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> |
88c18630 | 47 | #include <asm/barrier.h> |
1da177e4 | 48 | |
7a754743 | 49 | extern int rcu_expedited; /* for sysctl */ |
e5ab6772 | 50 | |
ad0dc7f9 PM |
51 | enum rcutorture_type { |
52 | RCU_FLAVOR, | |
53 | RCU_BH_FLAVOR, | |
54 | RCU_SCHED_FLAVOR, | |
69c60455 | 55 | RCU_TASKS_FLAVOR, |
ad0dc7f9 PM |
56 | SRCU_FLAVOR, |
57 | INVALID_RCU_FLAVOR | |
58 | }; | |
59 | ||
28f6569a | 60 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) |
ad0dc7f9 PM |
61 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, |
62 | unsigned long *gpnum, unsigned long *completed); | |
584dc4ce TB |
63 | void rcutorture_record_test_transition(void); |
64 | void rcutorture_record_progress(unsigned long vernum); | |
65 | void do_trace_rcu_torture_read(const char *rcutorturename, | |
66 | struct rcu_head *rhp, | |
67 | unsigned long secs, | |
68 | unsigned long c_old, | |
69 | unsigned long c); | |
4a298656 | 70 | #else |
ad0dc7f9 PM |
71 | static inline void rcutorture_get_gp_data(enum rcutorture_type test_type, |
72 | int *flags, | |
73 | unsigned long *gpnum, | |
74 | unsigned long *completed) | |
75 | { | |
76 | *flags = 0; | |
77 | *gpnum = 0; | |
78 | *completed = 0; | |
79 | } | |
4a298656 PM |
80 | static inline void rcutorture_record_test_transition(void) |
81 | { | |
82 | } | |
83 | static inline void rcutorture_record_progress(unsigned long vernum) | |
84 | { | |
85 | } | |
91afaf30 | 86 | #ifdef CONFIG_RCU_TRACE |
584dc4ce TB |
87 | void do_trace_rcu_torture_read(const char *rcutorturename, |
88 | struct rcu_head *rhp, | |
89 | unsigned long secs, | |
90 | unsigned long c_old, | |
91 | unsigned long c); | |
91afaf30 | 92 | #else |
52494535 PM |
93 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ |
94 | do { } while (0) | |
91afaf30 | 95 | #endif |
4a298656 PM |
96 | #endif |
97 | ||
e27fc964 TH |
98 | #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) |
99 | #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) | |
a3dc3fb1 PM |
100 | #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) |
101 | #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) | |
c0f4dfd4 | 102 | #define ulong2long(a) (*(long *)(&(a))) |
a3dc3fb1 | 103 | |
03b042bf | 104 | /* Exported common interfaces */ |
2c42818e PM |
105 | |
106 | #ifdef CONFIG_PREEMPT_RCU | |
107 | ||
108 | /** | |
109 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
110 | * @head: structure to be used for queueing the RCU updates. | |
111 | * @func: actual callback function to be invoked after the grace period | |
112 | * | |
113 | * The callback function will be invoked some time after a full grace | |
114 | * period elapses, in other words after all pre-existing RCU read-side | |
115 | * critical sections have completed. However, the callback function | |
116 | * might well execute concurrently with RCU read-side critical sections | |
117 | * that started after call_rcu() was invoked. RCU read-side critical | |
118 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
119 | * and may be nested. | |
f0a0e6f2 PM |
120 | * |
121 | * Note that all CPUs must agree that the grace period extended beyond | |
122 | * all pre-existing RCU read-side critical section. On systems with more | |
123 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
124 | * guaranteed to have executed a full memory barrier since the end of its | |
125 | * last RCU read-side critical section whose beginning preceded the call | |
126 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
127 | * critical section that continues beyond the start of "func()" must have | |
128 | * executed a memory barrier after the call_rcu() but before the beginning | |
129 | * of that RCU read-side critical section. Note that these guarantees | |
130 | * include CPUs that are offline, idle, or executing in user mode, as | |
131 | * well as CPUs that are executing in the kernel. | |
132 | * | |
133 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
134 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
135 | * guaranteed to execute a full memory barrier during the time interval | |
136 | * between the call to call_rcu() and the invocation of "func()" -- even | |
137 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
138 | * more than one CPU). | |
2c42818e | 139 | */ |
584dc4ce TB |
140 | void call_rcu(struct rcu_head *head, |
141 | void (*func)(struct rcu_head *head)); | |
2c42818e PM |
142 | |
143 | #else /* #ifdef CONFIG_PREEMPT_RCU */ | |
144 | ||
145 | /* In classic RCU, call_rcu() is just call_rcu_sched(). */ | |
146 | #define call_rcu call_rcu_sched | |
147 | ||
148 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
149 | ||
150 | /** | |
151 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
152 | * @head: structure to be used for queueing the RCU updates. | |
153 | * @func: actual callback function to be invoked after the grace period | |
154 | * | |
155 | * The callback function will be invoked some time after a full grace | |
156 | * period elapses, in other words after all currently executing RCU | |
157 | * read-side critical sections have completed. call_rcu_bh() assumes | |
158 | * that the read-side critical sections end on completion of a softirq | |
159 | * handler. This means that read-side critical sections in process | |
160 | * context must not be interrupted by softirqs. This interface is to be | |
161 | * used when most of the read-side critical sections are in softirq context. | |
162 | * RCU read-side critical sections are delimited by : | |
163 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. | |
164 | * OR | |
165 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
166 | * These may be nested. | |
f0a0e6f2 PM |
167 | * |
168 | * See the description of call_rcu() for more detailed information on | |
169 | * memory ordering guarantees. | |
2c42818e | 170 | */ |
584dc4ce TB |
171 | void call_rcu_bh(struct rcu_head *head, |
172 | void (*func)(struct rcu_head *head)); | |
2c42818e PM |
173 | |
174 | /** | |
175 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
176 | * @head: structure to be used for queueing the RCU updates. | |
177 | * @func: actual callback function to be invoked after the grace period | |
178 | * | |
179 | * The callback function will be invoked some time after a full grace | |
180 | * period elapses, in other words after all currently executing RCU | |
181 | * read-side critical sections have completed. call_rcu_sched() assumes | |
182 | * that the read-side critical sections end on enabling of preemption | |
183 | * or on voluntary preemption. | |
184 | * RCU read-side critical sections are delimited by : | |
185 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), | |
186 | * OR | |
187 | * anything that disables preemption. | |
