]>
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 DY |
50 | #ifdef CONFIG_RCU_TORTURE_TEST |
51 | extern int rcutorture_runnable; /* for sysctl */ | |
52 | #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */ | |
53 | ||
ad0dc7f9 PM |
54 | enum rcutorture_type { |
55 | RCU_FLAVOR, | |
56 | RCU_BH_FLAVOR, | |
57 | RCU_SCHED_FLAVOR, | |
69c60455 | 58 | RCU_TASKS_FLAVOR, |
ad0dc7f9 PM |
59 | SRCU_FLAVOR, |
60 | INVALID_RCU_FLAVOR | |
61 | }; | |
62 | ||
4a298656 | 63 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) |
ad0dc7f9 PM |
64 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, |
65 | unsigned long *gpnum, unsigned long *completed); | |
584dc4ce TB |
66 | void rcutorture_record_test_transition(void); |
67 | void rcutorture_record_progress(unsigned long vernum); | |
68 | void do_trace_rcu_torture_read(const char *rcutorturename, | |
69 | struct rcu_head *rhp, | |
70 | unsigned long secs, | |
71 | unsigned long c_old, | |
72 | unsigned long c); | |
4a298656 | 73 | #else |
ad0dc7f9 PM |
74 | static inline void rcutorture_get_gp_data(enum rcutorture_type test_type, |
75 | int *flags, | |
76 | unsigned long *gpnum, | |
77 | unsigned long *completed) | |
78 | { | |
79 | *flags = 0; | |
80 | *gpnum = 0; | |
81 | *completed = 0; | |
82 | } | |
4a298656 PM |
83 | static inline void rcutorture_record_test_transition(void) |
84 | { | |
85 | } | |
86 | static inline void rcutorture_record_progress(unsigned long vernum) | |
87 | { | |
88 | } | |
91afaf30 | 89 | #ifdef CONFIG_RCU_TRACE |
584dc4ce TB |
90 | void do_trace_rcu_torture_read(const char *rcutorturename, |
91 | struct rcu_head *rhp, | |
92 | unsigned long secs, | |
93 | unsigned long c_old, | |
94 | unsigned long c); | |
91afaf30 | 95 | #else |
52494535 PM |
96 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ |
97 | do { } while (0) | |
91afaf30 | 98 | #endif |
4a298656 PM |
99 | #endif |
100 | ||
e27fc964 TH |
101 | #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) |
102 | #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) | |
a3dc3fb1 PM |
103 | #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) |
104 | #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) | |
c0f4dfd4 | 105 | #define ulong2long(a) (*(long *)(&(a))) |
a3dc3fb1 | 106 | |
03b042bf | 107 | /* Exported common interfaces */ |
2c42818e PM |
108 | |
109 | #ifdef CONFIG_PREEMPT_RCU | |
110 | ||
111 | /** | |
112 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
113 | * @head: structure to be used for queueing the RCU updates. | |
114 | * @func: actual callback function to be invoked after the grace period | |
115 | * | |
116 | * The callback function will be invoked some time after a full grace | |
117 | * period elapses, in other words after all pre-existing RCU read-side | |
118 | * critical sections have completed. However, the callback function | |
119 | * might well execute concurrently with RCU read-side critical sections | |
120 | * that started after call_rcu() was invoked. RCU read-side critical | |
121 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
122 | * and may be nested. | |
f0a0e6f2 PM |
123 | * |
124 | * Note that all CPUs must agree that the grace period extended beyond | |
125 | * all pre-existing RCU read-side critical section. On systems with more | |
126 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
127 | * guaranteed to have executed a full memory barrier since the end of its | |
128 | * last RCU read-side critical section whose beginning preceded the call | |
129 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
130 | * critical section that continues beyond the start of "func()" must have | |
131 | * executed a memory barrier after the call_rcu() but before the beginning | |
132 | * of that RCU read-side critical section. Note that these guarantees | |
133 | * include CPUs that are offline, idle, or executing in user mode, as | |
134 | * well as CPUs that are executing in the kernel. | |
135 | * | |
136 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
137 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
138 | * guaranteed to execute a full memory barrier during the time interval | |
139 | * between the call to call_rcu() and the invocation of "func()" -- even | |
140 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
141 | * more than one CPU). | |
2c42818e | 142 | */ |
584dc4ce TB |
143 | void call_rcu(struct rcu_head *head, |
144 | void (*func)(struct rcu_head *head)); | |
2c42818e PM |
145 | |
146 | #else /* #ifdef CONFIG_PREEMPT_RCU */ | |
147 | ||
148 | /* In classic RCU, call_rcu() is just call_rcu_sched(). */ | |
149 | #define call_rcu call_rcu_sched | |
150 | ||
151 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
152 | ||
153 | /** | |
154 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
155 | * @head: structure to be used for queueing the RCU updates. | |
156 | * @func: actual callback function to be invoked after the grace period | |
157 | * | |
158 | * The callback function will be invoked some time after a full grace | |
159 | * period elapses, in other words after all currently executing RCU | |
160 | * read-side critical sections have completed. call_rcu_bh() assumes | |
161 | * that the read-side critical sections end on completion of a softirq | |
162 | * handler. This means that read-side critical sections in process | |
163 | * context must not be interrupted by softirqs. This interface is to be | |
164 | * used when most of the read-side critical sections are in softirq context. | |
165 | * RCU read-side critical sections are delimited by : | |
166 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. | |
167 | * OR | |
168 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
169 | * These may be nested. | |
f0a0e6f2 PM |
170 | * |
171 | * See the description of call_rcu() for more detailed information on | |
172 | * memory ordering guarantees. | |
2c42818e | 173 | */ |
584dc4ce TB |
174 | void call_rcu_bh(struct rcu_head *head, |
175 | void (*func)(struct rcu_head *head)); | |
2c42818e PM |
176 | |
177 | /** | |
178 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
179 | * @head: structure to be used for queueing the RCU updates. | |
180 | * @func: actual callback function to be invoked after the grace period | |
181 | * | |
182 | * The callback function will be invoked some time after a full grace | |
183 | * period elapses, in other words after all currently executing RCU | |
184 | * read-side critical sections have completed. call_rcu_sched() assumes | |
185 | * that the read-side critical sections end on enabling of preemption | |
186 | * or on voluntary preemption. | |
187 | * RCU read-side critical sections are delimited by : | |
188 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), | |
189 | * OR | |
