ref:`Answers to Quick Quiz <9_whatisRCU>`
780 761
781 .. _quiz_3: 762 .. _quiz_3:
782 763
783 Quick Quiz #3: 764 Quick Quiz #3:
784 If it is illegal to block in a 765 If it is illegal to block in an RCU read-side
785 critical section, what the hec 766 critical section, what the heck do you do in
786 CONFIG_PREEMPT_RT, where norma 767 CONFIG_PREEMPT_RT, where normal spinlocks can block???
787 768
788 :ref:`Answers to Quick Quiz <9_whatisRCU>` 769 :ref:`Answers to Quick Quiz <9_whatisRCU>`
789 770
790 .. _6_whatisRCU: 771 .. _6_whatisRCU:
791 772
792 6. ANALOGY WITH READER-WRITER LOCKING 773 6. ANALOGY WITH READER-WRITER LOCKING
793 -------------------------------------- 774 --------------------------------------
794 775
795 Although RCU can be used in many different way 776 Although RCU can be used in many different ways, a very common use of
796 RCU is analogous to reader-writer locking. Th 777 RCU is analogous to reader-writer locking. The following unified
797 diff shows how closely related RCU and reader- 778 diff shows how closely related RCU and reader-writer locking can be.
798 :: 779 ::
799 780
800 @@ -5,5 +5,5 @@ struct el { 781 @@ -5,5 +5,5 @@ struct el {
801 int data; 782 int data;
802 /* Other data fields */ 783 /* Other data fields */
803 }; 784 };
804 -rwlock_t listmutex; 785 -rwlock_t listmutex;
805 +spinlock_t listmutex; 786 +spinlock_t listmutex;
806 struct el head; 787 struct el head;
807 788
808 @@ -13,15 +14,15 @@ 789 @@ -13,15 +14,15 @@
809 struct list_head *lp; 790 struct list_head *lp;
810 struct el *p; 791 struct el *p;
811 792
812 - read_lock(&listmutex); 793 - read_lock(&listmutex);
813 - list_for_each_entry(p, head, l 794 - list_for_each_entry(p, head, lp) {
814 + rcu_read_lock(); 795 + rcu_read_lock();
815 + list_for_each_entry_rcu(p, hea 796 + list_for_each_entry_rcu(p, head, lp) {
816 if (p->key == key) { 797 if (p->key == key) {
817 *result = p->d 798 *result = p->data;
818 - read_unlock(&l 799 - read_unlock(&listmutex);
819 + rcu_read_unloc 800 + rcu_read_unlock();
820 return 1; 801 return 1;
821 } 802 }
822 } 803 }
823 - read_unlock(&listmutex); 804 - read_unlock(&listmutex);
824 + rcu_read_unlock(); 805 + rcu_read_unlock();
825 return 0; 806 return 0;
826 } 807 }
827 808
828 @@ -29,15 +30,16 @@ 809 @@ -29,15 +30,16 @@
829 { 810 {
830 struct el *p; 811 struct el *p;
831 812
832 - write_lock(&listmutex); 813 - write_lock(&listmutex);
833 + spin_lock(&listmutex); 814 + spin_lock(&listmutex);
834 list_for_each_entry(p, head, l 815 list_for_each_entry(p, head, lp) {
835 if (p->key == key) { 816 if (p->key == key) {
836 - list_del(&p->l 817 - list_del(&p->list);
837 - write_unlock(& 818 - write_unlock(&listmutex);
838 + list_del_rcu(& 819 + list_del_rcu(&p->list);
839 + spin_unlock(&l 820 + spin_unlock(&listmutex);
840 + synchronize_rc 821 + synchronize_rcu();
841 kfree(p); 822 kfree(p);
842 return 1; 823 return 1;
843 } 824 }
844 } 825 }
845 - write_unlock(&listmutex); 826 - write_unlock(&listmutex);
846 + spin_unlock(&listmutex); 827 + spin_unlock(&listmutex);
847 return 0; 828 return 0;
848 } 829 }
849 830
850 Or, for those who prefer a side-by-side listin 831 Or, for those who prefer a side-by-side listing::
851 832
852 1 struct el { 1 stru 833 1 struct el { 1 struct el {
853 2 struct list_head list; 2 st 834 2 struct list_head list; 2 struct list_head list;
854 3 long key; 3 lo 835 3 long key; 3 long key;
855 4 spinlock_t mutex; 4 sp 836 4 spinlock_t mutex; 4 spinlock_t mutex;
856 5 int data; 5 in 837 5 int data; 5 int data;
857 6 /* Other data fields */ 6 /* 838 6 /* Other data fields */ 6 /* Other data fields */
858 7 }; 7 }; 839 7 }; 7 };
859 8 rwlock_t listmutex; 8 spin 840 8 rwlock_t listmutex; 8 spinlock_t listmutex;
860 9 struct el head; 9 stru 841 9 struct el head; 9 struct el head;
861 842
862 :: 843 ::
863 844
864 1 int search(long key, int *result) 1 int 845 1 int search(long key, int *result) 1 int search(long key, int *result)
865 2 { 2 { 846 2 { 2 {
866 3 struct list_head *lp; 3 s 847 3 struct list_head *lp; 3 struct list_head *lp;