188 | * These may be nested. | |
f0a0e6f2 PM |
189 | * |
190 | * See the description of call_rcu() for more detailed information on | |
191 | * memory ordering guarantees. | |
2c42818e | 192 | */ |
584dc4ce TB |
193 | void call_rcu_sched(struct rcu_head *head, |
194 | void (*func)(struct rcu_head *rcu)); | |
2c42818e | 195 | |
584dc4ce | 196 | void synchronize_sched(void); |
03b042bf | 197 | |
ee376dbd PM |
198 | /* |
199 | * Structure allowing asynchronous waiting on RCU. | |
200 | */ | |
201 | struct rcu_synchronize { | |
202 | struct rcu_head head; | |
203 | struct completion completion; | |
204 | }; | |
205 | void wakeme_after_rcu(struct rcu_head *head); | |
206 | ||
8315f422 PM |
207 | /** |
208 | * call_rcu_tasks() - Queue an RCU for invocation task-based grace period | |
209 | * @head: structure to be used for queueing the RCU updates. | |
210 | * @func: actual callback function to be invoked after the grace period | |
211 | * | |
212 | * The callback function will be invoked some time after a full grace | |
213 | * period elapses, in other words after all currently executing RCU | |
214 | * read-side critical sections have completed. call_rcu_tasks() assumes | |
215 | * that the read-side critical sections end at a voluntary context | |
216 | * switch (not a preemption!), entry into idle, or transition to usermode | |
217 | * execution. As such, there are no read-side primitives analogous to | |
218 | * rcu_read_lock() and rcu_read_unlock() because this primitive is intended | |
219 | * to determine that all tasks have passed through a safe state, not so | |
220 | * much for data-strcuture synchronization. | |
221 | * | |
222 | * See the description of call_rcu() for more detailed information on | |
223 | * memory ordering guarantees. | |
224 | */ | |
225 | void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head)); | |
53c6d4ed PM |
226 | void synchronize_rcu_tasks(void); |
227 | void rcu_barrier_tasks(void); | |
8315f422 | 228 | |
a3dc3fb1 PM |
229 | #ifdef CONFIG_PREEMPT_RCU |
230 | ||
584dc4ce TB |
231 | void __rcu_read_lock(void); |
232 | void __rcu_read_unlock(void); | |
233 | void rcu_read_unlock_special(struct task_struct *t); | |
7b0b759b PM |
234 | void synchronize_rcu(void); |
235 | ||
a3dc3fb1 PM |
236 | /* |
237 | * Defined as a macro as it is a very low level header included from | |
238 | * areas that don't even know about current. This gives the rcu_read_lock() | |
239 | * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other | |
240 | * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. | |
241 | */ | |
242 | #define rcu_preempt_depth() (current->rcu_read_lock_nesting) | |
243 | ||
7b0b759b PM |
244 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
245 | ||
246 | static inline void __rcu_read_lock(void) | |
247 | { | |
248 | preempt_disable(); | |
249 | } | |
250 | ||
251 | static inline void __rcu_read_unlock(void) | |
252 | { | |
253 | preempt_enable(); | |
254 | } | |
255 | ||
256 | static inline void synchronize_rcu(void) | |
257 | { | |
258 | synchronize_sched(); | |
259 | } | |
260 | ||
261 | static inline int rcu_preempt_depth(void) | |
262 | { | |
263 | return 0; | |
264 | } | |
265 | ||
266 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
267 | ||
268 | /* Internal to kernel */ | |
584dc4ce | 269 | void rcu_init(void); |
284a8c93 PM |
270 | void rcu_sched_qs(void); |
271 | void rcu_bh_qs(void); | |
c3377c2d | 272 | void rcu_check_callbacks(int user); |
7b0b759b | 273 | struct notifier_block; |
584dc4ce TB |
274 | void rcu_idle_enter(void); |
275 | void rcu_idle_exit(void); | |
276 | void rcu_irq_enter(void); | |
277 | void rcu_irq_exit(void); | |
2b1d5024 | 278 | |
61f38db3 RR |
279 | #ifdef CONFIG_RCU_STALL_COMMON |
280 | void rcu_sysrq_start(void); | |
281 | void rcu_sysrq_end(void); | |
282 | #else /* #ifdef CONFIG_RCU_STALL_COMMON */ | |
283 | static inline void rcu_sysrq_start(void) | |
284 | { | |
285 | } | |
286 | static inline void rcu_sysrq_end(void) | |
287 | { | |
288 | } | |
289 | #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ | |
290 | ||
2b1d5024 | 291 | #ifdef CONFIG_RCU_USER_QS |
584dc4ce TB |
292 | void rcu_user_enter(void); |
293 | void rcu_user_exit(void); | |
2b1d5024 FW |
294 | #else |
295 | static inline void rcu_user_enter(void) { } | |
296 | static inline void rcu_user_exit(void) { } | |
4d9a5d43 FW |
297 | static inline void rcu_user_hooks_switch(struct task_struct *prev, |
298 | struct task_struct *next) { } | |
2b1d5024 FW |
299 | #endif /* CONFIG_RCU_USER_QS */ |
300 | ||
f4579fc5 PM |
301 | #ifdef CONFIG_RCU_NOCB_CPU |
302 | void rcu_init_nohz(void); | |
303 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
304 | static inline void rcu_init_nohz(void) | |
305 | { | |
306 | } | |
307 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
308 | ||
8a2ecf47 PM |
309 | /** |
310 | * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers | |
311 | * @a: Code that RCU needs to pay attention to. | |
312 | * | |
313 | * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden | |
314 | * in the inner idle loop, that is, between the rcu_idle_enter() and | |
315 | * the rcu_idle_exit() -- RCU will happily ignore any such read-side | |
316 | * critical sections. However, things like powertop need tracepoints | |
317 | * in the inner idle loop. | |
318 | * | |
319 | * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) | |
320 | * will tell RCU that it needs to pay attending, invoke its argument | |
321 | * (in this example, a call to the do_something_with_RCU() function), | |
322 | * and then tell RCU to go back to ignoring this CPU. It is permissible | |
323 | * to nest RCU_NONIDLE() wrappers, but the nesting level is currently | |
324 | * quite limited. If deeper nesting is required, it will be necessary | |
325 | * to adjust DYNTICK_TASK_NESTING_VALUE accordingly. | |
8a2ecf47 PM |
326 | */ |
327 | #define RCU_NONIDLE(a) \ | |
328 | do { \ | |
b4270ee3 | 329 | rcu_irq_enter(); \ |
8a2ecf47 | 330 | do { a; } while (0); \ |
b4270ee3 | 331 | rcu_irq_exit(); \ |
8a2ecf47 PM |
332 | } while (0) |
333 | ||
8315f422 PM |
334 | /* |
335 | * Note a voluntary context switch for RCU-tasks benefit. This is a | |
336 | * macro rather than an inline function to avoid #include hell. | |
337 | */ | |
338 | #ifdef CONFIG_TASKS_RCU | |
3f95aa81 PM |
339 | #define TASKS_RCU(x) x |
340 | extern struct srcu_struct tasks_rcu_exit_srcu; | |
8315f422 PM |
341 | #define rcu_note_voluntary_context_switch(t) \ |
342 | do { \ | |
5cd37193 | 343 | rcu_all_qs(); \ |
8315f422 PM |
344 | if (ACCESS_ONCE((t)->rcu_tasks_holdout)) \ |
345 | ACCESS_ONCE((t)->rcu_tasks_holdout) = false; \ | |
8315f422 PM |
346 | } while (0) |
347 | #else /* #ifdef CONFIG_TASKS_RCU */ | |
