]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
a71fca58 | 2 | * Read-Copy Update mechanism for mutual exclusion |
1da177e4 LT |
3 | * |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
01c1c660 | 18 | * Copyright IBM Corporation, 2001 |
1da177e4 LT |
19 | * |
20 | * Author: Dipankar Sarma <dipankar@in.ibm.com> | |
a71fca58 | 21 | * |
595182bc | 22 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> |
1da177e4 LT |
23 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. |
24 | * Papers: | |
25 | * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf | |
26 | * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) | |
27 | * | |
28 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 29 | * http://lse.sourceforge.net/locking/rcupdate.html |
1da177e4 LT |
30 | * |
31 | */ | |
32 | ||
33 | #ifndef __LINUX_RCUPDATE_H | |
34 | #define __LINUX_RCUPDATE_H | |
35 | ||
99098751 | 36 | #include <linux/types.h> |
1da177e4 LT |
37 | #include <linux/cache.h> |
38 | #include <linux/spinlock.h> | |
39 | #include <linux/threads.h> | |
1da177e4 LT |
40 | #include <linux/cpumask.h> |
41 | #include <linux/seqlock.h> | |
851a67b8 | 42 | #include <linux/lockdep.h> |
4446a36f | 43 | #include <linux/completion.h> |
551d55a9 | 44 | #include <linux/debugobjects.h> |
ca5ecddf | 45 | #include <linux/compiler.h> |
1da177e4 | 46 | |
e5ab6772 DY |
47 | #ifdef CONFIG_RCU_TORTURE_TEST |
48 | extern int rcutorture_runnable; /* for sysctl */ | |
49 | #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */ | |
50 | ||
4a298656 PM |
51 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) |
52 | extern void rcutorture_record_test_transition(void); | |
53 | extern void rcutorture_record_progress(unsigned long vernum); | |
91afaf30 PM |
54 | extern void do_trace_rcu_torture_read(char *rcutorturename, |
55 | struct rcu_head *rhp); | |
4a298656 PM |
56 | #else |
57 | static inline void rcutorture_record_test_transition(void) | |
58 | { | |
59 | } | |
60 | static inline void rcutorture_record_progress(unsigned long vernum) | |
61 | { | |
62 | } | |
91afaf30 PM |
63 | #ifdef CONFIG_RCU_TRACE |
64 | extern void do_trace_rcu_torture_read(char *rcutorturename, | |
65 | struct rcu_head *rhp); | |
66 | #else | |
67 | #define do_trace_rcu_torture_read(rcutorturename, rhp) do { } while (0) | |
68 | #endif | |
4a298656 PM |
69 | #endif |
70 | ||
e27fc964 TH |
71 | #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) |
72 | #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) | |
a3dc3fb1 PM |
73 | #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) |
74 | #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) | |
75 | ||
03b042bf | 76 | /* Exported common interfaces */ |
2c42818e PM |
77 | |
78 | #ifdef CONFIG_PREEMPT_RCU | |
79 | ||
80 | /** | |
81 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
82 | * @head: structure to be used for queueing the RCU updates. | |
83 | * @func: actual callback function to be invoked after the grace period | |
84 | * | |
85 | * The callback function will be invoked some time after a full grace | |
86 | * period elapses, in other words after all pre-existing RCU read-side | |
87 | * critical sections have completed. However, the callback function | |
88 | * might well execute concurrently with RCU read-side critical sections | |
89 | * that started after call_rcu() was invoked. RCU read-side critical | |
90 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
91 | * and may be nested. | |
92 | */ | |
93 | extern void call_rcu(struct rcu_head *head, | |
94 | void (*func)(struct rcu_head *head)); | |
95 | ||
96 | #else /* #ifdef CONFIG_PREEMPT_RCU */ | |
97 | ||
98 | /* In classic RCU, call_rcu() is just call_rcu_sched(). */ | |
99 | #define call_rcu call_rcu_sched | |
100 | ||
101 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
102 | ||
103 | /** | |
104 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
105 | * @head: structure to be used for queueing the RCU updates. | |
106 | * @func: actual callback function to be invoked after the grace period | |
107 | * | |
108 | * The callback function will be invoked some time after a full grace | |
109 | * period elapses, in other words after all currently executing RCU | |
110 | * read-side critical sections have completed. call_rcu_bh() assumes | |
111 | * that the read-side critical sections end on completion of a softirq | |
112 | * handler. This means that read-side critical sections in process | |
113 | * context must not be interrupted by softirqs. This interface is to be | |
114 | * used when most of the read-side critical sections are in softirq context. | |
115 | * RCU read-side critical sections are delimited by : | |
116 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. | |
117 | * OR | |
118 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
119 | * These may be nested. | |
120 | */ | |
121 | extern void call_rcu_bh(struct rcu_head *head, | |
122 | void (*func)(struct rcu_head *head)); | |
123 | ||
124 | /** | |
125 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
