]>
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 | |
87de1cfd PM |
15 | * along with this program; if not, you can access it online at |
16 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
1da177e4 | 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> |
187f1882 | 45 | #include <linux/bug.h> |
ca5ecddf | 46 | #include <linux/compiler.h> |
c1ad348b | 47 | #include <linux/ktime.h> |
4929c913 | 48 | #include <linux/irqflags.h> |
c1ad348b | 49 | |
88c18630 | 50 | #include <asm/barrier.h> |
1da177e4 | 51 | |
79cfea02 | 52 | #ifndef CONFIG_TINY_RCU |
7a754743 | 53 | extern int rcu_expedited; /* for sysctl */ |
5a9be7c6 | 54 | extern int rcu_normal; /* also for sysctl */ |
79cfea02 | 55 | #endif /* #ifndef CONFIG_TINY_RCU */ |
e5ab6772 | 56 | |
0d39482c PM |
57 | #ifdef CONFIG_TINY_RCU |
58 | /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */ | |
5a9be7c6 PM |
59 | static inline bool rcu_gp_is_normal(void) /* Internal RCU use. */ |
60 | { | |
61 | return true; | |
62 | } | |
0d39482c PM |
63 | static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */ |
64 | { | |
65 | return false; | |
66 | } | |
67 | ||
68 | static inline void rcu_expedite_gp(void) | |
69 | { | |
70 | } | |
71 | ||
72 | static inline void rcu_unexpedite_gp(void) | |
73 | { | |
74 | } | |
75 | #else /* #ifdef CONFIG_TINY_RCU */ | |
5a9be7c6 | 76 | bool rcu_gp_is_normal(void); /* Internal RCU use. */ |
0d39482c PM |
77 | bool rcu_gp_is_expedited(void); /* Internal RCU use. */ |
78 | void rcu_expedite_gp(void); | |
79 | void rcu_unexpedite_gp(void); | |
80 | #endif /* #else #ifdef CONFIG_TINY_RCU */ | |
81 | ||
ad0dc7f9 PM |
82 | enum rcutorture_type { |
83 | RCU_FLAVOR, | |
84 | RCU_BH_FLAVOR, | |
85 | RCU_SCHED_FLAVOR, | |
69c60455 | 86 | RCU_TASKS_FLAVOR, |
ad0dc7f9 PM |
87 | SRCU_FLAVOR, |
88 | INVALID_RCU_FLAVOR | |
89 | }; | |
90 | ||
28f6569a | 91 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) |
ad0dc7f9 PM |
92 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, |
93 | unsigned long *gpnum, unsigned long *completed); | |
584dc4ce TB |
94 | void rcutorture_record_test_transition(void); |
95 | void rcutorture_record_progress(unsigned long vernum); | |
96 | void do_trace_rcu_torture_read(const char *rcutorturename, | |
97 | struct rcu_head *rhp, | |
98 | unsigned long secs, | |
99 | unsigned long c_old, | |
100 | unsigned long c); | |
4a298656 | 101 | #else |
ad0dc7f9 PM |
102 | static inline void rcutorture_get_gp_data(enum rcutorture_type test_type, |
103 | int *flags, | |
104 | unsigned long *gpnum, | |
105 | unsigned long *completed) | |
106 | { | |
107 | *flags = 0; | |
108 | *gpnum = 0; | |
109 | *completed = 0; | |
110 | } | |
4a298656 PM |
111 | static inline void rcutorture_record_test_transition(void) |
112 | { | |
113 | } | |
114 | static inline void rcutorture_record_progress(unsigned long vernum) | |
115 | { | |
116 | } | |
91afaf30 | 117 | #ifdef CONFIG_RCU_TRACE |
584dc4ce TB |
118 | void do_trace_rcu_torture_read(const char *rcutorturename, |
119 | struct rcu_head *rhp, | |
120 | unsigned long secs, | |
121 | unsigned long c_old, | |
122 | unsigned long c); | |
91afaf30 | 123 | #else |
52494535 PM |
124 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ |
125 | do { } while (0) | |
91afaf30 | 126 | #endif |
4a298656 PM |
127 | #endif |
128 | ||
e27fc964 TH |
129 | #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) |
130 | #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) | |
a3dc3fb1 PM |
131 | #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) |
132 | #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) | |
c0f4dfd4 | 133 | #define ulong2long(a) (*(long *)(&(a))) |
a3dc3fb1 | 134 | |
03b042bf | 135 | /* Exported common interfaces */ |
2c42818e PM |
136 | |
137 | #ifdef CONFIG_PREEMPT_RCU | |
138 | ||
139 | /** | |
140 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
141 | * @head: structure to be used for queueing the RCU updates. | |
142 | * @func: actual callback function to be invoked after the grace period | |
143 | * | |
144 | * The callback function will be invoked some time after a full grace | |
145 | * period elapses, in other words after all pre-existing RCU read-side | |
146 | * critical sections have completed. However, the callback function | |
147 | * might well execute concurrently with RCU read-side critical sections | |
148 | * that started after call_rcu() was invoked. RCU read-side critical | |
149 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
150 | * and may be nested. | |
f0a0e6f2 PM |
151 | * |
152 | * Note that all CPUs must agree that the grace period extended beyond | |
153 | * all pre-existing RCU read-side critical section. On systems with more | |
154 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
155 | * guaranteed to have executed a full memory barrier since the end of its | |
156 | * last RCU read-side critical section whose beginning preceded the call | |
157 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
158 | * critical section that continues beyond the start of "func()" must have | |
159 | * executed a memory barrier after the call_rcu() but before the beginning | |
160 | * of that RCU read-side critical section. Note that these guarantees | |
161 | * include CPUs that are offline, idle, or executing in user mode, as | |
162 | * well as CPUs that are executing in the kernel. | |
163 | * | |
164 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
165 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
166 | * guaranteed to execute a full memory barrier during the time interval | |
167 | * between the call to call_rcu() and the invocation of "func()" -- even | |
168 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
169 | * more than one CPU). | |
2c42818e | 170 | */ |
584dc4ce | 171 | void call_rcu(struct rcu_head *head, |
b6a4ae76 | 172 | rcu_callback_t func); |
2c42818e PM |
173 | |
174 | #else /* #ifdef CONFIG_PREEMPT_RCU */ | |
175 | ||
176 | /* In classic RCU, call_rcu() is just call_rcu_sched(). */ | |
177 | #define call_rcu call_rcu_sched | |
178 | ||
179 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
180 | ||
181 | /** | |
182 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
183 | * @head: structure to be used for queueing the RCU updates. | |
184 | * @func: actual callback function to be invoked after the grace period | |
185 | * | |
186 | * The callback function will be invoked some time after a full grace | |
187 | * period elapses, in other words after all currently executing RCU | |
188 | * read-side critical sections have completed. call_rcu_bh() assumes | |
189 | * that the read-side critical sections end on completion of a softirq | |