3f95aa81 | 348 | #define TASKS_RCU(x) do { } while (0) |
5cd37193 | 349 | #define rcu_note_voluntary_context_switch(t) rcu_all_qs() |
8315f422 PM |
350 | #endif /* #else #ifdef CONFIG_TASKS_RCU */ |
351 | ||
bde6c3aa PM |
352 | /** |
353 | * cond_resched_rcu_qs - Report potential quiescent states to RCU | |
354 | * | |
355 | * This macro resembles cond_resched(), except that it is defined to | |
356 | * report potential quiescent states to RCU-tasks even if the cond_resched() | |
357 | * machinery were to be shut off, as some advocate for PREEMPT kernels. | |
358 | */ | |
359 | #define cond_resched_rcu_qs() \ | |
360 | do { \ | |
b6331ae8 PM |
361 | if (!cond_resched()) \ |
362 | rcu_note_voluntary_context_switch(current); \ | |
bde6c3aa PM |
363 | } while (0) |
364 | ||
cc6783f7 | 365 | #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) |
584dc4ce | 366 | bool __rcu_is_watching(void); |
cc6783f7 PM |
367 | #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */ |
368 | ||
2c42818e PM |
369 | /* |
370 | * Infrastructure to implement the synchronize_() primitives in | |
371 | * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. | |
372 | */ | |
373 | ||
374 | typedef void call_rcu_func_t(struct rcu_head *head, | |
375 | void (*func)(struct rcu_head *head)); | |
376 | void wait_rcu_gp(call_rcu_func_t crf); | |
377 | ||
28f6569a | 378 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) |
64db4cff | 379 | #include <linux/rcutree.h> |
127781d1 | 380 | #elif defined(CONFIG_TINY_RCU) |
9b1d82fa | 381 | #include <linux/rcutiny.h> |
64db4cff PM |
382 | #else |
383 | #error "Unknown RCU implementation specified to kernel configuration" | |
6b3ef48a | 384 | #endif |
01c1c660 | 385 | |
551d55a9 MD |
386 | /* |
387 | * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic | |
388 | * initialization and destruction of rcu_head on the stack. rcu_head structures | |
389 | * allocated dynamically in the heap or defined statically don't need any | |
390 | * initialization. | |
391 | */ | |
392 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
546a9d85 PM |
393 | void init_rcu_head(struct rcu_head *head); |
394 | void destroy_rcu_head(struct rcu_head *head); | |
584dc4ce TB |
395 | void init_rcu_head_on_stack(struct rcu_head *head); |
396 | void destroy_rcu_head_on_stack(struct rcu_head *head); | |
551d55a9 | 397 | #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
546a9d85 PM |
398 | static inline void init_rcu_head(struct rcu_head *head) |
399 | { | |
400 | } | |
401 | ||
402 | static inline void destroy_rcu_head(struct rcu_head *head) | |
403 | { | |
404 | } | |
405 | ||
4376030a MD |
406 | static inline void init_rcu_head_on_stack(struct rcu_head *head) |
407 | { | |
408 | } | |
409 | ||
410 | static inline void destroy_rcu_head_on_stack(struct rcu_head *head) | |
411 | { | |
412 | } | |
551d55a9 | 413 | #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
4376030a | 414 | |
c0d6d01b PM |
415 | #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) |
416 | bool rcu_lockdep_current_cpu_online(void); | |
417 | #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
418 | static inline bool rcu_lockdep_current_cpu_online(void) | |
419 | { | |
521d24ee | 420 | return true; |
c0d6d01b PM |
421 | } |
422 | #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
423 | ||
bc33f24b | 424 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
632ee200 | 425 | |
00f49e57 FW |
426 | static inline void rcu_lock_acquire(struct lockdep_map *map) |
427 | { | |
fb9edbe9 | 428 | lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); |
00f49e57 FW |
429 | } |
430 | ||
431 | static inline void rcu_lock_release(struct lockdep_map *map) | |
432 | { | |
00f49e57 FW |
433 | lock_release(map, 1, _THIS_IP_); |
434 | } | |
435 | ||
bc33f24b | 436 | extern struct lockdep_map rcu_lock_map; |
632ee200 | 437 | extern struct lockdep_map rcu_bh_lock_map; |
632ee200 | 438 | extern struct lockdep_map rcu_sched_lock_map; |
24ef659a | 439 | extern struct lockdep_map rcu_callback_map; |
a235c091 | 440 | int debug_lockdep_rcu_enabled(void); |
54dbf96c | 441 | |
85b39d30 | 442 | int rcu_read_lock_held(void); |
584dc4ce | 443 | int rcu_read_lock_bh_held(void); |
632ee200 PM |
444 | |
445 | /** | |
ca5ecddf | 446 | * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? |
632ee200 | 447 | * |
d20200b5 PM |
448 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an |
449 | * RCU-sched read-side critical section. In absence of | |
450 | * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side | |
451 | * critical section unless it can prove otherwise. Note that disabling | |
452 | * of preemption (including disabling irqs) counts as an RCU-sched | |
ca5ecddf PM |
453 | * read-side critical section. This is useful for debug checks in functions |
454 | * that required that they be called within an RCU-sched read-side | |
455 | * critical section. | |
54dbf96c | 456 | * |
32c141a0 PM |
457 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot |
458 | * and while lockdep is disabled. | |
e6b80a3b FW |
459 | * |
460 | * Note that if the CPU is in the idle loop from an RCU point of | |
461 | * view (ie: that we are in the section between rcu_idle_enter() and | |
462 | * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU | |
463 | * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs | |
464 | * that are in such a section, considering these as in extended quiescent | |
465 | * state, so such a CPU is effectively never in an RCU read-side critical | |
466 | * section regardless of what RCU primitives it invokes. This state of | |
467 | * affairs is required --- we need to keep an RCU-free window in idle | |
468 | * where the CPU may possibly enter into low power mode. This way we can | |
469 | * notice an extended quiescent state to other CPUs that started a grace | |
470 | * period. Otherwise we would delay any grace period as long as we run in | |
471 | * the idle task. | |
c0d6d01b PM |
472 | * |
473 | * Similarly, we avoid claiming an SRCU read lock held if the current | |
474 | * CPU is offline. | |
632ee200 | 475 | */ |
bdd4e85d | 476 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
477 | static inline int rcu_read_lock_sched_held(void) |
478 | { | |
479 | int lockdep_opinion = 0; | |
480 | ||
54dbf96c PM |
481 | if (!debug_lockdep_rcu_enabled()) |
482 | return 1; | |
5c173eb8 | 483 | if (!rcu_is_watching()) |
e6b80a3b | 484 | return 0; |
c0d6d01b PM |
485 | if (!rcu_lockdep_current_cpu_online()) |
486 | return 0; | |
632ee200 PM |
487 | if (debug_locks) |
488 | lockdep_opinion = lock_is_held(&rcu_sched_lock_map); | |
0cff810f | 489 | return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); |
632ee200 | 490 | } |
bdd4e85d | 491 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
