1 .. _rcu_dereference_doc: 1 .. _rcu_dereference_doc:
2 2
3 PROPER CARE AND FEEDING OF RETURN VALUES FROM 3 PROPER CARE AND FEEDING OF RETURN VALUES FROM rcu_dereference()
4 ============================================== 4 ===============================================================
5 5
6 Proper care and feeding of address and data de 6 Proper care and feeding of address and data dependencies is critically
7 important to correct use of things like RCU. 7 important to correct use of things like RCU. To this end, the pointers
8 returned from the rcu_dereference() family of 8 returned from the rcu_dereference() family of primitives carry address and
9 data dependencies. These dependencies extend 9 data dependencies. These dependencies extend from the rcu_dereference()
10 macro's load of the pointer to the later use o 10 macro's load of the pointer to the later use of that pointer to compute
11 either the address of a later memory access (r 11 either the address of a later memory access (representing an address
12 dependency) or the value written by a later me 12 dependency) or the value written by a later memory access (representing
13 a data dependency). 13 a data dependency).
14 14
15 Most of the time, these dependencies are prese 15 Most of the time, these dependencies are preserved, permitting you to
16 freely use values from rcu_dereference(). For 16 freely use values from rcu_dereference(). For example, dereferencing
17 (prefix "*"), field selection ("->"), assignme 17 (prefix "*"), field selection ("->"), assignment ("="), address-of
18 ("&"), casts, and addition or subtraction of c 18 ("&"), casts, and addition or subtraction of constants all work quite
19 naturally and safely. However, because curren 19 naturally and safely. However, because current compilers do not take
20 either address or data dependencies into accou 20 either address or data dependencies into account it is still possible
21 to get into trouble. 21 to get into trouble.
22 22
23 Follow these rules to preserve the address and 23 Follow these rules to preserve the address and data dependencies emanating
24 from your calls to rcu_dereference() and frien 24 from your calls to rcu_dereference() and friends, thus keeping your RCU
25 readers working properly: 25 readers working properly:
26 26
27 - You must use one of the rcu_dereferenc 27 - You must use one of the rcu_dereference() family of primitives
28 to load an RCU-protected pointer, othe 28 to load an RCU-protected pointer, otherwise CONFIG_PROVE_RCU
29 will complain. Worse yet, your code c 29 will complain. Worse yet, your code can see random memory-corruption
30 bugs due to games that compilers and D 30 bugs due to games that compilers and DEC Alpha can play.
31 Without one of the rcu_dereference() p 31 Without one of the rcu_dereference() primitives, compilers
32 can reload the value, and won't your c 32 can reload the value, and won't your code have fun with two
33 different values for a single pointer! 33 different values for a single pointer! Without rcu_dereference(),
34 DEC Alpha can load a pointer, derefere 34 DEC Alpha can load a pointer, dereference that pointer, and
35 return data preceding initialization t 35 return data preceding initialization that preceded the store
36 of the pointer. (As noted later, in r 36 of the pointer. (As noted later, in recent kernels READ_ONCE()
37 also prevents DEC Alpha from playing t 37 also prevents DEC Alpha from playing these tricks.)
38 38
39 In addition, the volatile cast in rcu_ 39 In addition, the volatile cast in rcu_dereference() prevents the
40 compiler from deducing the resulting p 40 compiler from deducing the resulting pointer value. Please see
41 the section entitled "EXAMPLE WHERE TH 41 the section entitled "EXAMPLE WHERE THE COMPILER KNOWS TOO MUCH"
42 for an example where the compiler can 42 for an example where the compiler can in fact deduce the exact
43 value of the pointer, and thus cause m 43 value of the pointer, and thus cause misordering.
44 44
45 - In the special case where data is adde 45 - In the special case where data is added but is never removed
46 while readers are accessing the struct 46 while readers are accessing the structure, READ_ONCE() may be used
47 instead of rcu_dereference(). In this 47 instead of rcu_dereference(). In this case, use of READ_ONCE()
48 takes on the role of the lockless_dere 48 takes on the role of the lockless_dereference() primitive that
49 was removed in v4.15. 49 was removed in v4.15.
50 50
51 - You are only permitted to use rcu_dere 51 - You are only permitted to use rcu_dereference() on pointer values.