190 | * handler. This means that read-side critical sections in process | |
191 | * context must not be interrupted by softirqs. This interface is to be | |
192 | * used when most of the read-side critical sections are in softirq context. | |
193 | * RCU read-side critical sections are delimited by : | |
194 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. | |
195 | * OR | |
196 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
197 | * These may be nested. | |
f0a0e6f2 PM |
198 | * |
199 | * See the description of call_rcu() for more detailed information on | |
200 | * memory ordering guarantees. | |
2c42818e | 201 | */ |
584dc4ce | 202 | void call_rcu_bh(struct rcu_head *head, |
b6a4ae76 | 203 | rcu_callback_t func); |
2c42818e PM |
204 | |
205 | /** | |
206 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
207 | * @head: structure to be used for queueing the RCU updates. | |
208 | * @func: actual callback function to be invoked after the grace period | |
209 | * | |
210 | * The callback function will be invoked some time after a full grace | |
211 | * period elapses, in other words after all currently executing RCU | |
212 | * read-side critical sections have completed. call_rcu_sched() assumes | |
213 | * that the read-side critical sections end on enabling of preemption | |
214 | * or on voluntary preemption. | |
215 | * RCU read-side critical sections are delimited by : | |
216 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), | |
217 | * OR | |
218 | * anything that disables preemption. | |
219 | * These may be nested. | |
f0a0e6f2 PM |
220 | * |
221 | * See the description of call_rcu() for more detailed information on | |
222 | * memory ordering guarantees. | |
2c42818e | 223 | */ |
584dc4ce | 224 | void call_rcu_sched(struct rcu_head *head, |
b6a4ae76 | 225 | rcu_callback_t func); |
2c42818e | 226 | |
584dc4ce | 227 | void synchronize_sched(void); |
03b042bf | 228 | |
ee376dbd PM |
229 | /* |
230 | * Structure allowing asynchronous waiting on RCU. | |
231 | */ | |
232 | struct rcu_synchronize { | |
233 | struct rcu_head head; | |
234 | struct completion completion; | |
235 | }; | |
236 | void wakeme_after_rcu(struct rcu_head *head); | |
237 | ||
ec90a194 PM |
238 | void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array, |
239 | struct rcu_synchronize *rs_array); | |
240 | ||
241 | #define _wait_rcu_gp(checktiny, ...) \ | |
66e8c57d ON |
242 | do { \ |
243 | call_rcu_func_t __crcu_array[] = { __VA_ARGS__ }; \ | |
244 | struct rcu_synchronize __rs_array[ARRAY_SIZE(__crcu_array)]; \ | |
245 | __wait_rcu_gp(checktiny, ARRAY_SIZE(__crcu_array), \ | |
246 | __crcu_array, __rs_array); \ | |
ec90a194 PM |
247 | } while (0) |
248 | ||
249 | #define wait_rcu_gp(...) _wait_rcu_gp(false, __VA_ARGS__) | |
250 | ||
251 | /** | |
252 | * synchronize_rcu_mult - Wait concurrently for multiple grace periods | |
253 | * @...: List of call_rcu() functions for the flavors to wait on. | |
254 | * | |
255 | * This macro waits concurrently for multiple flavors of RCU grace periods. | |
256 | * For example, synchronize_rcu_mult(call_rcu, call_rcu_bh) would wait | |
257 | * on concurrent RCU and RCU-bh grace periods. Waiting on a give SRCU | |
258 | * domain requires you to write a wrapper function for that SRCU domain's | |
259 | * call_srcu() function, supplying the corresponding srcu_struct. | |
260 | * | |
261 | * If Tiny RCU, tell _wait_rcu_gp() not to bother waiting for RCU | |
262 | * or RCU-bh, given that anywhere synchronize_rcu_mult() can be called | |
263 | * is automatically a grace period. | |
264 | */ | |
265 | #define synchronize_rcu_mult(...) \ | |
266 | _wait_rcu_gp(IS_ENABLED(CONFIG_TINY_RCU), __VA_ARGS__) | |
267 | ||
8315f422 PM |
268 | /** |
269 | * call_rcu_tasks() - Queue an RCU for invocation task-based grace period | |
270 | * @head: structure to be used for queueing the RCU updates. | |
271 | * @func: actual callback function to be invoked after the grace period | |
272 | * | |
273 | * The callback function will be invoked some time after a full grace | |
274 | * period elapses, in other words after all currently executing RCU | |
275 | * read-side critical sections have completed. call_rcu_tasks() assumes | |
276 | * that the read-side critical sections end at a voluntary context | |
277 | * switch (not a preemption!), entry into idle, or transition to usermode | |
278 | * execution. As such, there are no read-side primitives analogous to | |
279 | * rcu_read_lock() and rcu_read_unlock() because this primitive is intended | |
280 | * to determine that all tasks have passed through a safe state, not so | |
281 | * much for data-strcuture synchronization. | |
282 | * | |
283 | * See the description of call_rcu() for more detailed information on | |
284 | * memory ordering guarantees. | |
285 | */ | |
b6a4ae76 | 286 | void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); |
53c6d4ed PM |
287 | void synchronize_rcu_tasks(void); |
288 | void rcu_barrier_tasks(void); | |
8315f422 | 289 | |
a3dc3fb1 PM |
290 | #ifdef CONFIG_PREEMPT_RCU |
291 | ||
584dc4ce TB |
292 | void __rcu_read_lock(void); |
293 | void __rcu_read_unlock(void); | |
294 | void rcu_read_unlock_special(struct task_struct *t); | |
7b0b759b PM |
295 | void synchronize_rcu(void); |
296 | ||
a3dc3fb1 PM |
297 | /* |
298 | * Defined as a macro as it is a very low level header included from | |
299 | * areas that don't even know about current. This gives the rcu_read_lock() | |
300 | * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other | |
301 | * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. | |
302 | */ | |
303 | #define rcu_preempt_depth() (current->rcu_read_lock_nesting) | |
304 | ||
7b0b759b PM |
305 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
306 | ||
307 | static inline void __rcu_read_lock(void) | |
308 | { | |
bb73c52b BF |
309 | if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) |
310 | preempt_disable(); | |
7b0b759b PM |
311 | } |
312 | ||
313 | static inline void __rcu_read_unlock(void) | |
314 | { | |
bb73c52b BF |
315 | if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) |
316 | preempt_enable(); | |
7b0b759b PM |
317 | } |
318 | ||
319 | static inline void synchronize_rcu(void) | |
320 | { | |
321 | synchronize_sched(); | |
322 | } | |
323 | ||
324 | static inline int rcu_preempt_depth(void) | |
325 | { | |
326 | return 0; | |
327 | } | |
328 | ||
329 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
330 | ||
331 | /* Internal to kernel */ | |
584dc4ce | 332 | void rcu_init(void); |
284a8c93 PM |
333 | void rcu_sched_qs(void); |
334 | void rcu_bh_qs(void); | |