126 | * @head: structure to be used for queueing the RCU updates. | |
127 | * @func: actual callback function to be invoked after the grace period | |
128 | * | |
129 | * The callback function will be invoked some time after a full grace | |
130 | * period elapses, in other words after all currently executing RCU | |
131 | * read-side critical sections have completed. call_rcu_sched() assumes | |
132 | * that the read-side critical sections end on enabling of preemption | |
133 | * or on voluntary preemption. | |
134 | * RCU read-side critical sections are delimited by : | |
135 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), | |
136 | * OR | |
137 | * anything that disables preemption. | |
138 | * These may be nested. | |
139 | */ | |
7b0b759b PM |
140 | extern void call_rcu_sched(struct rcu_head *head, |
141 | void (*func)(struct rcu_head *rcu)); | |
2c42818e | 142 | |
7b0b759b | 143 | extern void synchronize_sched(void); |
03b042bf | 144 | |
a3dc3fb1 PM |
145 | #ifdef CONFIG_PREEMPT_RCU |
146 | ||
7b0b759b PM |
147 | extern void __rcu_read_lock(void); |
148 | extern void __rcu_read_unlock(void); | |
149 | void synchronize_rcu(void); | |
150 | ||
a3dc3fb1 PM |
151 | /* |
152 | * Defined as a macro as it is a very low level header included from | |
153 | * areas that don't even know about current. This gives the rcu_read_lock() | |
154 | * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other | |
155 | * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. | |
156 | */ | |
157 | #define rcu_preempt_depth() (current->rcu_read_lock_nesting) | |
158 | ||
7b0b759b PM |
159 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
160 | ||
161 | static inline void __rcu_read_lock(void) | |
162 | { | |
163 | preempt_disable(); | |
164 | } | |
165 | ||
166 | static inline void __rcu_read_unlock(void) | |
167 | { | |
168 | preempt_enable(); | |
169 | } | |
170 | ||
171 | static inline void synchronize_rcu(void) | |
172 | { | |
173 | synchronize_sched(); | |
174 | } | |
175 | ||
176 | static inline int rcu_preempt_depth(void) | |
177 | { | |
178 | return 0; | |
179 | } | |
180 | ||
181 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
182 | ||
183 | /* Internal to kernel */ | |
7b0b759b PM |
184 | extern void rcu_sched_qs(int cpu); |
185 | extern void rcu_bh_qs(int cpu); | |
186 | extern void rcu_check_callbacks(int cpu, int user); | |
187 | struct notifier_block; | |
9b2e4f18 PM |
188 | extern void rcu_idle_enter(void); |
189 | extern void rcu_idle_exit(void); | |
190 | extern void rcu_irq_enter(void); | |
191 | extern void rcu_irq_exit(void); | |
a3dc3fb1 | 192 | |
2c42818e PM |
193 | /* |
194 | * Infrastructure to implement the synchronize_() primitives in | |
195 | * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. | |
196 | */ | |
197 | ||
198 | typedef void call_rcu_func_t(struct rcu_head *head, | |
199 | void (*func)(struct rcu_head *head)); | |
200 | void wait_rcu_gp(call_rcu_func_t crf); | |
201 | ||
f41d911f | 202 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) |
64db4cff | 203 | #include <linux/rcutree.h> |
a57eb940 | 204 | #elif defined(CONFIG_TINY_RCU) || defined(CONFIG_TINY_PREEMPT_RCU) |
9b1d82fa | 205 | #include <linux/rcutiny.h> |
64db4cff PM |
206 | #else |
207 | #error "Unknown RCU implementation specified to kernel configuration" | |
6b3ef48a | 208 | #endif |
01c1c660 | 209 | |
551d55a9 MD |
210 | /* |
211 | * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic | |
212 | * initialization and destruction of rcu_head on the stack. rcu_head structures | |
213 | * allocated dynamically in the heap or defined statically don't need any | |
214 | * initialization. | |
215 | */ | |
216 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
217 | extern void init_rcu_head_on_stack(struct rcu_head *head); | |
218 | extern void destroy_rcu_head_on_stack(struct rcu_head *head); | |
219 | #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | |
4376030a MD |
220 | static inline void init_rcu_head_on_stack(struct rcu_head *head) |
221 | { | |
222 | } | |
223 | ||
224 | static inline void destroy_rcu_head_on_stack(struct rcu_head *head) | |
225 | { | |
226 | } | |
551d55a9 | 227 | #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
4376030a | 228 | |
bc33f24b | 229 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
632ee200 | 230 | |
e6b80a3b FW |
231 | #ifdef CONFIG_PROVE_RCU |
232 | extern int rcu_is_cpu_idle(void); | |
233 | #else /* !CONFIG_PROVE_RCU */ | |
234 | static inline int rcu_is_cpu_idle(void) | |
235 | { | |
236 | return 0; | |
237 | } | |
238 | #endif /* else !CONFIG_PROVE_RCU */ | |
239 | ||
00f49e57 FW |
240 | static inline void rcu_lock_acquire(struct lockdep_map *map) |
241 | { | |
242 | WARN_ON_ONCE(rcu_is_cpu_idle()); | |
243 | lock_acquire(map, 0, 0, 2, 1, NULL, _THIS_IP_); | |
244 | } | |
245 | ||
246 | static inline void rcu_lock_release(struct lockdep_map *map) | |
247 | { | |
248 | WARN_ON_ONCE(rcu_is_cpu_idle()); | |
249 | lock_release(map, 1, _THIS_IP_); | |
250 | } | |
251 | ||
bc33f24b | 252 | extern struct lockdep_map rcu_lock_map; |
