1 /*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright IBM Corporation, 2001
19 *
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
21 *
22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
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 -
29 * http://lse.sourceforge.net/locking/rcupdate.html
30 *
31 */
32
33 #ifndef __LINUX_RCUPDATE_H
34 #define __LINUX_RCUPDATE_H
35
36 #include <linux/types.h>
37 #include <linux/cache.h>
38 #include <linux/spinlock.h>
39 #include <linux/threads.h>
40 #include <linux/cpumask.h>
41 #include <linux/seqlock.h>
42 #include <linux/lockdep.h>
43 #include <linux/completion.h>
44 #include <linux/debugobjects.h>
45 #include <linux/bug.h>
46 #include <linux/compiler.h>
47
48 #ifdef CONFIG_RCU_TORTURE_TEST
49 extern int rcutorture_runnable; /* for sysctl */
50 #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
51
52 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
53 extern void rcutorture_record_test_transition(void);
54 extern void rcutorture_record_progress(unsigned long vernum);
55 extern void do_trace_rcu_torture_read(char *rcutorturename,
56 struct rcu_head *rhp,
57 unsigned long secs,
58 unsigned long c_old,
59 unsigned long c);
60 #else
rcutorture_record_test_transition(void)61 static inline void rcutorture_record_test_transition(void)
62 {
63 }
rcutorture_record_progress(unsigned long vernum)64 static inline void rcutorture_record_progress(unsigned long vernum)
65 {
66 }
67 #ifdef CONFIG_RCU_TRACE
68 extern void do_trace_rcu_torture_read(char *rcutorturename,
69 struct rcu_head *rhp,
70 unsigned long secs,
71 unsigned long c_old,
72 unsigned long c);
73 #else
74 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
75 do { } while (0)
76 #endif
77 #endif
78
79 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
80 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
81 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
82 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
83 #define ulong2long(a) (*(long *)(&(a)))
84
85 /* Exported common interfaces */
86
87 #ifdef CONFIG_PREEMPT_RCU
88
89 /**
90 * call_rcu() - Queue an RCU callback for invocation after a grace period.
91 * @head: structure to be used for queueing the RCU updates.
92 * @func: actual callback function to be invoked after the grace period
93 *
94 * The callback function will be invoked some time after a full grace
95 * period elapses, in other words after all pre-existing RCU read-side
96 * critical sections have completed. However, the callback function
97 * might well execute concurrently with RCU read-side critical sections
98 * that started after call_rcu() was invoked. RCU read-side critical
99 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
100 * and may be nested.
101 *
102 * Note that all CPUs must agree that the grace period extended beyond
103 * all pre-existing RCU read-side critical section. On systems with more
104 * than one CPU, this means that when "func()" is invoked, each CPU is
105 * guaranteed to have executed a full memory barrier since the end of its
106 * last RCU read-side critical section whose beginning preceded the call
107 * to call_rcu(). It also means that each CPU executing an RCU read-side
108 * critical section that continues beyond the start of "func()" must have
109 * executed a memory barrier after the call_rcu() but before the beginning
110 * of that RCU read-side critical section. Note that these guarantees
111 * include CPUs that are offline, idle, or executing in user mode, as
112 * well as CPUs that are executing in the kernel.
113 *
114 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
115 * resulting RCU callback function "func()", then both CPU A and CPU B are
116 * guaranteed to execute a full memory barrier during the time interval
117 * between the call to call_rcu() and the invocation of "func()" -- even
118 * if CPU A and CPU B are the same CPU (but again only if the system has
119 * more than one CPU).
120 */
121 extern void call_rcu(struct rcu_head *head,
122 void (*func)(struct rcu_head *head));
123
124 #else /* #ifdef CONFIG_PREEMPT_RCU */
125
126 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
127 #define call_rcu call_rcu_sched
128
129 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
130
131 /**
132 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
133 * @head: structure to be used for queueing the RCU updates.
134 * @func: actual callback function to be invoked after the grace period
135 *
136 * The callback function will be invoked some time after a full grace
137 * period elapses, in other words after all currently executing RCU
138 * read-side critical sections have completed. call_rcu_bh() assumes
139 * that the read-side critical sections end on completion of a softirq
140 * handler. This means that read-side critical sections in process
141 * context must not be interrupted by softirqs. This interface is to be
142 * used when most of the read-side critical sections are in softirq context.