492 | static inline int rcu_read_lock_sched_held(void) |
493 | { | |
494 | return 1; | |
632ee200 | 495 | } |
bdd4e85d | 496 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
497 | |
498 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
499 | ||
d8ab29f8 PM |
500 | # define rcu_lock_acquire(a) do { } while (0) |
501 | # define rcu_lock_release(a) do { } while (0) | |
632ee200 PM |
502 | |
503 | static inline int rcu_read_lock_held(void) | |
504 | { | |
505 | return 1; | |
506 | } | |
507 | ||
508 | static inline int rcu_read_lock_bh_held(void) | |
509 | { | |
510 | return 1; | |
511 | } | |
512 | ||
bdd4e85d | 513 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
514 | static inline int rcu_read_lock_sched_held(void) |
515 | { | |
bbad9379 | 516 | return preempt_count() != 0 || irqs_disabled(); |
632ee200 | 517 | } |
bdd4e85d | 518 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
519 | static inline int rcu_read_lock_sched_held(void) |
520 | { | |
521 | return 1; | |
632ee200 | 522 | } |
bdd4e85d | 523 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
524 | |
525 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
526 | ||
527 | #ifdef CONFIG_PROVE_RCU | |
528 | ||
4221a991 TH |
529 | /** |
530 | * rcu_lockdep_assert - emit lockdep splat if specified condition not met | |
531 | * @c: condition to check | |
b3fbab05 | 532 | * @s: informative message |
4221a991 | 533 | */ |
b3fbab05 | 534 | #define rcu_lockdep_assert(c, s) \ |
2b3fc35f | 535 | do { \ |
7ccaba53 | 536 | static bool __section(.data.unlikely) __warned; \ |
2b3fc35f LJ |
537 | if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ |
538 | __warned = true; \ | |
b3fbab05 | 539 | lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ |
2b3fc35f LJ |
540 | } \ |
541 | } while (0) | |
542 | ||
50406b98 PM |
543 | #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) |
544 | static inline void rcu_preempt_sleep_check(void) | |
545 | { | |
546 | rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), | |
5cf05ad7 | 547 | "Illegal context switch in RCU read-side critical section"); |
50406b98 PM |
548 | } |
549 | #else /* #ifdef CONFIG_PROVE_RCU */ | |
550 | static inline void rcu_preempt_sleep_check(void) | |
551 | { | |
552 | } | |
553 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
554 | ||
b3fbab05 PM |
555 | #define rcu_sleep_check() \ |
556 | do { \ | |
50406b98 | 557 | rcu_preempt_sleep_check(); \ |
b3fbab05 | 558 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \ |
41f4abd9 | 559 | "Illegal context switch in RCU-bh read-side critical section"); \ |
b3fbab05 | 560 | rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \ |
41f4abd9 | 561 | "Illegal context switch in RCU-sched read-side critical section"); \ |
b3fbab05 PM |
562 | } while (0) |
563 | ||
ca5ecddf PM |
564 | #else /* #ifdef CONFIG_PROVE_RCU */ |
565 | ||
b3fbab05 PM |
566 | #define rcu_lockdep_assert(c, s) do { } while (0) |
567 | #define rcu_sleep_check() do { } while (0) | |
ca5ecddf PM |
568 | |
569 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
570 | ||
571 | /* | |
572 | * Helper functions for rcu_dereference_check(), rcu_dereference_protected() | |
573 | * and rcu_assign_pointer(). Some of these could be folded into their | |
574 | * callers, but they are left separate in order to ease introduction of | |
575 | * multiple flavors of pointers to match the multiple flavors of RCU | |
576 | * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in | |
577 | * the future. | |
578 | */ | |
53ecfba2 PM |
579 | |
580 | #ifdef __CHECKER__ | |
581 | #define rcu_dereference_sparse(p, space) \ | |
582 | ((void)(((typeof(*p) space *)p) == p)) | |
583 | #else /* #ifdef __CHECKER__ */ | |
584 | #define rcu_dereference_sparse(p, space) | |
585 | #endif /* #else #ifdef __CHECKER__ */ | |
586 | ||
ca5ecddf | 587 | #define __rcu_access_pointer(p, space) \ |
0adab9b9 JP |
588 | ({ \ |
589 | typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \ | |
590 | rcu_dereference_sparse(p, space); \ | |
591 | ((typeof(*p) __force __kernel *)(_________p1)); \ | |
592 | }) | |
ca5ecddf | 593 | #define __rcu_dereference_check(p, c, space) \ |
0adab9b9 | 594 | ({ \ |
ac59853c PK |
595 | /* Dependency order vs. p above. */ \ |
596 | typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \ | |
0adab9b9 JP |
597 | rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \ |
598 | rcu_dereference_sparse(p, space); \ | |
ac59853c | 599 | ((typeof(*p) __force __kernel *)(________p1)); \ |
0adab9b9 | 600 | }) |
ca5ecddf | 601 | #define __rcu_dereference_protected(p, c, space) \ |
0adab9b9 JP |
602 | ({ \ |
603 | rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \ | |
604 | rcu_dereference_sparse(p, space); \ | |
605 | ((typeof(*p) __force __kernel *)(p)); \ | |
606 | }) | |
ca5ecddf | 607 | |
a4dd9925 | 608 | #define __rcu_access_index(p, space) \ |
0adab9b9 JP |
609 | ({ \ |
610 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
611 | rcu_dereference_sparse(p, space); \ | |
612 | (_________p1); \ | |
613 | }) | |
ca5ecddf | 614 | #define __rcu_dereference_index_check(p, c) \ |
0adab9b9 | 615 | ({ \ |
ac59853c PK |
616 | /* Dependency order vs. p above. */ \ |
617 | typeof(p) _________p1 = lockless_dereference(p); \ | |
0adab9b9 JP |
618 | rcu_lockdep_assert(c, \ |
619 | "suspicious rcu_dereference_index_check() usage"); \ | |
0adab9b9 JP |
620 | (_________p1); \ |
621 | }) | |
462225ae PM |
622 | |
623 | /** | |
624 | * RCU_INITIALIZER() - statically initialize an RCU-protected global variable | |
625 | * @v: The value to statically initialize with. | |
626 | */ | |
627 | #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) | |
628 | ||
54ef6df3 PM |
629 | /** |
630 | * lockless_dereference() - safely load a pointer for later dereference | |
631 | * @p: The pointer to load | |
632 | * | |
633 | * Similar to rcu_dereference(), but for situations where the pointed-to | |
634 | * object's lifetime is managed by something other than RCU. That | |
635 | * "something other" might be reference counting or simple immortality. | |
636 | */ | |
637 | #define lockless_dereference(p) \ | |
638 | ({ \ | |
639 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
640 | smp_read_barrier_depends(); /* Dependency order vs. p above. */ \ | |
641 | (_________p1); \ | |
642 | }) | |
643 | ||
462225ae PM |
644 | /** |
645 | * rcu_assign_pointer() - assign to RCU-protected pointer | |
646 | * @p: pointer to assign to | |
647 | * @v: value to assign (publish) | |
648 | * | |
649 | * Assigns the specified value to the specified RCU-protected | |
650 | * pointer, ensuring that any concurrent RCU readers will see | |
651 | * any prior initialization. | |
652 | * | |
653 | * Inserts memory barriers on architectures that require them | |
