1 .. SPDX-License-Identifier: GPL-2.0 1 .. SPDX-License-Identifier: GPL-2.0
2 2
3 ================================ 3 ================================
4 Review Checklist for RCU Patches 4 Review Checklist for RCU Patches
5 ================================ 5 ================================
6 6
7 7
8 This document contains a checklist for produci 8 This document contains a checklist for producing and reviewing patches
9 that make use of RCU. Violating any of the ru 9 that make use of RCU. Violating any of the rules listed below will
10 result in the same sorts of problems that leav 10 result in the same sorts of problems that leaving out a locking primitive
11 would cause. This list is based on experience 11 would cause. This list is based on experiences reviewing such patches
12 over a rather long period of time, but improve 12 over a rather long period of time, but improvements are always welcome!
13 13
14 0. Is RCU being applied to a read-mostly 14 0. Is RCU being applied to a read-mostly situation? If the data
15 structure is updated more than about 1 15 structure is updated more than about 10% of the time, then you
16 should strongly consider some other ap 16 should strongly consider some other approach, unless detailed
17 performance measurements show that RCU 17 performance measurements show that RCU is nonetheless the right
18 tool for the job. Yes, RCU does reduc 18 tool for the job. Yes, RCU does reduce read-side overhead by
19 increasing write-side overhead, which 19 increasing write-side overhead, which is exactly why normal uses
20 of RCU will do much more reading than 20 of RCU will do much more reading than updating.
21 21
22 Another exception is where performance 22 Another exception is where performance is not an issue, and RCU
23 provides a simpler implementation. An 23 provides a simpler implementation. An example of this situation
24 is the dynamic NMI code in the Linux 2 24 is the dynamic NMI code in the Linux 2.6 kernel, at least on
25 architectures where NMIs are rare. 25 architectures where NMIs are rare.
26 26
27 Yet another exception is where the low 27 Yet another exception is where the low real-time latency of RCU's
28 read-side primitives is critically imp 28 read-side primitives is critically important.
29 29
30 One final exception is where RCU reade 30 One final exception is where RCU readers are used to prevent
31 the ABA problem (https://en.wikipedia. 31 the ABA problem (https://en.wikipedia.org/wiki/ABA_problem)
32 for lockless updates. This does resul 32 for lockless updates. This does result in the mildly
33 counter-intuitive situation where rcu_ 33 counter-intuitive situation where rcu_read_lock() and
34 rcu_read_unlock() are used to protect 34 rcu_read_unlock() are used to protect updates, however, this
35 approach can provide the same simplifi 35 approach can provide the same simplifications to certain types
36 of lockless algorithms that garbage co 36 of lockless algorithms that garbage collectors do.
37 37
38 1. Does the update code have proper mutua 38 1. Does the update code have proper mutual exclusion?
39 39
40 RCU does allow *readers* to run (almos 40 RCU does allow *readers* to run (almost) naked, but *writers* must
41 still use some sort of mutual exclusio 41 still use some sort of mutual exclusion, such as:
42 42
43 a. locking, 43 a. locking,
44 b. atomic operations, or 44 b. atomic operations, or
45 c. restricting updates to a singl 45 c. restricting updates to a single task.
46 46
47 If you choose #b, be prepared to descr 47 If you choose #b, be prepared to describe how you have handled
48 memory barriers on weakly ordered mach 48 memory barriers on weakly ordered machines (pretty much all of
49 them -- even x86 allows later loads to 49 them -- even x86 allows later loads to be reordered to precede
50 earlier stores), and be prepared to ex 50 earlier stores), and be prepared to explain why this added
51 complexity is worthwhile. If you choo 51 complexity is worthwhile. If you choose #c, be prepared to
52 explain how this single task does not 52 explain how this single task does not become a major bottleneck
53 on large systems (for example, if the 53 on large systems (for example, if the task is updating information
54 relating to itself that other tasks ca 54 relating to itself that other tasks can read, there by definition
55 can be no bottleneck). Note that the 55 can be no bottleneck). Note that the definition of "large" has
56 changed significantly: Eight CPUs was 56 changed significantly: Eight CPUs was "large" in the year 2000,
57 but a hundred CPUs was unremarkable in 57 but a hundred CPUs was unremarkable in 2017.
58 58
59 2. Do the RCU read-side critical sections 59 2. Do the RCU read-side critical sections make proper use of
60 rcu_read_lock() and friends? These pr 60 rcu_read_lock() and friends? These primitives are needed
61 to prevent grace periods from ending p 61 to prevent grace periods from ending prematurely, which
62 could result in data being unceremonio 62 could result in data being unceremoniously freed out from
63 under your read-side code, which can g 63 under your read-side code, which can greatly increase the
64 actuarial risk of your kernel. 64 actuarial risk of your kernel.
65 65
66 As a rough rule of thumb, any derefere 66 As a rough rule of thumb, any dereference of an RCU-protected
67 pointer must be covered by rcu_read_lo 67 pointer must be covered by rcu_read_lock(), rcu_read_lock_bh(),
68 rcu_read_lock_sched(), or by the appro 68 rcu_read_lock_sched(), or by the appropriate update-side lock.
69 Explicit disabling of preemption (pree 69 Explicit disabling of preemption (preempt_disable(), for example)
70 can serve as rcu_read_lock_sched(), bu 70 can serve as rcu_read_lock_sched(), but is less readable and
71 prevents lockdep from detecting lockin 71 prevents lockdep from detecting locking issues. Acquiring a
72 spinlock also enters an RCU read-side 72 spinlock also enters an RCU read-side critical section.