143 * RCU read-side critical sections are delimited by :
144 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
145 * OR
146 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
147 * These may be nested.
148 *
149 * See the description of call_rcu() for more detailed information on
150 * memory ordering guarantees.
151 */
152 extern void call_rcu_bh(struct rcu_head *head,
153 void (*func)(struct rcu_head *head));
154
155 /**
156 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
157 * @head: structure to be used for queueing the RCU updates.
158 * @func: actual callback function to be invoked after the grace period
159 *
160 * The callback function will be invoked some time after a full grace
161 * period elapses, in other words after all currently executing RCU
162 * read-side critical sections have completed. call_rcu_sched() assumes
163 * that the read-side critical sections end on enabling of preemption
164 * or on voluntary preemption.
165 * RCU read-side critical sections are delimited by :
166 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
167 * OR
168 * anything that disables preemption.
169 * These may be nested.
170 *
171 * See the description of call_rcu() for more detailed information on
172 * memory ordering guarantees.
173 */
174 extern void call_rcu_sched(struct rcu_head *head,
175 void (*func)(struct rcu_head *rcu));
176
177 extern void synchronize_sched(void);
178
179 #ifdef CONFIG_PREEMPT_RCU
180
181 extern void __rcu_read_lock(void);
182 extern void __rcu_read_unlock(void);
183 extern void rcu_read_unlock_special(struct task_struct *t);
184 void synchronize_rcu(void);
185
186 /*
187 * Defined as a macro as it is a very low level header included from
188 * areas that don't even know about current. This gives the rcu_read_lock()
189 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
190 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
191 */
192 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
193
194 #else /* #ifdef CONFIG_PREEMPT_RCU */
195
__rcu_read_lock(void)196 static inline void __rcu_read_lock(void)
197 {
198 preempt_disable();
199 }
200
__rcu_read_unlock(void)201 static inline void __rcu_read_unlock(void)
202 {
203 preempt_enable();
204 }
205
synchronize_rcu(void)206 static inline void synchronize_rcu(void)
207 {
208 synchronize_sched();
209 }
210
rcu_preempt_depth(void)211 static inline int rcu_preempt_depth(void)
212 {
213 return 0;
214 }
215
216 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
217
218 /* Internal to kernel */
219 extern void rcu_sched_qs(int cpu);
220 extern void rcu_bh_qs(int cpu);
221 extern void rcu_check_callbacks(int cpu, int user);
222 struct notifier_block;
223 extern void rcu_idle_enter(void);
224 extern void rcu_idle_exit(void);
225 extern void rcu_irq_enter(void);
226 extern void rcu_irq_exit(void);
227
228 #ifdef CONFIG_RCU_USER_QS
229 extern void rcu_user_enter(void);
230 extern void rcu_user_exit(void);
231 extern void rcu_user_enter_after_irq(void);
232 extern void rcu_user_exit_after_irq(void);
233 #else
rcu_user_enter(void)234 static inline void rcu_user_enter(void) { }
rcu_user_exit(void)235 static inline void rcu_user_exit(void) { }
rcu_user_enter_after_irq(void)236 static inline void rcu_user_enter_after_irq(void) { }
rcu_user_exit_after_irq(void)237 static inline void rcu_user_exit_after_irq(void) { }
rcu_user_hooks_switch(struct task_struct * prev,struct task_struct * next)238 static inline void rcu_user_hooks_switch(struct task_struct *prev,
239 struct task_struct *next) { }
240 #endif /* CONFIG_RCU_USER_QS */
241
242 extern void exit_rcu(void);
243
244 /**
245 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
246 * @a: Code that RCU needs to pay attention to.
247 *
248 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
249 * in the inner idle loop, that is, between the rcu_idle_enter() and
250 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
251 * critical sections. However, things like powertop need tracepoints
252 * in the inner idle loop.