654 | * (which is most of them), and also prevents the compiler from | |
655 | * reordering the code that initializes the structure after the pointer | |
656 | * assignment. More importantly, this call documents which pointers | |
657 | * will be dereferenced by RCU read-side code. | |
658 | * | |
659 | * In some special cases, you may use RCU_INIT_POINTER() instead | |
660 | * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due | |
661 | * to the fact that it does not constrain either the CPU or the compiler. | |
662 | * That said, using RCU_INIT_POINTER() when you should have used | |
663 | * rcu_assign_pointer() is a very bad thing that results in | |
664 | * impossible-to-diagnose memory corruption. So please be careful. | |
665 | * See the RCU_INIT_POINTER() comment header for details. | |
666 | * | |
667 | * Note that rcu_assign_pointer() evaluates each of its arguments only | |
668 | * once, appearances notwithstanding. One of the "extra" evaluations | |
669 | * is in typeof() and the other visible only to sparse (__CHECKER__), | |
670 | * neither of which actually execute the argument. As with most cpp | |
671 | * macros, this execute-arguments-only-once property is important, so | |
672 | * please be careful when making changes to rcu_assign_pointer() and the | |
673 | * other macros that it invokes. | |
674 | */ | |
88c18630 | 675 | #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v)) |
ca5ecddf PM |
676 | |
677 | /** | |
678 | * rcu_access_pointer() - fetch RCU pointer with no dereferencing | |
679 | * @p: The pointer to read | |
680 | * | |
681 | * Return the value of the specified RCU-protected pointer, but omit the | |
682 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
683 | * when the value of this pointer is accessed, but the pointer is not | |
684 | * dereferenced, for example, when testing an RCU-protected pointer against | |
685 | * NULL. Although rcu_access_pointer() may also be used in cases where | |
686 | * update-side locks prevent the value of the pointer from changing, you | |
687 | * should instead use rcu_dereference_protected() for this use case. | |
5e1ee6e1 PM |
688 | * |
689 | * It is also permissible to use rcu_access_pointer() when read-side | |
690 | * access to the pointer was removed at least one grace period ago, as | |
691 | * is the case in the context of the RCU callback that is freeing up | |
692 | * the data, or after a synchronize_rcu() returns. This can be useful | |
693 | * when tearing down multi-linked structures after a grace period | |
694 | * has elapsed. | |
ca5ecddf PM |
695 | */ |
696 | #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) | |
697 | ||
632ee200 | 698 | /** |
ca5ecddf | 699 | * rcu_dereference_check() - rcu_dereference with debug checking |
c08c68dd DH |
700 | * @p: The pointer to read, prior to dereferencing |
701 | * @c: The conditions under which the dereference will take place | |
632ee200 | 702 | * |
c08c68dd | 703 | * Do an rcu_dereference(), but check that the conditions under which the |
ca5ecddf PM |
704 | * dereference will take place are correct. Typically the conditions |
705 | * indicate the various locking conditions that should be held at that | |
706 | * point. The check should return true if the conditions are satisfied. | |
707 | * An implicit check for being in an RCU read-side critical section | |
708 | * (rcu_read_lock()) is included. | |
c08c68dd DH |
709 | * |
710 | * For example: | |
711 | * | |
ca5ecddf | 712 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); |
c08c68dd DH |
713 | * |
714 | * could be used to indicate to lockdep that foo->bar may only be dereferenced | |
ca5ecddf | 715 | * if either rcu_read_lock() is held, or that the lock required to replace |
c08c68dd DH |
716 | * the bar struct at foo->bar is held. |
717 | * | |
718 | * Note that the list of conditions may also include indications of when a lock | |
719 | * need not be held, for example during initialisation or destruction of the | |
720 | * target struct: | |
721 | * | |
ca5ecddf | 722 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || |
c08c68dd | 723 | * atomic_read(&foo->usage) == 0); |
ca5ecddf PM |
724 | * |
725 | * Inserts memory barriers on architectures that require them | |
726 | * (currently only the Alpha), prevents the compiler from refetching | |
727 | * (and from merging fetches), and, more importantly, documents exactly | |
728 | * which pointers are protected by RCU and checks that the pointer is | |
729 | * annotated as __rcu. | |
632ee200 PM |
730 | */ |
731 | #define rcu_dereference_check(p, c) \ | |
b826565a | 732 | __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) |
ca5ecddf PM |
733 | |
734 | /** | |
735 | * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking | |
736 | * @p: The pointer to read, prior to dereferencing | |
737 | * @c: The conditions under which the dereference will take place | |
738 | * | |
739 | * This is the RCU-bh counterpart to rcu_dereference_check(). | |
740 | */ | |
741 | #define rcu_dereference_bh_check(p, c) \ | |
b826565a | 742 | __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) |
632ee200 | 743 | |
b62730ba | 744 | /** |
ca5ecddf PM |
745 | * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking |
746 | * @p: The pointer to read, prior to dereferencing | |
747 | * @c: The conditions under which the dereference will take place | |
748 | * | |
749 | * This is the RCU-sched counterpart to rcu_dereference_check(). | |
750 | */ | |
751 | #define rcu_dereference_sched_check(p, c) \ | |
b826565a | 752 | __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ |
ca5ecddf PM |
753 | __rcu) |
754 | ||
755 | #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ | |
756 | ||
12bcbe66 SR |
757 | /* |
758 | * The tracing infrastructure traces RCU (we want that), but unfortunately | |
759 | * some of the RCU checks causes tracing to lock up the system. | |
760 | * | |
761 | * The tracing version of rcu_dereference_raw() must not call | |
762 | * rcu_read_lock_held(). | |
763 | */ | |
764 | #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) | |
765 | ||
a4dd9925 PM |
766 | /** |
767 | * rcu_access_index() - fetch RCU index with no dereferencing | |
768 | * @p: The index to read | |
769 | * | |
770 | * Return the value of the specified RCU-protected index, but omit the | |
771 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
772 | * when the value of this index is accessed, but the index is not | |
773 | * dereferenced, for example, when testing an RCU-protected index against | |
774 | * -1. Although rcu_access_index() may also be used in cases where | |
775 | * update-side locks prevent the value of the index from changing, you | |
776 | * should instead use rcu_dereference_index_protected() for this use case. | |