632ee200 | 253 | extern struct lockdep_map rcu_bh_lock_map; |
632ee200 | 254 | extern struct lockdep_map rcu_sched_lock_map; |
bc293d62 | 255 | extern int debug_lockdep_rcu_enabled(void); |
54dbf96c | 256 | |
632ee200 | 257 | /** |
ca5ecddf | 258 | * rcu_read_lock_held() - might we be in RCU read-side critical section? |
632ee200 | 259 | * |
d20200b5 PM |
260 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU |
261 | * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC, | |
632ee200 | 262 | * this assumes we are in an RCU read-side critical section unless it can |
ca5ecddf PM |
263 | * prove otherwise. This is useful for debug checks in functions that |
264 | * require that they be called within an RCU read-side critical section. | |
54dbf96c | 265 | * |
ca5ecddf | 266 | * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot |
32c141a0 | 267 | * and while lockdep is disabled. |
3842a083 PM |
268 | * |
269 | * Note that rcu_read_lock() and the matching rcu_read_unlock() must | |
270 | * occur in the same context, for example, it is illegal to invoke | |
271 | * rcu_read_unlock() in process context if the matching rcu_read_lock() | |
272 | * was invoked from within an irq handler. | |
632ee200 PM |
273 | */ |
274 | static inline int rcu_read_lock_held(void) | |
275 | { | |
54dbf96c PM |
276 | if (!debug_lockdep_rcu_enabled()) |
277 | return 1; | |
e6b80a3b FW |
278 | if (rcu_is_cpu_idle()) |
279 | return 0; | |
54dbf96c | 280 | return lock_is_held(&rcu_lock_map); |
632ee200 PM |
281 | } |
282 | ||
e3818b8d PM |
283 | /* |
284 | * rcu_read_lock_bh_held() is defined out of line to avoid #include-file | |
285 | * hell. | |
632ee200 | 286 | */ |
e3818b8d | 287 | extern int rcu_read_lock_bh_held(void); |
632ee200 PM |
288 | |
289 | /** | |
ca5ecddf | 290 | * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? |
632ee200 | 291 | * |
d20200b5 PM |
292 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an |
293 | * RCU-sched read-side critical section. In absence of | |
294 | * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side | |
295 | * critical section unless it can prove otherwise. Note that disabling | |
296 | * of preemption (including disabling irqs) counts as an RCU-sched | |
ca5ecddf PM |
297 | * read-side critical section. This is useful for debug checks in functions |
298 | * that required that they be called within an RCU-sched read-side | |
299 | * critical section. | |
54dbf96c | 300 | * |
32c141a0 PM |
301 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot |
302 | * and while lockdep is disabled. | |
e6b80a3b FW |
303 | * |
304 | * Note that if the CPU is in the idle loop from an RCU point of | |
305 | * view (ie: that we are in the section between rcu_idle_enter() and | |
306 | * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU | |
307 | * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs | |
308 | * that are in such a section, considering these as in extended quiescent | |
309 | * state, so such a CPU is effectively never in an RCU read-side critical | |
310 | * section regardless of what RCU primitives it invokes. This state of | |
311 | * affairs is required --- we need to keep an RCU-free window in idle | |
312 | * where the CPU may possibly enter into low power mode. This way we can | |
313 | * notice an extended quiescent state to other CPUs that started a grace | |
314 | * period. Otherwise we would delay any grace period as long as we run in | |
315 | * the idle task. | |
632ee200 | 316 | */ |
bdd4e85d | 317 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
318 | static inline int rcu_read_lock_sched_held(void) |
319 | { | |
320 | int lockdep_opinion = 0; | |
321 | ||
54dbf96c PM |
322 | if (!debug_lockdep_rcu_enabled()) |
323 | return 1; | |
e6b80a3b FW |
324 | if (rcu_is_cpu_idle()) |
325 | return 0; | |
632ee200 PM |
326 | if (debug_locks) |
327 | lockdep_opinion = lock_is_held(&rcu_sched_lock_map); | |
0cff810f | 328 | return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); |
632ee200 | 329 | } |
bdd4e85d | 330 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
331 | static inline int rcu_read_lock_sched_held(void) |
332 | { | |
333 | return 1; | |
632ee200 | 334 | } |
bdd4e85d | 335 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
336 | |
337 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
338 | ||
d8ab29f8 PM |
339 | # define rcu_lock_acquire(a) do { } while (0) |
340 | # define rcu_lock_release(a) do { } while (0) | |
632ee200 PM |
341 | |
342 | static inline int rcu_read_lock_held(void) | |
343 | { | |
344 | return 1; | |
345 | } | |
346 | ||
347 | static inline int rcu_read_lock_bh_held(void) | |
348 | { | |
349 | return 1; | |
350 | } | |
351 | ||
bdd4e85d | 352 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
353 | static inline int rcu_read_lock_sched_held(void) |
354 | { | |
bbad9379 | 355 | return preempt_count() != 0 || irqs_disabled(); |
632ee200 | 356 | } |
bdd4e85d | 357 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
358 | static inline int rcu_read_lock_sched_held(void) |
359 | { | |