52 The compiler simply knows too much abo 52 The compiler simply knows too much about integral values to
53 trust it to carry dependencies through 53 trust it to carry dependencies through integer operations.
54 There are a very few exceptions, namel 54 There are a very few exceptions, namely that you can temporarily
55 cast the pointer to uintptr_t in order 55 cast the pointer to uintptr_t in order to:
56 56
57 - Set bits and clear bits down i 57 - Set bits and clear bits down in the must-be-zero low-order
58 bits of that pointer. This cl 58 bits of that pointer. This clearly means that the pointer
59 must have alignment constraint 59 must have alignment constraints, for example, this does
60 *not* work in general for char 60 *not* work in general for char* pointers.
61 61
62 - XOR bits to translate pointers 62 - XOR bits to translate pointers, as is done in some
63 classic buddy-allocator algori 63 classic buddy-allocator algorithms.
64 64
65 It is important to cast the value back 65 It is important to cast the value back to pointer before
66 doing much of anything else with it. 66 doing much of anything else with it.
67 67
68 - Avoid cancellation when using the "+" 68 - Avoid cancellation when using the "+" and "-" infix arithmetic
69 operators. For example, for a given v 69 operators. For example, for a given variable "x", avoid
70 "(x-(uintptr_t)x)" for char* pointers. 70 "(x-(uintptr_t)x)" for char* pointers. The compiler is within its
71 rights to substitute zero for this sor 71 rights to substitute zero for this sort of expression, so that
72 subsequent accesses no longer depend o 72 subsequent accesses no longer depend on the rcu_dereference(),
73 again possibly resulting in bugs due t 73 again possibly resulting in bugs due to misordering.
74 74
75 Of course, if "p" is a pointer from rc 75 Of course, if "p" is a pointer from rcu_dereference(), and "a"
76 and "b" are integers that happen to be 76 and "b" are integers that happen to be equal, the expression
77 "p+a-b" is safe because its value stil 77 "p+a-b" is safe because its value still necessarily depends on
78 the rcu_dereference(), thus maintainin 78 the rcu_dereference(), thus maintaining proper ordering.
79 79
80 - If you are using RCU to protect JITed 80 - If you are using RCU to protect JITed functions, so that the
81 "()" function-invocation operator is a 81 "()" function-invocation operator is applied to a value obtained
82 (directly or indirectly) from rcu_dere 82 (directly or indirectly) from rcu_dereference(), you may need to
83 interact directly with the hardware to 83 interact directly with the hardware to flush instruction caches.
84 This issue arises on some systems when 84 This issue arises on some systems when a newly JITed function is
85 using the same memory that was used by 85 using the same memory that was used by an earlier JITed function.
86 86
87 - Do not use the results from relational 87 - Do not use the results from relational operators ("==", "!=",
88 ">", ">=", "<", or "<=") when derefere 88 ">", ">=", "<", or "<=") when dereferencing. For example,
89 the following (quite strange) code is 89 the following (quite strange) code is buggy::
90 90
91 int *p; 91 int *p;
92 int *q; 92 int *q;
93 93
94 ... 94 ...
95 95
96 p = rcu_dereference(gp) 96 p = rcu_dereference(gp)
97 q = &global_q; 97 q = &global_q;
98 q += p > &oom_p; 98 q += p > &oom_p;
99 r1 = *q; /* BUGGY!!! */ 99 r1 = *q; /* BUGGY!!! */
100 100
101 As before, the reason this is buggy is 101 As before, the reason this is buggy is that relational operators
102 are often compiled using branches. An 102 are often compiled using branches. And as before, although
103 weak-memory machines such as ARM or Po 103 weak-memory machines such as ARM or PowerPC do order stores
104 after such branches, but can speculate 104 after such branches, but can speculate loads, which can again
105 result in misordering bugs. 105 result in misordering bugs.