c3377c2d | 335 | void rcu_check_callbacks(int user); |
27d50c7e | 336 | void rcu_report_dead(unsigned int cpu); |
7ec99de3 | 337 | void rcu_cpu_starting(unsigned int cpu); |
2b1d5024 | 338 | |
79cfea02 PM |
339 | #ifndef CONFIG_TINY_RCU |
340 | void rcu_end_inkernel_boot(void); | |
341 | #else /* #ifndef CONFIG_TINY_RCU */ | |
342 | static inline void rcu_end_inkernel_boot(void) { } | |
343 | #endif /* #ifndef CONFIG_TINY_RCU */ | |
344 | ||
61f38db3 RR |
345 | #ifdef CONFIG_RCU_STALL_COMMON |
346 | void rcu_sysrq_start(void); | |
347 | void rcu_sysrq_end(void); | |
348 | #else /* #ifdef CONFIG_RCU_STALL_COMMON */ | |
349 | static inline void rcu_sysrq_start(void) | |
350 | { | |
351 | } | |
352 | static inline void rcu_sysrq_end(void) | |
353 | { | |
354 | } | |
355 | #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ | |
356 | ||
d1ec4c34 | 357 | #ifdef CONFIG_NO_HZ_FULL |
584dc4ce TB |
358 | void rcu_user_enter(void); |
359 | void rcu_user_exit(void); | |
2b1d5024 FW |
360 | #else |
361 | static inline void rcu_user_enter(void) { } | |
362 | static inline void rcu_user_exit(void) { } | |
d1ec4c34 | 363 | #endif /* CONFIG_NO_HZ_FULL */ |
2b1d5024 | 364 | |
f4579fc5 PM |
365 | #ifdef CONFIG_RCU_NOCB_CPU |
366 | void rcu_init_nohz(void); | |
367 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
368 | static inline void rcu_init_nohz(void) | |
369 | { | |
370 | } | |
371 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
372 | ||
8a2ecf47 PM |
373 | /** |
374 | * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers | |
375 | * @a: Code that RCU needs to pay attention to. | |
376 | * | |
377 | * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden | |
378 | * in the inner idle loop, that is, between the rcu_idle_enter() and | |
379 | * the rcu_idle_exit() -- RCU will happily ignore any such read-side | |
380 | * critical sections. However, things like powertop need tracepoints | |
381 | * in the inner idle loop. | |
382 | * | |
383 | * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) | |
810ce8b5 PM |
384 | * will tell RCU that it needs to pay attention, invoke its argument |
385 | * (in this example, calling the do_something_with_RCU() function), | |
8a2ecf47 | 386 | * and then tell RCU to go back to ignoring this CPU. It is permissible |
810ce8b5 PM |
387 | * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is |
388 | * on the order of a million or so, even on 32-bit systems). It is | |
389 | * not legal to block within RCU_NONIDLE(), nor is it permissible to | |
390 | * transfer control either into or out of RCU_NONIDLE()'s statement. | |
8a2ecf47 PM |
391 | */ |
392 | #define RCU_NONIDLE(a) \ | |
393 | do { \ | |
7c9906ca | 394 | rcu_irq_enter_irqson(); \ |
8a2ecf47 | 395 | do { a; } while (0); \ |
7c9906ca | 396 | rcu_irq_exit_irqson(); \ |
8a2ecf47 PM |
397 | } while (0) |
398 | ||
8315f422 PM |
399 | /* |
400 | * Note a voluntary context switch for RCU-tasks benefit. This is a | |
401 | * macro rather than an inline function to avoid #include hell. | |
402 | */ | |
403 | #ifdef CONFIG_TASKS_RCU | |
3f95aa81 PM |
404 | #define TASKS_RCU(x) x |
405 | extern struct srcu_struct tasks_rcu_exit_srcu; | |
8315f422 PM |
406 | #define rcu_note_voluntary_context_switch(t) \ |
407 | do { \ | |
5cd37193 | 408 | rcu_all_qs(); \ |
7d0ae808 PM |
409 | if (READ_ONCE((t)->rcu_tasks_holdout)) \ |
410 | WRITE_ONCE((t)->rcu_tasks_holdout, false); \ | |
8315f422 PM |
411 | } while (0) |
412 | #else /* #ifdef CONFIG_TASKS_RCU */ | |
3f95aa81 | 413 | #define TASKS_RCU(x) do { } while (0) |
5cd37193 | 414 | #define rcu_note_voluntary_context_switch(t) rcu_all_qs() |
8315f422 PM |
415 | #endif /* #else #ifdef CONFIG_TASKS_RCU */ |
416 | ||
bde6c3aa PM |
417 | /** |
418 | * cond_resched_rcu_qs - Report potential quiescent states to RCU | |
419 | * | |
420 | * This macro resembles cond_resched(), except that it is defined to | |
421 | * report potential quiescent states to RCU-tasks even if the cond_resched() | |
422 | * machinery were to be shut off, as some advocate for PREEMPT kernels. | |
423 | */ | |
424 | #define cond_resched_rcu_qs() \ | |
425 | do { \ | |
b6331ae8 PM |
426 | if (!cond_resched()) \ |
427 | rcu_note_voluntary_context_switch(current); \ | |
bde6c3aa PM |
428 | } while (0) |
429 | ||
cc6783f7 | 430 | #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) |
584dc4ce | 431 | bool __rcu_is_watching(void); |
cc6783f7 PM |
432 | #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */ |
433 | ||
2c42818e PM |
434 | /* |
435 | * Infrastructure to implement the synchronize_() primitives in | |
436 | * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. | |
437 | */ | |
438 | ||
28f6569a | 439 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) |
64db4cff | 440 | #include <linux/rcutree.h> |
127781d1 | 441 | #elif defined(CONFIG_TINY_RCU) |
9b1d82fa | 442 | #include <linux/rcutiny.h> |
64db4cff PM |
443 | #else |
444 | #error "Unknown RCU implementation specified to kernel configuration" | |
6b3ef48a | 445 | #endif |
01c1c660 | 446 | |
551d55a9 MD |
447 | /* |
448 | * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic | |
449 | * initialization and destruction of rcu_head on the stack. rcu_head structures | |
450 | * allocated dynamically in the heap or defined statically don't need any | |
451 | * initialization. | |
452 | */ | |
453 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
546a9d85 PM |
454 | void init_rcu_head(struct rcu_head *head); |
455 | void destroy_rcu_head(struct rcu_head *head); | |
584dc4ce TB |
456 | void init_rcu_head_on_stack(struct rcu_head *head); |
457 | void destroy_rcu_head_on_stack(struct rcu_head *head); | |
551d55a9 | 458 | #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
546a9d85 PM |
459 | static inline void init_rcu_head(struct rcu_head *head) |
460 | { | |
461 | } | |
462 | ||
463 | static inline void destroy_rcu_head(struct rcu_head *head) | |
464 | { | |
465 | } | |
466 | ||
4376030a MD |
467 | static inline void init_rcu_head_on_stack(struct rcu_head *head) |
468 | { | |
469 | } | |
470 | ||
471 | static inline void destroy_rcu_head_on_stack(struct rcu_head *head) | |
472 | { | |
473 | } | |
551d55a9 | 474 | #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
4376030a | 475 | |
c0d6d01b PM |
476 | #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) |
477 | bool rcu_lockdep_current_cpu_online(void); | |
478 | #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
479 | static inline bool rcu_lockdep_current_cpu_online(void) | |
480 | { | |
521d24ee | 481 | return true; |
c0d6d01b PM |
482 | } |
483 | #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
484 | ||