73 73
74 Please note that you *cannot* rely on 74 Please note that you *cannot* rely on code known to be built
75 only in non-preemptible kernels. Such 75 only in non-preemptible kernels. Such code can and will break,
76 especially in kernels built with CONFI 76 especially in kernels built with CONFIG_PREEMPT_COUNT=y.
77 77
78 Letting RCU-protected pointers "leak" 78 Letting RCU-protected pointers "leak" out of an RCU read-side
79 critical section is every bit as bad a 79 critical section is every bit as bad as letting them leak out
80 from under a lock. Unless, of course, 80 from under a lock. Unless, of course, you have arranged some
81 other means of protection, such as a l 81 other means of protection, such as a lock or a reference count
82 *before* letting them out of the RCU r 82 *before* letting them out of the RCU read-side critical section.
83 83
84 3. Does the update code tolerate concurre 84 3. Does the update code tolerate concurrent accesses?
85 85
86 The whole point of RCU is to permit re 86 The whole point of RCU is to permit readers to run without
87 any locks or atomic operations. This 87 any locks or atomic operations. This means that readers will
88 be running while updates are in progre 88 be running while updates are in progress. There are a number
89 of ways to handle this concurrency, de 89 of ways to handle this concurrency, depending on the situation:
90 90
91 a. Use the RCU variants of the li 91 a. Use the RCU variants of the list and hlist update
92 primitives to add, remove, and 92 primitives to add, remove, and replace elements on
93 an RCU-protected list. Altern 93 an RCU-protected list. Alternatively, use the other
94 RCU-protected data structures 94 RCU-protected data structures that have been added to
95 the Linux kernel. 95 the Linux kernel.
96 96
97 This is almost always the best 97 This is almost always the best approach.
98 98
99 b. Proceed as in (a) above, but a 99 b. Proceed as in (a) above, but also maintain per-element
100 locks (that are acquired by bo 100 locks (that are acquired by both readers and writers)
101 that guard per-element state. 101 that guard per-element state. Fields that the readers
102 refrain from accessing can be 102 refrain from accessing can be guarded by some other lock
103 acquired only by updaters, if 103 acquired only by updaters, if desired.
104 104
105 This also works quite well. 105 This also works quite well.
106 106
107 c. Make updates appear atomic to 107 c. Make updates appear atomic to readers. For example,
108 pointer updates to properly al 108 pointer updates to properly aligned fields will
109 appear atomic, as will individ 109 appear atomic, as will individual atomic primitives.
110 Sequences of operations perfor 110 Sequences of operations performed under a lock will *not*
111 appear to be atomic to RCU rea 111 appear to be atomic to RCU readers, nor will sequences
112 of multiple atomic primitives. 112 of multiple atomic primitives. One alternative is to
113 move multiple individual field 113 move multiple individual fields to a separate structure,
114 thus solving the multiple-fiel 114 thus solving the multiple-field problem by imposing an
115 additional level of indirectio 115 additional level of indirection.
116 116
117 This can work, but is starting 117 This can work, but is starting to get a bit tricky.
118 118
119 d. Carefully order the updates an 119 d. Carefully order the updates and the reads so that readers
120 see valid data at all phases o 120 see valid data at all phases of the update. This is often
121 more difficult than it sounds, 121 more difficult than it sounds, especially given modern
122 CPUs' tendency to reorder memo 122 CPUs' tendency to reorder memory references. One must
123 usually liberally sprinkle mem 123 usually liberally sprinkle memory-ordering operations
124 through the code, making it di 124 through the code, making it difficult to understand and
125 to test. Where it works, it i 125 to test. Where it works, it is better to use things
126 like smp_store_release() and s 126 like smp_store_release() and smp_load_acquire(), but in
127 some cases the smp_mb() full m 127 some cases the smp_mb() full memory barrier is required.
128 128
129 As noted earlier, it is usuall 129 As noted earlier, it is usually better to group the
130 changing data into a separate 130 changing data into a separate structure, so that the
131 change may be made to appear a 131 change may be made to appear atomic by updating a pointer
132 to reference a new structure c 132 to reference a new structure containing updated values.
133 133
134 4. Weakly ordered CPUs pose special chall 134 4. Weakly ordered CPUs pose special challenges. Almost all CPUs
135 are weakly ordered -- even x86 CPUs al 135 are weakly ordered -- even x86 CPUs allow later loads to be
136 reordered to precede earlier stores. 136 reordered to precede earlier stores. RCU code must take all of
137 the following measures to prevent memo 137 the following measures to prevent memory-corruption problems:
138 138
139 a. Readers must maintain proper o 139 a. Readers must maintain proper ordering of their memory
140 accesses. The rcu_dereference 140 accesses. The rcu_dereference() primitive ensures that
141 the CPU picks up the pointer b 141 the CPU picks up the pointer before it picks up the data
142 that the pointer points to. T 142 that the pointer points to. This really is necessary
143 on Alpha CPUs. 143 on Alpha CPUs.
144 144
145 The rcu_dereference() primitiv 145 The rcu_dereference() primitive is also an excellent
146 documentation aid, letting the 146 documentation aid, letting the person reading the
147 code know exactly which pointe 147 code know exactly which pointers are protected by RCU.