190 | * anything that disables preemption. | |
191 | * These may be nested. | |
f0a0e6f2 PM |
192 | * |
193 | * See the description of call_rcu() for more detailed information on | |
194 | * memory ordering guarantees. | |
2c42818e | 195 | */ |
584dc4ce TB |
196 | void call_rcu_sched(struct rcu_head *head, |
197 | void (*func)(struct rcu_head *rcu)); | |
2c42818e | 198 | |
584dc4ce | 199 | void synchronize_sched(void); |
03b042bf | 200 | |
8315f422 PM |
201 | /** |
202 | * call_rcu_tasks() - Queue an RCU for invocation task-based grace period | |
203 | * @head: structure to be used for queueing the RCU updates. | |
204 | * @func: actual callback function to be invoked after the grace period | |
205 | * | |
206 | * The callback function will be invoked some time after a full grace | |
207 | * period elapses, in other words after all currently executing RCU | |
208 | * read-side critical sections have completed. call_rcu_tasks() assumes | |
209 | * that the read-side critical sections end at a voluntary context | |
210 | * switch (not a preemption!), entry into idle, or transition to usermode | |
211 | * execution. As such, there are no read-side primitives analogous to | |
212 | * rcu_read_lock() and rcu_read_unlock() because this primitive is intended | |
213 | * to determine that all tasks have passed through a safe state, not so | |
214 | * much for data-strcuture synchronization. | |
215 | * | |
216 | * See the description of call_rcu() for more detailed information on | |
217 | * memory ordering guarantees. | |
218 | */ | |
219 | void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head)); | |
53c6d4ed PM |
220 | void synchronize_rcu_tasks(void); |
221 | void rcu_barrier_tasks(void); | |
8315f422 | 222 | |
a3dc3fb1 PM |
223 | #ifdef CONFIG_PREEMPT_RCU |
224 | ||
584dc4ce TB |
225 | void __rcu_read_lock(void); |
226 | void __rcu_read_unlock(void); | |
227 | void rcu_read_unlock_special(struct task_struct *t); | |
7b0b759b PM |
228 | void synchronize_rcu(void); |
229 | ||
a3dc3fb1 PM |
230 | /* |
231 | * Defined as a macro as it is a very low level header included from | |
232 | * areas that don't even know about current. This gives the rcu_read_lock() | |
233 | * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other | |
234 | * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. | |
235 | */ | |
236 | #define rcu_preempt_depth() (current->rcu_read_lock_nesting) | |
237 | ||
7b0b759b PM |
238 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
239 | ||
240 | static inline void __rcu_read_lock(void) | |
241 | { | |
242 | preempt_disable(); | |
243 | } | |
244 | ||
245 | static inline void __rcu_read_unlock(void) | |
246 | { | |
247 | preempt_enable(); | |
248 | } | |
249 | ||
250 | static inline void synchronize_rcu(void) | |
251 | { | |
252 | synchronize_sched(); | |
253 | } | |
254 | ||
255 | static inline int rcu_preempt_depth(void) | |
256 | { | |
257 | return 0; | |
258 | } | |
259 | ||
260 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ | |
261 | ||
262 | /* Internal to kernel */ | |
584dc4ce TB |
263 | void rcu_init(void); |
264 | void rcu_sched_qs(int cpu); | |
265 | void rcu_bh_qs(int cpu); | |
266 | void rcu_check_callbacks(int cpu, int user); | |
7b0b759b | 267 | struct notifier_block; |
584dc4ce TB |
268 | void rcu_idle_enter(void); |
269 | void rcu_idle_exit(void); | |
270 | void rcu_irq_enter(void); | |
271 | void rcu_irq_exit(void); | |
2b1d5024 | 272 | |
61f38db3 RR |
273 | #ifdef CONFIG_RCU_STALL_COMMON |
274 | void rcu_sysrq_start(void); | |
275 | void rcu_sysrq_end(void); | |
276 | #else /* #ifdef CONFIG_RCU_STALL_COMMON */ | |
277 | static inline void rcu_sysrq_start(void) | |
278 | { | |
279 | } | |
280 | static inline void rcu_sysrq_end(void) | |
281 | { | |
282 | } | |
283 | #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ | |
284 | ||
2b1d5024 | 285 | #ifdef CONFIG_RCU_USER_QS |
584dc4ce TB |
286 | void rcu_user_enter(void); |
287 | void rcu_user_exit(void); | |
2b1d5024 FW |
288 | #else |
289 | static inline void rcu_user_enter(void) { } | |
290 | static inline void rcu_user_exit(void) { } | |
4d9a5d43 FW |
291 | static inline void rcu_user_hooks_switch(struct task_struct *prev, |
292 | struct task_struct *next) { } | |
2b1d5024 FW |
293 | #endif /* CONFIG_RCU_USER_QS */ |
294 | ||
8a2ecf47 PM |
295 | /** |
296 | * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers | |
297 | * @a: Code that RCU needs to pay attention to. | |
298 | * | |
299 | * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden | |
300 | * in the inner idle loop, that is, between the rcu_idle_enter() and | |
301 | * the rcu_idle_exit() -- RCU will happily ignore any such read-side | |
302 | * critical sections. However, things like powertop need tracepoints | |
303 | * in the inner idle loop. | |
304 | * | |
305 | * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) | |
306 | * will tell RCU that it needs to pay attending, invoke its argument | |
307 | * (in this example, a call to the do_something_with_RCU() function), | |
308 | * and then tell RCU to go back to ignoring this CPU. It is permissible | |
309 | * to nest RCU_NONIDLE() wrappers, but the nesting level is currently | |
310 | * quite limited. If deeper nesting is required, it will be necessary | |
311 | * to adjust DYNTICK_TASK_NESTING_VALUE accordingly. | |
8a2ecf47 PM |
312 | */ |
313 | #define RCU_NONIDLE(a) \ | |
314 | do { \ | |
b4270ee3 | 315 | rcu_irq_enter(); \ |
8a2ecf47 | 316 | do { a; } while (0); \ |
b4270ee3 | 317 | rcu_irq_exit(); \ |
8a2ecf47 PM |
318 | } while (0) |
319 | ||
8315f422 PM |
320 | /* |
321 | * Note a voluntary context switch for RCU-tasks benefit. This is a | |
322 | * macro rather than an inline function to avoid #include hell. | |
323 | */ | |
324 | #ifdef CONFIG_TASKS_RCU | |
3f95aa81 PM |
325 | #define TASKS_RCU(x) x |
326 | extern struct srcu_struct tasks_rcu_exit_srcu; | |
8315f422 PM |
327 | #define rcu_note_voluntary_context_switch(t) \ |
328 | do { \ | |
8315f422 PM |
329 | if (ACCESS_ONCE((t)->rcu_tasks_holdout)) \ |
330 | ACCESS_ONCE((t)->rcu_tasks_holdout) = false; \ | |
8315f422 PM |
331 | } while (0) |
332 | #else /* #ifdef CONFIG_TASKS_RCU */ | |
3f95aa81 | 333 | #define TASKS_RCU(x) do { } while (0) |
8315f422 PM |
334 | #define rcu_note_voluntary_context_switch(t) do { } while (0) |
335 | #endif /* #else #ifdef CONFIG_TASKS_RCU */ | |
336 | ||
bde6c3aa PM |
337 | /** |
338 | * cond_resched_rcu_qs - Report potential quiescent states to RCU | |
339 | * | |
340 | * This macro resembles cond_resched(), except that it is defined to | |
341 | * report potential quiescent states to RCU-tasks even if the cond_resched() | |
342 | * machinery were to be shut off, as some advocate for PREEMPT kernels. | |
343 | */ | |
344 | #define cond_resched_rcu_qs() \ | |