bc33f24b | 485 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
632ee200 | 486 | |
00f49e57 FW |
487 | static inline void rcu_lock_acquire(struct lockdep_map *map) |
488 | { | |
fb9edbe9 | 489 | lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); |
00f49e57 FW |
490 | } |
491 | ||
492 | static inline void rcu_lock_release(struct lockdep_map *map) | |
493 | { | |
00f49e57 FW |
494 | lock_release(map, 1, _THIS_IP_); |
495 | } | |
496 | ||
bc33f24b | 497 | extern struct lockdep_map rcu_lock_map; |
632ee200 | 498 | extern struct lockdep_map rcu_bh_lock_map; |
632ee200 | 499 | extern struct lockdep_map rcu_sched_lock_map; |
24ef659a | 500 | extern struct lockdep_map rcu_callback_map; |
a235c091 | 501 | int debug_lockdep_rcu_enabled(void); |
54dbf96c | 502 | |
85b39d30 | 503 | int rcu_read_lock_held(void); |
584dc4ce | 504 | int rcu_read_lock_bh_held(void); |
632ee200 PM |
505 | |
506 | /** | |
ca5ecddf | 507 | * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? |
632ee200 | 508 | * |
d20200b5 PM |
509 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an |
510 | * RCU-sched read-side critical section. In absence of | |
511 | * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side | |
d5671f6b | 512 | * critical section unless it can prove otherwise. |
632ee200 | 513 | */ |
d5671f6b | 514 | int rcu_read_lock_sched_held(void); |
632ee200 PM |
515 | |
516 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
517 | ||
d8ab29f8 PM |
518 | # define rcu_lock_acquire(a) do { } while (0) |
519 | # define rcu_lock_release(a) do { } while (0) | |
632ee200 PM |
520 | |
521 | static inline int rcu_read_lock_held(void) | |
522 | { | |
523 | return 1; | |
524 | } | |
525 | ||
526 | static inline int rcu_read_lock_bh_held(void) | |
527 | { | |
528 | return 1; | |
529 | } | |
530 | ||
531 | static inline int rcu_read_lock_sched_held(void) | |
532 | { | |
293e2421 | 533 | return !preemptible(); |
632ee200 | 534 | } |
632ee200 PM |
535 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ |
536 | ||
537 | #ifdef CONFIG_PROVE_RCU | |
538 | ||
f78f5b90 PM |
539 | /** |
540 | * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met | |
541 | * @c: condition to check | |
542 | * @s: informative message | |
543 | */ | |
544 | #define RCU_LOCKDEP_WARN(c, s) \ | |
545 | do { \ | |
546 | static bool __section(.data.unlikely) __warned; \ | |
547 | if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ | |
548 | __warned = true; \ | |
549 | lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ | |
550 | } \ | |
551 | } while (0) | |
552 | ||
50406b98 PM |
553 | #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) |
554 | static inline void rcu_preempt_sleep_check(void) | |
555 | { | |
f78f5b90 PM |
556 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), |
557 | "Illegal context switch in RCU read-side critical section"); | |
50406b98 PM |
558 | } |
559 | #else /* #ifdef CONFIG_PROVE_RCU */ | |
560 | static inline void rcu_preempt_sleep_check(void) | |
561 | { | |
562 | } | |
563 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
564 | ||
b3fbab05 PM |
565 | #define rcu_sleep_check() \ |
566 | do { \ | |
50406b98 | 567 | rcu_preempt_sleep_check(); \ |
f78f5b90 PM |
568 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ |
569 | "Illegal context switch in RCU-bh read-side critical section"); \ | |
570 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ | |
571 | "Illegal context switch in RCU-sched read-side critical section"); \ | |
b3fbab05 PM |
572 | } while (0) |
573 | ||
ca5ecddf PM |
574 | #else /* #ifdef CONFIG_PROVE_RCU */ |
575 | ||
f78f5b90 | 576 | #define RCU_LOCKDEP_WARN(c, s) do { } while (0) |
b3fbab05 | 577 | #define rcu_sleep_check() do { } while (0) |
ca5ecddf PM |
578 | |
579 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
580 | ||
581 | /* | |
582 | * Helper functions for rcu_dereference_check(), rcu_dereference_protected() | |
583 | * and rcu_assign_pointer(). Some of these could be folded into their | |
584 | * callers, but they are left separate in order to ease introduction of | |
585 | * multiple flavors of pointers to match the multiple flavors of RCU | |
586 | * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in | |
587 | * the future. | |
588 | */ | |
53ecfba2 PM |
589 | |
590 | #ifdef __CHECKER__ | |
591 | #define rcu_dereference_sparse(p, space) \ | |
592 | ((void)(((typeof(*p) space *)p) == p)) | |
593 | #else /* #ifdef __CHECKER__ */ | |
594 | #define rcu_dereference_sparse(p, space) | |
595 | #endif /* #else #ifdef __CHECKER__ */ | |
596 | ||
ca5ecddf | 597 | #define __rcu_access_pointer(p, space) \ |
0adab9b9 | 598 | ({ \ |
7d0ae808 | 599 | typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ |
0adab9b9 JP |
600 | rcu_dereference_sparse(p, space); \ |
601 | ((typeof(*p) __force __kernel *)(_________p1)); \ | |
602 | }) | |
ca5ecddf | 603 | #define __rcu_dereference_check(p, c, space) \ |
0adab9b9 | 604 | ({ \ |
ac59853c PK |
605 | /* Dependency order vs. p above. */ \ |
606 | typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \ | |
f78f5b90 | 607 | RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ |
0adab9b9 | 608 | rcu_dereference_sparse(p, space); \ |
ac59853c | 609 | ((typeof(*p) __force __kernel *)(________p1)); \ |
0adab9b9 | 610 | }) |
ca5ecddf | 611 | #define __rcu_dereference_protected(p, c, space) \ |
0adab9b9 | 612 | ({ \ |
f78f5b90 | 613 | RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ |
0adab9b9 JP |
614 | rcu_dereference_sparse(p, space); \ |
615 | ((typeof(*p) __force __kernel *)(p)); \ | |
616 | }) | |
995f1405 PM |
617 | #define rcu_dereference_raw(p) \ |
618 | ({ \ | |
619 | /* Dependency order vs. p above. */ \ | |
620 | typeof(p) ________p1 = lockless_dereference(p); \ | |
621 | ((typeof(*p) __force __kernel *)(________p1)); \ | |
622 | }) | |
ca5ecddf | 623 | |
462225ae PM |
624 | /** |
625 | * RCU_INITIALIZER() - statically initialize an RCU-protected global variable | |
626 | * @v: The value to statically initialize with. | |
627 | */ | |
628 | #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) | |
629 | ||
630 | /** | |
631 | * rcu_assign_pointer() - assign to RCU-protected pointer | |
632 | * @p: pointer to assign to | |
633 | * @v: value to assign (publish) | |
634 | * | |
635 | * Assigns the specified value to the specified RCU-protected | |
636 | * pointer, ensuring that any concurrent RCU readers will see | |
637 | * any prior initialization. | |
638 | * | |
639 | * Inserts memory barriers on architectures that require them | |
640 | * (which is most of them), and also prevents the compiler from | |
641 | * reordering the code that initializes the structure after the pointer | |