777 | */ | |
778 | #define rcu_access_index(p) __rcu_access_index((p), __rcu) | |
779 | ||
ca5ecddf PM |
780 | /** |
781 | * rcu_dereference_index_check() - rcu_dereference for indices with debug checking | |
782 | * @p: The pointer to read, prior to dereferencing | |
783 | * @c: The conditions under which the dereference will take place | |
784 | * | |
785 | * Similar to rcu_dereference_check(), but omits the sparse checking. | |
786 | * This allows rcu_dereference_index_check() to be used on integers, | |
787 | * which can then be used as array indices. Attempting to use | |
788 | * rcu_dereference_check() on an integer will give compiler warnings | |
789 | * because the sparse address-space mechanism relies on dereferencing | |
790 | * the RCU-protected pointer. Dereferencing integers is not something | |
791 | * that even gcc will put up with. | |
792 | * | |
793 | * Note that this function does not implicitly check for RCU read-side | |
794 | * critical sections. If this function gains lots of uses, it might | |
795 | * make sense to provide versions for each flavor of RCU, but it does | |
796 | * not make sense as of early 2010. | |
797 | */ | |
798 | #define rcu_dereference_index_check(p, c) \ | |
799 | __rcu_dereference_index_check((p), (c)) | |
800 | ||
801 | /** | |
802 | * rcu_dereference_protected() - fetch RCU pointer when updates prevented | |
803 | * @p: The pointer to read, prior to dereferencing | |
804 | * @c: The conditions under which the dereference will take place | |
b62730ba PM |
805 | * |
806 | * Return the value of the specified RCU-protected pointer, but omit | |
807 | * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This | |
808 | * is useful in cases where update-side locks prevent the value of the | |
809 | * pointer from changing. Please note that this primitive does -not- | |
810 | * prevent the compiler from repeating this reference or combining it | |
811 | * with other references, so it should not be used without protection | |
812 | * of appropriate locks. | |
ca5ecddf PM |
813 | * |
814 | * This function is only for update-side use. Using this function | |
815 | * when protected only by rcu_read_lock() will result in infrequent | |
816 | * but very ugly failures. | |
b62730ba PM |
817 | */ |
818 | #define rcu_dereference_protected(p, c) \ | |
ca5ecddf | 819 | __rcu_dereference_protected((p), (c), __rcu) |
b62730ba | 820 | |
bc33f24b | 821 | |
b62730ba | 822 | /** |
ca5ecddf PM |
823 | * rcu_dereference() - fetch RCU-protected pointer for dereferencing |
824 | * @p: The pointer to read, prior to dereferencing | |
b62730ba | 825 | * |
ca5ecddf | 826 | * This is a simple wrapper around rcu_dereference_check(). |
b62730ba | 827 | */ |
ca5ecddf | 828 | #define rcu_dereference(p) rcu_dereference_check(p, 0) |
b62730ba | 829 | |
1da177e4 | 830 | /** |
ca5ecddf PM |
831 | * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing |
832 | * @p: The pointer to read, prior to dereferencing | |
833 | * | |
834 | * Makes rcu_dereference_check() do the dirty work. | |
835 | */ | |
836 | #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) | |
837 | ||
838 | /** | |
839 | * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing | |
840 | * @p: The pointer to read, prior to dereferencing | |
841 | * | |
842 | * Makes rcu_dereference_check() do the dirty work. | |
843 | */ | |
844 | #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) | |
845 | ||
846 | /** | |
847 | * rcu_read_lock() - mark the beginning of an RCU read-side critical section | |
1da177e4 | 848 | * |
9b06e818 | 849 | * When synchronize_rcu() is invoked on one CPU while other CPUs |
1da177e4 | 850 | * are within RCU read-side critical sections, then the |
9b06e818 | 851 | * synchronize_rcu() is guaranteed to block until after all the other |
1da177e4 LT |
852 | * CPUs exit their critical sections. Similarly, if call_rcu() is invoked |
853 | * on one CPU while other CPUs are within RCU read-side critical | |
854 | * sections, invocation of the corresponding RCU callback is deferred | |
855 | * until after the all the other CPUs exit their critical sections. | |
856 | * | |
857 | * Note, however, that RCU callbacks are permitted to run concurrently | |
77d8485a | 858 | * with new RCU read-side critical sections. One way that this can happen |
1da177e4 LT |
859 | * is via the following sequence of events: (1) CPU 0 enters an RCU |
860 | * read-side critical section, (2) CPU 1 invokes call_rcu() to register | |
861 | * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, | |
862 | * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU | |
863 | * callback is invoked. This is legal, because the RCU read-side critical | |
864 | * section that was running concurrently with the call_rcu() (and which | |
865 | * therefore might be referencing something that the corresponding RCU | |
866 | * callback would free up) has completed before the corresponding | |
867 | * RCU callback is invoked. | |
868 | * | |
869 | * RCU read-side critical sections may be nested. Any deferred actions | |
870 | * will be deferred until the outermost RCU read-side critical section | |
871 | * completes. | |
872 | * | |
9079fd7c PM |
873 | * You can avoid reading and understanding the next paragraph by |
874 | * following this rule: don't put anything in an rcu_read_lock() RCU | |
875 | * read-side critical section that would block in a !PREEMPT kernel. | |
876 | * But if you want the full story, read on! | |
877 | * | |
ab74fdfd PM |
878 | * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), |
879 | * it is illegal to block while in an RCU read-side critical section. | |
28f6569a | 880 | * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT |
ab74fdfd PM |
881 | * kernel builds, RCU read-side critical sections may be preempted, |
882 | * but explicit blocking is illegal. Finally, in preemptible RCU | |
883 | * implementations in real-time (with -rt patchset) kernel builds, RCU | |
884 | * read-side critical sections may be preempted and they may also block, but | |
885 | * only when acquiring spinlocks that are subject to priority inheritance. | |
1da177e4 | 886 | */ |
bc33f24b PM |
887 | static inline void rcu_read_lock(void) |
888 | { | |
889 | __rcu_read_lock(); | |
890 | __acquire(RCU); | |
d8ab29f8 | 891 | rcu_lock_acquire(&rcu_lock_map); |
5c173eb8 | 892 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 893 | "rcu_read_lock() used illegally while idle"); |
bc33f24b | 894 | } |
1da177e4 | 895 | |
1da177e4 LT |
896 | /* |
897 | * So where is rcu_write_lock()? It does not exist, as there is no | |
898 | * way for writers to lock out RCU readers. This is a feature, not | |
899 | * a bug -- this property is what provides RCU's performance benefits. | |