345 | do { \ | |
346 | rcu_note_voluntary_context_switch(current); \ | |
347 | cond_resched(); \ | |
348 | } while (0) | |
349 | ||
cc6783f7 | 350 | #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) |
584dc4ce | 351 | bool __rcu_is_watching(void); |
cc6783f7 PM |
352 | #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */ |
353 | ||
2c42818e PM |
354 | /* |
355 | * Infrastructure to implement the synchronize_() primitives in | |
356 | * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. | |
357 | */ | |
358 | ||
359 | typedef void call_rcu_func_t(struct rcu_head *head, | |
360 | void (*func)(struct rcu_head *head)); | |
361 | void wait_rcu_gp(call_rcu_func_t crf); | |
362 | ||
f41d911f | 363 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) |
64db4cff | 364 | #include <linux/rcutree.h> |
127781d1 | 365 | #elif defined(CONFIG_TINY_RCU) |
9b1d82fa | 366 | #include <linux/rcutiny.h> |
64db4cff PM |
367 | #else |
368 | #error "Unknown RCU implementation specified to kernel configuration" | |
6b3ef48a | 369 | #endif |
01c1c660 | 370 | |
551d55a9 MD |
371 | /* |
372 | * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic | |
373 | * initialization and destruction of rcu_head on the stack. rcu_head structures | |
374 | * allocated dynamically in the heap or defined statically don't need any | |
375 | * initialization. | |
376 | */ | |
377 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
546a9d85 PM |
378 | void init_rcu_head(struct rcu_head *head); |
379 | void destroy_rcu_head(struct rcu_head *head); | |
584dc4ce TB |
380 | void init_rcu_head_on_stack(struct rcu_head *head); |
381 | void destroy_rcu_head_on_stack(struct rcu_head *head); | |
551d55a9 | 382 | #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
546a9d85 PM |
383 | static inline void init_rcu_head(struct rcu_head *head) |
384 | { | |
385 | } | |
386 | ||
387 | static inline void destroy_rcu_head(struct rcu_head *head) | |
388 | { | |
389 | } | |
390 | ||
4376030a MD |
391 | static inline void init_rcu_head_on_stack(struct rcu_head *head) |
392 | { | |
393 | } | |
394 | ||
395 | static inline void destroy_rcu_head_on_stack(struct rcu_head *head) | |
396 | { | |
397 | } | |
551d55a9 | 398 | #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ |
4376030a | 399 | |
c0d6d01b PM |
400 | #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) |
401 | bool rcu_lockdep_current_cpu_online(void); | |
402 | #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
403 | static inline bool rcu_lockdep_current_cpu_online(void) | |
404 | { | |
405 | return 1; | |
406 | } | |
407 | #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ | |
408 | ||
bc33f24b | 409 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
632ee200 | 410 | |
00f49e57 FW |
411 | static inline void rcu_lock_acquire(struct lockdep_map *map) |
412 | { | |
fb9edbe9 | 413 | lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); |
00f49e57 FW |
414 | } |
415 | ||
416 | static inline void rcu_lock_release(struct lockdep_map *map) | |
417 | { | |
00f49e57 FW |
418 | lock_release(map, 1, _THIS_IP_); |
419 | } | |
420 | ||
bc33f24b | 421 | extern struct lockdep_map rcu_lock_map; |
632ee200 | 422 | extern struct lockdep_map rcu_bh_lock_map; |
632ee200 | 423 | extern struct lockdep_map rcu_sched_lock_map; |
24ef659a | 424 | extern struct lockdep_map rcu_callback_map; |
a235c091 | 425 | int debug_lockdep_rcu_enabled(void); |
54dbf96c | 426 | |
632ee200 | 427 | /** |
ca5ecddf | 428 | * rcu_read_lock_held() - might we be in RCU read-side critical section? |
632ee200 | 429 | * |
d20200b5 PM |
430 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU |
431 | * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC, | |
632ee200 | 432 | * this assumes we are in an RCU read-side critical section unless it can |
ca5ecddf PM |
433 | * prove otherwise. This is useful for debug checks in functions that |
434 | * require that they be called within an RCU read-side critical section. | |
54dbf96c | 435 | * |
ca5ecddf | 436 | * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot |
32c141a0 | 437 | * and while lockdep is disabled. |
3842a083 PM |
438 | * |
439 | * Note that rcu_read_lock() and the matching rcu_read_unlock() must | |
440 | * occur in the same context, for example, it is illegal to invoke | |
441 | * rcu_read_unlock() in process context if the matching rcu_read_lock() | |
442 | * was invoked from within an irq handler. | |
c0d6d01b PM |
443 | * |
444 | * Note that rcu_read_lock() is disallowed if the CPU is either idle or | |
445 | * offline from an RCU perspective, so check for those as well. | |
632ee200 PM |
446 | */ |
447 | static inline int rcu_read_lock_held(void) | |
448 | { | |
54dbf96c PM |
449 | if (!debug_lockdep_rcu_enabled()) |
450 | return 1; | |
5c173eb8 | 451 | if (!rcu_is_watching()) |
e6b80a3b | 452 | return 0; |
c0d6d01b PM |
453 | if (!rcu_lockdep_current_cpu_online()) |
454 | return 0; | |
54dbf96c | 455 | return lock_is_held(&rcu_lock_map); |
632ee200 PM |
456 | } |
457 | ||
e3818b8d PM |
458 | /* |
459 | * rcu_read_lock_bh_held() is defined out of line to avoid #include-file | |
460 | * hell. | |
632ee200 | 461 | */ |
584dc4ce | 462 | int rcu_read_lock_bh_held(void); |
632ee200 PM |
463 | |
464 | /** | |
ca5ecddf | 465 | * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? |
632ee200 | 466 | * |
d20200b5 PM |
467 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an |
468 | * RCU-sched read-side critical section. In absence of | |
469 | * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side | |
470 | * critical section unless it can prove otherwise. Note that disabling | |
471 | * of preemption (including disabling irqs) counts as an RCU-sched | |
ca5ecddf PM |
472 | * read-side critical section. This is useful for debug checks in functions |
473 | * that required that they be called within an RCU-sched read-side | |
474 | * critical section. | |
54dbf96c | 475 | * |
32c141a0 PM |
476 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot |
477 | * and while lockdep is disabled. | |
e6b80a3b FW |
478 | * |
479 | * Note that if the CPU is in the idle loop from an RCU point of | |
480 | * view (ie: that we are in the section between rcu_idle_enter() and | |
481 | * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU | |
482 | * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs | |
483 | * that are in such a section, considering these as in extended quiescent | |
484 | * state, so such a CPU is effectively never in an RCU read-side critical | |
485 | * section regardless of what RCU primitives it invokes. This state of | |
486 | * affairs is required --- we need to keep an RCU-free window in idle | |
487 | * where the CPU may possibly enter into low power mode. This way we can | |