642 | * assignment. More importantly, this call documents which pointers | |
643 | * will be dereferenced by RCU read-side code. | |
644 | * | |
645 | * In some special cases, you may use RCU_INIT_POINTER() instead | |
646 | * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due | |
647 | * to the fact that it does not constrain either the CPU or the compiler. | |
648 | * That said, using RCU_INIT_POINTER() when you should have used | |
649 | * rcu_assign_pointer() is a very bad thing that results in | |
650 | * impossible-to-diagnose memory corruption. So please be careful. | |
651 | * See the RCU_INIT_POINTER() comment header for details. | |
652 | * | |
653 | * Note that rcu_assign_pointer() evaluates each of its arguments only | |
654 | * once, appearances notwithstanding. One of the "extra" evaluations | |
655 | * is in typeof() and the other visible only to sparse (__CHECKER__), | |
656 | * neither of which actually execute the argument. As with most cpp | |
657 | * macros, this execute-arguments-only-once property is important, so | |
658 | * please be careful when making changes to rcu_assign_pointer() and the | |
659 | * other macros that it invokes. | |
660 | */ | |
3a37f727 PM |
661 | #define rcu_assign_pointer(p, v) \ |
662 | ({ \ | |
663 | uintptr_t _r_a_p__v = (uintptr_t)(v); \ | |
664 | \ | |
665 | if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ | |
666 | WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ | |
667 | else \ | |
668 | smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ | |
669 | _r_a_p__v; \ | |
670 | }) | |
ca5ecddf PM |
671 | |
672 | /** | |
673 | * rcu_access_pointer() - fetch RCU pointer with no dereferencing | |
674 | * @p: The pointer to read | |
675 | * | |
676 | * Return the value of the specified RCU-protected pointer, but omit the | |
7d0ae808 | 677 | * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful |
ca5ecddf PM |
678 | * when the value of this pointer is accessed, but the pointer is not |
679 | * dereferenced, for example, when testing an RCU-protected pointer against | |
680 | * NULL. Although rcu_access_pointer() may also be used in cases where | |
681 | * update-side locks prevent the value of the pointer from changing, you | |
682 | * should instead use rcu_dereference_protected() for this use case. | |
5e1ee6e1 PM |
683 | * |
684 | * It is also permissible to use rcu_access_pointer() when read-side | |
685 | * access to the pointer was removed at least one grace period ago, as | |
686 | * is the case in the context of the RCU callback that is freeing up | |
687 | * the data, or after a synchronize_rcu() returns. This can be useful | |
688 | * when tearing down multi-linked structures after a grace period | |
689 | * has elapsed. | |
ca5ecddf PM |
690 | */ |
691 | #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) | |
692 | ||
632ee200 | 693 | /** |
ca5ecddf | 694 | * rcu_dereference_check() - rcu_dereference with debug checking |
c08c68dd DH |
695 | * @p: The pointer to read, prior to dereferencing |
696 | * @c: The conditions under which the dereference will take place | |
632ee200 | 697 | * |
c08c68dd | 698 | * Do an rcu_dereference(), but check that the conditions under which the |
ca5ecddf PM |
699 | * dereference will take place are correct. Typically the conditions |
700 | * indicate the various locking conditions that should be held at that | |
701 | * point. The check should return true if the conditions are satisfied. | |
702 | * An implicit check for being in an RCU read-side critical section | |
703 | * (rcu_read_lock()) is included. | |
c08c68dd DH |
704 | * |
705 | * For example: | |
706 | * | |
ca5ecddf | 707 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); |
c08c68dd DH |
708 | * |
709 | * could be used to indicate to lockdep that foo->bar may only be dereferenced | |
ca5ecddf | 710 | * if either rcu_read_lock() is held, or that the lock required to replace |
c08c68dd DH |
711 | * the bar struct at foo->bar is held. |
712 | * | |
713 | * Note that the list of conditions may also include indications of when a lock | |
714 | * need not be held, for example during initialisation or destruction of the | |
715 | * target struct: | |
716 | * | |
ca5ecddf | 717 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || |
c08c68dd | 718 | * atomic_read(&foo->usage) == 0); |
ca5ecddf PM |
719 | * |
720 | * Inserts memory barriers on architectures that require them | |
721 | * (currently only the Alpha), prevents the compiler from refetching | |
722 | * (and from merging fetches), and, more importantly, documents exactly | |
723 | * which pointers are protected by RCU and checks that the pointer is | |
724 | * annotated as __rcu. | |
632ee200 PM |
725 | */ |
726 | #define rcu_dereference_check(p, c) \ | |
b826565a | 727 | __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) |
ca5ecddf PM |
728 | |
729 | /** | |
730 | * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking | |
731 | * @p: The pointer to read, prior to dereferencing | |
732 | * @c: The conditions under which the dereference will take place | |
733 | * | |
734 | * This is the RCU-bh counterpart to rcu_dereference_check(). | |
735 | */ | |
736 | #define rcu_dereference_bh_check(p, c) \ | |
b826565a | 737 | __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) |
632ee200 | 738 | |
b62730ba | 739 | /** |
ca5ecddf PM |
740 | * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking |
741 | * @p: The pointer to read, prior to dereferencing | |
742 | * @c: The conditions under which the dereference will take place | |
743 | * | |
744 | * This is the RCU-sched counterpart to rcu_dereference_check(). | |
745 | */ | |
746 | #define rcu_dereference_sched_check(p, c) \ | |
b826565a | 747 | __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ |
ca5ecddf PM |
748 | __rcu) |
749 | ||
12bcbe66 SR |
750 | /* |
751 | * The tracing infrastructure traces RCU (we want that), but unfortunately | |
752 | * some of the RCU checks causes tracing to lock up the system. | |
753 | * | |
f039f0af | 754 | * The no-tracing version of rcu_dereference_raw() must not call |
12bcbe66 SR |
755 | * rcu_read_lock_held(). |
756 | */ | |
757 | #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) | |
758 | ||
ca5ecddf PM |
759 | /** |
760 | * rcu_dereference_protected() - fetch RCU pointer when updates prevented | |
761 | * @p: The pointer to read, prior to dereferencing | |
762 | * @c: The conditions under which the dereference will take place | |
b62730ba PM |
763 | * |
764 | * Return the value of the specified RCU-protected pointer, but omit | |
7d0ae808 | 765 | * both the smp_read_barrier_depends() and the READ_ONCE(). This |