148 Please note that compilers can 148 Please note that compilers can also reorder code, and
149 they are becoming increasingly 149 they are becoming increasingly aggressive about doing
150 just that. The rcu_dereferenc 150 just that. The rcu_dereference() primitive therefore also
151 prevents destructive compiler 151 prevents destructive compiler optimizations. However,
152 with a bit of devious creativi 152 with a bit of devious creativity, it is possible to
153 mishandle the return value fro 153 mishandle the return value from rcu_dereference().
154 Please see rcu_dereference.rst 154 Please see rcu_dereference.rst for more information.
155 155
156 The rcu_dereference() primitiv 156 The rcu_dereference() primitive is used by the
157 various "_rcu()" list-traversa 157 various "_rcu()" list-traversal primitives, such
158 as the list_for_each_entry_rcu 158 as the list_for_each_entry_rcu(). Note that it is
159 perfectly legal (if redundant) 159 perfectly legal (if redundant) for update-side code to
160 use rcu_dereference() and the 160 use rcu_dereference() and the "_rcu()" list-traversal
161 primitives. This is particula 161 primitives. This is particularly useful in code that
162 is common to readers and updat 162 is common to readers and updaters. However, lockdep
163 will complain if you access rc 163 will complain if you access rcu_dereference() outside
164 of an RCU read-side critical s 164 of an RCU read-side critical section. See lockdep.rst
165 to learn what to do about this 165 to learn what to do about this.
166 166
167 Of course, neither rcu_derefer 167 Of course, neither rcu_dereference() nor the "_rcu()"
168 list-traversal primitives can 168 list-traversal primitives can substitute for a good
169 concurrency design coordinatin 169 concurrency design coordinating among multiple updaters.
170 170
171 b. If the list macros are being u 171 b. If the list macros are being used, the list_add_tail_rcu()
172 and list_add_rcu() primitives 172 and list_add_rcu() primitives must be used in order
173 to prevent weakly ordered mach 173 to prevent weakly ordered machines from misordering
174 structure initialization and p 174 structure initialization and pointer planting.
175 Similarly, if the hlist macros 175 Similarly, if the hlist macros are being used, the
176 hlist_add_head_rcu() primitive 176 hlist_add_head_rcu() primitive is required.
177 177
178 c. If the list macros are being u 178 c. If the list macros are being used, the list_del_rcu()
179 primitive must be used to keep 179 primitive must be used to keep list_del()'s pointer
180 poisoning from inflicting toxi 180 poisoning from inflicting toxic effects on concurrent
181 readers. Similarly, if the hl 181 readers. Similarly, if the hlist macros are being used,
182 the hlist_del_rcu() primitive 182 the hlist_del_rcu() primitive is required.
183 183
184 The list_replace_rcu() and hli 184 The list_replace_rcu() and hlist_replace_rcu() primitives
185 may be used to replace an old 185 may be used to replace an old structure with a new one
186 in their respective types of R 186 in their respective types of RCU-protected lists.
187 187
188 d. Rules similar to (4b) and (4c) 188 d. Rules similar to (4b) and (4c) apply to the "hlist_nulls"
189 type of RCU-protected linked l 189 type of RCU-protected linked lists.
190 190
191 e. Updates must ensure that initi 191 e. Updates must ensure that initialization of a given
192 structure happens before point 192 structure happens before pointers to that structure are
193 publicized. Use the rcu_assig 193 publicized. Use the rcu_assign_pointer() primitive
194 when publicizing a pointer to 194 when publicizing a pointer to a structure that can
195 be traversed by an RCU read-si 195 be traversed by an RCU read-side critical section.
196 196
197 5. If any of call_rcu(), call_srcu(), cal 197 5. If any of call_rcu(), call_srcu(), call_rcu_tasks(), or
198 call_rcu_tasks_trace() is used, the ca 198 call_rcu_tasks_trace() is used, the callback function may be
199 invoked from softirq context, and in a 199 invoked from softirq context, and in any case with bottom halves
200 disabled. In particular, this callbac 200 disabled. In particular, this callback function cannot block.
201 If you need the callback to block, run 201 If you need the callback to block, run that code in a workqueue
202 handler scheduled from the callback. 202 handler scheduled from the callback. The queue_rcu_work()
203 function does this for you in the case 203 function does this for you in the case of call_rcu().
204 204
205 6. Since synchronize_rcu() can block, it 205 6. Since synchronize_rcu() can block, it cannot be called
206 from any sort of irq context. The sam 206 from any sort of irq context. The same rule applies
207 for synchronize_srcu(), synchronize_rc 207 for synchronize_srcu(), synchronize_rcu_expedited(),
208 synchronize_srcu_expedited(), synchron 208 synchronize_srcu_expedited(), synchronize_rcu_tasks(),
209 synchronize_rcu_tasks_rude(), and sync 209 synchronize_rcu_tasks_rude(), and synchronize_rcu_tasks_trace().
210 210
211 The expedited forms of these primitive 211 The expedited forms of these primitives have the same semantics
212 as the non-expedited forms, but expedi 212 as the non-expedited forms, but expediting is more CPU intensive.
213 Use of the expedited primitives should 213 Use of the expedited primitives should be restricted to rare
214 configuration-change operations that w 214 configuration-change operations that would not normally be
215 undertaken while a real-time workload 215 undertaken while a real-time workload is running. Note that
216 IPI-sensitive real-time workloads can 216 IPI-sensitive real-time workloads can use the rcupdate.rcu_normal
217 kernel boot parameter to completely di 217 kernel boot parameter to completely disable expedited grace
218 periods, though this might have perfor 218 periods, though this might have performance implications.