360 | return 1; | |
632ee200 | 361 | } |
bdd4e85d | 362 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
363 | |
364 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
365 | ||
366 | #ifdef CONFIG_PROVE_RCU | |
367 | ||
ee84b824 PM |
368 | extern int rcu_my_thread_group_empty(void); |
369 | ||
4221a991 TH |
370 | /** |
371 | * rcu_lockdep_assert - emit lockdep splat if specified condition not met | |
372 | * @c: condition to check | |
b3fbab05 | 373 | * @s: informative message |
4221a991 | 374 | */ |
b3fbab05 | 375 | #define rcu_lockdep_assert(c, s) \ |
2b3fc35f LJ |
376 | do { \ |
377 | static bool __warned; \ | |
378 | if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ | |
379 | __warned = true; \ | |
b3fbab05 | 380 | lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ |
2b3fc35f LJ |
381 | } \ |
382 | } while (0) | |
383 | ||
b3fbab05 PM |
384 | #define rcu_sleep_check() \ |
385 | do { \ | |
386 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \ | |
387 | "Illegal context switch in RCU-bh" \ | |
388 | " read-side critical section"); \ | |
389 | rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \ | |
390 | "Illegal context switch in RCU-sched"\ | |
391 | " read-side critical section"); \ | |
392 | } while (0) | |
393 | ||
ca5ecddf PM |
394 | #else /* #ifdef CONFIG_PROVE_RCU */ |
395 | ||
b3fbab05 PM |
396 | #define rcu_lockdep_assert(c, s) do { } while (0) |
397 | #define rcu_sleep_check() do { } while (0) | |
ca5ecddf PM |
398 | |
399 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
400 | ||
401 | /* | |
402 | * Helper functions for rcu_dereference_check(), rcu_dereference_protected() | |
403 | * and rcu_assign_pointer(). Some of these could be folded into their | |
404 | * callers, but they are left separate in order to ease introduction of | |
405 | * multiple flavors of pointers to match the multiple flavors of RCU | |
406 | * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in | |
407 | * the future. | |
408 | */ | |
53ecfba2 PM |
409 | |
410 | #ifdef __CHECKER__ | |
411 | #define rcu_dereference_sparse(p, space) \ | |
412 | ((void)(((typeof(*p) space *)p) == p)) | |
413 | #else /* #ifdef __CHECKER__ */ | |
414 | #define rcu_dereference_sparse(p, space) | |
415 | #endif /* #else #ifdef __CHECKER__ */ | |
416 | ||
ca5ecddf PM |
417 | #define __rcu_access_pointer(p, space) \ |
418 | ({ \ | |
419 | typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \ | |
53ecfba2 | 420 | rcu_dereference_sparse(p, space); \ |
ca5ecddf PM |
421 | ((typeof(*p) __force __kernel *)(_________p1)); \ |
422 | }) | |
423 | #define __rcu_dereference_check(p, c, space) \ | |
424 | ({ \ | |
425 | typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \ | |
b3fbab05 PM |
426 | rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \ |
427 | " usage"); \ | |
53ecfba2 | 428 | rcu_dereference_sparse(p, space); \ |
ca5ecddf PM |
429 | smp_read_barrier_depends(); \ |
430 | ((typeof(*p) __force __kernel *)(_________p1)); \ | |
431 | }) | |
432 | #define __rcu_dereference_protected(p, c, space) \ | |
433 | ({ \ | |
b3fbab05 PM |
434 | rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \ |
435 | " usage"); \ | |
53ecfba2 | 436 | rcu_dereference_sparse(p, space); \ |
ca5ecddf PM |
437 | ((typeof(*p) __force __kernel *)(p)); \ |
438 | }) | |
439 | ||
a4dd9925 PM |
440 | #define __rcu_access_index(p, space) \ |
441 | ({ \ | |
442 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
443 | rcu_dereference_sparse(p, space); \ | |
444 | (_________p1); \ | |
445 | }) | |
ca5ecddf PM |
446 | #define __rcu_dereference_index_check(p, c) \ |
447 | ({ \ | |
448 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
b3fbab05 PM |
449 | rcu_lockdep_assert(c, \ |
450 | "suspicious rcu_dereference_index_check()" \ | |
451 | " usage"); \ | |
ca5ecddf PM |
452 | smp_read_barrier_depends(); \ |
453 | (_________p1); \ | |
454 | }) | |
455 | #define __rcu_assign_pointer(p, v, space) \ | |
456 | ({ \ | |
d322f45c | 457 | smp_wmb(); \ |
ca5ecddf PM |
458 | (p) = (typeof(*v) __force space *)(v); \ |
459 | }) | |
460 | ||
461 | ||
462 | /** | |
463 | * rcu_access_pointer() - fetch RCU pointer with no dereferencing | |
464 | * @p: The pointer to read | |
465 | * | |
466 | * Return the value of the specified RCU-protected pointer, but omit the | |
467 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
468 | * when the value of this pointer is accessed, but the pointer is not | |
469 | * dereferenced, for example, when testing an RCU-protected pointer against | |
470 | * NULL. Although rcu_access_pointer() may also be used in cases where | |
471 | * update-side locks prevent the value of the pointer from changing, you | |
472 | * should instead use rcu_dereference_protected() for this use case. | |
473 | */ | |
474 | #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) | |
475 | ||
632ee200 | 476 | /** |
ca5ecddf | 477 | * rcu_dereference_check() - rcu_dereference with debug checking |
c08c68dd DH |
478 | * @p: The pointer to read, prior to dereferencing |
479 | * @c: The conditions under which the dereference will take place | |