253 *
254 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
255 * will tell RCU that it needs to pay attending, invoke its argument
256 * (in this example, a call to the do_something_with_RCU() function),
257 * and then tell RCU to go back to ignoring this CPU. It is permissible
258 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
259 * quite limited. If deeper nesting is required, it will be necessary
260 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
261 */
262 #define RCU_NONIDLE(a) \
263 do { \
264 rcu_irq_enter(); \
265 do { a; } while (0); \
266 rcu_irq_exit(); \
267 } while (0)
268
269 /*
270 * Infrastructure to implement the synchronize_() primitives in
271 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
272 */
273
274 typedef void call_rcu_func_t(struct rcu_head *head,
275 void (*func)(struct rcu_head *head));
276 void wait_rcu_gp(call_rcu_func_t crf);
277
278 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
279 #include <linux/rcutree.h>
280 #elif defined(CONFIG_TINY_RCU) || defined(CONFIG_TINY_PREEMPT_RCU)
281 #include <linux/rcutiny.h>
282 #else
283 #error "Unknown RCU implementation specified to kernel configuration"
284 #endif
285
286 /*
287 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
288 * initialization and destruction of rcu_head on the stack. rcu_head structures
289 * allocated dynamically in the heap or defined statically don't need any
290 * initialization.
291 */
292 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
293 extern void init_rcu_head_on_stack(struct rcu_head *head);
294 extern void destroy_rcu_head_on_stack(struct rcu_head *head);
295 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
init_rcu_head_on_stack(struct rcu_head * head)296 static inline void init_rcu_head_on_stack(struct rcu_head *head)
297 {
298 }
299
destroy_rcu_head_on_stack(struct rcu_head * head)300 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
301 {
302 }
303 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
304
305 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP)
306 extern int rcu_is_cpu_idle(void);
307 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP) */
308
309 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
310 bool rcu_lockdep_current_cpu_online(void);
311 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
rcu_lockdep_current_cpu_online(void)312 static inline bool rcu_lockdep_current_cpu_online(void)
313 {
314 return 1;
315 }
316 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
317
318 #ifdef CONFIG_DEBUG_LOCK_ALLOC
319
rcu_lock_acquire(struct lockdep_map * map)320 static inline void rcu_lock_acquire(struct lockdep_map *map)
321 {
322 lock_acquire(map, 0, 0, 2, 1, NULL, _THIS_IP_);
323 }
324
rcu_lock_release(struct lockdep_map * map)325 static inline void rcu_lock_release(struct lockdep_map *map)
326 {
327 lock_release(map, 1, _THIS_IP_);
328 }
329
330 extern struct lockdep_map rcu_lock_map;
331 extern struct lockdep_map rcu_bh_lock_map;
332 extern struct lockdep_map rcu_sched_lock_map;
333 extern int debug_lockdep_rcu_enabled(void);
334
335 /**
336 * rcu_read_lock_held() - might we be in RCU read-side critical section?
337 *
338 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
339 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
340 * this assumes we are in an RCU read-side critical section unless it can
341 * prove otherwise. This is useful for debug checks in functions that
342 * require that they be called within an RCU read-side critical section.
343 *
344 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
345 * and while lockdep is disabled.
346 *
347 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
348 * occur in the same context, for example, it is illegal to invoke
349 * rcu_read_unlock() in process context if the matching rcu_read_lock()
350 * was invoked from within an irq handler.
351 *
352 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
353 * offline from an RCU perspective, so check for those as well.
354 */
rcu_read_lock_held(void)355 static inline int rcu_read_lock_held(void)
356 {
357 if (!debug_lockdep_rcu_enabled())
358 return 1;
359 if (rcu_is_cpu_idle())
360 return 0;
361 if (!rcu_lockdep_current_cpu_online())
362 return 0;
363 return lock_is_held(&rcu_lock_map);
364 }
365
366 /*
367 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
368 * hell.
369 */
370 extern int rcu_read_lock_bh_held(void);
371
372 /**
373 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
374 *
375 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
376 * RCU-sched read-side critical section. In absence of
377 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
378 * critical section unless it can prove otherwise. Note that disabling
379 * of preemption (including disabling irqs) counts as an RCU-sched
380 * read-side critical section. This is useful for debug checks in functions
381 * that required that they be called within an RCU-sched read-side
382 * critical section.
383 *
384 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
385 * and while lockdep is disabled.
386 *
387 * Note that if the CPU is in the idle loop from an RCU point of
388 * view (ie: that we are in the section between rcu_idle_enter() and
389 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
390 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
391 * that are in such a section, considering these as in extended quiescent
392 * state, so such a CPU is effectively never in an RCU read-side critical
393 * section regardless of what RCU primitives it invokes. This state of
394 * affairs is required --- we need to keep an RCU-free window in idle
395 * where the CPU may possibly enter into low power mode. This way we can
396 * notice an extended quiescent state to other CPUs that started a grace
397 * period. Otherwise we would delay any grace period as long as we run in
398 * the idle task.