488 | * notice an extended quiescent state to other CPUs that started a grace | |
489 | * period. Otherwise we would delay any grace period as long as we run in | |
490 | * the idle task. | |
c0d6d01b PM |
491 | * |
492 | * Similarly, we avoid claiming an SRCU read lock held if the current | |
493 | * CPU is offline. | |
632ee200 | 494 | */ |
bdd4e85d | 495 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
496 | static inline int rcu_read_lock_sched_held(void) |
497 | { | |
498 | int lockdep_opinion = 0; | |
499 | ||
54dbf96c PM |
500 | if (!debug_lockdep_rcu_enabled()) |
501 | return 1; | |
5c173eb8 | 502 | if (!rcu_is_watching()) |
e6b80a3b | 503 | return 0; |
c0d6d01b PM |
504 | if (!rcu_lockdep_current_cpu_online()) |
505 | return 0; | |
632ee200 PM |
506 | if (debug_locks) |
507 | lockdep_opinion = lock_is_held(&rcu_sched_lock_map); | |
0cff810f | 508 | return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); |
632ee200 | 509 | } |
bdd4e85d | 510 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
511 | static inline int rcu_read_lock_sched_held(void) |
512 | { | |
513 | return 1; | |
632ee200 | 514 | } |
bdd4e85d | 515 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
516 | |
517 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
518 | ||
d8ab29f8 PM |
519 | # define rcu_lock_acquire(a) do { } while (0) |
520 | # define rcu_lock_release(a) do { } while (0) | |
632ee200 PM |
521 | |
522 | static inline int rcu_read_lock_held(void) | |
523 | { | |
524 | return 1; | |
525 | } | |
526 | ||
527 | static inline int rcu_read_lock_bh_held(void) | |
528 | { | |
529 | return 1; | |
530 | } | |
531 | ||
bdd4e85d | 532 | #ifdef CONFIG_PREEMPT_COUNT |
632ee200 PM |
533 | static inline int rcu_read_lock_sched_held(void) |
534 | { | |
bbad9379 | 535 | return preempt_count() != 0 || irqs_disabled(); |
632ee200 | 536 | } |
bdd4e85d | 537 | #else /* #ifdef CONFIG_PREEMPT_COUNT */ |
e6033e3b PM |
538 | static inline int rcu_read_lock_sched_held(void) |
539 | { | |
540 | return 1; | |
632ee200 | 541 | } |
bdd4e85d | 542 | #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ |
632ee200 PM |
543 | |
544 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
545 | ||
546 | #ifdef CONFIG_PROVE_RCU | |
547 | ||
4221a991 TH |
548 | /** |
549 | * rcu_lockdep_assert - emit lockdep splat if specified condition not met | |
550 | * @c: condition to check | |
b3fbab05 | 551 | * @s: informative message |
4221a991 | 552 | */ |
b3fbab05 | 553 | #define rcu_lockdep_assert(c, s) \ |
2b3fc35f | 554 | do { \ |
7ccaba53 | 555 | static bool __section(.data.unlikely) __warned; \ |
2b3fc35f LJ |
556 | if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ |
557 | __warned = true; \ | |
b3fbab05 | 558 | lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ |
2b3fc35f LJ |
559 | } \ |
560 | } while (0) | |
561 | ||
50406b98 PM |
562 | #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) |
563 | static inline void rcu_preempt_sleep_check(void) | |
564 | { | |
565 | rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), | |
5cf05ad7 | 566 | "Illegal context switch in RCU read-side critical section"); |
50406b98 PM |
567 | } |
568 | #else /* #ifdef CONFIG_PROVE_RCU */ | |
569 | static inline void rcu_preempt_sleep_check(void) | |
570 | { | |
571 | } | |
572 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
573 | ||
b3fbab05 PM |
574 | #define rcu_sleep_check() \ |
575 | do { \ | |
50406b98 | 576 | rcu_preempt_sleep_check(); \ |
b3fbab05 | 577 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \ |
41f4abd9 | 578 | "Illegal context switch in RCU-bh read-side critical section"); \ |
b3fbab05 | 579 | rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \ |
41f4abd9 | 580 | "Illegal context switch in RCU-sched read-side critical section"); \ |
b3fbab05 PM |
581 | } while (0) |
582 | ||
ca5ecddf PM |
583 | #else /* #ifdef CONFIG_PROVE_RCU */ |
584 | ||
b3fbab05 PM |
585 | #define rcu_lockdep_assert(c, s) do { } while (0) |
586 | #define rcu_sleep_check() do { } while (0) | |
ca5ecddf PM |
587 | |
588 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
589 | ||
590 | /* | |
591 | * Helper functions for rcu_dereference_check(), rcu_dereference_protected() | |
592 | * and rcu_assign_pointer(). Some of these could be folded into their | |
593 | * callers, but they are left separate in order to ease introduction of | |
594 | * multiple flavors of pointers to match the multiple flavors of RCU | |
595 | * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in | |
596 | * the future. | |
597 | */ | |
53ecfba2 PM |
598 | |
599 | #ifdef __CHECKER__ | |
600 | #define rcu_dereference_sparse(p, space) \ | |
601 | ((void)(((typeof(*p) space *)p) == p)) | |
602 | #else /* #ifdef __CHECKER__ */ | |
603 | #define rcu_dereference_sparse(p, space) | |
604 | #endif /* #else #ifdef __CHECKER__ */ | |
605 | ||
ca5ecddf | 606 | #define __rcu_access_pointer(p, space) \ |
0adab9b9 JP |
607 | ({ \ |
608 | typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \ | |
609 | rcu_dereference_sparse(p, space); \ | |
610 | ((typeof(*p) __force __kernel *)(_________p1)); \ | |
611 | }) | |
ca5ecddf | 612 | #define __rcu_dereference_check(p, c, space) \ |
0adab9b9 JP |
613 | ({ \ |
614 | typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \ | |
615 | rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \ | |
616 | rcu_dereference_sparse(p, space); \ | |
617 | smp_read_barrier_depends(); /* Dependency order vs. p above. */ \ | |
618 | ((typeof(*p) __force __kernel *)(_________p1)); \ | |
619 | }) | |
ca5ecddf | 620 | #define __rcu_dereference_protected(p, c, space) \ |
0adab9b9 JP |
621 | ({ \ |
622 | rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \ | |
623 | rcu_dereference_sparse(p, space); \ | |
624 | ((typeof(*p) __force __kernel *)(p)); \ | |
625 | }) | |
ca5ecddf | 626 | |
a4dd9925 | 627 | #define __rcu_access_index(p, space) \ |
0adab9b9 JP |
628 | ({ \ |
629 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
630 | rcu_dereference_sparse(p, space); \ | |
631 | (_________p1); \ | |
632 | }) | |
ca5ecddf | 633 | #define __rcu_dereference_index_check(p, c) \ |
0adab9b9 JP |
634 | ({ \ |
635 | typeof(p) _________p1 = ACCESS_ONCE(p); \ | |
636 | rcu_lockdep_assert(c, \ | |
637 | "suspicious rcu_dereference_index_check() usage"); \ | |
638 | smp_read_barrier_depends(); /* Dependency order vs. p above. */ \ | |
639 | (_________p1); \ | |
640 | }) | |
462225ae PM |
641 | |
642 | /** | |
643 | * RCU_INITIALIZER() - statically initialize an RCU-protected global variable | |
644 | * @v: The value to statically initialize with. | |
645 | */ | |
646 | #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) | |
647 | ||
648 | /** | |
649 | * rcu_assign_pointer() - assign to RCU-protected pointer | |