632ee200 | 480 | * |
c08c68dd | 481 | * Do an rcu_dereference(), but check that the conditions under which the |
ca5ecddf PM |
482 | * dereference will take place are correct. Typically the conditions |
483 | * indicate the various locking conditions that should be held at that | |
484 | * point. The check should return true if the conditions are satisfied. | |
485 | * An implicit check for being in an RCU read-side critical section | |
486 | * (rcu_read_lock()) is included. | |
c08c68dd DH |
487 | * |
488 | * For example: | |
489 | * | |
ca5ecddf | 490 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); |
c08c68dd DH |
491 | * |
492 | * could be used to indicate to lockdep that foo->bar may only be dereferenced | |
ca5ecddf | 493 | * if either rcu_read_lock() is held, or that the lock required to replace |
c08c68dd DH |
494 | * the bar struct at foo->bar is held. |
495 | * | |
496 | * Note that the list of conditions may also include indications of when a lock | |
497 | * need not be held, for example during initialisation or destruction of the | |
498 | * target struct: | |
499 | * | |
ca5ecddf | 500 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || |
c08c68dd | 501 | * atomic_read(&foo->usage) == 0); |
ca5ecddf PM |
502 | * |
503 | * Inserts memory barriers on architectures that require them | |
504 | * (currently only the Alpha), prevents the compiler from refetching | |
505 | * (and from merging fetches), and, more importantly, documents exactly | |
506 | * which pointers are protected by RCU and checks that the pointer is | |
507 | * annotated as __rcu. | |
632ee200 PM |
508 | */ |
509 | #define rcu_dereference_check(p, c) \ | |
ca5ecddf PM |
510 | __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu) |
511 | ||
512 | /** | |
513 | * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking | |
514 | * @p: The pointer to read, prior to dereferencing | |
515 | * @c: The conditions under which the dereference will take place | |
516 | * | |
517 | * This is the RCU-bh counterpart to rcu_dereference_check(). | |
518 | */ | |
519 | #define rcu_dereference_bh_check(p, c) \ | |
520 | __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu) | |
632ee200 | 521 | |
b62730ba | 522 | /** |
ca5ecddf PM |
523 | * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking |
524 | * @p: The pointer to read, prior to dereferencing | |
525 | * @c: The conditions under which the dereference will take place | |
526 | * | |
527 | * This is the RCU-sched counterpart to rcu_dereference_check(). | |
528 | */ | |
529 | #define rcu_dereference_sched_check(p, c) \ | |
530 | __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \ | |
531 | __rcu) | |
532 | ||
533 | #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ | |
534 | ||
a4dd9925 PM |
535 | /** |
536 | * rcu_access_index() - fetch RCU index with no dereferencing | |
537 | * @p: The index to read | |
538 | * | |
539 | * Return the value of the specified RCU-protected index, but omit the | |
540 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
541 | * when the value of this index is accessed, but the index is not | |
542 | * dereferenced, for example, when testing an RCU-protected index against | |
543 | * -1. Although rcu_access_index() may also be used in cases where | |
544 | * update-side locks prevent the value of the index from changing, you | |
545 | * should instead use rcu_dereference_index_protected() for this use case. | |
546 | */ | |
547 | #define rcu_access_index(p) __rcu_access_index((p), __rcu) | |
548 | ||
ca5ecddf PM |
549 | /** |
550 | * rcu_dereference_index_check() - rcu_dereference for indices with debug checking | |
551 | * @p: The pointer to read, prior to dereferencing | |
552 | * @c: The conditions under which the dereference will take place | |
553 | * | |
554 | * Similar to rcu_dereference_check(), but omits the sparse checking. | |
555 | * This allows rcu_dereference_index_check() to be used on integers, | |
556 | * which can then be used as array indices. Attempting to use | |
557 | * rcu_dereference_check() on an integer will give compiler warnings | |
558 | * because the sparse address-space mechanism relies on dereferencing | |
559 | * the RCU-protected pointer. Dereferencing integers is not something | |
560 | * that even gcc will put up with. | |
561 | * | |
562 | * Note that this function does not implicitly check for RCU read-side | |
563 | * critical sections. If this function gains lots of uses, it might | |
564 | * make sense to provide versions for each flavor of RCU, but it does | |
565 | * not make sense as of early 2010. | |
566 | */ | |
567 | #define rcu_dereference_index_check(p, c) \ | |
568 | __rcu_dereference_index_check((p), (c)) | |
569 | ||
570 | /** | |
571 | * rcu_dereference_protected() - fetch RCU pointer when updates prevented | |
572 | * @p: The pointer to read, prior to dereferencing | |
573 | * @c: The conditions under which the dereference will take place | |
b62730ba PM |
574 | * |
575 | * Return the value of the specified RCU-protected pointer, but omit | |
