• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0+ */
2 /*
3  * Read-Copy Update mechanism for mutual exclusion
4  *
5  * Copyright IBM Corporation, 2001
6  *
7  * Author: Dipankar Sarma <dipankar@in.ibm.com>
8  *
9  * Based on the original work by Paul McKenney <paulmck@vnet.ibm.com>
10  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
11  * Papers:
12  * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
13  * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
14  *
15  * For detailed explanation of Read-Copy Update mechanism see -
16  *		http://lse.sourceforge.net/locking/rcupdate.html
17  *
18  */
19 
20 #ifndef __LINUX_RCUPDATE_H
21 #define __LINUX_RCUPDATE_H
22 
23 #include <linux/types.h>
24 #include <linux/compiler.h>
25 #include <linux/atomic.h>
26 #include <linux/irqflags.h>
27 #include <linux/preempt.h>
28 #include <linux/bottom_half.h>
29 #include <linux/lockdep.h>
30 #include <asm/processor.h>
31 #include <linux/cpumask.h>
32 
33 #define ULONG_CMP_GE(a, b)	(ULONG_MAX / 2 >= (a) - (b))
34 #define ULONG_CMP_LT(a, b)	(ULONG_MAX / 2 < (a) - (b))
35 #define ulong2long(a)		(*(long *)(&(a)))
36 #define USHORT_CMP_GE(a, b)	(USHRT_MAX / 2 >= (unsigned short)((a) - (b)))
37 #define USHORT_CMP_LT(a, b)	(USHRT_MAX / 2 < (unsigned short)((a) - (b)))
38 
39 /* Exported common interfaces */
40 void call_rcu(struct rcu_head *head, rcu_callback_t func);
41 void rcu_barrier_tasks(void);
42 void rcu_barrier_tasks_rude(void);
43 void synchronize_rcu(void);
44 
45 #ifdef CONFIG_PREEMPT_RCU
46 
47 void __rcu_read_lock(void);
48 void __rcu_read_unlock(void);
49 
50 /*
51  * Defined as a macro as it is a very low level header included from
52  * areas that don't even know about current.  This gives the rcu_read_lock()
53  * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
54  * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
55  */
56 #define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
57 
58 #else /* #ifdef CONFIG_PREEMPT_RCU */
59 
60 #ifdef CONFIG_TINY_RCU
61 #define rcu_read_unlock_strict() do { } while (0)
62 #else
63 void rcu_read_unlock_strict(void);
64 #endif
65 
__rcu_read_lock(void)66 static inline void __rcu_read_lock(void)
67 {
68 	preempt_disable();
69 }
70 
__rcu_read_unlock(void)71 static inline void __rcu_read_unlock(void)
72 {
73 	preempt_enable();
74 	rcu_read_unlock_strict();
75 }
76 
rcu_preempt_depth(void)77 static inline int rcu_preempt_depth(void)
78 {
79 	return 0;
80 }
81 
82 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
83 
84 /* Internal to kernel */
85 void rcu_init(void);
86 extern int rcu_scheduler_active;
87 void rcu_sched_clock_irq(int user);
88 void rcu_report_dead(unsigned int cpu);
89 void rcutree_migrate_callbacks(int cpu);
90 
91 #ifdef CONFIG_TASKS_RCU_GENERIC
92 void rcu_init_tasks_generic(void);
93 #else
rcu_init_tasks_generic(void)94 static inline void rcu_init_tasks_generic(void) { }
95 #endif
96 
97 #ifdef CONFIG_RCU_STALL_COMMON
98 void rcu_sysrq_start(void);
99 void rcu_sysrq_end(void);
100 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
rcu_sysrq_start(void)101 static inline void rcu_sysrq_start(void) { }
rcu_sysrq_end(void)102 static inline void rcu_sysrq_end(void) { }
103 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
104 
105 #ifdef CONFIG_NO_HZ_FULL
106 void rcu_user_enter(void);
107 void rcu_user_exit(void);
108 #else
rcu_user_enter(void)109 static inline void rcu_user_enter(void) { }
rcu_user_exit(void)110 static inline void rcu_user_exit(void) { }
111 #endif /* CONFIG_NO_HZ_FULL */
112 
113 #ifdef CONFIG_RCU_NOCB_CPU
114 void rcu_init_nohz(void);
115 int rcu_nocb_cpu_offload(int cpu);
116 int rcu_nocb_cpu_deoffload(int cpu);
117 void rcu_nocb_flush_deferred_wakeup(void);
118 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
rcu_init_nohz(void)119 static inline void rcu_init_nohz(void) { }
rcu_nocb_cpu_offload(int cpu)120 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
rcu_nocb_cpu_deoffload(int cpu)121 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
rcu_nocb_flush_deferred_wakeup(void)122 static inline void rcu_nocb_flush_deferred_wakeup(void) { }
123 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
124 
125 /**
126  * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
127  * @a: Code that RCU needs to pay attention to.
128  *
129  * RCU read-side critical sections are forbidden in the inner idle loop,
130  * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU
131  * will happily ignore any such read-side critical sections.  However,
132  * things like powertop need tracepoints in the inner idle loop.