650 | * @p: pointer to assign to | |
651 | * @v: value to assign (publish) | |
652 | * | |
653 | * Assigns the specified value to the specified RCU-protected | |
654 | * pointer, ensuring that any concurrent RCU readers will see | |
655 | * any prior initialization. | |
656 | * | |
657 | * Inserts memory barriers on architectures that require them | |
658 | * (which is most of them), and also prevents the compiler from | |
659 | * reordering the code that initializes the structure after the pointer | |
660 | * assignment. More importantly, this call documents which pointers | |
661 | * will be dereferenced by RCU read-side code. | |
662 | * | |
663 | * In some special cases, you may use RCU_INIT_POINTER() instead | |
664 | * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due | |
665 | * to the fact that it does not constrain either the CPU or the compiler. | |
666 | * That said, using RCU_INIT_POINTER() when you should have used | |
667 | * rcu_assign_pointer() is a very bad thing that results in | |
668 | * impossible-to-diagnose memory corruption. So please be careful. | |
669 | * See the RCU_INIT_POINTER() comment header for details. | |
670 | * | |
671 | * Note that rcu_assign_pointer() evaluates each of its arguments only | |
672 | * once, appearances notwithstanding. One of the "extra" evaluations | |
673 | * is in typeof() and the other visible only to sparse (__CHECKER__), | |
674 | * neither of which actually execute the argument. As with most cpp | |
675 | * macros, this execute-arguments-only-once property is important, so | |
676 | * please be careful when making changes to rcu_assign_pointer() and the | |
677 | * other macros that it invokes. | |
678 | */ | |
88c18630 | 679 | #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v)) |
ca5ecddf PM |
680 | |
681 | /** | |
682 | * rcu_access_pointer() - fetch RCU pointer with no dereferencing | |
683 | * @p: The pointer to read | |
684 | * | |
685 | * Return the value of the specified RCU-protected pointer, but omit the | |
686 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
687 | * when the value of this pointer is accessed, but the pointer is not | |
688 | * dereferenced, for example, when testing an RCU-protected pointer against | |
689 | * NULL. Although rcu_access_pointer() may also be used in cases where | |
690 | * update-side locks prevent the value of the pointer from changing, you | |
691 | * should instead use rcu_dereference_protected() for this use case. | |
5e1ee6e1 PM |
692 | * |
693 | * It is also permissible to use rcu_access_pointer() when read-side | |
694 | * access to the pointer was removed at least one grace period ago, as | |
695 | * is the case in the context of the RCU callback that is freeing up | |
696 | * the data, or after a synchronize_rcu() returns. This can be useful | |
697 | * when tearing down multi-linked structures after a grace period | |
698 | * has elapsed. | |
ca5ecddf PM |
699 | */ |
700 | #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) | |
701 | ||
632ee200 | 702 | /** |
ca5ecddf | 703 | * rcu_dereference_check() - rcu_dereference with debug checking |
c08c68dd DH |
704 | * @p: The pointer to read, prior to dereferencing |
705 | * @c: The conditions under which the dereference will take place | |
632ee200 | 706 | * |
c08c68dd | 707 | * Do an rcu_dereference(), but check that the conditions under which the |
ca5ecddf PM |
708 | * dereference will take place are correct. Typically the conditions |
709 | * indicate the various locking conditions that should be held at that | |
710 | * point. The check should return true if the conditions are satisfied. | |
711 | * An implicit check for being in an RCU read-side critical section | |
712 | * (rcu_read_lock()) is included. | |
c08c68dd DH |
713 | * |
714 | * For example: | |
715 | * | |
ca5ecddf | 716 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); |
c08c68dd DH |
717 | * |
718 | * could be used to indicate to lockdep that foo->bar may only be dereferenced | |
ca5ecddf | 719 | * if either rcu_read_lock() is held, or that the lock required to replace |
c08c68dd DH |
720 | * the bar struct at foo->bar is held. |
721 | * | |
722 | * Note that the list of conditions may also include indications of when a lock | |
723 | * need not be held, for example during initialisation or destruction of the | |
724 | * target struct: | |
725 | * | |
ca5ecddf | 726 | * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || |
c08c68dd | 727 | * atomic_read(&foo->usage) == 0); |
ca5ecddf PM |
728 | * |
729 | * Inserts memory barriers on architectures that require them | |
730 | * (currently only the Alpha), prevents the compiler from refetching | |
731 | * (and from merging fetches), and, more importantly, documents exactly | |
732 | * which pointers are protected by RCU and checks that the pointer is | |
733 | * annotated as __rcu. | |
632ee200 PM |
734 | */ |
735 | #define rcu_dereference_check(p, c) \ | |
ca5ecddf PM |
736 | __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu) |
737 | ||
738 | /** | |
739 | * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking | |
740 | * @p: The pointer to read, prior to dereferencing | |
741 | * @c: The conditions under which the dereference will take place | |
742 | * | |
743 | * This is the RCU-bh counterpart to rcu_dereference_check(). | |
744 | */ | |
745 | #define rcu_dereference_bh_check(p, c) \ | |
746 | __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu) | |
632ee200 | 747 | |
b62730ba | 748 | /** |
ca5ecddf PM |
749 | * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking |
750 | * @p: The pointer to read, prior to dereferencing | |
751 | * @c: The conditions under which the dereference will take place | |
752 | * | |
753 | * This is the RCU-sched counterpart to rcu_dereference_check(). | |
754 | */ | |
755 | #define rcu_dereference_sched_check(p, c) \ | |
756 | __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \ | |
757 | __rcu) | |
758 | ||
759 | #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ | |
760 | ||
12bcbe66 SR |
761 | /* |
762 | * The tracing infrastructure traces RCU (we want that), but unfortunately | |
763 | * some of the RCU checks causes tracing to lock up the system. | |
764 | * | |
765 | * The tracing version of rcu_dereference_raw() must not call | |
766 | * rcu_read_lock_held(). | |
767 | */ | |
768 | #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) | |
769 | ||
a4dd9925 PM |
770 | /** |
771 | * rcu_access_index() - fetch RCU index with no dereferencing | |
772 | * @p: The index to read | |
773 | * | |
774 | * Return the value of the specified RCU-protected index, but omit the | |
775 | * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful | |
776 | * when the value of this index is accessed, but the index is not | |