867 4 struct el *p; 4 s 848 4 struct el *p; 4 struct el *p;
868 5 5 849 5 5
869 6 read_lock(&listmutex); 6 r 850 6 read_lock(&listmutex); 6 rcu_read_lock();
870 7 list_for_each_entry(p, head, lp) { 7 l 851 7 list_for_each_entry(p, head, lp) { 7 list_for_each_entry_rcu(p, head, lp) {
871 8 if (p->key == key) { 8 852 8 if (p->key == key) { 8 if (p->key == key) {
872 9 *result = p->data; 9 853 9 *result = p->data; 9 *result = p->data;
873 10 read_unlock(&listmutex); 10 854 10 read_unlock(&listmutex); 10 rcu_read_unlock();
874 11 return 1; 11 855 11 return 1; 11 return 1;
875 12 } 12 856 12 } 12 }
876 13 } 13 } 857 13 } 13 }
877 14 read_unlock(&listmutex); 14 r 858 14 read_unlock(&listmutex); 14 rcu_read_unlock();
878 15 return 0; 15 r 859 15 return 0; 15 return 0;
879 16 } 16 } 860 16 } 16 }
880 861
881 :: 862 ::
882 863
883 1 int delete(long key) 1 int 864 1 int delete(long key) 1 int delete(long key)
884 2 { 2 { 865 2 { 2 {
885 3 struct el *p; 3 s 866 3 struct el *p; 3 struct el *p;
886 4 4 867 4 4
887 5 write_lock(&listmutex); 5 s 868 5 write_lock(&listmutex); 5 spin_lock(&listmutex);
888 6 list_for_each_entry(p, head, lp) { 6 l 869 6 list_for_each_entry(p, head, lp) { 6 list_for_each_entry(p, head, lp) {
889 7 if (p->key == key) { 7 870 7 if (p->key == key) { 7 if (p->key == key) {
890 8 list_del(&p->list); 8 871 8 list_del(&p->list); 8 list_del_rcu(&p->list);
891 9 write_unlock(&listmutex); 9 872 9 write_unlock(&listmutex); 9 spin_unlock(&listmutex);
892 10 873 10 synchronize_rcu();
893 10 kfree(p); 11 874 10 kfree(p); 11 kfree(p);
894 11 return 1; 12 875 11 return 1; 12 return 1;
895 12 } 13 876 12 } 13 }
896 13 } 14 } 877 13 } 14 }
897 14 write_unlock(&listmutex); 15 s 878 14 write_unlock(&listmutex); 15 spin_unlock(&listmutex);
898 15 return 0; 16 r 879 15 return 0; 16 return 0;
899 16 } 17 } 880 16 } 17 }
900 881
901 Either way, the differences are quite small. 882 Either way, the differences are quite small. Read-side locking moves
902 to rcu_read_lock() and rcu_read_unlock, update 883 to rcu_read_lock() and rcu_read_unlock, update-side locking moves from
903 a reader-writer lock to a simple spinlock, and 884 a reader-writer lock to a simple spinlock, and a synchronize_rcu()
904 precedes the kfree(). 885 precedes the kfree().
905 886
906 However, there is one potential catch: the rea 887 However, there is one potential catch: the read-side and update-side
907 critical sections can now run concurrently. I 888 critical sections can now run concurrently. In many cases, this will
908 not be a problem, but it is necessary to check 889 not be a problem, but it is necessary to check carefully regardless.
909 For example, if multiple independent list upda 890 For example, if multiple independent list updates must be seen as
910 a single atomic update, converting to RCU will 891 a single atomic update, converting to RCU will require special care.
911 892
912 Also, the presence of synchronize_rcu() means 893 Also, the presence of synchronize_rcu() means that the RCU version of
913 delete() can now block. If this is a problem, 894 delete() can now block. If this is a problem, there is a callback-based
914 mechanism that never blocks, namely call_rcu() 895 mechanism that never blocks, namely call_rcu() or kfree_rcu(), that can
915 be used in place of synchronize_rcu(). 896 be used in place of synchronize_rcu().
916 897
917 .. _7_whatisRCU: 898 .. _7_whatisRCU:
918 899
919 7. ANALOGY WITH REFERENCE COUNTING 900 7. ANALOGY WITH REFERENCE COUNTING
920 ----------------------------------- 901 -----------------------------------
921 902
922 The reader-writer analogy (illustrated by the 903 The reader-writer analogy (illustrated by the previous section) is not
923 always the best way to think about using RCU. 904 always the best way to think about using RCU. Another helpful analogy
924 considers RCU an effective reference count on 905 considers RCU an effective reference count on everything which is
925 protected by RCU. 906 protected by RCU.