106 106
107 - Be very careful about comparing pointe 107 - Be very careful about comparing pointers obtained from
108 rcu_dereference() against non-NULL val 108 rcu_dereference() against non-NULL values. As Linus Torvalds
109 explained, if the two pointers are equ 109 explained, if the two pointers are equal, the compiler could
110 substitute the pointer you are compari 110 substitute the pointer you are comparing against for the pointer
111 obtained from rcu_dereference(). For 111 obtained from rcu_dereference(). For example::
112 112
113 p = rcu_dereference(gp); 113 p = rcu_dereference(gp);
114 if (p == &default_struct) 114 if (p == &default_struct)
115 do_default(p->a); 115 do_default(p->a);
116 116
117 Because the compiler now knows that th 117 Because the compiler now knows that the value of "p" is exactly
118 the address of the variable "default_s 118 the address of the variable "default_struct", it is free to
119 transform this code into the following 119 transform this code into the following::
120 120
121 p = rcu_dereference(gp); 121 p = rcu_dereference(gp);
122 if (p == &default_struct) 122 if (p == &default_struct)
123 do_default(default_str 123 do_default(default_struct.a);
124 124
125 On ARM and Power hardware, the load fr 125 On ARM and Power hardware, the load from "default_struct.a"
126 can now be speculated, such that it mi 126 can now be speculated, such that it might happen before the
127 rcu_dereference(). This could result 127 rcu_dereference(). This could result in bugs due to misordering.
128 128
129 However, comparisons are OK in the fol 129 However, comparisons are OK in the following cases:
130 130
131 - The comparison was against the 131 - The comparison was against the NULL pointer. If the
132 compiler knows that the pointe 132 compiler knows that the pointer is NULL, you had better
133 not be dereferencing it anyway 133 not be dereferencing it anyway. If the comparison is
134 non-equal, the compiler is non 134 non-equal, the compiler is none the wiser. Therefore,
135 it is safe to compare pointers 135 it is safe to compare pointers from rcu_dereference()
136 against NULL pointers. 136 against NULL pointers.
137 137
138 - The pointer is never dereferen 138 - The pointer is never dereferenced after being compared.
139 Since there are no subsequent 139 Since there are no subsequent dereferences, the compiler
140 cannot use anything it learned 140 cannot use anything it learned from the comparison
141 to reorder the non-existent su 141 to reorder the non-existent subsequent dereferences.
142 This sort of comparison occurs 142 This sort of comparison occurs frequently when scanning
143 RCU-protected circular linked 143 RCU-protected circular linked lists.
144 144
145 Note that if the pointer compa 145 Note that if the pointer comparison is done outside
146 of an RCU read-side critical s 146 of an RCU read-side critical section, and the pointer
147 is never dereferenced, rcu_acc 147 is never dereferenced, rcu_access_pointer() should be
148 used in place of rcu_dereferen 148 used in place of rcu_dereference(). In most cases,
149 it is best to avoid accidental 149 it is best to avoid accidental dereferences by testing
150 the rcu_access_pointer() retur 150 the rcu_access_pointer() return value directly, without
151 assigning it to a variable. 151 assigning it to a variable.
152 152
153 Within an RCU read-side critic 153 Within an RCU read-side critical section, there is little
154 reason to use rcu_access_point 154 reason to use rcu_access_pointer().
155 155
156 - The comparison is against a po 156 - The comparison is against a pointer that references memory
157 that was initialized "a long t 157 that was initialized "a long time ago." The reason
158 this is safe is that even if m 158 this is safe is that even if misordering occurs, the
159 misordering will not affect th 159 misordering will not affect the accesses that follow
160 the comparison. So exactly ho 160 the comparison. So exactly how long ago is "a long
161 time ago"? Here are some poss 161 time ago"? Here are some possibilities:
162 162
163 - Compile time. 163 - Compile time.
164 164
165 - Boot time. 165 - Boot time.
166 166
167 - Module-init time for m 167 - Module-init time for module code.
168 168
169 - Prior to kthread creat 169 - Prior to kthread creation for kthread code.
170 170
171 - During some prior acqu 171 - During some prior acquisition of the lock that
172 we now hold. 172 we now hold.
173 173
174 - Before mod_timer() tim 174 - Before mod_timer() time for a timer handler.
175 175
176 There are many other possibili 176 There are many other possibilities involving the Linux
177 kernel's wide array of primiti 177 kernel's wide array of primitives that cause code to
178 be invoked at a later time. 178 be invoked at a later time.