219 219
220 In particular, if you find yourself in 220 In particular, if you find yourself invoking one of the expedited
221 primitives repeatedly in a loop, pleas 221 primitives repeatedly in a loop, please do everyone a favor:
222 Restructure your code so that it batch 222 Restructure your code so that it batches the updates, allowing
223 a single non-expedited primitive to co 223 a single non-expedited primitive to cover the entire batch.
224 This will very likely be faster than t 224 This will very likely be faster than the loop containing the
225 expedited primitive, and will be much 225 expedited primitive, and will be much much easier on the rest
226 of the system, especially to real-time 226 of the system, especially to real-time workloads running on the
227 rest of the system. Alternatively, in 227 rest of the system. Alternatively, instead use asynchronous
228 primitives such as call_rcu(). 228 primitives such as call_rcu().
229 229
230 7. As of v4.20, a given kernel implements 230 7. As of v4.20, a given kernel implements only one RCU flavor, which
231 is RCU-sched for PREEMPTION=n and RCU- 231 is RCU-sched for PREEMPTION=n and RCU-preempt for PREEMPTION=y.
232 If the updater uses call_rcu() or sync 232 If the updater uses call_rcu() or synchronize_rcu(), then
233 the corresponding readers may use: (1 233 the corresponding readers may use: (1) rcu_read_lock() and
234 rcu_read_unlock(), (2) any pair of pri 234 rcu_read_unlock(), (2) any pair of primitives that disables
235 and re-enables softirq, for example, r 235 and re-enables softirq, for example, rcu_read_lock_bh() and
236 rcu_read_unlock_bh(), or (3) any pair 236 rcu_read_unlock_bh(), or (3) any pair of primitives that disables
237 and re-enables preemption, for example 237 and re-enables preemption, for example, rcu_read_lock_sched() and
238 rcu_read_unlock_sched(). If the updat 238 rcu_read_unlock_sched(). If the updater uses synchronize_srcu()
239 or call_srcu(), then the corresponding 239 or call_srcu(), then the corresponding readers must use
240 srcu_read_lock() and srcu_read_unlock( 240 srcu_read_lock() and srcu_read_unlock(), and with the same
241 srcu_struct. The rules for the expedi 241 srcu_struct. The rules for the expedited RCU grace-period-wait
242 primitives are the same as for their n 242 primitives are the same as for their non-expedited counterparts.
243 243
244 Similarly, it is necessary to correctl 244 Similarly, it is necessary to correctly use the RCU Tasks flavors:
245 245
246 a. If the updater uses synchroniz 246 a. If the updater uses synchronize_rcu_tasks() or
247 call_rcu_tasks(), then the rea 247 call_rcu_tasks(), then the readers must refrain from
248 executing voluntary context sw 248 executing voluntary context switches, that is, from
249 blocking. 249 blocking.
250 250
251 b. If the updater uses call_rcu_t 251 b. If the updater uses call_rcu_tasks_trace()
252 or synchronize_rcu_tasks_trace 252 or synchronize_rcu_tasks_trace(), then the
253 corresponding readers must use 253 corresponding readers must use rcu_read_lock_trace()
254 and rcu_read_unlock_trace(). 254 and rcu_read_unlock_trace().
255 255
256 c. If an updater uses synchronize 256 c. If an updater uses synchronize_rcu_tasks_rude(),
257 then the corresponding readers 257 then the corresponding readers must use anything that
258 disables preemption, for examp 258 disables preemption, for example, preempt_disable()
259 and preempt_enable(). 259 and preempt_enable().
260 260
261 Mixing things up will result in confus 261 Mixing things up will result in confusion and broken kernels, and
262 has even resulted in an exploitable se 262 has even resulted in an exploitable security issue. Therefore,
263 when using non-obvious pairs of primit 263 when using non-obvious pairs of primitives, commenting is
264 of course a must. One example of non- 264 of course a must. One example of non-obvious pairing is
265 the XDP feature in networking, which c 265 the XDP feature in networking, which calls BPF programs from
266 network-driver NAPI (softirq) context. 266 network-driver NAPI (softirq) context. BPF relies heavily on RCU
267 protection for its data structures, bu 267 protection for its data structures, but because the BPF program
268 invocation happens entirely within a s 268 invocation happens entirely within a single local_bh_disable()
269 section in a NAPI poll cycle, this usa 269 section in a NAPI poll cycle, this usage is safe. The reason
270 that this usage is safe is that reader 270 that this usage is safe is that readers can use anything that
271 disables BH when updaters use call_rcu 271 disables BH when updaters use call_rcu() or synchronize_rcu().
272 272
273 8. Although synchronize_rcu() is slower t 273 8. Although synchronize_rcu() is slower than is call_rcu(),
274 it usually results in simpler code. S 274 it usually results in simpler code. So, unless update
275 performance is critically important, t 275 performance is critically important, the updaters cannot block,
276 or the latency of synchronize_rcu() is 276 or the latency of synchronize_rcu() is visible from userspace,
277 synchronize_rcu() should be used in pr 277 synchronize_rcu() should be used in preference to call_rcu().