926 907
927 A reference count typically does not prevent t 908 A reference count typically does not prevent the referenced object's
928 values from changing, but does prevent changes 909 values from changing, but does prevent changes to type -- particularly the
929 gross change of type that happens when that ob 910 gross change of type that happens when that object's memory is freed and
930 re-allocated for some other purpose. Once a t 911 re-allocated for some other purpose. Once a type-safe reference to the
931 object is obtained, some other mechanism is ne 912 object is obtained, some other mechanism is needed to ensure consistent
932 access to the data in the object. This could 913 access to the data in the object. This could involve taking a spinlock,
933 but with RCU the typical approach is to perfor 914 but with RCU the typical approach is to perform reads with SMP-aware
934 operations such as smp_load_acquire(), to perf 915 operations such as smp_load_acquire(), to perform updates with atomic
935 read-modify-write operations, and to provide t 916 read-modify-write operations, and to provide the necessary ordering.
936 RCU provides a number of support functions tha 917 RCU provides a number of support functions that embed the required
937 operations and ordering, such as the list_for_ 918 operations and ordering, such as the list_for_each_entry_rcu() macro
938 used in the previous section. 919 used in the previous section.
939 920
940 A more focused view of the reference counting 921 A more focused view of the reference counting behavior is that,
941 between rcu_read_lock() and rcu_read_unlock(), 922 between rcu_read_lock() and rcu_read_unlock(), any reference taken with
942 rcu_dereference() on a pointer marked as ``__r 923 rcu_dereference() on a pointer marked as ``__rcu`` can be treated as
943 though a reference-count on that object has be 924 though a reference-count on that object has been temporarily increased.
944 This prevents the object from changing type. 925 This prevents the object from changing type. Exactly what this means
945 will depend on normal expectations of objects 926 will depend on normal expectations of objects of that type, but it
946 typically includes that spinlocks can still be 927 typically includes that spinlocks can still be safely locked, normal
947 reference counters can be safely manipulated, 928 reference counters can be safely manipulated, and ``__rcu`` pointers
948 can be safely dereferenced. 929 can be safely dereferenced.
949 930
950 Some operations that one might expect to see o 931 Some operations that one might expect to see on an object for
951 which an RCU reference is held include: 932 which an RCU reference is held include:
952 933
953 - Copying out data that is guaranteed to be s 934 - Copying out data that is guaranteed to be stable by the object's type.
954 - Using kref_get_unless_zero() or similar to 935 - Using kref_get_unless_zero() or similar to get a longer-term
955 reference. This may fail of course. 936 reference. This may fail of course.
956 - Acquiring a spinlock in the object, and che 937 - Acquiring a spinlock in the object, and checking if the object still
957 is the expected object and if so, manipulat 938 is the expected object and if so, manipulating it freely.
958 939
959 The understanding that RCU provides a referenc 940 The understanding that RCU provides a reference that only prevents a
960 change of type is particularly visible with ob 941 change of type is particularly visible with objects allocated from a
961 slab cache marked ``SLAB_TYPESAFE_BY_RCU``. R 942 slab cache marked ``SLAB_TYPESAFE_BY_RCU``. RCU operations may yield a
962 reference to an object from such a cache that 943 reference to an object from such a cache that has been concurrently freed
963 and the memory reallocated to a completely dif 944 and the memory reallocated to a completely different object, though of
964 the same type. In this case RCU doesn't even 945 the same type. In this case RCU doesn't even protect the identity of the
965 object from changing, only its type. So the o 946 object from changing, only its type. So the object found may not be the
966 one expected, but it will be one where it is s 947 one expected, but it will be one where it is safe to take a reference
967 (and then potentially acquiring a spinlock), a 948 (and then potentially acquiring a spinlock), allowing subsequent code
968 to check whether the identity matches expectat 949 to check whether the identity matches expectations. It is tempting
969 to simply acquire the spinlock without first t 950 to simply acquire the spinlock without first taking the reference, but
970 unfortunately any spinlock in a ``SLAB_TYPESAF 951 unfortunately any spinlock in a ``SLAB_TYPESAFE_BY_RCU`` object must be
971 initialized after each and every call to kmem_ 952 initialized after each and every call to kmem_cache_alloc(), which renders
972 reference-free spinlock acquisition completely 953 reference-free spinlock acquisition completely unsafe. Therefore, when
973 using ``SLAB_TYPESAFE_BY_RCU``, make proper us 954 using ``SLAB_TYPESAFE_BY_RCU``, make proper use of a reference counter.
974 (Those willing to initialize their locks in a !! 955 (Those willing to use a kmem_cache constructor may also use locking,
975 may also use locking, including cache-friendly !! 956 including cache-friendly sequence locking.)