576 | * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This | |
577 | * is useful in cases where update-side locks prevent the value of the | |
578 | * pointer from changing. Please note that this primitive does -not- | |
579 | * prevent the compiler from repeating this reference or combining it | |
580 | * with other references, so it should not be used without protection | |
581 | * of appropriate locks. | |
ca5ecddf PM |
582 | * |
583 | * This function is only for update-side use. Using this function | |
584 | * when protected only by rcu_read_lock() will result in infrequent | |
585 | * but very ugly failures. | |
b62730ba PM |
586 | */ |
587 | #define rcu_dereference_protected(p, c) \ | |
ca5ecddf | 588 | __rcu_dereference_protected((p), (c), __rcu) |
b62730ba | 589 | |
bc33f24b | 590 | |
b62730ba | 591 | /** |
ca5ecddf PM |
592 | * rcu_dereference() - fetch RCU-protected pointer for dereferencing |
593 | * @p: The pointer to read, prior to dereferencing | |
b62730ba | 594 | * |
ca5ecddf | 595 | * This is a simple wrapper around rcu_dereference_check(). |
b62730ba | 596 | */ |
ca5ecddf | 597 | #define rcu_dereference(p) rcu_dereference_check(p, 0) |
b62730ba | 598 | |
1da177e4 | 599 | /** |
ca5ecddf PM |
600 | * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing |
601 | * @p: The pointer to read, prior to dereferencing | |
602 | * | |
603 | * Makes rcu_dereference_check() do the dirty work. | |
604 | */ | |
605 | #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) | |
606 | ||
607 | /** | |
608 | * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing | |
609 | * @p: The pointer to read, prior to dereferencing | |
610 | * | |
611 | * Makes rcu_dereference_check() do the dirty work. | |
612 | */ | |
613 | #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) | |
614 | ||
615 | /** | |
616 | * rcu_read_lock() - mark the beginning of an RCU read-side critical section | |
1da177e4 | 617 | * |
9b06e818 | 618 | * When synchronize_rcu() is invoked on one CPU while other CPUs |
1da177e4 | 619 | * are within RCU read-side critical sections, then the |
9b06e818 | 620 | * synchronize_rcu() is guaranteed to block until after all the other |
1da177e4 LT |
621 | * CPUs exit their critical sections. Similarly, if call_rcu() is invoked |
622 | * on one CPU while other CPUs are within RCU read-side critical | |
623 | * sections, invocation of the corresponding RCU callback is deferred | |
624 | * until after the all the other CPUs exit their critical sections. | |
625 | * | |
626 | * Note, however, that RCU callbacks are permitted to run concurrently | |
77d8485a | 627 | * with new RCU read-side critical sections. One way that this can happen |
1da177e4 LT |
628 | * is via the following sequence of events: (1) CPU 0 enters an RCU |
629 | * read-side critical section, (2) CPU 1 invokes call_rcu() to register | |
630 | * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, | |
631 | * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU | |
632 | * callback is invoked. This is legal, because the RCU read-side critical | |
633 | * section that was running concurrently with the call_rcu() (and which | |
634 | * therefore might be referencing something that the corresponding RCU | |
635 | * callback would free up) has completed before the corresponding | |
636 | * RCU callback is invoked. | |
637 | * | |
638 | * RCU read-side critical sections may be nested. Any deferred actions | |
639 | * will be deferred until the outermost RCU read-side critical section | |
640 | * completes. | |
641 | * | |
9079fd7c PM |
642 | * You can avoid reading and understanding the next paragraph by |
643 | * following this rule: don't put anything in an rcu_read_lock() RCU | |
644 | * read-side critical section that would block in a !PREEMPT kernel. | |
645 | * But if you want the full story, read on! | |
646 | * | |
647 | * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it | |
648 | * is illegal to block while in an RCU read-side critical section. In | |
649 | * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU) | |
650 | * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may | |
651 | * be preempted, but explicit blocking is illegal. Finally, in preemptible | |
652 | * RCU implementations in real-time (CONFIG_PREEMPT_RT) kernel builds, | |
653 | * RCU read-side critical sections may be preempted and they may also | |
654 | * block, but only when acquiring spinlocks that are subject to priority | |
655 | * inheritance. | |
1da177e4 | 656 | */ |
bc33f24b PM |
657 | static inline void rcu_read_lock(void) |
658 | { | |
659 | __rcu_read_lock(); | |
660 | __acquire(RCU); | |
d8ab29f8 | 661 | rcu_lock_acquire(&rcu_lock_map); |
bc33f24b | 662 | } |
1da177e4 | 663 | |
1da177e4 LT |
664 | /* |
665 | * So where is rcu_write_lock()? It does not exist, as there is no | |
666 | * way for writers to lock out RCU readers. This is a feature, not | |
667 | * a bug -- this property is what provides RCU's performance benefits. | |
668 | * Of course, writers must coordinate with each other. The normal | |