900 | * Of course, writers must coordinate with each other. The normal | |
901 | * spinlock primitives work well for this, but any other technique may be | |
902 | * used as well. RCU does not care how the writers keep out of each | |
903 | * others' way, as long as they do so. | |
904 | */ | |
3d76c082 PM |
905 | |
906 | /** | |
ca5ecddf | 907 | * rcu_read_unlock() - marks the end of an RCU read-side critical section. |
3d76c082 | 908 | * |
f27bc487 PM |
909 | * In most situations, rcu_read_unlock() is immune from deadlock. |
910 | * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() | |
911 | * is responsible for deboosting, which it does via rt_mutex_unlock(). | |
912 | * Unfortunately, this function acquires the scheduler's runqueue and | |
913 | * priority-inheritance spinlocks. This means that deadlock could result | |
914 | * if the caller of rcu_read_unlock() already holds one of these locks or | |
ce36f2f3 ON |
915 | * any lock that is ever acquired while holding them; or any lock which |
916 | * can be taken from interrupt context because rcu_boost()->rt_mutex_lock() | |
917 | * does not disable irqs while taking ->wait_lock. | |
f27bc487 PM |
918 | * |
919 | * That said, RCU readers are never priority boosted unless they were | |
920 | * preempted. Therefore, one way to avoid deadlock is to make sure | |
921 | * that preemption never happens within any RCU read-side critical | |
922 | * section whose outermost rcu_read_unlock() is called with one of | |
923 | * rt_mutex_unlock()'s locks held. Such preemption can be avoided in | |
924 | * a number of ways, for example, by invoking preempt_disable() before | |
925 | * critical section's outermost rcu_read_lock(). | |
926 | * | |
927 | * Given that the set of locks acquired by rt_mutex_unlock() might change | |
928 | * at any time, a somewhat more future-proofed approach is to make sure | |
929 | * that that preemption never happens within any RCU read-side critical | |
930 | * section whose outermost rcu_read_unlock() is called with irqs disabled. | |
931 | * This approach relies on the fact that rt_mutex_unlock() currently only | |
932 | * acquires irq-disabled locks. | |
933 | * | |
934 | * The second of these two approaches is best in most situations, | |
935 | * however, the first approach can also be useful, at least to those | |
936 | * developers willing to keep abreast of the set of locks acquired by | |
937 | * rt_mutex_unlock(). | |
938 | * | |
3d76c082 PM |
939 | * See rcu_read_lock() for more information. |
940 | */ | |
bc33f24b PM |
941 | static inline void rcu_read_unlock(void) |
942 | { | |
5c173eb8 | 943 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 944 | "rcu_read_unlock() used illegally while idle"); |
bc33f24b PM |
945 | __release(RCU); |
946 | __rcu_read_unlock(); | |
d24209bb | 947 | rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ |
bc33f24b | 948 | } |
1da177e4 LT |
949 | |
950 | /** | |
ca5ecddf | 951 | * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section |
1da177e4 LT |
952 | * |
953 | * This is equivalent of rcu_read_lock(), but to be used when updates | |
ca5ecddf PM |
954 | * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since |
955 | * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a | |
956 | * softirq handler to be a quiescent state, a process in RCU read-side | |
957 | * critical section must be protected by disabling softirqs. Read-side | |
958 | * critical sections in interrupt context can use just rcu_read_lock(), | |
959 | * though this should at least be commented to avoid confusing people | |
960 | * reading the code. | |
3842a083 PM |
961 | * |
962 | * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() | |
963 | * must occur in the same context, for example, it is illegal to invoke | |
964 | * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() | |
965 | * was invoked from some other task. | |
1da177e4 | 966 | */ |
bc33f24b PM |
967 | static inline void rcu_read_lock_bh(void) |
968 | { | |
6206ab9b | 969 | local_bh_disable(); |
bc33f24b | 970 | __acquire(RCU_BH); |
d8ab29f8 | 971 | rcu_lock_acquire(&rcu_bh_lock_map); |
5c173eb8 | 972 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 973 | "rcu_read_lock_bh() used illegally while idle"); |
bc33f24b | 974 | } |
1da177e4 LT |
975 | |
976 | /* | |
977 | * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section | |
978 | * | |
979 | * See rcu_read_lock_bh() for more information. | |
980 | */ | |
bc33f24b PM |
981 | static inline void rcu_read_unlock_bh(void) |
982 | { | |
5c173eb8 | 983 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 984 | "rcu_read_unlock_bh() used illegally while idle"); |
d8ab29f8 | 985 | rcu_lock_release(&rcu_bh_lock_map); |
bc33f24b | 986 | __release(RCU_BH); |
6206ab9b | 987 | local_bh_enable(); |
bc33f24b | 988 | } |
1da177e4 | 989 | |
1c50b728 | 990 | /** |
ca5ecddf | 991 | * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section |
1c50b728 | 992 | * |
ca5ecddf PM |
993 | * This is equivalent of rcu_read_lock(), but to be used when updates |
994 | * are being done using call_rcu_sched() or synchronize_rcu_sched(). | |
995 | * Read-side critical sections can also be introduced by anything that | |
996 | * disables preemption, including local_irq_disable() and friends. | |
3842a083 PM |
997 | * |
998 | * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() | |
999 | * must occur in the same context, for example, it is illegal to invoke | |
1000 | * rcu_read_unlock_sched() from process context if the matching | |
1001 | * rcu_read_lock_sched() was invoked from an NMI handler. | |
1c50b728 | 1002 | */ |
d6714c22 PM |
1003 | static inline void rcu_read_lock_sched(void) |
1004 | { | |
1005 | preempt_disable(); | |
bc33f24b | 1006 | __acquire(RCU_SCHED); |
d8ab29f8 | 1007 | rcu_lock_acquire(&rcu_sched_lock_map); |
5c173eb8 | 1008 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 1009 | "rcu_read_lock_sched() used illegally while idle"); |
d6714c22 | 1010 | } |
1eba8f84 PM |
1011 | |
1012 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 1013 | static inline notrace void rcu_read_lock_sched_notrace(void) |
d6714c22 PM |
1014 | { |
1015 | preempt_disable_notrace(); | |
bc33f24b | 1016 | __acquire(RCU_SCHED); |
d6714c22 | 1017 | } |
1c50b728 MD |
1018 | |
1019 | /* | |
1020 | * rcu_read_unlock_sched - marks the end of a RCU-classic critical section | |
1021 | * | |
1022 | * See rcu_read_lock_sched for more information. | |
1023 | */ | |
d6714c22 PM |
1024 | static inline void rcu_read_unlock_sched(void) |
1025 | { | |
5c173eb8 | 1026 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 1027 | "rcu_read_unlock_sched() used illegally while idle"); |
d8ab29f8 | 1028 | rcu_lock_release(&rcu_sched_lock_map); |
bc33f24b | 1029 | __release(RCU_SCHED); |