777 | * dereferenced, for example, when testing an RCU-protected index against | |
778 | * -1. Although rcu_access_index() may also be used in cases where | |
779 | * update-side locks prevent the value of the index from changing, you | |
780 | * should instead use rcu_dereference_index_protected() for this use case. | |
781 | */ | |
782 | #define rcu_access_index(p) __rcu_access_index((p), __rcu) | |
783 | ||
ca5ecddf PM |
784 | /** |
785 | * rcu_dereference_index_check() - rcu_dereference for indices with debug checking | |
786 | * @p: The pointer to read, prior to dereferencing | |
787 | * @c: The conditions under which the dereference will take place | |
788 | * | |
789 | * Similar to rcu_dereference_check(), but omits the sparse checking. | |
790 | * This allows rcu_dereference_index_check() to be used on integers, | |
791 | * which can then be used as array indices. Attempting to use | |
792 | * rcu_dereference_check() on an integer will give compiler warnings | |
793 | * because the sparse address-space mechanism relies on dereferencing | |
794 | * the RCU-protected pointer. Dereferencing integers is not something | |
795 | * that even gcc will put up with. | |
796 | * | |
797 | * Note that this function does not implicitly check for RCU read-side | |
798 | * critical sections. If this function gains lots of uses, it might | |
799 | * make sense to provide versions for each flavor of RCU, but it does | |
800 | * not make sense as of early 2010. | |
801 | */ | |
802 | #define rcu_dereference_index_check(p, c) \ | |
803 | __rcu_dereference_index_check((p), (c)) | |
804 | ||
805 | /** | |
806 | * rcu_dereference_protected() - fetch RCU pointer when updates prevented | |
807 | * @p: The pointer to read, prior to dereferencing | |
808 | * @c: The conditions under which the dereference will take place | |
b62730ba PM |
809 | * |
810 | * Return the value of the specified RCU-protected pointer, but omit | |
811 | * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This | |
812 | * is useful in cases where update-side locks prevent the value of the | |
813 | * pointer from changing. Please note that this primitive does -not- | |
814 | * prevent the compiler from repeating this reference or combining it | |
815 | * with other references, so it should not be used without protection | |
816 | * of appropriate locks. | |
ca5ecddf PM |
817 | * |
818 | * This function is only for update-side use. Using this function | |
819 | * when protected only by rcu_read_lock() will result in infrequent | |
820 | * but very ugly failures. | |
b62730ba PM |
821 | */ |
822 | #define rcu_dereference_protected(p, c) \ | |
ca5ecddf | 823 | __rcu_dereference_protected((p), (c), __rcu) |
b62730ba | 824 | |
bc33f24b | 825 | |
b62730ba | 826 | /** |
ca5ecddf PM |
827 | * rcu_dereference() - fetch RCU-protected pointer for dereferencing |
828 | * @p: The pointer to read, prior to dereferencing | |
b62730ba | 829 | * |
ca5ecddf | 830 | * This is a simple wrapper around rcu_dereference_check(). |
b62730ba | 831 | */ |
ca5ecddf | 832 | #define rcu_dereference(p) rcu_dereference_check(p, 0) |
b62730ba | 833 | |
1da177e4 | 834 | /** |
ca5ecddf PM |
835 | * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing |
836 | * @p: The pointer to read, prior to dereferencing | |
837 | * | |
838 | * Makes rcu_dereference_check() do the dirty work. | |
839 | */ | |
840 | #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) | |
841 | ||
842 | /** | |
843 | * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing | |
844 | * @p: The pointer to read, prior to dereferencing | |
845 | * | |
846 | * Makes rcu_dereference_check() do the dirty work. | |
847 | */ | |
848 | #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) | |
849 | ||
850 | /** | |
851 | * rcu_read_lock() - mark the beginning of an RCU read-side critical section | |
1da177e4 | 852 | * |
9b06e818 | 853 | * When synchronize_rcu() is invoked on one CPU while other CPUs |
1da177e4 | 854 | * are within RCU read-side critical sections, then the |
9b06e818 | 855 | * synchronize_rcu() is guaranteed to block until after all the other |
1da177e4 LT |
856 | * CPUs exit their critical sections. Similarly, if call_rcu() is invoked |
857 | * on one CPU while other CPUs are within RCU read-side critical | |
858 | * sections, invocation of the corresponding RCU callback is deferred | |
859 | * until after the all the other CPUs exit their critical sections. | |
860 | * | |
861 | * Note, however, that RCU callbacks are permitted to run concurrently | |
77d8485a | 862 | * with new RCU read-side critical sections. One way that this can happen |
1da177e4 LT |
863 | * is via the following sequence of events: (1) CPU 0 enters an RCU |
864 | * read-side critical section, (2) CPU 1 invokes call_rcu() to register | |
865 | * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, | |
866 | * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU | |
867 | * callback is invoked. This is legal, because the RCU read-side critical | |
868 | * section that was running concurrently with the call_rcu() (and which | |
869 | * therefore might be referencing something that the corresponding RCU | |
870 | * callback would free up) has completed before the corresponding | |
871 | * RCU callback is invoked. | |
872 | * | |
873 | * RCU read-side critical sections may be nested. Any deferred actions | |
874 | * will be deferred until the outermost RCU read-side critical section | |
875 | * completes. | |
876 | * | |
9079fd7c PM |
877 | * You can avoid reading and understanding the next paragraph by |
878 | * following this rule: don't put anything in an rcu_read_lock() RCU | |
879 | * read-side critical section that would block in a !PREEMPT kernel. | |
880 | * But if you want the full story, read on! | |
881 | * | |
ab74fdfd PM |
882 | * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), |
883 | * it is illegal to block while in an RCU read-side critical section. | |
884 | * In preemptible RCU implementations (TREE_PREEMPT_RCU) in CONFIG_PREEMPT | |
885 | * kernel builds, RCU read-side critical sections may be preempted, | |
886 | * but explicit blocking is illegal. Finally, in preemptible RCU | |
887 | * implementations in real-time (with -rt patchset) kernel builds, RCU | |
888 | * read-side critical sections may be preempted and they may also block, but | |
889 | * only when acquiring spinlocks that are subject to priority inheritance. | |
1da177e4 | 890 | */ |
bc33f24b PM |
891 | static inline void rcu_read_lock(void) |
892 | { | |
893 | __rcu_read_lock(); | |
894 | __acquire(RCU); | |
d8ab29f8 | 895 | rcu_lock_acquire(&rcu_lock_map); |
5c173eb8 | 896 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 897 | "rcu_read_lock() used illegally while idle"); |
bc33f24b | 898 | } |