b62730ba PM |
766 | * is useful in cases where update-side locks prevent the value of the |
767 | * pointer from changing. Please note that this primitive does -not- | |
768 | * prevent the compiler from repeating this reference or combining it | |
769 | * with other references, so it should not be used without protection | |
770 | * of appropriate locks. | |
ca5ecddf PM |
771 | * |
772 | * This function is only for update-side use. Using this function | |
773 | * when protected only by rcu_read_lock() will result in infrequent | |
774 | * but very ugly failures. | |
b62730ba PM |
775 | */ |
776 | #define rcu_dereference_protected(p, c) \ | |
ca5ecddf | 777 | __rcu_dereference_protected((p), (c), __rcu) |
b62730ba | 778 | |
bc33f24b | 779 | |
b62730ba | 780 | /** |
ca5ecddf PM |
781 | * rcu_dereference() - fetch RCU-protected pointer for dereferencing |
782 | * @p: The pointer to read, prior to dereferencing | |
b62730ba | 783 | * |
ca5ecddf | 784 | * This is a simple wrapper around rcu_dereference_check(). |
b62730ba | 785 | */ |
ca5ecddf | 786 | #define rcu_dereference(p) rcu_dereference_check(p, 0) |
b62730ba | 787 | |
1da177e4 | 788 | /** |
ca5ecddf PM |
789 | * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing |
790 | * @p: The pointer to read, prior to dereferencing | |
791 | * | |
792 | * Makes rcu_dereference_check() do the dirty work. | |
793 | */ | |
794 | #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) | |
795 | ||
796 | /** | |
797 | * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing | |
798 | * @p: The pointer to read, prior to dereferencing | |
799 | * | |
800 | * Makes rcu_dereference_check() do the dirty work. | |
801 | */ | |
802 | #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) | |
803 | ||
c3ac7cf1 PM |
804 | /** |
805 | * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism | |
806 | * @p: The pointer to hand off | |
807 | * | |
808 | * This is simply an identity function, but it documents where a pointer | |
809 | * is handed off from RCU to some other synchronization mechanism, for | |
810 | * example, reference counting or locking. In C11, it would map to | |
811 | * kill_dependency(). It could be used as follows: | |
812 | * | |
813 | * rcu_read_lock(); | |
814 | * p = rcu_dereference(gp); | |
815 | * long_lived = is_long_lived(p); | |
816 | * if (long_lived) { | |
817 | * if (!atomic_inc_not_zero(p->refcnt)) | |
818 | * long_lived = false; | |
819 | * else | |
820 | * p = rcu_pointer_handoff(p); | |
821 | * } | |
822 | * rcu_read_unlock(); | |
823 | */ | |
824 | #define rcu_pointer_handoff(p) (p) | |
825 | ||
ca5ecddf PM |
826 | /** |
827 | * rcu_read_lock() - mark the beginning of an RCU read-side critical section | |
1da177e4 | 828 | * |
9b06e818 | 829 | * When synchronize_rcu() is invoked on one CPU while other CPUs |
1da177e4 | 830 | * are within RCU read-side critical sections, then the |
9b06e818 | 831 | * synchronize_rcu() is guaranteed to block until after all the other |
1da177e4 LT |
832 | * CPUs exit their critical sections. Similarly, if call_rcu() is invoked |
833 | * on one CPU while other CPUs are within RCU read-side critical | |
834 | * sections, invocation of the corresponding RCU callback is deferred | |
835 | * until after the all the other CPUs exit their critical sections. | |
836 | * | |
837 | * Note, however, that RCU callbacks are permitted to run concurrently | |
77d8485a | 838 | * with new RCU read-side critical sections. One way that this can happen |
1da177e4 LT |
839 | * is via the following sequence of events: (1) CPU 0 enters an RCU |
840 | * read-side critical section, (2) CPU 1 invokes call_rcu() to register | |
841 | * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, | |
842 | * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU | |
843 | * callback is invoked. This is legal, because the RCU read-side critical | |
844 | * section that was running concurrently with the call_rcu() (and which | |
845 | * therefore might be referencing something that the corresponding RCU | |
846 | * callback would free up) has completed before the corresponding | |
847 | * RCU callback is invoked. | |
848 | * | |
849 | * RCU read-side critical sections may be nested. Any deferred actions | |
850 | * will be deferred until the outermost RCU read-side critical section | |
851 | * completes. | |
852 | * | |
9079fd7c PM |
853 | * You can avoid reading and understanding the next paragraph by |
854 | * following this rule: don't put anything in an rcu_read_lock() RCU | |
855 | * read-side critical section that would block in a !PREEMPT kernel. | |
856 | * But if you want the full story, read on! | |
857 | * | |
ab74fdfd PM |
858 | * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), |
859 | * it is illegal to block while in an RCU read-side critical section. | |
28f6569a | 860 | * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT |
ab74fdfd PM |
861 | * kernel builds, RCU read-side critical sections may be preempted, |
862 | * but explicit blocking is illegal. Finally, in preemptible RCU | |
863 | * implementations in real-time (with -rt patchset) kernel builds, RCU | |
864 | * read-side critical sections may be preempted and they may also block, but | |
865 | * only when acquiring spinlocks that are subject to priority inheritance. | |
1da177e4 | 866 | */ |
bc33f24b PM |
867 | static inline void rcu_read_lock(void) |
868 | { | |
869 | __rcu_read_lock(); | |
870 | __acquire(RCU); | |
d8ab29f8 | 871 | rcu_lock_acquire(&rcu_lock_map); |
f78f5b90 PM |
872 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
873 | "rcu_read_lock() used illegally while idle"); | |
bc33f24b | 874 | } |
1da177e4 | 875 | |
1da177e4 LT |
876 | /* |
877 | * So where is rcu_write_lock()? It does not exist, as there is no | |
878 | * way for writers to lock out RCU readers. This is a feature, not | |
879 | * a bug -- this property is what provides RCU's performance benefits. | |
880 | * Of course, writers must coordinate with each other. The normal | |
881 | * spinlock primitives work well for this, but any other technique may be | |
882 | * used as well. RCU does not care how the writers keep out of each | |
883 | * others' way, as long as they do so. | |
884 | */ | |
3d76c082 PM |
885 | |
886 | /** | |
ca5ecddf | 887 | * rcu_read_unlock() - marks the end of an RCU read-side critical section. |
3d76c082 | 888 | * |
f27bc487 PM |
889 | * In most situations, rcu_read_unlock() is immune from deadlock. |
890 | * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() | |
891 | * is responsible for deboosting, which it does via rt_mutex_unlock(). | |