179 179
180 - The pointer being compared aga 180 - The pointer being compared against also came from
181 rcu_dereference(). In this ca 181 rcu_dereference(). In this case, both pointers depend
182 on one rcu_dereference() or an 182 on one rcu_dereference() or another, so you get proper
183 ordering either way. 183 ordering either way.
184 184
185 That said, this situation can 185 That said, this situation can make certain RCU usage
186 bugs more likely to happen. W 186 bugs more likely to happen. Which can be a good thing,
187 at least if they happen during 187 at least if they happen during testing. An example
188 of such an RCU usage bug is sh 188 of such an RCU usage bug is shown in the section titled
189 "EXAMPLE OF AMPLIFIED RCU-USAG 189 "EXAMPLE OF AMPLIFIED RCU-USAGE BUG".
190 190
191 - All of the accesses following 191 - All of the accesses following the comparison are stores,
192 so that a control dependency p 192 so that a control dependency preserves the needed ordering.
193 That said, it is easy to get c 193 That said, it is easy to get control dependencies wrong.
194 Please see the "CONTROL DEPEND 194 Please see the "CONTROL DEPENDENCIES" section of
195 Documentation/memory-barriers. 195 Documentation/memory-barriers.txt for more details.
196 196
197 - The pointers are not equal *an 197 - The pointers are not equal *and* the compiler does
198 not have enough information to 198 not have enough information to deduce the value of the
199 pointer. Note that the volati 199 pointer. Note that the volatile cast in rcu_dereference()
200 will normally prevent the comp 200 will normally prevent the compiler from knowing too much.
201 201
202 However, please note that if t 202 However, please note that if the compiler knows that the
203 pointer takes on only one of t 203 pointer takes on only one of two values, a not-equal
204 comparison will provide exactl 204 comparison will provide exactly the information that the
205 compiler needs to deduce the v 205 compiler needs to deduce the value of the pointer.
206 206
207 - Disable any value-speculation optimiza 207 - Disable any value-speculation optimizations that your compiler
208 might provide, especially if you are m 208 might provide, especially if you are making use of feedback-based
209 optimizations that take data collected 209 optimizations that take data collected from prior runs. Such
210 value-speculation optimizations reorde 210 value-speculation optimizations reorder operations by design.
211 211
212 There is one exception to this rule: 212 There is one exception to this rule: Value-speculation
213 optimizations that leverage the branch 213 optimizations that leverage the branch-prediction hardware are
214 safe on strongly ordered systems (such 214 safe on strongly ordered systems (such as x86), but not on weakly
215 ordered systems (such as ARM or Power) 215 ordered systems (such as ARM or Power). Choose your compiler
216 command-line options wisely! 216 command-line options wisely!
217 217
218 218
219 EXAMPLE OF AMPLIFIED RCU-USAGE BUG 219 EXAMPLE OF AMPLIFIED RCU-USAGE BUG
220 ---------------------------------- 220 ----------------------------------
221 221
222 Because updaters can run concurrently with RCU 222 Because updaters can run concurrently with RCU readers, RCU readers can
223 see stale and/or inconsistent values. If RCU 223 see stale and/or inconsistent values. If RCU readers need fresh or
224 consistent values, which they sometimes do, th 224 consistent values, which they sometimes do, they need to take proper
225 precautions. To see this, consider the follow 225 precautions. To see this, consider the following code fragment::
226 226
227 struct foo { 227 struct foo {
228 int a; 228 int a;
229 int b; 229 int b;
230 int c; 230 int c;
231 }; 231 };
232 struct foo *gp1; 232 struct foo *gp1;
233 struct foo *gp2; 233 struct foo *gp2;
234 234
235 void updater(void) 235 void updater(void)
236 { 236 {
237 struct foo *p; 237 struct foo *p;
238 238
239 p = kmalloc(...); 239 p = kmalloc(...);
240 if (p == NULL) 240 if (p == NULL)
241 deal_with_it(); 241 deal_with_it();
242 p->a = 42; /* Each field in i 242 p->a = 42; /* Each field in its own cache line. */
243 p->b = 43; 243 p->b = 43;