d6714c22 PM |
1030 | preempt_enable(); |
1031 | } | |
1eba8f84 PM |
1032 | |
1033 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 1034 | static inline notrace void rcu_read_unlock_sched_notrace(void) |
d6714c22 | 1035 | { |
bc33f24b | 1036 | __release(RCU_SCHED); |
d6714c22 PM |
1037 | preempt_enable_notrace(); |
1038 | } | |
1c50b728 | 1039 | |
ca5ecddf PM |
1040 | /** |
1041 | * RCU_INIT_POINTER() - initialize an RCU protected pointer | |
1042 | * | |
6846c0c5 PM |
1043 | * Initialize an RCU-protected pointer in special cases where readers |
1044 | * do not need ordering constraints on the CPU or the compiler. These | |
1045 | * special cases are: | |
1046 | * | |
1047 | * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- | |
1048 | * 2. The caller has taken whatever steps are required to prevent | |
1049 | * RCU readers from concurrently accessing this pointer -or- | |
1050 | * 3. The referenced data structure has already been exposed to | |
1051 | * readers either at compile time or via rcu_assign_pointer() -and- | |
1052 | * a. You have not made -any- reader-visible changes to | |
1053 | * this structure since then -or- | |
1054 | * b. It is OK for readers accessing this structure from its | |
1055 | * new location to see the old state of the structure. (For | |
1056 | * example, the changes were to statistical counters or to | |
1057 | * other state where exact synchronization is not required.) | |
1058 | * | |
1059 | * Failure to follow these rules governing use of RCU_INIT_POINTER() will | |
1060 | * result in impossible-to-diagnose memory corruption. As in the structures | |
1061 | * will look OK in crash dumps, but any concurrent RCU readers might | |
1062 | * see pre-initialized values of the referenced data structure. So | |
1063 | * please be very careful how you use RCU_INIT_POINTER()!!! | |
1064 | * | |
1065 | * If you are creating an RCU-protected linked structure that is accessed | |
1066 | * by a single external-to-structure RCU-protected pointer, then you may | |
1067 | * use RCU_INIT_POINTER() to initialize the internal RCU-protected | |
1068 | * pointers, but you must use rcu_assign_pointer() to initialize the | |
1069 | * external-to-structure pointer -after- you have completely initialized | |
1070 | * the reader-accessible portions of the linked structure. | |
71a9b269 PM |
1071 | * |
1072 | * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no | |
1073 | * ordering guarantees for either the CPU or the compiler. | |
ca5ecddf PM |
1074 | */ |
1075 | #define RCU_INIT_POINTER(p, v) \ | |
d1b88eb9 | 1076 | do { \ |
1a6c9b26 | 1077 | rcu_dereference_sparse(p, __rcu); \ |
462225ae | 1078 | p = RCU_INITIALIZER(v); \ |
d1b88eb9 | 1079 | } while (0) |
9ab1544e | 1080 | |
172708d0 PM |
1081 | /** |
1082 | * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer | |
1083 | * | |
1084 | * GCC-style initialization for an RCU-protected pointer in a structure field. | |
1085 | */ | |
1086 | #define RCU_POINTER_INITIALIZER(p, v) \ | |
462225ae | 1087 | .p = RCU_INITIALIZER(v) |
9ab1544e | 1088 | |
d8169d4c JE |
1089 | /* |
1090 | * Does the specified offset indicate that the corresponding rcu_head | |
1091 | * structure can be handled by kfree_rcu()? | |
1092 | */ | |
1093 | #define __is_kfree_rcu_offset(offset) ((offset) < 4096) | |
1094 | ||
1095 | /* | |
1096 | * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. | |
1097 | */ | |
1098 | #define __kfree_rcu(head, offset) \ | |
1099 | do { \ | |
1100 | BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ | |
4fa3b6cb | 1101 | kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \ |
d8169d4c JE |
1102 | } while (0) |
1103 | ||
9ab1544e LJ |
1104 | /** |
1105 | * kfree_rcu() - kfree an object after a grace period. | |
1106 | * @ptr: pointer to kfree | |
1107 | * @rcu_head: the name of the struct rcu_head within the type of @ptr. | |
1108 | * | |
1109 | * Many rcu callbacks functions just call kfree() on the base structure. | |
1110 | * These functions are trivial, but their size adds up, and furthermore | |
1111 | * when they are used in a kernel module, that module must invoke the | |
1112 | * high-latency rcu_barrier() function at module-unload time. | |
1113 | * | |
1114 | * The kfree_rcu() function handles this issue. Rather than encoding a | |
1115 | * function address in the embedded rcu_head structure, kfree_rcu() instead | |
1116 | * encodes the offset of the rcu_head structure within the base structure. | |
1117 | * Because the functions are not allowed in the low-order 4096 bytes of | |
1118 | * kernel virtual memory, offsets up to 4095 bytes can be accommodated. | |
1119 | * If the offset is larger than 4095 bytes, a compile-time error will | |
1120 | * be generated in __kfree_rcu(). If this error is triggered, you can | |
1121 | * either fall back to use of call_rcu() or rearrange the structure to | |
1122 | * position the rcu_head structure into the first 4096 bytes. | |
1123 | * | |
1124 | * Note that the allowable offset might decrease in the future, for example, | |
1125 | * to allow something like kmem_cache_free_rcu(). | |
d8169d4c JE |
1126 | * |
1127 | * The BUILD_BUG_ON check must not involve any function calls, hence the | |
1128 | * checks are done in macros here. | |
9ab1544e LJ |
1129 | */ |
1130 | #define kfree_rcu(ptr, rcu_head) \ | |
1131 | __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) | |
1132 | ||
ffa83fb5 | 1133 | #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) |
aa6da514 | 1134 | static inline int rcu_needs_cpu(unsigned long *delta_jiffies) |
ffa83fb5 PM |
1135 | { |
1136 | *delta_jiffies = ULONG_MAX; | |
1137 | return 0; | |
1138 | } | |
1139 | #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */ | |
1140 | ||
2f33b512 PM |
1141 | #if defined(CONFIG_RCU_NOCB_CPU_ALL) |
1142 | static inline bool rcu_is_nocb_cpu(int cpu) { return true; } | |
1143 | #elif defined(CONFIG_RCU_NOCB_CPU) | |
584dc4ce | 1144 | bool rcu_is_nocb_cpu(int cpu); |
d1e43fa5 FW |
1145 | #else |
1146 | static inline bool rcu_is_nocb_cpu(int cpu) { return false; } | |
2f33b512 | 1147 | #endif |
d1e43fa5 FW |
1148 | |
1149 | ||
0edd1b17 PM |
1150 | /* Only for use by adaptive-ticks code. */ |
1151 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
584dc4ce TB |
1152 | bool rcu_sys_is_idle(void); |
1153 | void rcu_sysidle_force_exit(void); | |
0edd1b17 PM |
1154 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ |
1155 | ||
1156 | static inline bool rcu_sys_is_idle(void) | |
1157 | { | |
1158 | return false; | |
1159 | } | |
1160 | ||
1161 | static inline void rcu_sysidle_force_exit(void) | |
1162 | { | |
1163 | } | |
1164 | ||
1165 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
1166 | ||
1167 | ||
1da177e4 | 1168 | #endif /* __LINUX_RCUPDATE_H */ |