976 957
977 With traditional reference counting -- such as 958 With traditional reference counting -- such as that implemented by the
978 kref library in Linux -- there is typically co 959 kref library in Linux -- there is typically code that runs when the last
979 reference to an object is dropped. With kref, 960 reference to an object is dropped. With kref, this is the function
980 passed to kref_put(). When RCU is being used, 961 passed to kref_put(). When RCU is being used, such finalization code
981 must not be run until all ``__rcu`` pointers r 962 must not be run until all ``__rcu`` pointers referencing the object have
982 been updated, and then a grace period has pass 963 been updated, and then a grace period has passed. Every remaining
983 globally visible pointer to the object must be 964 globally visible pointer to the object must be considered to be a
984 potential counted reference, and the finalizat 965 potential counted reference, and the finalization code is typically run
985 using call_rcu() only after all those pointers 966 using call_rcu() only after all those pointers have been changed.
986 967
987 To see how to choose between these two analogi 968 To see how to choose between these two analogies -- of RCU as a
988 reader-writer lock and RCU as a reference coun 969 reader-writer lock and RCU as a reference counting system -- it is useful
989 to reflect on the scale of the thing being pro 970 to reflect on the scale of the thing being protected. The reader-writer
990 lock analogy looks at larger multi-part object 971 lock analogy looks at larger multi-part objects such as a linked list
991 and shows how RCU can facilitate concurrency w 972 and shows how RCU can facilitate concurrency while elements are added
992 to, and removed from, the list. The reference 973 to, and removed from, the list. The reference-count analogy looks at
993 the individual objects and looks at how they c 974 the individual objects and looks at how they can be accessed safely
994 within whatever whole they are a part of. 975 within whatever whole they are a part of.
995 976
996 .. _8_whatisRCU: 977 .. _8_whatisRCU:
997 978
998 8. FULL LIST OF RCU APIs 979 8. FULL LIST OF RCU APIs
999 ------------------------- 980 -------------------------
1000 981
1001 The RCU APIs are documented in docbook-format 982 The RCU APIs are documented in docbook-format header comments in the
1002 Linux-kernel source code, but it helps to hav 983 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 984 APIs, since there does not appear to be a way to categorize them
1004 in docbook. Here is the list, by category. 985 in docbook. Here is the list, by category.
1005 986
1006 RCU list traversal:: 987 RCU list traversal::
1007 988
1008 list_entry_rcu 989 list_entry_rcu
1009 list_entry_lockless 990 list_entry_lockless
1010 list_first_entry_rcu 991 list_first_entry_rcu
1011 list_next_rcu 992 list_next_rcu
1012 list_for_each_entry_rcu 993 list_for_each_entry_rcu
1013 list_for_each_entry_continue_rcu 994 list_for_each_entry_continue_rcu
1014 list_for_each_entry_from_rcu 995 list_for_each_entry_from_rcu
1015 list_first_or_null_rcu 996 list_first_or_null_rcu
1016 list_next_or_null_rcu 997 list_next_or_null_rcu
1017 hlist_first_rcu 998 hlist_first_rcu
1018 hlist_next_rcu 999 hlist_next_rcu
1019 hlist_pprev_rcu 1000 hlist_pprev_rcu
1020 hlist_for_each_entry_rcu 1001 hlist_for_each_entry_rcu
1021 hlist_for_each_entry_rcu_bh 1002 hlist_for_each_entry_rcu_bh
1022 hlist_for_each_entry_from_rcu 1003 hlist_for_each_entry_from_rcu
1023 hlist_for_each_entry_continue_rcu 1004 hlist_for_each_entry_continue_rcu
1024 hlist_for_each_entry_continue_rcu_bh 1005 hlist_for_each_entry_continue_rcu_bh
1025 hlist_nulls_first_rcu 1006 hlist_nulls_first_rcu
1026 hlist_nulls_for_each_entry_rcu 1007 hlist_nulls_for_each_entry_rcu
1027 hlist_bl_first_rcu 1008 hlist_bl_first_rcu
1028 hlist_bl_for_each_entry_rcu 1009 hlist_bl_for_each_entry_rcu
1029 1010
1030 RCU pointer/list update:: 1011 RCU pointer/list update::
1031 1012
1032 rcu_assign_pointer 1013 rcu_assign_pointer