669 | * spinlock primitives work well for this, but any other technique may be | |
670 | * used as well. RCU does not care how the writers keep out of each | |
671 | * others' way, as long as they do so. | |
672 | */ | |
3d76c082 PM |
673 | |
674 | /** | |
ca5ecddf | 675 | * rcu_read_unlock() - marks the end of an RCU read-side critical section. |
3d76c082 PM |
676 | * |
677 | * See rcu_read_lock() for more information. | |
678 | */ | |
bc33f24b PM |
679 | static inline void rcu_read_unlock(void) |
680 | { | |
d8ab29f8 | 681 | rcu_lock_release(&rcu_lock_map); |
bc33f24b PM |
682 | __release(RCU); |
683 | __rcu_read_unlock(); | |
684 | } | |
1da177e4 LT |
685 | |
686 | /** | |
ca5ecddf | 687 | * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section |
1da177e4 LT |
688 | * |
689 | * This is equivalent of rcu_read_lock(), but to be used when updates | |
ca5ecddf PM |
690 | * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since |
691 | * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a | |
692 | * softirq handler to be a quiescent state, a process in RCU read-side | |
693 | * critical section must be protected by disabling softirqs. Read-side | |
694 | * critical sections in interrupt context can use just rcu_read_lock(), | |
695 | * though this should at least be commented to avoid confusing people | |
696 | * reading the code. | |
3842a083 PM |
697 | * |
698 | * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() | |
699 | * must occur in the same context, for example, it is illegal to invoke | |
700 | * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() | |
701 | * was invoked from some other task. | |
1da177e4 | 702 | */ |
bc33f24b PM |
703 | static inline void rcu_read_lock_bh(void) |
704 | { | |
6206ab9b | 705 | local_bh_disable(); |
bc33f24b | 706 | __acquire(RCU_BH); |
d8ab29f8 | 707 | rcu_lock_acquire(&rcu_bh_lock_map); |
bc33f24b | 708 | } |
1da177e4 LT |
709 | |
710 | /* | |
711 | * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section | |
712 | * | |
713 | * See rcu_read_lock_bh() for more information. | |
714 | */ | |
bc33f24b PM |
715 | static inline void rcu_read_unlock_bh(void) |
716 | { | |
d8ab29f8 | 717 | rcu_lock_release(&rcu_bh_lock_map); |
bc33f24b | 718 | __release(RCU_BH); |
6206ab9b | 719 | local_bh_enable(); |
bc33f24b | 720 | } |
1da177e4 | 721 | |
1c50b728 | 722 | /** |
ca5ecddf | 723 | * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section |
1c50b728 | 724 | * |
ca5ecddf PM |
725 | * This is equivalent of rcu_read_lock(), but to be used when updates |
726 | * are being done using call_rcu_sched() or synchronize_rcu_sched(). | |
727 | * Read-side critical sections can also be introduced by anything that | |
728 | * disables preemption, including local_irq_disable() and friends. | |
3842a083 PM |
729 | * |
730 | * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() | |
731 | * must occur in the same context, for example, it is illegal to invoke | |
732 | * rcu_read_unlock_sched() from process context if the matching | |
733 | * rcu_read_lock_sched() was invoked from an NMI handler. | |
1c50b728 | 734 | */ |
d6714c22 PM |
735 | static inline void rcu_read_lock_sched(void) |
736 | { | |
737 | preempt_disable(); | |
bc33f24b | 738 | __acquire(RCU_SCHED); |
d8ab29f8 | 739 | rcu_lock_acquire(&rcu_sched_lock_map); |
d6714c22 | 740 | } |
1eba8f84 PM |
741 | |
742 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 743 | static inline notrace void rcu_read_lock_sched_notrace(void) |
d6714c22 PM |
744 | { |
745 | preempt_disable_notrace(); | |
bc33f24b | 746 | __acquire(RCU_SCHED); |
d6714c22 | 747 | } |
1c50b728 MD |
748 | |
749 | /* | |
750 | * rcu_read_unlock_sched - marks the end of a RCU-classic critical section | |
751 | * | |
752 | * See rcu_read_lock_sched for more information. | |
753 | */ | |
d6714c22 PM |
754 | static inline void rcu_read_unlock_sched(void) |
755 | { | |
d8ab29f8 | 756 | rcu_lock_release(&rcu_sched_lock_map); |
bc33f24b | 757 | __release(RCU_SCHED); |
d6714c22 PM |
758 | preempt_enable(); |
759 | } | |
1eba8f84 PM |
760 | |
761 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 762 | static inline notrace void rcu_read_unlock_sched_notrace(void) |
d6714c22 | 763 | { |
bc33f24b | 764 | __release(RCU_SCHED); |
d6714c22 PM |
765 | preempt_enable_notrace(); |
766 | } | |
1c50b728 | 767 | |
1da177e4 | 768 | /** |
ca5ecddf PM |
769 | * rcu_assign_pointer() - assign to RCU-protected pointer |
770 | * @p: pointer to assign to | |
771 | * @v: value to assign (publish) | |
c26d34a5 | 772 | * |
ca5ecddf PM |
773 | * Assigns the specified value to the specified RCU-protected |
774 | * pointer, ensuring that any concurrent RCU readers will see | |
775 | * any prior initialization. Returns the value assigned. | |
1da177e4 LT |
776 | * |
777 | * Inserts memory barriers on architectures that require them | |
6846c0c5 PM |
778 | * (which is most of them), and also prevents the compiler from |
779 | * reordering the code that initializes the structure after the pointer | |