1da177e4 | 899 | |
1da177e4 LT |
900 | /* |
901 | * So where is rcu_write_lock()? It does not exist, as there is no | |
902 | * way for writers to lock out RCU readers. This is a feature, not | |
903 | * a bug -- this property is what provides RCU's performance benefits. | |
904 | * Of course, writers must coordinate with each other. The normal | |
905 | * spinlock primitives work well for this, but any other technique may be | |
906 | * used as well. RCU does not care how the writers keep out of each | |
907 | * others' way, as long as they do so. | |
908 | */ | |
3d76c082 PM |
909 | |
910 | /** | |
ca5ecddf | 911 | * rcu_read_unlock() - marks the end of an RCU read-side critical section. |
3d76c082 | 912 | * |
f27bc487 PM |
913 | * In most situations, rcu_read_unlock() is immune from deadlock. |
914 | * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() | |
915 | * is responsible for deboosting, which it does via rt_mutex_unlock(). | |
916 | * Unfortunately, this function acquires the scheduler's runqueue and | |
917 | * priority-inheritance spinlocks. This means that deadlock could result | |
918 | * if the caller of rcu_read_unlock() already holds one of these locks or | |
919 | * any lock that is ever acquired while holding them. | |
920 | * | |
921 | * That said, RCU readers are never priority boosted unless they were | |
922 | * preempted. Therefore, one way to avoid deadlock is to make sure | |
923 | * that preemption never happens within any RCU read-side critical | |
924 | * section whose outermost rcu_read_unlock() is called with one of | |
925 | * rt_mutex_unlock()'s locks held. Such preemption can be avoided in | |
926 | * a number of ways, for example, by invoking preempt_disable() before | |
927 | * critical section's outermost rcu_read_lock(). | |
928 | * | |
929 | * Given that the set of locks acquired by rt_mutex_unlock() might change | |
930 | * at any time, a somewhat more future-proofed approach is to make sure | |
931 | * that that preemption never happens within any RCU read-side critical | |
932 | * section whose outermost rcu_read_unlock() is called with irqs disabled. | |
933 | * This approach relies on the fact that rt_mutex_unlock() currently only | |
934 | * acquires irq-disabled locks. | |
935 | * | |
936 | * The second of these two approaches is best in most situations, | |
937 | * however, the first approach can also be useful, at least to those | |
938 | * developers willing to keep abreast of the set of locks acquired by | |
939 | * rt_mutex_unlock(). | |
940 | * | |
3d76c082 PM |
941 | * See rcu_read_lock() for more information. |
942 | */ | |
bc33f24b PM |
943 | static inline void rcu_read_unlock(void) |
944 | { | |
5c173eb8 | 945 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 946 | "rcu_read_unlock() used illegally while idle"); |
d8ab29f8 | 947 | rcu_lock_release(&rcu_lock_map); |
bc33f24b PM |
948 | __release(RCU); |
949 | __rcu_read_unlock(); | |
950 | } | |
1da177e4 LT |
951 | |
952 | /** | |
ca5ecddf | 953 | * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section |
1da177e4 LT |
954 | * |
955 | * This is equivalent of rcu_read_lock(), but to be used when updates | |
ca5ecddf PM |
956 | * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since |
957 | * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a | |
958 | * softirq handler to be a quiescent state, a process in RCU read-side | |
959 | * critical section must be protected by disabling softirqs. Read-side | |
960 | * critical sections in interrupt context can use just rcu_read_lock(), | |
961 | * though this should at least be commented to avoid confusing people | |
962 | * reading the code. | |
3842a083 PM |
963 | * |
964 | * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() | |
965 | * must occur in the same context, for example, it is illegal to invoke | |
966 | * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() | |
967 | * was invoked from some other task. | |
1da177e4 | 968 | */ |
bc33f24b PM |
969 | static inline void rcu_read_lock_bh(void) |
970 | { | |
6206ab9b | 971 | local_bh_disable(); |
bc33f24b | 972 | __acquire(RCU_BH); |
d8ab29f8 | 973 | rcu_lock_acquire(&rcu_bh_lock_map); |
5c173eb8 | 974 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 975 | "rcu_read_lock_bh() used illegally while idle"); |
bc33f24b | 976 | } |
1da177e4 LT |
977 | |
978 | /* | |
979 | * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section | |
980 | * | |
981 | * See rcu_read_lock_bh() for more information. | |
982 | */ | |
bc33f24b PM |
983 | static inline void rcu_read_unlock_bh(void) |
984 | { | |
5c173eb8 | 985 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 986 | "rcu_read_unlock_bh() used illegally while idle"); |
d8ab29f8 | 987 | rcu_lock_release(&rcu_bh_lock_map); |
bc33f24b | 988 | __release(RCU_BH); |
6206ab9b | 989 | local_bh_enable(); |
bc33f24b | 990 | } |
1da177e4 | 991 | |
1c50b728 | 992 | /** |
ca5ecddf | 993 | * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section |
1c50b728 | 994 | * |
ca5ecddf PM |
995 | * This is equivalent of rcu_read_lock(), but to be used when updates |
996 | * are being done using call_rcu_sched() or synchronize_rcu_sched(). | |
997 | * Read-side critical sections can also be introduced by anything that | |
998 | * disables preemption, including local_irq_disable() and friends. | |
3842a083 PM |
999 | * |
1000 | * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() | |
1001 | * must occur in the same context, for example, it is illegal to invoke | |
1002 | * rcu_read_unlock_sched() from process context if the matching | |
1003 | * rcu_read_lock_sched() was invoked from an NMI handler. | |
1c50b728 | 1004 | */ |
d6714c22 PM |
1005 | static inline void rcu_read_lock_sched(void) |
1006 | { | |
1007 | preempt_disable(); | |
bc33f24b | 1008 | __acquire(RCU_SCHED); |
d8ab29f8 | 1009 | rcu_lock_acquire(&rcu_sched_lock_map); |
5c173eb8 | 1010 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 1011 | "rcu_read_lock_sched() used illegally while idle"); |
d6714c22 | 1012 | } |
1eba8f84 PM |
1013 | |
1014 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 1015 | static inline notrace void rcu_read_lock_sched_notrace(void) |
d6714c22 PM |
1016 | { |
1017 | preempt_disable_notrace(); | |
bc33f24b | 1018 | __acquire(RCU_SCHED); |
d6714c22 | 1019 | } |
1c50b728 MD |
1020 | |
1021 | /* | |
1022 | * rcu_read_unlock_sched - marks the end of a RCU-classic critical section | |
1023 | * | |
1024 | * See rcu_read_lock_sched for more information. | |
1025 | */ | |
d6714c22 PM |
1026 | static inline void rcu_read_unlock_sched(void) |
1027 | { | |
5c173eb8 | 1028 | rcu_lockdep_assert(rcu_is_watching(), |
bde23c68 | 1029 | "rcu_read_unlock_sched() used illegally while idle"); |
d8ab29f8 | 1030 | rcu_lock_release(&rcu_sched_lock_map); |