1033 list_add_rcu 1014 list_add_rcu
1034 list_add_tail_rcu 1015 list_add_tail_rcu
1035 list_del_rcu 1016 list_del_rcu
1036 list_replace_rcu 1017 list_replace_rcu
1037 hlist_add_behind_rcu 1018 hlist_add_behind_rcu
1038 hlist_add_before_rcu 1019 hlist_add_before_rcu
1039 hlist_add_head_rcu 1020 hlist_add_head_rcu
1040 hlist_add_tail_rcu 1021 hlist_add_tail_rcu
1041 hlist_del_rcu 1022 hlist_del_rcu
1042 hlist_del_init_rcu 1023 hlist_del_init_rcu
1043 hlist_replace_rcu 1024 hlist_replace_rcu
1044 list_splice_init_rcu 1025 list_splice_init_rcu
1045 list_splice_tail_init_rcu 1026 list_splice_tail_init_rcu
1046 hlist_nulls_del_init_rcu 1027 hlist_nulls_del_init_rcu
1047 hlist_nulls_del_rcu 1028 hlist_nulls_del_rcu
1048 hlist_nulls_add_head_rcu 1029 hlist_nulls_add_head_rcu
1049 hlist_bl_add_head_rcu 1030 hlist_bl_add_head_rcu
1050 hlist_bl_del_init_rcu 1031 hlist_bl_del_init_rcu
1051 hlist_bl_del_rcu 1032 hlist_bl_del_rcu
1052 hlist_bl_set_first_rcu 1033 hlist_bl_set_first_rcu
1053 1034
1054 RCU:: 1035 RCU::
1055 1036
1056 Critical sections Grace period 1037 Critical sections Grace period Barrier
1057 1038
1058 rcu_read_lock synchronize_n 1039 rcu_read_lock synchronize_net rcu_barrier
1059 rcu_read_unlock synchronize_r 1040 rcu_read_unlock synchronize_rcu
1060 rcu_dereference synchronize_r 1041 rcu_dereference synchronize_rcu_expedited
1061 rcu_read_lock_held call_rcu 1042 rcu_read_lock_held call_rcu
1062 rcu_dereference_check kfree_rcu 1043 rcu_dereference_check kfree_rcu
1063 rcu_dereference_protected 1044 rcu_dereference_protected
1064 1045
1065 bh:: 1046 bh::
1066 1047
1067 Critical sections Grace period 1048 Critical sections Grace period Barrier
1068 1049
1069 rcu_read_lock_bh call_rcu 1050 rcu_read_lock_bh call_rcu rcu_barrier
1070 rcu_read_unlock_bh synchronize_r 1051 rcu_read_unlock_bh synchronize_rcu
1071 [local_bh_disable] synchronize_r 1052 [local_bh_disable] synchronize_rcu_expedited
1072 [and friends] 1053 [and friends]
1073 rcu_dereference_bh 1054 rcu_dereference_bh
1074 rcu_dereference_bh_check 1055 rcu_dereference_bh_check
1075 rcu_dereference_bh_protected 1056 rcu_dereference_bh_protected
1076 rcu_read_lock_bh_held 1057 rcu_read_lock_bh_held
1077 1058
1078 sched:: 1059 sched::
1079 1060
1080 Critical sections Grace period 1061 Critical sections Grace period Barrier
1081 1062
1082 rcu_read_lock_sched call_rcu 1063 rcu_read_lock_sched call_rcu rcu_barrier
1083 rcu_read_unlock_sched synchronize_r 1064 rcu_read_unlock_sched synchronize_rcu
1084 [preempt_disable] synchronize_r 1065 [preempt_disable] synchronize_rcu_expedited
1085 [and friends] 1066 [and friends]
1086 rcu_read_lock_sched_notrace 1067 rcu_read_lock_sched_notrace
1087 rcu_read_unlock_sched_notrace 1068 rcu_read_unlock_sched_notrace
1088 rcu_dereference_sched 1069 rcu_dereference_sched
1089 rcu_dereference_sched_check 1070 rcu_dereference_sched_check
1090 rcu_dereference_sched_protected 1071 rcu_dereference_sched_protected
1091 rcu_read_lock_sched_held 1072 rcu_read_lock_sched_held
1092 1073
1093 1074
1094 RCU-Tasks:: 1075 RCU-Tasks::
1095 1076
1096 Critical sections Grace period 1077 Critical sections Grace period Barrier
1097 1078
1098 N/A call_rcu_task 1079 N/A call_rcu_tasks rcu_barrier_tasks
1099 synchronize_r 1080 synchronize_rcu_tasks
1100 1081
1101 1082
1102 RCU-Tasks-Rude:: 1083 RCU-Tasks-Rude::
1103 1084
1104 Critical sections Grace period 1085 Critical sections Grace period Barrier
1105 1086
1106 N/A call_rcu_task 1087 N/A call_rcu_tasks_rude rcu_barrier_tasks_rude
1107 synchronize_r 1088 synchronize_rcu_tasks_rude
1108 1089
1109 1090
1110 RCU-Tasks-Trace:: 1091 RCU-Tasks-Trace::
1111 1092
1112 Critical sections Grace period 1093 Critical sections Grace period Barrier
1113 1094
1114 rcu_read_lock_trace call_rcu_task 1095 rcu_read_lock_trace call_rcu_tasks_trace rcu_barrier_tasks_trace
1115 rcu_read_unlock_trace synchronize_r 1096 rcu_read_unlock_trace synchronize_rcu_tasks_trace
1116 1097
1117 1098
1118 SRCU:: 1099 SRCU::
1119 1100
1120 Critical sections Grace period 1101 Critical sections Grace period Barrier
