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