bc33f24b | 1031 | __release(RCU_SCHED); |
d6714c22 PM |
1032 | preempt_enable(); |
1033 | } | |
1eba8f84 PM |
1034 | |
1035 | /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ | |
7c614d64 | 1036 | static inline notrace void rcu_read_unlock_sched_notrace(void) |
d6714c22 | 1037 | { |
bc33f24b | 1038 | __release(RCU_SCHED); |
d6714c22 PM |
1039 | preempt_enable_notrace(); |
1040 | } | |
1c50b728 | 1041 | |
ca5ecddf PM |
1042 | /** |
1043 | * RCU_INIT_POINTER() - initialize an RCU protected pointer | |
1044 | * | |
6846c0c5 PM |
1045 | * Initialize an RCU-protected pointer in special cases where readers |
1046 | * do not need ordering constraints on the CPU or the compiler. These | |
1047 | * special cases are: | |
1048 | * | |
1049 | * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- | |
1050 | * 2. The caller has taken whatever steps are required to prevent | |
1051 | * RCU readers from concurrently accessing this pointer -or- | |
1052 | * 3. The referenced data structure has already been exposed to | |
1053 | * readers either at compile time or via rcu_assign_pointer() -and- | |
1054 | * a. You have not made -any- reader-visible changes to | |
1055 | * this structure since then -or- | |
1056 | * b. It is OK for readers accessing this structure from its | |
1057 | * new location to see the old state of the structure. (For | |
1058 | * example, the changes were to statistical counters or to | |
1059 | * other state where exact synchronization is not required.) | |
1060 | * | |
1061 | * Failure to follow these rules governing use of RCU_INIT_POINTER() will | |
1062 | * result in impossible-to-diagnose memory corruption. As in the structures | |
1063 | * will look OK in crash dumps, but any concurrent RCU readers might | |
1064 | * see pre-initialized values of the referenced data structure. So | |
1065 | * please be very careful how you use RCU_INIT_POINTER()!!! | |
1066 | * | |
1067 | * If you are creating an RCU-protected linked structure that is accessed | |
1068 | * by a single external-to-structure RCU-protected pointer, then you may | |
1069 | * use RCU_INIT_POINTER() to initialize the internal RCU-protected | |
1070 | * pointers, but you must use rcu_assign_pointer() to initialize the | |
1071 | * external-to-structure pointer -after- you have completely initialized | |
1072 | * the reader-accessible portions of the linked structure. | |
71a9b269 PM |
1073 | * |
1074 | * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no | |
1075 | * ordering guarantees for either the CPU or the compiler. | |
ca5ecddf PM |
1076 | */ |
1077 | #define RCU_INIT_POINTER(p, v) \ | |
d1b88eb9 | 1078 | do { \ |
462225ae | 1079 | p = RCU_INITIALIZER(v); \ |
d1b88eb9 | 1080 | } while (0) |
9ab1544e | 1081 | |
172708d0 PM |
1082 | /** |
1083 | * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer | |
1084 | * | |
1085 | * GCC-style initialization for an RCU-protected pointer in a structure field. | |
1086 | */ | |
1087 | #define RCU_POINTER_INITIALIZER(p, v) \ | |
462225ae | 1088 | .p = RCU_INITIALIZER(v) |
9ab1544e | 1089 | |
d8169d4c JE |
1090 | /* |
1091 | * Does the specified offset indicate that the corresponding rcu_head | |
1092 | * structure can be handled by kfree_rcu()? | |
1093 | */ | |
1094 | #define __is_kfree_rcu_offset(offset) ((offset) < 4096) | |
1095 | ||
1096 | /* | |
1097 | * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. | |
1098 | */ | |
1099 | #define __kfree_rcu(head, offset) \ | |
1100 | do { \ | |
1101 | BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ | |
4fa3b6cb | 1102 | kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \ |
d8169d4c JE |
1103 | } while (0) |
1104 | ||
9ab1544e LJ |
1105 | /** |
1106 | * kfree_rcu() - kfree an object after a grace period. | |
1107 | * @ptr: pointer to kfree | |
1108 | * @rcu_head: the name of the struct rcu_head within the type of @ptr. | |
1109 | * | |
1110 | * Many rcu callbacks functions just call kfree() on the base structure. | |
1111 | * These functions are trivial, but their size adds up, and furthermore | |
1112 | * when they are used in a kernel module, that module must invoke the | |
1113 | * high-latency rcu_barrier() function at module-unload time. | |
1114 | * | |
1115 | * The kfree_rcu() function handles this issue. Rather than encoding a | |
1116 | * function address in the embedded rcu_head structure, kfree_rcu() instead | |
1117 | * encodes the offset of the rcu_head structure within the base structure. | |
1118 | * Because the functions are not allowed in the low-order 4096 bytes of | |
1119 | * kernel virtual memory, offsets up to 4095 bytes can be accommodated. | |
1120 | * If the offset is larger than 4095 bytes, a compile-time error will | |
1121 | * be generated in __kfree_rcu(). If this error is triggered, you can | |
1122 | * either fall back to use of call_rcu() or rearrange the structure to | |
1123 | * position the rcu_head structure into the first 4096 bytes. | |
1124 | * | |
1125 | * Note that the allowable offset might decrease in the future, for example, | |
1126 | * to allow something like kmem_cache_free_rcu(). | |
d8169d4c JE |
1127 | * |
1128 | * The BUILD_BUG_ON check must not involve any function calls, hence the | |
1129 | * checks are done in macros here. | |
9ab1544e LJ |
1130 | */ |
1131 | #define kfree_rcu(ptr, rcu_head) \ | |
1132 | __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) | |
1133 | ||
ffa83fb5 PM |
1134 | #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) |
1135 | static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) | |
1136 | { | |
1137 | *delta_jiffies = ULONG_MAX; | |
1138 | return 0; | |
1139 | } | |
1140 | #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */ | |
1141 | ||
2f33b512 PM |
1142 | #if defined(CONFIG_RCU_NOCB_CPU_ALL) |
1143 | static inline bool rcu_is_nocb_cpu(int cpu) { return true; } | |
1144 | #elif defined(CONFIG_RCU_NOCB_CPU) | |
584dc4ce | 1145 | bool rcu_is_nocb_cpu(int cpu); |
d1e43fa5 FW |
1146 | #else |
1147 | static inline bool rcu_is_nocb_cpu(int cpu) { return false; } | |
2f33b512 | 1148 | #endif |
d1e43fa5 FW |
1149 | |
1150 | ||
0edd1b17 PM |
1151 | /* Only for use by adaptive-ticks code. */ |
1152 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
584dc4ce TB |
1153 | bool rcu_sys_is_idle(void); |
1154 | void rcu_sysidle_force_exit(void); | |
0edd1b17 PM |
1155 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ |
1156 | ||
1157 | static inline bool rcu_sys_is_idle(void) | |
1158 | { | |
1159 | return false; | |
1160 | } | |
1161 | ||
1162 | static inline void rcu_sysidle_force_exit(void) | |
1163 | { | |
1164 | } | |
1165 | ||
1166 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
1167 | ||
1168 | ||
1da177e4 | 1169 | #endif /* __LINUX_RCUPDATE_H */ |