244 p->c = 44; 244 p->c = 44;
245 rcu_assign_pointer(gp1, p); 245 rcu_assign_pointer(gp1, p);
246 p->b = 143; 246 p->b = 143;
247 p->c = 144; 247 p->c = 144;
248 rcu_assign_pointer(gp2, p); 248 rcu_assign_pointer(gp2, p);
249 } 249 }
250 250
251 void reader(void) 251 void reader(void)
252 { 252 {
253 struct foo *p; 253 struct foo *p;
254 struct foo *q; 254 struct foo *q;
255 int r1, r2; 255 int r1, r2;
256 256
257 rcu_read_lock(); 257 rcu_read_lock();
258 p = rcu_dereference(gp2); 258 p = rcu_dereference(gp2);
259 if (p == NULL) 259 if (p == NULL)
260 return; 260 return;
261 r1 = p->b; /* Guaranteed to g 261 r1 = p->b; /* Guaranteed to get 143. */
262 q = rcu_dereference(gp1); /* 262 q = rcu_dereference(gp1); /* Guaranteed non-NULL. */
263 if (p == q) { 263 if (p == q) {
264 /* The compiler decide 264 /* The compiler decides that q->c is same as p->c. */
265 r2 = p->c; /* Could ge 265 r2 = p->c; /* Could get 44 on weakly order system. */
266 } else { 266 } else {
267 r2 = p->c - r1; /* Unc 267 r2 = p->c - r1; /* Unconditional access to p->c. */
268 } 268 }
269 rcu_read_unlock(); 269 rcu_read_unlock();
270 do_something_with(r1, r2); 270 do_something_with(r1, r2);
271 } 271 }
272 272
273 You might be surprised that the outcome (r1 == 273 You might be surprised that the outcome (r1 == 143 && r2 == 44) is possible,
274 but you should not be. After all, the updater 274 but you should not be. After all, the updater might have been invoked
275 a second time between the time reader() loaded 275 a second time between the time reader() loaded into "r1" and the time
276 that it loaded into "r2". The fact that this 276 that it loaded into "r2". The fact that this same result can occur due
277 to some reordering from the compiler and CPUs 277 to some reordering from the compiler and CPUs is beside the point.
278 278
279 But suppose that the reader needs a consistent 279 But suppose that the reader needs a consistent view?
280 280
281 Then one approach is to use locking, for examp 281 Then one approach is to use locking, for example, as follows::
282 282
283 struct foo { 283 struct foo {
284 int a; 284 int a;
285 int b; 285 int b;
286 int c; 286 int c;
287 spinlock_t lock; 287 spinlock_t lock;
288 }; 288 };
289 struct foo *gp1; 289 struct foo *gp1;
290 struct foo *gp2; 290 struct foo *gp2;
291 291
292 void updater(void) 292 void updater(void)
293 { 293 {
294 struct foo *p; 294 struct foo *p;
295 295
296 p = kmalloc(...); 296 p = kmalloc(...);
297 if (p == NULL) 297 if (p == NULL)
298 deal_with_it(); 298 deal_with_it();
299 spin_lock(&p->lock); 299 spin_lock(&p->lock);
300 p->a = 42; /* Each field in i 300 p->a = 42; /* Each field in its own cache line. */
301 p->b = 43; 301 p->b = 43;
302 p->c = 44; 302 p->c = 44;
303 spin_unlock(&p->lock); 303 spin_unlock(&p->lock);
304 rcu_assign_pointer(gp1, p); 304 rcu_assign_pointer(gp1, p);
305 spin_lock(&p->lock); 305 spin_lock(&p->lock);
306 p->b = 143; 306 p->b = 143;
307 p->c = 144; 307 p->c = 144;
308 spin_unlock(&p->lock); 308 spin_unlock(&p->lock);
309 rcu_assign_pointer(gp2, p); 309 rcu_assign_pointer(gp2, p);
310 } 310 }
311 311
312 void reader(void) 312 void reader(void)
313 { 313 {
314 struct foo *p; 314 struct foo *p;
315 struct foo *q; 315 struct foo *q;
316 int r1, r2; 316 int r1, r2;
317 317
318 rcu_read_lock(); 318 rcu_read_lock();
319 p = rcu_dereference(gp2); 319 p = rcu_dereference(gp2);
320 if (p == NULL) 320 if (p == NULL)
321 return; 321 return;
322 spin_lock(&p->lock); 322 spin_lock(&p->lock);
323 r1 = p->b; /* Guaranteed to g 323 r1 = p->b; /* Guaranteed to get 143. */
324 q = rcu_dereference(gp1); /* 324 q = rcu_dereference(gp1); /* Guaranteed non-NULL. */
325 if (p == q) { 325 if (p == q) {
326 /* The compiler decide 326 /* The compiler decides that q->c is same as p->c. */
327 r2 = p->c; /* Locking 327 r2 = p->c; /* Locking guarantees r2 == 144. */
328 } else { 328 } else {
329 spin_lock(&q->lock); 329 spin_lock(&q->lock);
330 r2 = q->c - r1; 330 r2 = q->c - r1;
331 spin_unlock(&q->lock); 331 spin_unlock(&q->lock);
332 } 332 }
333 rcu_read_unlock(); 333 rcu_read_unlock();
334 spin_unlock(&p->lock); 334 spin_unlock(&p->lock);
335 do_something_with(r1, r2); 335 do_something_with(r1, r2);