1121 1102
1122 srcu_read_lock call_srcu 1103 srcu_read_lock call_srcu srcu_barrier
1123 srcu_read_unlock synchronize_s 1104 srcu_read_unlock synchronize_srcu
1124 srcu_dereference synchronize_s 1105 srcu_dereference synchronize_srcu_expedited
1125 srcu_dereference_check 1106 srcu_dereference_check
1126 srcu_read_lock_held 1107 srcu_read_lock_held
1127 1108
1128 SRCU: Initialization/cleanup:: 1109 SRCU: Initialization/cleanup::
1129 1110
1130 DEFINE_SRCU 1111 DEFINE_SRCU
1131 DEFINE_STATIC_SRCU 1112 DEFINE_STATIC_SRCU
1132 init_srcu_struct 1113 init_srcu_struct
1133 cleanup_srcu_struct 1114 cleanup_srcu_struct
1134 1115
1135 All: lockdep-checked RCU utility APIs:: 1116 All: lockdep-checked RCU utility APIs::
1136 1117
1137 RCU_LOCKDEP_WARN 1118 RCU_LOCKDEP_WARN
1138 rcu_sleep_check 1119 rcu_sleep_check
1139 1120
1140 All: Unchecked RCU-protected pointer access:: 1121 All: Unchecked RCU-protected pointer access::
1141 1122
1142 rcu_dereference_raw 1123 rcu_dereference_raw
1143 1124
1144 All: Unchecked RCU-protected pointer access w 1125 All: Unchecked RCU-protected pointer access with dereferencing prohibited::
1145 1126
1146 rcu_access_pointer 1127 rcu_access_pointer
1147 1128
1148 See the comment headers in the source code (o 1129 See the comment headers in the source code (or the docbook generated
1149 from them) for more information. 1130 from them) for more information.
1150 1131
1151 However, given that there are no fewer than f 1132 However, given that there are no fewer than four families of RCU APIs
1152 in the Linux kernel, how do you choose which 1133 in the Linux kernel, how do you choose which one to use? The following
1153 list can be helpful: 1134 list can be helpful:
1154 1135
1155 a. Will readers need to block? If so, y 1136 a. Will readers need to block? If so, you need SRCU.
1156 1137
1157 b. Will readers need to block and are yo 1138 b. Will readers need to block and are you doing tracing, for
1158 example, ftrace or BPF? If so, you n 1139 example, ftrace or BPF? If so, you need RCU-tasks,
1159 RCU-tasks-rude, and/or RCU-tasks-trac 1140 RCU-tasks-rude, and/or RCU-tasks-trace.
1160 1141
1161 c. What about the -rt patchset? If read 1142 c. What about the -rt patchset? If readers would need to block in
1162 an non-rt kernel, you need SRCU. If 1143 an non-rt kernel, you need SRCU. If readers would block when
1163 acquiring spinlocks in a -rt kernel, 1144 acquiring spinlocks in a -rt kernel, but not in a non-rt kernel,
1164 SRCU is not necessary. (The -rt patc 1145 SRCU is not necessary. (The -rt patchset turns spinlocks into
1165 sleeplocks, hence this distinction.) 1146 sleeplocks, hence this distinction.)
1166 1147
1167 d. Do you need to treat NMI handlers, ha 1148 d. Do you need to treat NMI handlers, hardirq handlers,
1168 and code segments with preemption dis 1149 and code segments with preemption disabled (whether
1169 via preempt_disable(), local_irq_save 1150 via preempt_disable(), local_irq_save(), local_bh_disable(),
1170 or some other mechanism) as if they w 1151 or some other mechanism) as if they were explicit RCU readers?
1171 If so, RCU-sched readers are the only 1152 If so, RCU-sched readers are the only choice that will work
1172 for you, but since about v4.20 you us 1153 for you, but since about v4.20 you use can use the vanilla RCU
1173 update primitives. 1154 update primitives.
1174 1155
1175 e. Do you need RCU grace periods to comp 1156 e. Do you need RCU grace periods to complete even in the face of
1176 softirq monopolization of one or more 1157 softirq monopolization of one or more of the CPUs? For example,
1177 is your code subject to network-based 1158 is your code subject to network-based denial-of-service attacks?
1178 If so, you should disable softirq acr 1159 If so, you should disable softirq across your readers, for
1179 example, by using rcu_read_lock_bh(). 1160 example, by using rcu_read_lock_bh(). Since about v4.20 you
1180 use can use the vanilla RCU update pr 1161 use can use the vanilla RCU update primitives.