278 Furthermore, kfree_rcu() and kvfree_rc 278 Furthermore, kfree_rcu() and kvfree_rcu() usually result
279 in even simpler code than does synchro 279 in even simpler code than does synchronize_rcu() without
280 synchronize_rcu()'s multi-millisecond 280 synchronize_rcu()'s multi-millisecond latency. So please take
281 advantage of kfree_rcu()'s and kvfree_ 281 advantage of kfree_rcu()'s and kvfree_rcu()'s "fire and forget"
282 memory-freeing capabilities where it a 282 memory-freeing capabilities where it applies.
283 283
284 An especially important property of th 284 An especially important property of the synchronize_rcu()
285 primitive is that it automatically sel 285 primitive is that it automatically self-limits: if grace periods
286 are delayed for whatever reason, then 286 are delayed for whatever reason, then the synchronize_rcu()
287 primitive will correspondingly delay u 287 primitive will correspondingly delay updates. In contrast,
288 code using call_rcu() should explicitl 288 code using call_rcu() should explicitly limit update rate in
289 cases where grace periods are delayed, 289 cases where grace periods are delayed, as failing to do so can
290 result in excessive realtime latencies 290 result in excessive realtime latencies or even OOM conditions.
291 291
292 Ways of gaining this self-limiting pro 292 Ways of gaining this self-limiting property when using call_rcu(),
293 kfree_rcu(), or kvfree_rcu() include: 293 kfree_rcu(), or kvfree_rcu() include:
294 294
295 a. Keeping a count of the number 295 a. Keeping a count of the number of data-structure elements
296 used by the RCU-protected data 296 used by the RCU-protected data structure, including
297 those waiting for a grace peri 297 those waiting for a grace period to elapse. Enforce a
298 limit on this number, stalling 298 limit on this number, stalling updates as needed to allow
299 previously deferred frees to c 299 previously deferred frees to complete. Alternatively,
300 limit only the number awaiting 300 limit only the number awaiting deferred free rather than
301 the total number of elements. 301 the total number of elements.
302 302
303 One way to stall the updates i 303 One way to stall the updates is to acquire the update-side
304 mutex. (Don't try this with a 304 mutex. (Don't try this with a spinlock -- other CPUs
305 spinning on the lock could pre 305 spinning on the lock could prevent the grace period
306 from ever ending.) Another wa 306 from ever ending.) Another way to stall the updates
307 is for the updates to use a wr 307 is for the updates to use a wrapper function around
308 the memory allocator, so that 308 the memory allocator, so that this wrapper function
309 simulates OOM when there is to 309 simulates OOM when there is too much memory awaiting an
310 RCU grace period. There are o 310 RCU grace period. There are of course many other
311 variations on this theme. 311 variations on this theme.
312 312
313 b. Limiting update rate. For exa 313 b. Limiting update rate. For example, if updates occur only
314 once per hour, then no explici 314 once per hour, then no explicit rate limiting is
315 required, unless your system i 315 required, unless your system is already badly broken.
316 Older versions of the dcache s 316 Older versions of the dcache subsystem take this approach,
317 guarding updates with a global 317 guarding updates with a global lock, limiting their rate.
318 318
319 c. Trusted update -- if updates c 319 c. Trusted update -- if updates can only be done manually by
320 superuser or some other truste 320 superuser or some other trusted user, then it might not
321 be necessary to automatically 321 be necessary to automatically limit them. The theory
322 here is that superuser already 322 here is that superuser already has lots of ways to crash
323 the machine. 323 the machine.
324 324
325 d. Periodically invoke rcu_barrie 325 d. Periodically invoke rcu_barrier(), permitting a limited
326 number of updates per grace pe 326 number of updates per grace period.
327 327
328 The same cautions apply to call_srcu() 328 The same cautions apply to call_srcu(), call_rcu_tasks(), and
329 call_rcu_tasks_trace(). This is why t 329 call_rcu_tasks_trace(). This is why there is an srcu_barrier(),
330 rcu_barrier_tasks(), and rcu_barrier_t 330 rcu_barrier_tasks(), and rcu_barrier_tasks_trace(), respectively.
331 331
332 Note that although these primitives do 332 Note that although these primitives do take action to avoid
333 memory exhaustion when any given CPU h 333 memory exhaustion when any given CPU has too many callbacks,
334 a determined user or administrator can 334 a determined user or administrator can still exhaust memory.
335 This is especially the case if a syste 335 This is especially the case if a system with a large number of
336 CPUs has been configured to offload al 336 CPUs has been configured to offload all of its RCU callbacks onto
337 a single CPU, or if the system has rel 337 a single CPU, or if the system has relatively little free memory.
338 338
339 9. All RCU list-traversal primitives, whi 339 9. All RCU list-traversal primitives, which include
340 rcu_dereference(), list_for_each_entry 340 rcu_dereference(), list_for_each_entry_rcu(), and
341 list_for_each_safe_rcu(), must be eith 341 list_for_each_safe_rcu(), must be either within an RCU read-side
342 critical section or must be protected 342 critical section or must be protected by appropriate update-side
343 locks. RCU read-side critical section 343 locks. RCU read-side critical sections are delimited by
344 rcu_read_lock() and rcu_read_unlock(), 344 rcu_read_lock() and rcu_read_unlock(), or by similar primitives
345 such as rcu_read_lock_bh() and rcu_rea 345 such as rcu_read_lock_bh() and rcu_read_unlock_bh(), in which
346 case the matching rcu_dereference() pr 346 case the matching rcu_dereference() primitive must be used in
347 order to keep lockdep happy, in this c 347 order to keep lockdep happy, in this case, rcu_dereference_bh().