399 *
400 * Similarly, we avoid claiming an SRCU read lock held if the current
401 * CPU is offline.
402 */
403 #ifdef CONFIG_PREEMPT_COUNT
rcu_read_lock_sched_held(void)404 static inline int rcu_read_lock_sched_held(void)
405 {
406 int lockdep_opinion = 0;
407
408 if (!debug_lockdep_rcu_enabled())
409 return 1;
410 if (rcu_is_cpu_idle())
411 return 0;
412 if (!rcu_lockdep_current_cpu_online())
413 return 0;
414 if (debug_locks)
415 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
416 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
417 }
418 #else /* #ifdef CONFIG_PREEMPT_COUNT */
rcu_read_lock_sched_held(void)419 static inline int rcu_read_lock_sched_held(void)
420 {
421 return 1;
422 }
423 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
424
425 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
426
427 # define rcu_lock_acquire(a) do { } while (0)
428 # define rcu_lock_release(a) do { } while (0)
429
rcu_read_lock_held(void)430 static inline int rcu_read_lock_held(void)
431 {
432 return 1;
433 }
434
rcu_read_lock_bh_held(void)435 static inline int rcu_read_lock_bh_held(void)
436 {
437 return 1;
438 }
439
440 #ifdef CONFIG_PREEMPT_COUNT
rcu_read_lock_sched_held(void)441 static inline int rcu_read_lock_sched_held(void)
442 {
443 return preempt_count() != 0 || irqs_disabled();
444 }
445 #else /* #ifdef CONFIG_PREEMPT_COUNT */
rcu_read_lock_sched_held(void)446 static inline int rcu_read_lock_sched_held(void)
447 {
448 return 1;
449 }
450 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
451
452 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
453
454 #ifdef CONFIG_PROVE_RCU
455
456 extern int rcu_my_thread_group_empty(void);
457
458 /**
459 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
460 * @c: condition to check
461 * @s: informative message
462 */
463 #define rcu_lockdep_assert(c, s) \
464 do { \
465 static bool __section(.data.unlikely) __warned; \
466 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
467 __warned = true; \
468 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
469 } \
470 } while (0)
471
472 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
rcu_preempt_sleep_check(void)473 static inline void rcu_preempt_sleep_check(void)
474 {
475 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
476 "Illegal context switch in RCU read-side critical section");
477 }
478 #else /* #ifdef CONFIG_PROVE_RCU */
rcu_preempt_sleep_check(void)479 static inline void rcu_preempt_sleep_check(void)
480 {
481 }
482 #endif /* #else #ifdef CONFIG_PROVE_RCU */
483
484 #define rcu_sleep_check() \
485 do { \
486 rcu_preempt_sleep_check(); \
487 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
488 "Illegal context switch in RCU-bh" \
489 " read-side critical section"); \
490 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
491 "Illegal context switch in RCU-sched"\
492 " read-side critical section"); \
493 } while (0)
494
495 #else /* #ifdef CONFIG_PROVE_RCU */
496
497 #define rcu_lockdep_assert(c, s) do { } while (0)
498 #define rcu_sleep_check() do { } while (0)
499
500 #endif /* #else #ifdef CONFIG_PROVE_RCU */
501
502 /*
503 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
504 * and rcu_assign_pointer(). Some of these could be folded into their
505 * callers, but they are left separate in order to ease introduction of
506 * multiple flavors of pointers to match the multiple flavors of RCU
507 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
508 * the future.