1181 1162
1182 f. Is your workload too update-intensive 1163 f. Is your workload too update-intensive for normal use of
1183 RCU, but inappropriate for other sync 1164 RCU, but inappropriate for other synchronization mechanisms?
1184 If so, consider SLAB_TYPESAFE_BY_RCU 1165 If so, consider SLAB_TYPESAFE_BY_RCU (which was originally
1185 named SLAB_DESTROY_BY_RCU). But plea 1166 named SLAB_DESTROY_BY_RCU). But please be careful!
1186 1167
1187 g. Do you need read-side critical sectio 1168 g. Do you need read-side critical sections that are respected even
1188 on CPUs that are deep in the idle loo 1169 on CPUs that are deep in the idle loop, during entry to or exit
1189 from user-mode execution, or on an of 1170 from user-mode execution, or on an offlined CPU? If so, SRCU
1190 and RCU Tasks Trace are the only choi 1171 and RCU Tasks Trace are the only choices that will work for you,
1191 with SRCU being strongly preferred in 1172 with SRCU being strongly preferred in almost all cases.
1192 1173
1193 h. Otherwise, use RCU. 1174 h. Otherwise, use RCU.
1194 1175
1195 Of course, this all assumes that you have det 1176 Of course, this all assumes that you have determined that RCU is in fact
1196 the right tool for your job. 1177 the right tool for your job.
1197 1178
1198 .. _9_whatisRCU: 1179 .. _9_whatisRCU:
1199 1180
1200 9. ANSWERS TO QUICK QUIZZES 1181 9. ANSWERS TO QUICK QUIZZES
1201 ---------------------------- 1182 ----------------------------
1202 1183
1203 Quick Quiz #1: 1184 Quick Quiz #1:
1204 Why is this argument naive? 1185 Why is this argument naive? How could a deadlock
1205 occur when using this algorit 1186 occur when using this algorithm in a real-world Linux
1206 kernel? [Referring to the lo 1187 kernel? [Referring to the lock-based "toy" RCU
1207 algorithm.] 1188 algorithm.]
1208 1189
1209 Answer: 1190 Answer:
1210 Consider the following sequen 1191 Consider the following sequence of events:
1211 1192
1212 1. CPU 0 acquires some u 1193 1. CPU 0 acquires some unrelated lock, call it
1213 "problematic_lock", d 1194 "problematic_lock", disabling irq via
1214 spin_lock_irqsave(). 1195 spin_lock_irqsave().
1215 1196
1216 2. CPU 1 enters synchron 1197 2. CPU 1 enters synchronize_rcu(), write-acquiring
1217 rcu_gp_mutex. 1198 rcu_gp_mutex.
1218 1199
1219 3. CPU 0 enters rcu_read 1200 3. CPU 0 enters rcu_read_lock(), but must wait
1220 because CPU 1 holds r 1201 because CPU 1 holds rcu_gp_mutex.
1221 1202
1222 4. CPU 1 is interrupted, 1203 4. CPU 1 is interrupted, and the irq handler
1223 attempts to acquire p 1204 attempts to acquire problematic_lock.
1224 1205
1225 The system is now deadlocked. 1206 The system is now deadlocked.
1226 1207
1227 One way to avoid this deadloc 1208 One way to avoid this deadlock is to use an approach like
1228 that of CONFIG_PREEMPT_RT, wh 1209 that of CONFIG_PREEMPT_RT, where all normal spinlocks
1229 become blocking locks, and al 1210 become blocking locks, and all irq handlers execute in
1230 the context of special tasks. 1211 the context of special tasks. In this case, in step 4
1231 above, the irq handler would 1212 above, the irq handler would block, allowing CPU 1 to
1232 release rcu_gp_mutex, avoidin 1213 release rcu_gp_mutex, avoiding the deadlock.
1233 1214
1234 Even in the absence of deadlo 1215 Even in the absence of deadlock, this RCU implementation
1235 allows latency to "bleed" fro 1216 allows latency to "bleed" from readers to other
1236 readers through synchronize_r 1217 readers through synchronize_rcu(). To see this,
1237 consider task A in an RCU rea 1218 consider task A in an RCU read-side critical section
1238 (thus read-holding rcu_gp_mut 1219 (thus read-holding rcu_gp_mutex), task B blocked
1239 attempting to write-acquire r 1220 attempting to write-acquire rcu_gp_mutex, and
1240 task C blocked in rcu_read_lo 1221 task C blocked in rcu_read_lock() attempting to
1241 read_acquire rcu_gp_mutex. T 1222 read_acquire rcu_gp_mutex. Task A's RCU read-side
1242 latency is holding up task C, 1223 latency is holding up task C, albeit indirectly via
1243 task B. 1224 task B.
1244 1225
1245 Realtime RCU implementations 1226 Realtime RCU implementations therefore use a counter-based
1246 approach where tasks in RCU r 1227 approach where tasks in RCU read-side critical sections
1247 cannot be blocked by tasks ex 1228 cannot be blocked by tasks executing synchronize_rcu().
