1 .. _rcu_dereference_doc:
3 PROPER CARE AND FEEDING OF RETURN VALUES FROM rcu_dereference()
4 ===============================================================
6 Most of the time, you can use values from rcu_dereference() or one of
7 the similar primitives without worries. Dereferencing (prefix "*"),
8 field selection ("->"), assignment ("="), address-of ("&"), addition and
9 subtraction of constants, and casts all work quite naturally and safely.
11 It is nevertheless possible to get into trouble with other operations.
12 Follow these rules to keep your RCU code working properly:
14 - You must use one of the rcu_dereference() family of primitives
15 to load an RCU-protected pointer, otherwise CONFIG_PROVE_RCU
16 will complain. Worse yet, your code can see random memory-corruption
17 bugs due to games that compilers and DEC Alpha can play.
18 Without one of the rcu_dereference() primitives, compilers
19 can reload the value, and won't your code have fun with two
20 different values for a single pointer! Without rcu_dereference(),
21 DEC Alpha can load a pointer, dereference that pointer, and
22 return data preceding initialization that preceded the store of
25 In addition, the volatile cast in rcu_dereference() prevents the
26 compiler from deducing the resulting pointer value. Please see
27 the section entitled "EXAMPLE WHERE THE COMPILER KNOWS TOO MUCH"
28 for an example where the compiler can in fact deduce the exact
29 value of the pointer, and thus cause misordering.
31 - In the special case where data is added but is never removed
32 while readers are accessing the structure, READ_ONCE() may be used
33 instead of rcu_dereference(). In this case, use of READ_ONCE()
34 takes on the role of the lockless_dereference() primitive that
37 - You are only permitted to use rcu_dereference on pointer values.
38 The compiler simply knows too much about integral values to
39 trust it to carry dependencies through integer operations.
40 There are a very few exceptions, namely that you can temporarily
41 cast the pointer to uintptr_t in order to:
43 - Set bits and clear bits down in the must-be-zero low-order
44 bits of that pointer. This clearly means that the pointer
45 must have alignment constraints, for example, this does
46 *not* work in general for char* pointers.
48 - XOR bits to translate pointers, as is done in some
49 classic buddy-allocator algorithms.
51 It is important to cast the value back to pointer before
52 doing much of anything else with it.
54 - Avoid cancellation when using the "+" and "-" infix arithmetic
55 operators. For example, for a given variable "x", avoid
56 "(x-(uintptr_t)x)" for char* pointers. The compiler is within its
57 rights to substitute zero for this sort of expression, so that
58 subsequent accesses no longer depend on the rcu_dereference(),
59 again possibly resulting in bugs due to misordering.
61 Of course, if "p" is a pointer from rcu_dereference(), and "a"
62 and "b" are integers that happen to be equal, the expression
63 "p+a-b" is safe because its value still necessarily depends on
64 the rcu_dereference(), thus maintaining proper ordering.
66 - If you are using RCU to protect JITed functions, so that the
67 "()" function-invocation operator is applied to a value obtained
68 (directly or indirectly) from rcu_dereference(), you may need to
69 interact directly with the hardware to flush instruction caches.
70 This issue arises on some systems when a newly JITed function is
71 using the same memory that was used by an earlier JITed function.
73 - Do not use the results from relational operators ("==", "!=",
74 ">", ">=", "<", or "<=") when dereferencing. For example,
75 the following (quite strange) code is buggy::
82 p = rcu_dereference(gp)
85 r1 = *q; /* BUGGY!!! */
87 As before, the reason this is buggy is that relational operators
88 are often compiled using branches. And as before, although
89 weak-memory machines such as ARM or PowerPC do order stores
90 after such branches, but can speculate loads, which can again
91 result in misordering bugs.
93 - Be very careful about comparing pointers obtained from
94 rcu_dereference() against non-NULL values. As Linus Torvalds
95 explained, if the two pointers are equal, the compiler could
96 substitute the pointer you are comparing against for the pointer
97 obtained from rcu_dereference(). For example::
99 p = rcu_dereference(gp);
100 if (p == &default_struct)
103 Because the compiler now knows that the value of "p" is exactly
104 the address of the variable "default_struct", it is free to
105 transform this code into the following::
107 p = rcu_dereference(gp);
108 if (p == &default_struct)
109 do_default(default_struct.a);
111 On ARM and Power hardware, the load from "default_struct.a"
112 can now be speculated, such that it might happen before the
113 rcu_dereference(). This could result in bugs due to misordering.
115 However, comparisons are OK in the following cases:
117 - The comparison was against the NULL pointer. If the
118 compiler knows that the pointer is NULL, you had better
119 not be dereferencing it anyway. If the comparison is
120 non-equal, the compiler is none the wiser. Therefore,
121 it is safe to compare pointers from rcu_dereference()
122 against NULL pointers.
124 - The pointer is never dereferenced after being compared.
125 Since there are no subsequent dereferences, the compiler
126 cannot use anything it learned from the comparison
127 to reorder the non-existent subsequent dereferences.