892 | * Unfortunately, this function acquires the scheduler's runqueue and | |
893 | * priority-inheritance spinlocks. This means that deadlock could result | |
894 | * if the caller of rcu_read_unlock() already holds one of these locks or | |
ce36f2f3 ON |
895 | * any lock that is ever acquired while holding them; or any lock which |
896 | * can be taken from interrupt context because rcu_boost()->rt_mutex_lock() | |
897 | * does not disable irqs while taking ->wait_lock. | |
f27bc487 PM |
898 | * |
899 | * That said, RCU readers are never priority boosted unless they were | |
900 | * preempted. Therefore, one way to avoid deadlock is to make sure | |
901 | * that preemption never happens within any RCU read-side critical | |
902 | * section whose outermost rcu_read_unlock() is called with one of | |
903 | * rt_mutex_unlock()'s locks held. Such preemption can be avoided in | |
904 | * a number of ways, for example, by invoking preempt_disable() before | |
905 | * critical section's outermost rcu_read_lock(). | |
906 | * | |
907 | * Given that the set of locks acquired by rt_mutex_unlock() might change | |
908 | * at any time, a somewhat more future-proofed approach is to make sure | |
909 | * that that preemption never happens within any RCU read-side critical | |
910 | * section whose outermost rcu_read_unlock() is called with irqs disabled. | |
911 | * This approach relies on the fact that rt_mutex_unlock() currently only | |
912 | * acquires irq-disabled locks. | |
913 | * | |
914 | * The second of these two approaches is best in most situations, | |
915 | * however, the first approach can also be useful, at least to those | |
916 | * developers willing to keep abreast of the set of locks acquired by | |
917 | * rt_mutex_unlock(). | |
918 | * | |
3d76c082 PM |
919 | * See rcu_read_lock() for more information. |
920 | */ | |
bc33f24b PM |
921 | static inline void rcu_read_unlock(void) |
922 | { | |
f78f5b90 PM |
923 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
924 | "rcu_read_unlock() used illegally while idle"); | |
bc33f24b PM |
925 | __release(RCU); |
926 | __rcu_read_unlock(); | |
d24209bb | 927 | rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ |
bc33f24b | 928 | } |
1da177e4 LT |
929 | |
930 | /** | |
ca5ecddf | 931 | * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section |
1da177e4 LT |
932 | * |
933 | * This is equivalent of rcu_read_lock(), but to be used when updates | |
ca5ecddf PM |
934 | * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since |
935 | * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a | |
936 | * softirq handler to be a quiescent state, a process in RCU read-side | |
937 | * critical section must be protected by disabling softirqs. Read-side | |
938 | * critical sections in interrupt context can use just rcu_read_lock(), | |
939 | * though this should at least be commented to avoid confusing people | |
940 | * reading the code. | |
3842a083 PM |
941 | * |
942 | * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() | |
943 | * must occur in the same context, for example, it is illegal to invoke | |
944 | * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() | |
945 | * was invoked from some other task. | |
1da177e4 | 946 | */ |
bc33f24b PM |
947 | static inline void rcu_read_lock_bh(void) |
948 | { | |
6206ab9b | 949 | local_bh_disable(); |
bc33f24b | 950 | __acquire(RCU_BH); |
d8ab29f8 | 951 | rcu_lock_acquire(&rcu_bh_lock_map); |
f78f5b90 PM |
952 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
953 | "rcu_read_lock_bh() used illegally while idle"); | |
bc33f24b | 954 | } |
1da177e4 LT |
955 | |
956 | /* | |
957 | * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section | |
958 | * | |
959 | * See rcu_read_lock_bh() for more information. | |
960 | */ | |
bc33f24b PM |
961 | static inline void rcu_read_unlock_bh(void) |
962 | { | |
f78f5b90 PM |
963 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
964 | "rcu_read_unlock_bh() used illegally while idle"); | |
d8ab29f8 | 965 | rcu_lock_release(&rcu_bh_lock_map); |
bc33f24b | 966 | __release(RCU_BH); |
6206ab9b | 967 | local_bh_enable(); |
bc33f24b | 968 | } |
1da177e4 | 969 | |
1c50b728 | 970 | /** |
ca5ecddf | 971 | * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section |
1c50b728 | 972 | * |
ca5ecddf PM |
973 | * This is equivalent of rcu_read_lock(), but to be used when updates |
974 | * are being done using call_rcu_sched() or synchronize_rcu_sched(). | |
975 | * Read-side critical sections can also be introduced by anything that | |
976 | * disables preemption, including local_irq_disable() and friends. | |
3842a083 PM |
977 | * |
978 | * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() | |
979 | * must occur in the same context, for example, it is illegal to invoke | |
980 | * rcu_read_unlock_sched() from process context if the matching | |
981 | * rcu_read_lock_sched() was invoked from an NMI handler. | |
1c50b728 | 982 | */ |
d6714c22 PM |
983 | static inline void rcu_read_lock_sched(void) |
984 | { | |
985 | preempt_disable(); | |
bc33f24b | 986 | __acquire(RCU_SCHED); |
d8ab29f8 | 987 | rcu_lock_acquire(&rcu_sched_lock_map); |
f78f5b90 PM |
988 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
989 | "rcu_read_lock_sched() used illegally while idle"); | |
d6714c22 | 990 | } |
1eba8f84 PM |
991 | |
992 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 993 | static inline notrace void rcu_read_lock_sched_notrace(void) |
d6714c22 PM |
994 | { |
995 | preempt_disable_notrace(); | |
bc33f24b | 996 | __acquire(RCU_SCHED); |
d6714c22 | 997 | } |
1c50b728 MD |
998 | |
999 | /* | |
1000 | * rcu_read_unlock_sched - marks the end of a RCU-classic critical section | |
1001 | * | |
1002 | * See rcu_read_lock_sched for more information. | |
1003 | */ | |
d6714c22 PM |
1004 | static inline void rcu_read_unlock_sched(void) |
1005 | { | |
f78f5b90 PM |
1006 | RCU_LOCKDEP_WARN(!rcu_is_watching(), |
1007 | "rcu_read_unlock_sched() used illegally while idle"); | |
d8ab29f8 | 1008 | rcu_lock_release(&rcu_sched_lock_map); |
bc33f24b | 1009 | __release(RCU_SCHED); |
d6714c22 PM |
1010 | preempt_enable(); |
1011 | } | |
1eba8f84 PM |
1012 | |
1013 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 1014 | static inline notrace void rcu_read_unlock_sched_notrace(void) |
d6714c22 | 1015 | { |
bc33f24b | 1016 | __release(RCU_SCHED); |
d6714c22 PM |
1017 | preempt_enable_notrace(); |
1018 | } | |
1c50b728 | 1019 | |
ca5ecddf PM |
1020 | /** |
1021 | * RCU_INIT_POINTER() - initialize an RCU protected pointer | |
1022 | * | |
6846c0c5 PM |
1023 | * Initialize an RCU-protected pointer in special cases where readers |