509 */
510
511 #ifdef __CHECKER__
512 #define rcu_dereference_sparse(p, space) \
513 ((void)(((typeof(*p) space *)p) == p))
514 #else /* #ifdef __CHECKER__ */
515 #define rcu_dereference_sparse(p, space)
516 #endif /* #else #ifdef __CHECKER__ */
517
518 #define __rcu_access_pointer(p, space) \
519 ({ \
520 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
521 rcu_dereference_sparse(p, space); \
522 ((typeof(*p) __force __kernel *)(_________p1)); \
523 })
524 #define __rcu_dereference_check(p, c, space) \
525 ({ \
526 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
527 rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \
528 " usage"); \
529 rcu_dereference_sparse(p, space); \
530 smp_read_barrier_depends(); \
531 ((typeof(*p) __force __kernel *)(_________p1)); \
532 })
533 #define __rcu_dereference_protected(p, c, space) \
534 ({ \
535 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \
536 " usage"); \
537 rcu_dereference_sparse(p, space); \
538 ((typeof(*p) __force __kernel *)(p)); \
539 })
540
541 #define __rcu_access_index(p, space) \
542 ({ \
543 typeof(p) _________p1 = ACCESS_ONCE(p); \
544 rcu_dereference_sparse(p, space); \
545 (_________p1); \
546 })
547 #define __rcu_dereference_index_check(p, c) \
548 ({ \
549 typeof(p) _________p1 = ACCESS_ONCE(p); \
550 rcu_lockdep_assert(c, \
551 "suspicious rcu_dereference_index_check()" \
552 " usage"); \
553 smp_read_barrier_depends(); \
554 (_________p1); \
555 })
556 #define __rcu_assign_pointer(p, v, space) \
557 do { \
558 smp_wmb(); \
559 (p) = (typeof(*v) __force space *)(v); \
560 } while (0)
561
562
563 /**
564 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
565 * @p: The pointer to read
566 *
567 * Return the value of the specified RCU-protected pointer, but omit the
568 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
569 * when the value of this pointer is accessed, but the pointer is not
570 * dereferenced, for example, when testing an RCU-protected pointer against
571 * NULL. Although rcu_access_pointer() may also be used in cases where
572 * update-side locks prevent the value of the pointer from changing, you
573 * should instead use rcu_dereference_protected() for this use case.
574 *
575 * It is also permissible to use rcu_access_pointer() when read-side
576 * access to the pointer was removed at least one grace period ago, as
577 * is the case in the context of the RCU callback that is freeing up
578 * the data, or after a synchronize_rcu() returns. This can be useful
579 * when tearing down multi-linked structures after a grace period
580 * has elapsed.
581 */
582 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
583
584 /**
585 * rcu_dereference_check() - rcu_dereference with debug checking
586 * @p: The pointer to read, prior to dereferencing
587 * @c: The conditions under which the dereference will take place
588 *
589 * Do an rcu_dereference(), but check that the conditions under which the
590 * dereference will take place are correct. Typically the conditions
591 * indicate the various locking conditions that should be held at that
592 * point. The check should return true if the conditions are satisfied.
593 * An implicit check for being in an RCU read-side critical section
594 * (rcu_read_lock()) is included.
595 *
596 * For example:
597 *
598 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
599 *
600 * could be used to indicate to lockdep that foo->bar may only be dereferenced
601 * if either rcu_read_lock() is held, or that the lock required to replace
602 * the bar struct at foo->bar is held.
603 *
604 * Note that the list of conditions may also include indications of when a lock
605 * need not be held, for example during initialisation or destruction of the
606 * target struct:
607 *
608 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
609 * atomic_read(&foo->usage) == 0);
610 *
611 * Inserts memory barriers on architectures that require them
612 * (currently only the Alpha), prevents the compiler from refetching
613 * (and from merging fetches), and, more importantly, documents exactly
614 * which pointers are protected by RCU and checks that the pointer is
615 * annotated as __rcu.
616 */
617 #define rcu_dereference_check(p, c) \
618 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
619
620 /**
621 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
622 * @p: The pointer to read, prior to dereferencing
623 * @c: The conditions under which the dereference will take place
624 *
625 * This is the RCU-bh counterpart to rcu_dereference_check().
626 */
627 #define rcu_dereference_bh_check(p, c) \
628 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
629
630 /**
631 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
632 * @p: The pointer to read, prior to dereferencing
633 * @c: The conditions under which the dereference will take place
634 *
635 * This is the RCU-sched counterpart to rcu_dereference_check().
636 */
637 #define rcu_dereference_sched_check(p, c) \
638 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
639 __rcu)
640
641 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
642
643 /*
644 * The tracing infrastructure traces RCU (we want that), but unfortunately
645 * some of the RCU checks causes tracing to lock up the system.
646 *
647 * The tracing version of rcu_dereference_raw() must not call
648 * rcu_read_lock_held().
649 */
650 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
651
652 /**
653 * rcu_access_index() - fetch RCU index with no dereferencing
654 * @p: The index to read
655 *
656 * Return the value of the specified RCU-protected index, but omit the
657 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
658 * when the value of this index is accessed, but the index is not
659 * dereferenced, for example, when testing an RCU-protected index against
660 * -1. Although rcu_access_index() may also be used in cases where
661 * update-side locks prevent the value of the index from changing, you
662 * should instead use rcu_dereference_index_protected() for this use case.