336 } 336 }
337 337
338 As always, use the right tool for the job! 338 As always, use the right tool for the job!
339 339
340 340
341 EXAMPLE WHERE THE COMPILER KNOWS TOO MUCH 341 EXAMPLE WHERE THE COMPILER KNOWS TOO MUCH
342 ----------------------------------------- 342 -----------------------------------------
343 343
344 If a pointer obtained from rcu_dereference() c 344 If a pointer obtained from rcu_dereference() compares not-equal to some
345 other pointer, the compiler normally has no cl 345 other pointer, the compiler normally has no clue what the value of the
346 first pointer might be. This lack of knowledg 346 first pointer might be. This lack of knowledge prevents the compiler
347 from carrying out optimizations that otherwise 347 from carrying out optimizations that otherwise might destroy the ordering
348 guarantees that RCU depends on. And the volat 348 guarantees that RCU depends on. And the volatile cast in rcu_dereference()
349 should prevent the compiler from guessing the 349 should prevent the compiler from guessing the value.
350 350
351 But without rcu_dereference(), the compiler kn 351 But without rcu_dereference(), the compiler knows more than you might
352 expect. Consider the following code fragment: 352 expect. Consider the following code fragment::
353 353
354 struct foo { 354 struct foo {
355 int a; 355 int a;
356 int b; 356 int b;
357 }; 357 };
358 static struct foo variable1; 358 static struct foo variable1;
359 static struct foo variable2; 359 static struct foo variable2;
360 static struct foo *gp = &variable1; 360 static struct foo *gp = &variable1;
361 361
362 void updater(void) 362 void updater(void)
363 { 363 {
364 initialize_foo(&variable2); 364 initialize_foo(&variable2);
365 rcu_assign_pointer(gp, &variab 365 rcu_assign_pointer(gp, &variable2);
366 /* 366 /*
367 * The above is the only store 367 * The above is the only store to gp in this translation unit,
368 * and the address of gp is no 368 * and the address of gp is not exported in any way.
369 */ 369 */
370 } 370 }
371 371
372 int reader(void) 372 int reader(void)
373 { 373 {
374 struct foo *p; 374 struct foo *p;
375 375
376 p = gp; 376 p = gp;
377 barrier(); 377 barrier();
378 if (p == &variable1) 378 if (p == &variable1)
379 return p->a; /* Must b 379 return p->a; /* Must be variable1.a. */
380 else 380 else
381 return p->b; /* Must b 381 return p->b; /* Must be variable2.b. */
382 } 382 }
383 383
384 Because the compiler can see all stores to "gp 384 Because the compiler can see all stores to "gp", it knows that the only
385 possible values of "gp" are "variable1" on the 385 possible values of "gp" are "variable1" on the one hand and "variable2"
386 on the other. The comparison in reader() ther 386 on the other. The comparison in reader() therefore tells the compiler
387 the exact value of "p" even in the not-equals 387 the exact value of "p" even in the not-equals case. This allows the
388 compiler to make the return values independent 388 compiler to make the return values independent of the load from "gp",
389 in turn destroying the ordering between this l 389 in turn destroying the ordering between this load and the loads of the
390 return values. This can result in "p->b" retu 390 return values. This can result in "p->b" returning pre-initialization
391 garbage values on weakly ordered systems. 391 garbage values on weakly ordered systems.