348 348
349 The reason that it is permissible to u 349 The reason that it is permissible to use RCU list-traversal
350 primitives when the update-side lock i 350 primitives when the update-side lock is held is that doing so
351 can be quite helpful in reducing code 351 can be quite helpful in reducing code bloat when common code is
352 shared between readers and updaters. 352 shared between readers and updaters. Additional primitives
353 are provided for this case, as discuss 353 are provided for this case, as discussed in lockdep.rst.
354 354
355 One exception to this rule is when dat 355 One exception to this rule is when data is only ever added to
356 the linked data structure, and is neve 356 the linked data structure, and is never removed during any
357 time that readers might be accessing t 357 time that readers might be accessing that structure. In such
358 cases, READ_ONCE() may be used in plac 358 cases, READ_ONCE() may be used in place of rcu_dereference()
359 and the read-side markers (rcu_read_lo 359 and the read-side markers (rcu_read_lock() and rcu_read_unlock(),
360 for example) may be omitted. 360 for example) may be omitted.
361 361
362 10. Conversely, if you are in an RCU read- 362 10. Conversely, if you are in an RCU read-side critical section,
363 and you don't hold the appropriate upd 363 and you don't hold the appropriate update-side lock, you *must*
364 use the "_rcu()" variants of the list 364 use the "_rcu()" variants of the list macros. Failing to do so
365 will break Alpha, cause aggressive com 365 will break Alpha, cause aggressive compilers to generate bad code,
366 and confuse people trying to understan 366 and confuse people trying to understand your code.
367 367
368 11. Any lock acquired by an RCU callback m 368 11. Any lock acquired by an RCU callback must be acquired elsewhere
369 with softirq disabled, e.g., via spin_ 369 with softirq disabled, e.g., via spin_lock_bh(). Failing to
370 disable softirq on a given acquisition 370 disable softirq on a given acquisition of that lock will result
371 in deadlock as soon as the RCU softirq 371 in deadlock as soon as the RCU softirq handler happens to run
372 your RCU callback while interrupting t 372 your RCU callback while interrupting that acquisition's critical
373 section. 373 section.
374 374
375 12. RCU callbacks can be and are executed 375 12. RCU callbacks can be and are executed in parallel. In many cases,
376 the callback code simply wrappers arou 376 the callback code simply wrappers around kfree(), so that this
377 is not an issue (or, more accurately, 377 is not an issue (or, more accurately, to the extent that it is
378 an issue, the memory-allocator locking 378 an issue, the memory-allocator locking handles it). However,
379 if the callbacks do manipulate a share 379 if the callbacks do manipulate a shared data structure, they
380 must use whatever locking or other syn 380 must use whatever locking or other synchronization is required
381 to safely access and/or modify that da 381 to safely access and/or modify that data structure.
382 382
383 Do not assume that RCU callbacks will 383 Do not assume that RCU callbacks will be executed on the same
384 CPU that executed the corresponding ca 384 CPU that executed the corresponding call_rcu(), call_srcu(),
385 call_rcu_tasks(), or call_rcu_tasks_tr 385 call_rcu_tasks(), or call_rcu_tasks_trace(). For example, if
386 a given CPU goes offline while having 386 a given CPU goes offline while having an RCU callback pending,
387 then that RCU callback will execute on 387 then that RCU callback will execute on some surviving CPU.
388 (If this was not the case, a self-spaw 388 (If this was not the case, a self-spawning RCU callback would
389 prevent the victim CPU from ever going 389 prevent the victim CPU from ever going offline.) Furthermore,
390 CPUs designated by rcu_nocbs= might we 390 CPUs designated by rcu_nocbs= might well *always* have their
391 RCU callbacks executed on some other C 391 RCU callbacks executed on some other CPUs, in fact, for some
392 real-time workloads, this is the whole 392 real-time workloads, this is the whole point of using the
393 rcu_nocbs= kernel boot parameter. 393 rcu_nocbs= kernel boot parameter.
394 394
395 In addition, do not assume that callba 395 In addition, do not assume that callbacks queued in a given order
396 will be invoked in that order, even if 396 will be invoked in that order, even if they all are queued on the
397 same CPU. Furthermore, do not assume 397 same CPU. Furthermore, do not assume that same-CPU callbacks will
398 be invoked serially. For example, in 398 be invoked serially. For example, in recent kernels, CPUs can be
399 switched between offloaded and de-offl 399 switched between offloaded and de-offloaded callback invocation,
400 and while a given CPU is undergoing su 400 and while a given CPU is undergoing such a switch, its callbacks
401 might be concurrently invoked by that 401 might be concurrently invoked by that CPU's softirq handler and
402 that CPU's rcuo kthread. At such time 402 that CPU's rcuo kthread. At such times, that CPU's callbacks
403 might be executed both concurrently an 403 might be executed both concurrently and out of order.
404 404
405 13. Unlike most flavors of RCU, it *is* pe 405 13. Unlike most flavors of RCU, it *is* permissible to block in an
406 SRCU read-side critical section (demar 406 SRCU read-side critical section (demarked by srcu_read_lock()
407 and srcu_read_unlock()), hence the "SR 407 and srcu_read_unlock()), hence the "SRCU": "sleepable RCU".
408 Please note that if you don't need to 408 Please note that if you don't need to sleep in read-side critical
409 sections, you should be using RCU rath 409 sections, you should be using RCU rather than SRCU, because RCU
410 is almost always faster and easier to 410 is almost always faster and easier to use than is SRCU.