1024 | * do not need ordering constraints on the CPU or the compiler. These | |
1025 | * special cases are: | |
1026 | * | |
1027 | * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- | |
1028 | * 2. The caller has taken whatever steps are required to prevent | |
1029 | * RCU readers from concurrently accessing this pointer -or- | |
1030 | * 3. The referenced data structure has already been exposed to | |
1031 | * readers either at compile time or via rcu_assign_pointer() -and- | |
1032 | * a. You have not made -any- reader-visible changes to | |
1033 | * this structure since then -or- | |
1034 | * b. It is OK for readers accessing this structure from its | |
1035 | * new location to see the old state of the structure. (For | |
1036 | * example, the changes were to statistical counters or to | |
1037 | * other state where exact synchronization is not required.) | |
1038 | * | |
1039 | * Failure to follow these rules governing use of RCU_INIT_POINTER() will | |
1040 | * result in impossible-to-diagnose memory corruption. As in the structures | |
1041 | * will look OK in crash dumps, but any concurrent RCU readers might | |
1042 | * see pre-initialized values of the referenced data structure. So | |
1043 | * please be very careful how you use RCU_INIT_POINTER()!!! | |
1044 | * | |
1045 | * If you are creating an RCU-protected linked structure that is accessed | |
1046 | * by a single external-to-structure RCU-protected pointer, then you may | |
1047 | * use RCU_INIT_POINTER() to initialize the internal RCU-protected | |
1048 | * pointers, but you must use rcu_assign_pointer() to initialize the | |
1049 | * external-to-structure pointer -after- you have completely initialized | |
1050 | * the reader-accessible portions of the linked structure. | |
71a9b269 PM |
1051 | * |
1052 | * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no | |
1053 | * ordering guarantees for either the CPU or the compiler. | |
ca5ecddf PM |
1054 | */ |
1055 | #define RCU_INIT_POINTER(p, v) \ | |
d1b88eb9 | 1056 | do { \ |
1a6c9b26 | 1057 | rcu_dereference_sparse(p, __rcu); \ |
155d1d12 | 1058 | WRITE_ONCE(p, RCU_INITIALIZER(v)); \ |
d1b88eb9 | 1059 | } while (0) |
9ab1544e | 1060 | |
172708d0 PM |
1061 | /** |
1062 | * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer | |
1063 | * | |
1064 | * GCC-style initialization for an RCU-protected pointer in a structure field. | |
1065 | */ | |
1066 | #define RCU_POINTER_INITIALIZER(p, v) \ | |
462225ae | 1067 | .p = RCU_INITIALIZER(v) |
9ab1544e | 1068 | |
d8169d4c JE |
1069 | /* |
1070 | * Does the specified offset indicate that the corresponding rcu_head | |
1071 | * structure can be handled by kfree_rcu()? | |
1072 | */ | |
1073 | #define __is_kfree_rcu_offset(offset) ((offset) < 4096) | |
1074 | ||
1075 | /* | |
1076 | * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. | |
1077 | */ | |
1078 | #define __kfree_rcu(head, offset) \ | |
1079 | do { \ | |
1080 | BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ | |
b6a4ae76 | 1081 | kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ |
d8169d4c JE |
1082 | } while (0) |
1083 | ||
9ab1544e LJ |
1084 | /** |
1085 | * kfree_rcu() - kfree an object after a grace period. | |
1086 | * @ptr: pointer to kfree | |
1087 | * @rcu_head: the name of the struct rcu_head within the type of @ptr. | |
1088 | * | |
1089 | * Many rcu callbacks functions just call kfree() on the base structure. | |
1090 | * These functions are trivial, but their size adds up, and furthermore | |
1091 | * when they are used in a kernel module, that module must invoke the | |
1092 | * high-latency rcu_barrier() function at module-unload time. | |
1093 | * | |
1094 | * The kfree_rcu() function handles this issue. Rather than encoding a | |
1095 | * function address in the embedded rcu_head structure, kfree_rcu() instead | |
1096 | * encodes the offset of the rcu_head structure within the base structure. | |
1097 | * Because the functions are not allowed in the low-order 4096 bytes of | |
1098 | * kernel virtual memory, offsets up to 4095 bytes can be accommodated. | |
1099 | * If the offset is larger than 4095 bytes, a compile-time error will | |
1100 | * be generated in __kfree_rcu(). If this error is triggered, you can | |
1101 | * either fall back to use of call_rcu() or rearrange the structure to | |
1102 | * position the rcu_head structure into the first 4096 bytes. | |
1103 | * | |
1104 | * Note that the allowable offset might decrease in the future, for example, | |
1105 | * to allow something like kmem_cache_free_rcu(). | |
d8169d4c JE |
1106 | * |
1107 | * The BUILD_BUG_ON check must not involve any function calls, hence the | |
1108 | * checks are done in macros here. | |
9ab1544e LJ |
1109 | */ |
1110 | #define kfree_rcu(ptr, rcu_head) \ | |
1111 | __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) | |
1112 | ||
3382adbc | 1113 | #ifdef CONFIG_TINY_RCU |
c1ad348b | 1114 | static inline int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
ffa83fb5 | 1115 | { |
c1ad348b | 1116 | *nextevt = KTIME_MAX; |
ffa83fb5 PM |
1117 | return 0; |
1118 | } | |
3382adbc | 1119 | #endif /* #ifdef CONFIG_TINY_RCU */ |
ffa83fb5 | 1120 | |
2f33b512 PM |
1121 | #if defined(CONFIG_RCU_NOCB_CPU_ALL) |
1122 | static inline bool rcu_is_nocb_cpu(int cpu) { return true; } | |
1123 | #elif defined(CONFIG_RCU_NOCB_CPU) | |
584dc4ce | 1124 | bool rcu_is_nocb_cpu(int cpu); |
d1e43fa5 FW |
1125 | #else |
1126 | static inline bool rcu_is_nocb_cpu(int cpu) { return false; } | |
2f33b512 | 1127 | #endif |
d1e43fa5 FW |
1128 | |
1129 | ||
0edd1b17 PM |
1130 | /* Only for use by adaptive-ticks code. */ |
1131 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
584dc4ce TB |
1132 | bool rcu_sys_is_idle(void); |
1133 | void rcu_sysidle_force_exit(void); | |
0edd1b17 PM |
1134 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ |
1135 | ||
1136 | static inline bool rcu_sys_is_idle(void) | |
1137 | { | |
1138 | return false; | |
1139 | } | |
1140 | ||
1141 | static inline void rcu_sysidle_force_exit(void) | |
1142 | { | |
1143 | } | |
1144 | ||
1145 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
1146 | ||
1147 | ||
274529ba PM |
1148 | /* |
1149 | * Dump the ftrace buffer, but only one time per callsite per boot. | |
1150 | */ | |
1151 | #define rcu_ftrace_dump(oops_dump_mode) \ | |
1152 | do { \ | |
1153 | static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \ | |
1154 | \ | |
1155 | if (!atomic_read(&___rfd_beenhere) && \ | |
1156 | !atomic_xchg(&___rfd_beenhere, 1)) \ | |
1157 | ftrace_dump(oops_dump_mode); \ | |
1158 | } while (0) | |
1159 | ||
1160 | ||
1da177e4 | 1161 | #endif /* __LINUX_RCUPDATE_H */ |