128 This sort of comparison occurs frequently when scanning
129 RCU-protected circular linked lists.
131 Note that if the pointer comparison is done outside
132 of an RCU read-side critical section, and the pointer
133 is never dereferenced, rcu_access_pointer() should be
134 used in place of rcu_dereference(). In most cases,
135 it is best to avoid accidental dereferences by testing
136 the rcu_access_pointer() return value directly, without
137 assigning it to a variable.
139 Within an RCU read-side critical section, there is little
140 reason to use rcu_access_pointer().
142 - The comparison is against a pointer that references memory
143 that was initialized "a long time ago." The reason
144 this is safe is that even if misordering occurs, the
145 misordering will not affect the accesses that follow
146 the comparison. So exactly how long ago is "a long
147 time ago"? Here are some possibilities:
153 - Module-init time for module code.
155 - Prior to kthread creation for kthread code.
157 - During some prior acquisition of the lock that
160 - Before mod_timer() time for a timer handler.
162 There are many other possibilities involving the Linux
163 kernel's wide array of primitives that cause code to
164 be invoked at a later time.
166 - The pointer being compared against also came from
167 rcu_dereference(). In this case, both pointers depend
168 on one rcu_dereference() or another, so you get proper
171 That said, this situation can make certain RCU usage
172 bugs more likely to happen. Which can be a good thing,
173 at least if they happen during testing. An example
174 of such an RCU usage bug is shown in the section titled
175 "EXAMPLE OF AMPLIFIED RCU-USAGE BUG".
177 - All of the accesses following the comparison are stores,
178 so that a control dependency preserves the needed ordering.
179 That said, it is easy to get control dependencies wrong.
180 Please see the "CONTROL DEPENDENCIES" section of
181 Documentation/memory-barriers.txt for more details.
183 - The pointers are not equal *and* the compiler does
184 not have enough information to deduce the value of the
185 pointer. Note that the volatile cast in rcu_dereference()
186 will normally prevent the compiler from knowing too much.
188 However, please note that if the compiler knows that the
189 pointer takes on only one of two values, a not-equal
190 comparison will provide exactly the information that the
191 compiler needs to deduce the value of the pointer.
193 - Disable any value-speculation optimizations that your compiler
194 might provide, especially if you are making use of feedback-based
195 optimizations that take data collected from prior runs. Such
196 value-speculation optimizations reorder operations by design.
198 There is one exception to this rule: Value-speculation
199 optimizations that leverage the branch-prediction hardware are
200 safe on strongly ordered systems (such as x86), but not on weakly
201 ordered systems (such as ARM or Power). Choose your compiler
202 command-line options wisely!
205 EXAMPLE OF AMPLIFIED RCU-USAGE BUG
206 ----------------------------------
208 Because updaters can run concurrently with RCU readers, RCU readers can
209 see stale and/or inconsistent values. If RCU readers need fresh or
210 consistent values, which they sometimes do, they need to take proper
211 precautions. To see this, consider the following code fragment::
228 p->a = 42; /* Each field in its own cache line. */
231 rcu_assign_pointer(gp1, p);
234 rcu_assign_pointer(gp2, p);
243 p = rcu_dereference(gp2);
246 r1 = p->b; /* Guaranteed to get 143. */
247 q = rcu_dereference(gp1); /* Guaranteed non-NULL. */
249 /* The compiler decides that q->c is same as p->c. */
250 r2 = p->c; /* Could get 44 on weakly order system. */
252 do_something_with(r1, r2);
255 You might be surprised that the outcome (r1 == 143 && r2 == 44) is possible,
256 but you should not be. After all, the updater might have been invoked
257 a second time between the time reader() loaded into "r1" and the time
258 that it loaded into "r2". The fact that this same result can occur due
259 to some reordering from the compiler and CPUs is beside the point.
261 But suppose that the reader needs a consistent view?
263 Then one approach is to use locking, for example, as follows::
282 p->a = 42; /* Each field in its own cache line. */
285 spin_unlock(&p->lock);
286 rcu_assign_pointer(gp1, p);
290 spin_unlock(&p->lock);
291 rcu_assign_pointer(gp2, p);
300 p = rcu_dereference(gp2);
304 r1 = p->b; /* Guaranteed to get 143. */
305 q = rcu_dereference(gp1); /* Guaranteed non-NULL. */
307 /* The compiler decides that q->c is same as p->c. */
308 r2 = p->c; /* Locking guarantees r2 == 144. */
310 spin_unlock(&p->lock);
311 do_something_with(r1, r2);
314 As always, use the right tool for the job!
317 EXAMPLE WHERE THE COMPILER KNOWS TOO MUCH
318 -----------------------------------------
320 If a pointer obtained from rcu_dereference() compares not-equal to some
321 other pointer, the compiler normally has no clue what the value of the
322 first pointer might be. This lack of knowledge prevents the compiler
323 from carrying out optimizations that otherwise might destroy the ordering
324 guarantees that RCU depends on. And the volatile cast in rcu_dereference()
325 should prevent the compiler from guessing the value.