411 411
412 Also unlike other forms of RCU, explic 412 Also unlike other forms of RCU, explicit initialization and
413 cleanup is required either at build ti 413 cleanup is required either at build time via DEFINE_SRCU()
414 or DEFINE_STATIC_SRCU() or at runtime 414 or DEFINE_STATIC_SRCU() or at runtime via init_srcu_struct()
415 and cleanup_srcu_struct(). These last 415 and cleanup_srcu_struct(). These last two are passed a
416 "struct srcu_struct" that defines the 416 "struct srcu_struct" that defines the scope of a given
417 SRCU domain. Once initialized, the sr 417 SRCU domain. Once initialized, the srcu_struct is passed
418 to srcu_read_lock(), srcu_read_unlock( 418 to srcu_read_lock(), srcu_read_unlock() synchronize_srcu(),
419 synchronize_srcu_expedited(), and call 419 synchronize_srcu_expedited(), and call_srcu(). A given
420 synchronize_srcu() waits only for SRCU 420 synchronize_srcu() waits only for SRCU read-side critical
421 sections governed by srcu_read_lock() 421 sections governed by srcu_read_lock() and srcu_read_unlock()
422 calls that have been passed the same s 422 calls that have been passed the same srcu_struct. This property
423 is what makes sleeping read-side criti 423 is what makes sleeping read-side critical sections tolerable --
424 a given subsystem delays only its own 424 a given subsystem delays only its own updates, not those of other
425 subsystems using SRCU. Therefore, SRC 425 subsystems using SRCU. Therefore, SRCU is less prone to OOM the
426 system than RCU would be if RCU's read 426 system than RCU would be if RCU's read-side critical sections
427 were permitted to sleep. 427 were permitted to sleep.
428 428
429 The ability to sleep in read-side crit 429 The ability to sleep in read-side critical sections does not
430 come for free. First, corresponding s 430 come for free. First, corresponding srcu_read_lock() and
431 srcu_read_unlock() calls must be passe 431 srcu_read_unlock() calls must be passed the same srcu_struct.
432 Second, grace-period-detection overhea 432 Second, grace-period-detection overhead is amortized only
433 over those updates sharing a given src 433 over those updates sharing a given srcu_struct, rather than
434 being globally amortized as they are f 434 being globally amortized as they are for other forms of RCU.
435 Therefore, SRCU should be used in pref 435 Therefore, SRCU should be used in preference to rw_semaphore
436 only in extremely read-intensive situa 436 only in extremely read-intensive situations, or in situations
437 requiring SRCU's read-side deadlock im 437 requiring SRCU's read-side deadlock immunity or low read-side
438 realtime latency. You should also con 438 realtime latency. You should also consider percpu_rw_semaphore
439 when you need lightweight readers. 439 when you need lightweight readers.
440 440
441 SRCU's expedited primitive (synchroniz 441 SRCU's expedited primitive (synchronize_srcu_expedited())
442 never sends IPIs to other CPUs, so it 442 never sends IPIs to other CPUs, so it is easier on
443 real-time workloads than is synchroniz 443 real-time workloads than is synchronize_rcu_expedited().
444 444
445 It is also permissible to sleep in RCU 445 It is also permissible to sleep in RCU Tasks Trace read-side
446 critical section, which are delimited 446 critical section, which are delimited by rcu_read_lock_trace() and
447 rcu_read_unlock_trace(). However, thi 447 rcu_read_unlock_trace(). However, this is a specialized flavor
448 of RCU, and you should not use it with 448 of RCU, and you should not use it without first checking with
449 its current users. In most cases, you 449 its current users. In most cases, you should instead use SRCU.
450 450
451 Note that rcu_assign_pointer() relates 451 Note that rcu_assign_pointer() relates to SRCU just as it does to
452 other forms of RCU, but instead of rcu 452 other forms of RCU, but instead of rcu_dereference() you should
453 use srcu_dereference() in order to avo 453 use srcu_dereference() in order to avoid lockdep splats.
454 454
455 14. The whole point of call_rcu(), synchro 455 14. The whole point of call_rcu(), synchronize_rcu(), and friends
456 is to wait until all pre-existing read 456 is to wait until all pre-existing readers have finished before
457 carrying out some otherwise-destructiv 457 carrying out some otherwise-destructive operation. It is
458 therefore critically important to *fir 458 therefore critically important to *first* remove any path
459 that readers can follow that could be 459 that readers can follow that could be affected by the
460 destructive operation, and *only then* 460 destructive operation, and *only then* invoke call_rcu(),
461 synchronize_rcu(), or friends. 461 synchronize_rcu(), or friends.
462 462
463 Because these primitives only wait for 463 Because these primitives only wait for pre-existing readers, it
464 is the caller's responsibility to guar 464 is the caller's responsibility to guarantee that any subsequent
465 readers will execute safely. 465 readers will execute safely.
466 466
467 15. The various RCU read-side primitives d 467 15. The various RCU read-side primitives do *not* necessarily contain
468 memory barriers. You should therefore 468 memory barriers. You should therefore plan for the CPU
469 and the compiler to freely reorder cod 469 and the compiler to freely reorder code into and out of RCU
470 read-side critical sections. It is th 470 read-side critical sections. It is the responsibility of the
471 RCU update-side primitives to deal wit 471 RCU update-side primitives to deal with this.
472 472
473 For SRCU readers, you can use smp_mb__ 473 For SRCU readers, you can use smp_mb__after_srcu_read_unlock()
474 immediately after an srcu_read_unlock( 474 immediately after an srcu_read_unlock() to get a full barrier.