663 */
664 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
665
666 /**
667 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
668 * @p: The pointer to read, prior to dereferencing
669 * @c: The conditions under which the dereference will take place
670 *
671 * Similar to rcu_dereference_check(), but omits the sparse checking.
672 * This allows rcu_dereference_index_check() to be used on integers,
673 * which can then be used as array indices. Attempting to use
674 * rcu_dereference_check() on an integer will give compiler warnings
675 * because the sparse address-space mechanism relies on dereferencing
676 * the RCU-protected pointer. Dereferencing integers is not something
677 * that even gcc will put up with.
678 *
679 * Note that this function does not implicitly check for RCU read-side
680 * critical sections. If this function gains lots of uses, it might
681 * make sense to provide versions for each flavor of RCU, but it does
682 * not make sense as of early 2010.
683 */
684 #define rcu_dereference_index_check(p, c) \
685 __rcu_dereference_index_check((p), (c))
686
687 /**
688 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
689 * @p: The pointer to read, prior to dereferencing
690 * @c: The conditions under which the dereference will take place
691 *
692 * Return the value of the specified RCU-protected pointer, but omit
693 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
694 * is useful in cases where update-side locks prevent the value of the
695 * pointer from changing. Please note that this primitive does -not-
696 * prevent the compiler from repeating this reference or combining it
697 * with other references, so it should not be used without protection
698 * of appropriate locks.
699 *
700 * This function is only for update-side use. Using this function
701 * when protected only by rcu_read_lock() will result in infrequent
702 * but very ugly failures.
703 */
704 #define rcu_dereference_protected(p, c) \
705 __rcu_dereference_protected((p), (c), __rcu)
706
707
708 /**
709 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
710 * @p: The pointer to read, prior to dereferencing
711 *
712 * This is a simple wrapper around rcu_dereference_check().
713 */
714 #define rcu_dereference(p) rcu_dereference_check(p, 0)
715
716 /**
717 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
718 * @p: The pointer to read, prior to dereferencing
719 *
720 * Makes rcu_dereference_check() do the dirty work.
721 */
722 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
723
724 /**
725 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
726 * @p: The pointer to read, prior to dereferencing
727 *
728 * Makes rcu_dereference_check() do the dirty work.
729 */
730 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
731
732 /**
733 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
734 *
735 * When synchronize_rcu() is invoked on one CPU while other CPUs
736 * are within RCU read-side critical sections, then the
737 * synchronize_rcu() is guaranteed to block until after all the other
738 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
739 * on one CPU while other CPUs are within RCU read-side critical
740 * sections, invocation of the corresponding RCU callback is deferred
741 * until after the all the other CPUs exit their critical sections.
742 *
743 * Note, however, that RCU callbacks are permitted to run concurrently
744 * with new RCU read-side critical sections. One way that this can happen
745 * is via the following sequence of events: (1) CPU 0 enters an RCU
746 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
747 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
748 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
749 * callback is invoked. This is legal, because the RCU read-side critical
750 * section that was running concurrently with the call_rcu() (and which
751 * therefore might be referencing something that the corresponding RCU
752 * callback would free up) has completed before the corresponding
753 * RCU callback is invoked.
754 *
755 * RCU read-side critical sections may be nested. Any deferred actions
756 * will be deferred until the outermost RCU read-side critical section
757 * completes.
758 *
759 * You can avoid reading and understanding the next paragraph by
760 * following this rule: don't put anything in an rcu_read_lock() RCU
761 * read-side critical section that would block in a !PREEMPT kernel.
762 * But if you want the full story, read on!
763 *
764 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it
765 * is illegal to block while in an RCU read-side critical section. In
766 * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU)
767 * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may
768 * be preempted, but explicit blocking is illegal. Finally, in preemptible
769 * RCU implementations in real-time (with -rt patchset) kernel builds,
770 * RCU read-side critical sections may be preempted and they may also
771 * block, but only when acquiring spinlocks that are subject to priority
772 * inheritance.
773 */
rcu_read_lock(void)774 static inline void rcu_read_lock(void)
775 {
776 __rcu_read_lock();
777 __acquire(RCU);
778 rcu_lock_acquire(&rcu_lock_map);
779 rcu_lockdep_assert(!rcu_is_cpu_idle(),
780 "rcu_read_lock() used illegally while idle");
781 }
782
783 /*
784 * So where is rcu_write_lock()? It does not exist, as there is no
785 * way for writers to lock out RCU readers. This is a feature, not
786 * a bug -- this property is what provides RCU's performance benefits.