392 392
393 In short, rcu_dereference() is *not* optional 393 In short, rcu_dereference() is *not* optional when you are going to
394 dereference the resulting pointer. 394 dereference the resulting pointer.
395 395
396 396
397 WHICH MEMBER OF THE rcu_dereference() FAMILY S 397 WHICH MEMBER OF THE rcu_dereference() FAMILY SHOULD YOU USE?
398 ---------------------------------------------- 398 ------------------------------------------------------------
399 399
400 First, please avoid using rcu_dereference_raw( 400 First, please avoid using rcu_dereference_raw() and also please avoid
401 using rcu_dereference_check() and rcu_derefere 401 using rcu_dereference_check() and rcu_dereference_protected() with a
402 second argument with a constant value of 1 (or 402 second argument with a constant value of 1 (or true, for that matter).
403 With that caution out of the way, here is some 403 With that caution out of the way, here is some guidance for which
404 member of the rcu_dereference() to use in vari 404 member of the rcu_dereference() to use in various situations:
405 405
406 1. If the access needs to be within an RC 406 1. If the access needs to be within an RCU read-side critical
407 section, use rcu_dereference(). With 407 section, use rcu_dereference(). With the new consolidated
408 RCU flavors, an RCU read-side critical 408 RCU flavors, an RCU read-side critical section is entered
409 using rcu_read_lock(), anything that d 409 using rcu_read_lock(), anything that disables bottom halves,
410 anything that disables interrupts, or 410 anything that disables interrupts, or anything that disables
411 preemption. Please note that spinlock 411 preemption. Please note that spinlock critical sections
412 are also implied RCU read-side critica 412 are also implied RCU read-side critical sections, even when
413 they are preemptible, as they are in k 413 they are preemptible, as they are in kernels built with
414 CONFIG_PREEMPT_RT=y. 414 CONFIG_PREEMPT_RT=y.
415 415
416 2. If the access might be within an RCU r 416 2. If the access might be within an RCU read-side critical section
417 on the one hand, or protected by (say) 417 on the one hand, or protected by (say) my_lock on the other,
418 use rcu_dereference_check(), for examp 418 use rcu_dereference_check(), for example::
419 419
420 p1 = rcu_dereference_check(p-> 420 p1 = rcu_dereference_check(p->rcu_protected_pointer,
421 loc 421 lockdep_is_held(&my_lock));
422 422
423 423
424 3. If the access might be within an RCU r 424 3. If the access might be within an RCU read-side critical section
425 on the one hand, or protected by eithe 425 on the one hand, or protected by either my_lock or your_lock on
426 the other, again use rcu_dereference_c 426 the other, again use rcu_dereference_check(), for example::
427 427
428 p1 = rcu_dereference_check(p-> 428 p1 = rcu_dereference_check(p->rcu_protected_pointer,
429 loc 429 lockdep_is_held(&my_lock) ||
430 loc 430 lockdep_is_held(&your_lock));
431 431
432 4. If the access is on the update side, s 432 4. If the access is on the update side, so that it is always protected
433 by my_lock, use rcu_dereference_protec 433 by my_lock, use rcu_dereference_protected()::
434 434
435 p1 = rcu_dereference_protected 435 p1 = rcu_dereference_protected(p->rcu_protected_pointer,
436 436 lockdep_is_held(&my_lock));
437 437
438 This can be extended to handle multipl 438 This can be extended to handle multiple locks as in #3 above,
439 and both can be extended to check othe 439 and both can be extended to check other conditions as well.
440 440
441 5. If the protection is supplied by the c 441 5. If the protection is supplied by the caller, and is thus unknown
442 to this code, that is the rare case wh 442 to this code, that is the rare case when rcu_dereference_raw()
443 is appropriate. In addition, rcu_dere 443 is appropriate. In addition, rcu_dereference_raw() might be
444 appropriate when the lockdep expressio 444 appropriate when the lockdep expression would be excessively
445 complex, except that a better approach 445 complex, except that a better approach in that case might be to
446 take a long hard look at your synchron 446 take a long hard look at your synchronization design. Still,
447 there are data-locking cases where any 447 there are data-locking cases where any one of a very large number
448 of locks or reference counters suffice 448 of locks or reference counters suffices to protect the pointer,
449 so rcu_dereference_raw() does have its 449 so rcu_dereference_raw() does have its place.