475 475
476 16. Use CONFIG_PROVE_LOCKING, CONFIG_DEBUG 476 16. Use CONFIG_PROVE_LOCKING, CONFIG_DEBUG_OBJECTS_RCU_HEAD, and the
477 __rcu sparse checks to validate your R 477 __rcu sparse checks to validate your RCU code. These can help
478 find problems as follows: 478 find problems as follows:
479 479
480 CONFIG_PROVE_LOCKING: 480 CONFIG_PROVE_LOCKING:
481 check that accesses to RCU-pro 481 check that accesses to RCU-protected data structures
482 are carried out under the prop 482 are carried out under the proper RCU read-side critical
483 section, while holding the rig 483 section, while holding the right combination of locks,
484 or whatever other conditions a 484 or whatever other conditions are appropriate.
485 485
486 CONFIG_DEBUG_OBJECTS_RCU_HEAD: 486 CONFIG_DEBUG_OBJECTS_RCU_HEAD:
487 check that you don't pass the 487 check that you don't pass the same object to call_rcu()
488 (or friends) before an RCU gra 488 (or friends) before an RCU grace period has elapsed
489 since the last time that you p 489 since the last time that you passed that same object to
490 call_rcu() (or friends). 490 call_rcu() (or friends).
491 491
492 CONFIG_RCU_STRICT_GRACE_PERIOD: 492 CONFIG_RCU_STRICT_GRACE_PERIOD:
493 combine with KASAN to check fo 493 combine with KASAN to check for pointers leaked out
494 of RCU read-side critical sect 494 of RCU read-side critical sections. This Kconfig
495 option is tough on both perfor 495 option is tough on both performance and scalability,
496 and so is limited to four-CPU 496 and so is limited to four-CPU systems.
497 497
498 __rcu sparse checks: 498 __rcu sparse checks:
499 tag the pointer to the RCU-pro 499 tag the pointer to the RCU-protected data structure
500 with __rcu, and sparse will wa 500 with __rcu, and sparse will warn you if you access that
501 pointer without the services o 501 pointer without the services of one of the variants
502 of rcu_dereference(). 502 of rcu_dereference().
503 503
504 These debugging aids can help you find 504 These debugging aids can help you find problems that are
505 otherwise extremely difficult to spot. 505 otherwise extremely difficult to spot.
506 506
507 17. If you pass a callback function define 507 17. If you pass a callback function defined within a module
508 to one of call_rcu(), call_srcu(), cal 508 to one of call_rcu(), call_srcu(), call_rcu_tasks(), or
509 call_rcu_tasks_trace(), then it is nec 509 call_rcu_tasks_trace(), then it is necessary to wait for all
510 pending callbacks to be invoked before 510 pending callbacks to be invoked before unloading that module.
511 Note that it is absolutely *not* suffi 511 Note that it is absolutely *not* sufficient to wait for a grace
512 period! For example, synchronize_rcu( 512 period! For example, synchronize_rcu() implementation is *not*
513 guaranteed to wait for callbacks regis 513 guaranteed to wait for callbacks registered on other CPUs via
514 call_rcu(). Or even on the current CP 514 call_rcu(). Or even on the current CPU if that CPU recently
515 went offline and came back online. 515 went offline and came back online.
516 516
517 You instead need to use one of the bar 517 You instead need to use one of the barrier functions:
518 518
519 - call_rcu() -> rcu_barrier() 519 - call_rcu() -> rcu_barrier()
520 - call_srcu() -> srcu_barrier() 520 - call_srcu() -> srcu_barrier()
521 - call_rcu_tasks() -> rcu_barrie 521 - call_rcu_tasks() -> rcu_barrier_tasks()
522 - call_rcu_tasks_trace() -> rcu_ 522 - call_rcu_tasks_trace() -> rcu_barrier_tasks_trace()
523 523
524 However, these barrier functions are a 524 However, these barrier functions are absolutely *not* guaranteed
525 to wait for a grace period. For examp 525 to wait for a grace period. For example, if there are no
526 call_rcu() callbacks queued anywhere i 526 call_rcu() callbacks queued anywhere in the system, rcu_barrier()
527 can and will return immediately. 527 can and will return immediately.
528 528
529 So if you need to wait for both a grac 529 So if you need to wait for both a grace period and for all
530 pre-existing callbacks, you will need 530 pre-existing callbacks, you will need to invoke both functions,
531 with the pair depending on the flavor 531 with the pair depending on the flavor of RCU:
532 532
533 - Either synchronize_rcu() or sy 533 - Either synchronize_rcu() or synchronize_rcu_expedited(),
534 together with rcu_barrier() 534 together with rcu_barrier()
535 - Either synchronize_srcu() or s 535 - Either synchronize_srcu() or synchronize_srcu_expedited(),
536 together with and srcu_barrier 536 together with and srcu_barrier()
537 - synchronize_rcu_tasks() and rc 537 - synchronize_rcu_tasks() and rcu_barrier_tasks()
538 - synchronize_tasks_trace() and 538 - synchronize_tasks_trace() and rcu_barrier_tasks_trace()
539 539
540 If necessary, you can use something li 540 If necessary, you can use something like workqueues to execute
541 the requisite pair of functions concur 541 the requisite pair of functions concurrently.
542 542
543 See rcubarrier.rst for more informatio 543 See rcubarrier.rst for more information.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.