780 | * assignment. More importantly, this call documents which pointers | |
781 | * will be dereferenced by RCU read-side code. | |
782 | * | |
783 | * In some special cases, you may use RCU_INIT_POINTER() instead | |
784 | * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due | |
785 | * to the fact that it does not constrain either the CPU or the compiler. | |
786 | * That said, using RCU_INIT_POINTER() when you should have used | |
787 | * rcu_assign_pointer() is a very bad thing that results in | |
788 | * impossible-to-diagnose memory corruption. So please be careful. | |
789 | * See the RCU_INIT_POINTER() comment header for details. | |
1da177e4 | 790 | */ |
d99c4f6b | 791 | #define rcu_assign_pointer(p, v) \ |
ca5ecddf PM |
792 | __rcu_assign_pointer((p), (v), __rcu) |
793 | ||
794 | /** | |
795 | * RCU_INIT_POINTER() - initialize an RCU protected pointer | |
796 | * | |
6846c0c5 PM |
797 | * Initialize an RCU-protected pointer in special cases where readers |
798 | * do not need ordering constraints on the CPU or the compiler. These | |
799 | * special cases are: | |
800 | * | |
801 | * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- | |
802 | * 2. The caller has taken whatever steps are required to prevent | |
803 | * RCU readers from concurrently accessing this pointer -or- | |
804 | * 3. The referenced data structure has already been exposed to | |
805 | * readers either at compile time or via rcu_assign_pointer() -and- | |
806 | * a. You have not made -any- reader-visible changes to | |
807 | * this structure since then -or- | |
808 | * b. It is OK for readers accessing this structure from its | |
809 | * new location to see the old state of the structure. (For | |
810 | * example, the changes were to statistical counters or to | |
811 | * other state where exact synchronization is not required.) | |
812 | * | |
813 | * Failure to follow these rules governing use of RCU_INIT_POINTER() will | |
814 | * result in impossible-to-diagnose memory corruption. As in the structures | |
815 | * will look OK in crash dumps, but any concurrent RCU readers might | |
816 | * see pre-initialized values of the referenced data structure. So | |
817 | * please be very careful how you use RCU_INIT_POINTER()!!! | |
818 | * | |
819 | * If you are creating an RCU-protected linked structure that is accessed | |
820 | * by a single external-to-structure RCU-protected pointer, then you may | |
821 | * use RCU_INIT_POINTER() to initialize the internal RCU-protected | |
822 | * pointers, but you must use rcu_assign_pointer() to initialize the | |
823 | * external-to-structure pointer -after- you have completely initialized | |
824 | * the reader-accessible portions of the linked structure. | |
ca5ecddf PM |
825 | */ |
826 | #define RCU_INIT_POINTER(p, v) \ | |
827 | p = (typeof(*v) __force __rcu *)(v) | |
1da177e4 | 828 | |
9ab1544e LJ |
829 | static __always_inline bool __is_kfree_rcu_offset(unsigned long offset) |
830 | { | |
831 | return offset < 4096; | |
832 | } | |
833 | ||
834 | static __always_inline | |
835 | void __kfree_rcu(struct rcu_head *head, unsigned long offset) | |
836 | { | |
837 | typedef void (*rcu_callback)(struct rcu_head *); | |
838 | ||
839 | BUILD_BUG_ON(!__builtin_constant_p(offset)); | |
840 | ||
841 | /* See the kfree_rcu() header comment. */ | |
842 | BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); | |
843 | ||
486e2593 | 844 | kfree_call_rcu(head, (rcu_callback)offset); |
9ab1544e LJ |
845 | } |
846 | ||
9ab1544e LJ |
847 | /** |
848 | * kfree_rcu() - kfree an object after a grace period. | |
849 | * @ptr: pointer to kfree | |
850 | * @rcu_head: the name of the struct rcu_head within the type of @ptr. | |
851 | * | |
852 | * Many rcu callbacks functions just call kfree() on the base structure. | |
853 | * These functions are trivial, but their size adds up, and furthermore | |
854 | * when they are used in a kernel module, that module must invoke the | |
855 | * high-latency rcu_barrier() function at module-unload time. | |
856 | * | |
857 | * The kfree_rcu() function handles this issue. Rather than encoding a | |
858 | * function address in the embedded rcu_head structure, kfree_rcu() instead | |
859 | * encodes the offset of the rcu_head structure within the base structure. | |
860 | * Because the functions are not allowed in the low-order 4096 bytes of | |
861 | * kernel virtual memory, offsets up to 4095 bytes can be accommodated. | |
862 | * If the offset is larger than 4095 bytes, a compile-time error will | |
863 | * be generated in __kfree_rcu(). If this error is triggered, you can | |
864 | * either fall back to use of call_rcu() or rearrange the structure to | |
865 | * position the rcu_head structure into the first 4096 bytes. | |
866 | * | |
867 | * Note that the allowable offset might decrease in the future, for example, | |
868 | * to allow something like kmem_cache_free_rcu(). | |
869 | */ | |
870 | #define kfree_rcu(ptr, rcu_head) \ | |
871 | __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) | |
872 | ||
1da177e4 | 873 | #endif /* __LINUX_RCUPDATE_H */ |