327 But without rcu_dereference(), the compiler knows more than you might
328 expect. Consider the following code fragment::
334 static struct foo variable1;
335 static struct foo variable2;
336 static struct foo *gp = &variable1;
340 initialize_foo(&variable2);
341 rcu_assign_pointer(gp, &variable2);
343 * The above is the only store to gp in this translation unit,
344 * and the address of gp is not exported in any way.
355 return p->a; /* Must be variable1.a. */
357 return p->b; /* Must be variable2.b. */
360 Because the compiler can see all stores to "gp", it knows that the only
361 possible values of "gp" are "variable1" on the one hand and "variable2"
362 on the other. The comparison in reader() therefore tells the compiler
363 the exact value of "p" even in the not-equals case. This allows the
364 compiler to make the return values independent of the load from "gp",
365 in turn destroying the ordering between this load and the loads of the
366 return values. This can result in "p->b" returning pre-initialization
369 In short, rcu_dereference() is *not* optional when you are going to
370 dereference the resulting pointer.
373 WHICH MEMBER OF THE rcu_dereference() FAMILY SHOULD YOU USE?
374 ------------------------------------------------------------
376 First, please avoid using rcu_dereference_raw() and also please avoid
377 using rcu_dereference_check() and rcu_dereference_protected() with a
378 second argument with a constant value of 1 (or true, for that matter).
379 With that caution out of the way, here is some guidance for which
380 member of the rcu_dereference() to use in various situations:
382 1. If the access needs to be within an RCU read-side critical
383 section, use rcu_dereference(). With the new consolidated
384 RCU flavors, an RCU read-side critical section is entered
385 using rcu_read_lock(), anything that disables bottom halves,
386 anything that disables interrupts, or anything that disables
389 2. If the access might be within an RCU read-side critical section
390 on the one hand, or protected by (say) my_lock on the other,
391 use rcu_dereference_check(), for example::
393 p1 = rcu_dereference_check(p->rcu_protected_pointer,
394 lockdep_is_held(&my_lock));
397 3. If the access might be within an RCU read-side critical section
398 on the one hand, or protected by either my_lock or your_lock on
399 the other, again use rcu_dereference_check(), for example::
401 p1 = rcu_dereference_check(p->rcu_protected_pointer,
402 lockdep_is_held(&my_lock) ||
403 lockdep_is_held(&your_lock));
405 4. If the access is on the update side, so that it is always protected
406 by my_lock, use rcu_dereference_protected()::
408 p1 = rcu_dereference_protected(p->rcu_protected_pointer,
409 lockdep_is_held(&my_lock));
411 This can be extended to handle multiple locks as in #3 above,
412 and both can be extended to check other conditions as well.
414 5. If the protection is supplied by the caller, and is thus unknown
415 to this code, that is the rare case when rcu_dereference_raw()
416 is appropriate. In addition, rcu_dereference_raw() might be
417 appropriate when the lockdep expression would be excessively
418 complex, except that a better approach in that case might be to
419 take a long hard look at your synchronization design. Still,
420 there are data-locking cases where any one of a very large number
421 of locks or reference counters suffices to protect the pointer,
422 so rcu_dereference_raw() does have its place.
424 However, its place is probably quite a bit smaller than one
425 might expect given the number of uses in the current kernel.
426 Ditto for its synonym, rcu_dereference_check( ... , 1), and
427 its close relative, rcu_dereference_protected(... , 1).
430 SPARSE CHECKING OF RCU-PROTECTED POINTERS
431 -----------------------------------------
433 The sparse static-analysis tool checks for direct access to RCU-protected
434 pointers, which can result in "interesting" bugs due to compiler
435 optimizations involving invented loads and perhaps also load tearing.
436 For example, suppose someone mistakenly does something like this::
438 p = q->rcu_protected_pointer;
439 do_something_with(p->a);
440 do_something_else_with(p->b);
442 If register pressure is high, the compiler might optimize "p" out
443 of existence, transforming the code to something like this::
445 do_something_with(q->rcu_protected_pointer->a);
446 do_something_else_with(q->rcu_protected_pointer->b);
448 This could fatally disappoint your code if q->rcu_protected_pointer
449 changed in the meantime. Nor is this a theoretical problem: Exactly
450 this sort of bug cost Paul E. McKenney (and several of his innocent
451 colleagues) a three-day weekend back in the early 1990s.
453 Load tearing could of course result in dereferencing a mashup of a pair
454 of pointers, which also might fatally disappoint your code.
456 These problems could have been avoided simply by making the code instead
459 p = rcu_dereference(q->rcu_protected_pointer);
460 do_something_with(p->a);
461 do_something_else_with(p->b);
463 Unfortunately, these sorts of bugs can be extremely hard to spot during
464 review. This is where the sparse tool comes into play, along with the
465 "__rcu" marker. If you mark a pointer declaration, whether in a structure
466 or as a formal parameter, with "__rcu", which tells sparse to complain if
467 this pointer is accessed directly. It will also cause sparse to complain
468 if a pointer not marked with "__rcu" is accessed using rcu_dereference()
469 and friends. For example, ->rcu_protected_pointer might be declared as
472 struct foo __rcu *rcu_protected_pointer;
474 Use of "__rcu" is opt-in. If you choose not to use it, then you should
475 ignore the sparse warnings.