1248 1229
1249 :ref:`Back to Quick Quiz #1 <quiz_1>` 1230 :ref:`Back to Quick Quiz #1 <quiz_1>`
1250 1231
1251 Quick Quiz #2: 1232 Quick Quiz #2:
1252 Give an example where Classic 1233 Give an example where Classic RCU's read-side
1253 overhead is **negative**. 1234 overhead is **negative**.
1254 1235
1255 Answer: 1236 Answer:
1256 Imagine a single-CPU system w 1237 Imagine a single-CPU system with a non-CONFIG_PREEMPTION
1257 kernel where a routing table 1238 kernel where a routing table is used by process-context
1258 code, but can be updated by i 1239 code, but can be updated by irq-context code (for example,
1259 by an "ICMP REDIRECT" packet) 1240 by an "ICMP REDIRECT" packet). The usual way of handling
1260 this would be to have the pro 1241 this would be to have the process-context code disable
1261 interrupts while searching th 1242 interrupts while searching the routing table. Use of
1262 RCU allows such interrupt-dis 1243 RCU allows such interrupt-disabling to be dispensed with.
1263 Thus, without RCU, you pay th 1244 Thus, without RCU, you pay the cost of disabling interrupts,
1264 and with RCU you don't. 1245 and with RCU you don't.
1265 1246
1266 One can argue that the overhe 1247 One can argue that the overhead of RCU in this
1267 case is negative with respect 1248 case is negative with respect to the single-CPU
1268 interrupt-disabling approach. 1249 interrupt-disabling approach. Others might argue that
1269 the overhead of RCU is merely 1250 the overhead of RCU is merely zero, and that replacing
1270 the positive overhead of the 1251 the positive overhead of the interrupt-disabling scheme
1271 with the zero-overhead RCU sc 1252 with the zero-overhead RCU scheme does not constitute
1272 negative overhead. 1253 negative overhead.
1273 1254
1274 In real life, of course, thin 1255 In real life, of course, things are more complex. But
1275 even the theoretical possibil 1256 even the theoretical possibility of negative overhead for
1276 a synchronization primitive i 1257 a synchronization primitive is a bit unexpected. ;-)
1277 1258
1278 :ref:`Back to Quick Quiz #2 <quiz_2>` 1259 :ref:`Back to Quick Quiz #2 <quiz_2>`
1279 1260
1280 Quick Quiz #3: 1261 Quick Quiz #3:
1281 If it is illegal to block in 1262 If it is illegal to block in an RCU read-side
1282 critical section, what the he 1263 critical section, what the heck do you do in
1283 CONFIG_PREEMPT_RT, where norm 1264 CONFIG_PREEMPT_RT, where normal spinlocks can block???
1284 1265
1285 Answer: 1266 Answer:
1286 Just as CONFIG_PREEMPT_RT per 1267 Just as CONFIG_PREEMPT_RT permits preemption of spinlock
1287 critical sections, it permits 1268 critical sections, it permits preemption of RCU
1288 read-side critical sections. 1269 read-side critical sections. It also permits
1289 spinlocks blocking while in R 1270 spinlocks blocking while in RCU read-side critical
1290 sections. 1271 sections.
1291 1272
1292 Why the apparent inconsistenc 1273 Why the apparent inconsistency? Because it is
1293 possible to use priority boos 1274 possible to use priority boosting to keep the RCU
1294 grace periods short if need b 1275 grace periods short if need be (for example, if running
1295 short of memory). In contras 1276 short of memory). In contrast, if blocking waiting
1296 for (say) network reception, 1277 for (say) network reception, there is no way to know
1297 what should be boosted. Espe 1278 what should be boosted. Especially given that the
1298 process we need to boost migh 1279 process we need to boost might well be a human being
1299 who just went out for a pizza 1280 who just went out for a pizza or something. And although
1300 a computer-operated cattle pr 1281 a computer-operated cattle prod might arouse serious
1301 interest, it might also provo 1282 interest, it might also provoke serious objections.
1302 Besides, how does the compute 1283 Besides, how does the computer know what pizza parlor
1303 the human being went to??? 1284 the human being went to???
1304 1285
1305 :ref:`Back to Quick Quiz #3 <quiz_3>` 1286 :ref:`Back to Quick Quiz #3 <quiz_3>`
1306 1287
1307 ACKNOWLEDGEMENTS 1288 ACKNOWLEDGEMENTS
1308 1289
1309 My thanks to the people who helped make this 1290 My thanks to the people who helped make this human-readable, including
1310 Jon Walpole, Josh Triplett, Serge Hallyn, Suz 1291 Jon Walpole, Josh Triplett, Serge Hallyn, Suzanne Wood, and Alan Stern.
1311 1292
1312 1293
1313 For more information, see http://www.rdrop.co 1294 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.