450 450
451 However, its place is probably quite a 451 However, its place is probably quite a bit smaller than one
452 might expect given the number of uses 452 might expect given the number of uses in the current kernel.
453 Ditto for its synonym, rcu_dereference 453 Ditto for its synonym, rcu_dereference_check( ... , 1), and
454 its close relative, rcu_dereference_pr 454 its close relative, rcu_dereference_protected(... , 1).
455 455
456 456
457 SPARSE CHECKING OF RCU-PROTECTED POINTERS 457 SPARSE CHECKING OF RCU-PROTECTED POINTERS
458 ----------------------------------------- 458 -----------------------------------------
459 459
460 The sparse static-analysis tool checks for non 460 The sparse static-analysis tool checks for non-RCU access to RCU-protected
461 pointers, which can result in "interesting" bu 461 pointers, which can result in "interesting" bugs due to compiler
462 optimizations involving invented loads and per 462 optimizations involving invented loads and perhaps also load tearing.
463 For example, suppose someone mistakenly does s 463 For example, suppose someone mistakenly does something like this::
464 464
465 p = q->rcu_protected_pointer; 465 p = q->rcu_protected_pointer;
466 do_something_with(p->a); 466 do_something_with(p->a);
467 do_something_else_with(p->b); 467 do_something_else_with(p->b);
468 468
469 If register pressure is high, the compiler mig 469 If register pressure is high, the compiler might optimize "p" out
470 of existence, transforming the code to somethi 470 of existence, transforming the code to something like this::
471 471
472 do_something_with(q->rcu_protected_poi 472 do_something_with(q->rcu_protected_pointer->a);
473 do_something_else_with(q->rcu_protecte 473 do_something_else_with(q->rcu_protected_pointer->b);
474 474
475 This could fatally disappoint your code if q-> 475 This could fatally disappoint your code if q->rcu_protected_pointer
476 changed in the meantime. Nor is this a theore 476 changed in the meantime. Nor is this a theoretical problem: Exactly
477 this sort of bug cost Paul E. McKenney (and se 477 this sort of bug cost Paul E. McKenney (and several of his innocent
478 colleagues) a three-day weekend back in the ea 478 colleagues) a three-day weekend back in the early 1990s.
479 479
480 Load tearing could of course result in derefer 480 Load tearing could of course result in dereferencing a mashup of a pair
481 of pointers, which also might fatally disappoi 481 of pointers, which also might fatally disappoint your code.
482 482
483 These problems could have been avoided simply 483 These problems could have been avoided simply by making the code instead
484 read as follows:: 484 read as follows::
485 485
486 p = rcu_dereference(q->rcu_protected_p 486 p = rcu_dereference(q->rcu_protected_pointer);
487 do_something_with(p->a); 487 do_something_with(p->a);
488 do_something_else_with(p->b); 488 do_something_else_with(p->b);
489 489
490 Unfortunately, these sorts of bugs can be extr 490 Unfortunately, these sorts of bugs can be extremely hard to spot during
491 review. This is where the sparse tool comes i 491 review. This is where the sparse tool comes into play, along with the
492 "__rcu" marker. If you mark a pointer declara 492 "__rcu" marker. If you mark a pointer declaration, whether in a structure
493 or as a formal parameter, with "__rcu", which 493 or as a formal parameter, with "__rcu", which tells sparse to complain if
494 this pointer is accessed directly. It will al 494 this pointer is accessed directly. It will also cause sparse to complain
495 if a pointer not marked with "__rcu" is access 495 if a pointer not marked with "__rcu" is accessed using rcu_dereference()
496 and friends. For example, ->rcu_protected_poi 496 and friends. For example, ->rcu_protected_pointer might be declared as
497 follows:: 497 follows::
498 498
499 struct foo __rcu *rcu_protected_pointe 499 struct foo __rcu *rcu_protected_pointer;
500 500
501 Use of "__rcu" is opt-in. If you choose not t 501 Use of "__rcu" is opt-in. If you choose not to use it, then you should
502 ignore the sparse warnings. 502 ignore the sparse warnings.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.