787 * Of course, writers must coordinate with each other. The normal
788 * spinlock primitives work well for this, but any other technique may be
789 * used as well. RCU does not care how the writers keep out of each
790 * others' way, as long as they do so.
791 */
792
793 /**
794 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
795 *
796 * See rcu_read_lock() for more information.
797 */
rcu_read_unlock(void)798 static inline void rcu_read_unlock(void)
799 {
800 rcu_lockdep_assert(!rcu_is_cpu_idle(),
801 "rcu_read_unlock() used illegally while idle");
802 rcu_lock_release(&rcu_lock_map);
803 __release(RCU);
804 __rcu_read_unlock();
805 }
806
807 /**
808 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
809 *
810 * This is equivalent of rcu_read_lock(), but to be used when updates
811 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
812 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
813 * softirq handler to be a quiescent state, a process in RCU read-side
814 * critical section must be protected by disabling softirqs. Read-side
815 * critical sections in interrupt context can use just rcu_read_lock(),
816 * though this should at least be commented to avoid confusing people
817 * reading the code.
818 *
819 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
820 * must occur in the same context, for example, it is illegal to invoke
821 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
822 * was invoked from some other task.
823 */
rcu_read_lock_bh(void)824 static inline void rcu_read_lock_bh(void)
825 {
826 local_bh_disable();
827 __acquire(RCU_BH);
828 rcu_lock_acquire(&rcu_bh_lock_map);
829 rcu_lockdep_assert(!rcu_is_cpu_idle(),
830 "rcu_read_lock_bh() used illegally while idle");
831 }
832
833 /*
834 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
835 *
836 * See rcu_read_lock_bh() for more information.
837 */
rcu_read_unlock_bh(void)838 static inline void rcu_read_unlock_bh(void)
839 {
840 rcu_lockdep_assert(!rcu_is_cpu_idle(),
841 "rcu_read_unlock_bh() used illegally while idle");
842 rcu_lock_release(&rcu_bh_lock_map);
843 __release(RCU_BH);
844 local_bh_enable();
845 }
846
847 /**
848 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
849 *
850 * This is equivalent of rcu_read_lock(), but to be used when updates
851 * are being done using call_rcu_sched() or synchronize_rcu_sched().
852 * Read-side critical sections can also be introduced by anything that
853 * disables preemption, including local_irq_disable() and friends.
854 *
855 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
856 * must occur in the same context, for example, it is illegal to invoke
857 * rcu_read_unlock_sched() from process context if the matching
858 * rcu_read_lock_sched() was invoked from an NMI handler.
859 */
rcu_read_lock_sched(void)860 static inline void rcu_read_lock_sched(void)
861 {
862 preempt_disable();
863 __acquire(RCU_SCHED);
864 rcu_lock_acquire(&rcu_sched_lock_map);
865 rcu_lockdep_assert(!rcu_is_cpu_idle(),
866 "rcu_read_lock_sched() used illegally while idle");
867 }
868
869 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_lock_sched_notrace(void)870 static inline notrace void rcu_read_lock_sched_notrace(void)
871 {
872 preempt_disable_notrace();
873 __acquire(RCU_SCHED);
874 }
875
876 /*
877 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
878 *
879 * See rcu_read_lock_sched for more information.
880 */
rcu_read_unlock_sched(void)881 static inline void rcu_read_unlock_sched(void)
882 {
883 rcu_lockdep_assert(!rcu_is_cpu_idle(),
884 "rcu_read_unlock_sched() used illegally while idle");
885 rcu_lock_release(&rcu_sched_lock_map);
886 __release(RCU_SCHED);
887 preempt_enable();
888 }
889
890 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_unlock_sched_notrace(void)891 static inline notrace void rcu_read_unlock_sched_notrace(void)
892 {
893 __release(RCU_SCHED);
894 preempt_enable_notrace();
895 }
896
897 /**
898 * rcu_assign_pointer() - assign to RCU-protected pointer
899 * @p: pointer to assign to
900 * @v: value to assign (publish)
901 *
902 * Assigns the specified value to the specified RCU-protected
903 * pointer, ensuring that any concurrent RCU readers will see
904 * any prior initialization.
905 *
906 * Inserts memory barriers on architectures that require them
907 * (which is most of them), and also prevents the compiler from
908 * reordering the code that initializes the structure after the pointer
909 * assignment. More importantly, this call documents which pointers
910 * will be dereferenced by RCU read-side code.
911 *
912 * In some special cases, you may use RCU_INIT_POINTER() instead
913 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
914 * to the fact that it does not constrain either the CPU or the compiler.
915 * That said, using RCU_INIT_POINTER() when you should have used
916 * rcu_assign_pointer() is a very bad thing that results in
917 * impossible-to-diagnose memory corruption. So please be careful.
918 * See the RCU_INIT_POINTER() comment header for details.
919 */
920 #define rcu_assign_pointer(p, v) \
921 __rcu_assign_pointer((p), (v), __rcu)
922
923 /**
924 * RCU_INIT_POINTER() - initialize an RCU protected pointer
925 *
926 * Initialize an RCU-protected pointer in special cases where readers
927 * do not need ordering constraints on the CPU or the compiler. These
928 * special cases are:
929 *
930 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
931 * 2. The caller has taken whatever steps are required to prevent
932 * RCU readers from concurrently accessing this pointer -or-
933 * 3. The referenced data structure has already been exposed to
934 * readers either at compile time or via rcu_assign_pointer() -and-
935 * a. You have not made -any- reader-visible changes to
936 * this structure since then -or-
937 * b. It is OK for readers accessing this structure from its
938 * new location to see the old state of the structure. (For
939 * example, the changes were to statistical counters or to
940 * other state where exact synchronization is not required.)
941 *
942 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
943 * result in impossible-to-diagnose memory corruption. As in the structures
944 * will look OK in crash dumps, but any concurrent RCU readers might
945 * see pre-initialized values of the referenced data structure. So
946 * please be very careful how you use RCU_INIT_POINTER()!!!
947 *
948 * If you are creating an RCU-protected linked structure that is accessed
949 * by a single external-to-structure RCU-protected pointer, then you may
950 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
951 * pointers, but you must use rcu_assign_pointer() to initialize the
952 * external-to-structure pointer -after- you have completely initialized
953 * the reader-accessible portions of the linked structure.
954 */
955 #define RCU_INIT_POINTER(p, v) \
956 do { \
957 p = (typeof(*v) __force __rcu *)(v); \
958 } while (0)
959
960 /**
961 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
962 *
963 * GCC-style initialization for an RCU-protected pointer in a structure field.
964 */
965 #define RCU_POINTER_INITIALIZER(p, v) \
966 .p = (typeof(*v) __force __rcu *)(v)
967
968 /*
969 * Does the specified offset indicate that the corresponding rcu_head
970 * structure can be handled by kfree_rcu()?
971 */
972 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
973
974 /*
975 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
976 */
977 #define __kfree_rcu(head, offset) \
978 do { \
979 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
980 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
981 } while (0)
982
983 /**
984 * kfree_rcu() - kfree an object after a grace period.
985 * @ptr: pointer to kfree
986 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
987 *
988 * Many rcu callbacks functions just call kfree() on the base structure.
989 * These functions are trivial, but their size adds up, and furthermore
990 * when they are used in a kernel module, that module must invoke the
991 * high-latency rcu_barrier() function at module-unload time.
992 *
993 * The kfree_rcu() function handles this issue. Rather than encoding a
994 * function address in the embedded rcu_head structure, kfree_rcu() instead
995 * encodes the offset of the rcu_head structure within the base structure.
996 * Because the functions are not allowed in the low-order 4096 bytes of
997 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
998 * If the offset is larger than 4095 bytes, a compile-time error will
999 * be generated in __kfree_rcu(). If this error is triggered, you can
1000 * either fall back to use of call_rcu() or rearrange the structure to
1001 * position the rcu_head structure into the first 4096 bytes.
1002 *
1003 * Note that the allowable offset might decrease in the future, for example,
1004 * to allow something like kmem_cache_free_rcu().
1005 *
1006 * The BUILD_BUG_ON check must not involve any function calls, hence the
1007 * checks are done in macros here.
1008 */
1009 #define kfree_rcu(ptr, rcu_head) \
1010 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1011
1012 #ifdef CONFIG_RCU_NOCB_CPU
1013 extern bool rcu_is_nocb_cpu(int cpu);
1014 #else
rcu_is_nocb_cpu(int cpu)1015 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1016 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
1017
1018
1019 #endif /* __LINUX_RCUPDATE_H */
1020