133  *
134  * This macro provides the way out:  RCU_NONIDLE(do_something_with_RCU())
135  * will tell RCU that it needs to pay attention, invoke its argument
136  * (in this example, calling the do_something_with_RCU() function),
137  * and then tell RCU to go back to ignoring this CPU.  It is permissible
138  * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
139  * on the order of a million or so, even on 32-bit systems).  It is
140  * not legal to block within RCU_NONIDLE(), nor is it permissible to
141  * transfer control either into or out of RCU_NONIDLE()'s statement.
142  */
143 #define RCU_NONIDLE(a) \
144 	do { \
145 		rcu_irq_enter_irqson(); \
146 		do { a; } while (0); \
147 		rcu_irq_exit_irqson(); \
148 	} while (0)
149 
150 /*
151  * Note a quasi-voluntary context switch for RCU-tasks's benefit.
152  * This is a macro rather than an inline function to avoid #include hell.
153  */
154 #ifdef CONFIG_TASKS_RCU_GENERIC
155 
156 # ifdef CONFIG_TASKS_RCU
157 # define rcu_tasks_classic_qs(t, preempt)				\
158 	do {								\
159 		if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout))	\
160 			WRITE_ONCE((t)->rcu_tasks_holdout, false);	\
161 	} while (0)
162 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
163 void synchronize_rcu_tasks(void);
164 # else
165 # define rcu_tasks_classic_qs(t, preempt) do { } while (0)
166 # define call_rcu_tasks call_rcu
167 # define synchronize_rcu_tasks synchronize_rcu
168 # endif
169 
170 # ifdef CONFIG_TASKS_TRACE_RCU
171 # define rcu_tasks_trace_qs(t)						\
172 	do {								\
173 		if (!likely(READ_ONCE((t)->trc_reader_checked)) &&	\
174 		    !unlikely(READ_ONCE((t)->trc_reader_nesting))) {	\
175 			smp_store_release(&(t)->trc_reader_checked, true); \
176 			smp_mb(); /* Readers partitioned by store. */	\
177 		}							\
178 	} while (0)
179 # else
180 # define rcu_tasks_trace_qs(t) do { } while (0)
181 # endif
182 
183 #define rcu_tasks_qs(t, preempt)					\
184 do {									\
185 	rcu_tasks_classic_qs((t), (preempt));				\
186 	rcu_tasks_trace_qs((t));					\
187 } while (0)
188 
189 # ifdef CONFIG_TASKS_RUDE_RCU
190 void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
191 void synchronize_rcu_tasks_rude(void);
192 # endif
193 
194 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
195 void exit_tasks_rcu_start(void);
196 void exit_tasks_rcu_stop(void);
197 void exit_tasks_rcu_finish(void);
198 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
199 #define rcu_tasks_qs(t, preempt) do { } while (0)
200 #define rcu_note_voluntary_context_switch(t) do { } while (0)
201 #define call_rcu_tasks call_rcu
202 #define synchronize_rcu_tasks synchronize_rcu
exit_tasks_rcu_start(void)203 static inline void exit_tasks_rcu_start(void) { }
exit_tasks_rcu_stop(void)204 static inline void exit_tasks_rcu_stop(void) { }
exit_tasks_rcu_finish(void)205 static inline void exit_tasks_rcu_finish(void) { }
206 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
207 
208 /**
209  * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
210  *
211  * This macro resembles cond_resched(), except that it is defined to
212  * report potential quiescent states to RCU-tasks even if the cond_resched()
213  * machinery were to be shut off, as some advocate for PREEMPTION kernels.
214  */
215 #define cond_resched_tasks_rcu_qs() \
216 do { \
217 	rcu_tasks_qs(current, false); \
218 	cond_resched(); \
219 } while (0)
220 
221 /*
222  * Infrastructure to implement the synchronize_() primitives in
223  * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
224  */
225 
226 #if defined(CONFIG_TREE_RCU)
227 #include <linux/rcutree.h>
228 #elif defined(CONFIG_TINY_RCU)
229 #include <linux/rcutiny.h>
230 #else
231 #error "Unknown RCU implementation specified to kernel configuration"
232 #endif
233 
234 /*
235  * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
236  * are needed for dynamic initialization and destruction of rcu_head
237  * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
238  * dynamic initialization and destruction of statically allocated rcu_head
239  * structures.  However, rcu_head structures allocated dynamically in the
240  * heap don't need any initialization.
241  */
242 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
243 void init_rcu_head(struct rcu_head *head);
244 void destroy_rcu_head(struct rcu_head *head);
245 void init_rcu_head_on_stack(struct rcu_head *head);
246 void destroy_rcu_head_on_stack(struct rcu_head *head);
247 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
init_rcu_head(struct rcu_head * head)248 static inline void init_rcu_head(struct rcu_head *head) { }
destroy_rcu_head(struct rcu_head * head)249 static inline void destroy_rcu_head(struct rcu_head *head) { }
init_rcu_head_on_stack(struct rcu_head * head)250 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
destroy_rcu_head_on_stack(struct rcu_head * head)251 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
252 #endif	/* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
253 
254 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
255 bool rcu_lockdep_current_cpu_online(void);
256 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
rcu_lockdep_current_cpu_online(void)257 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
258 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
259 
260 extern struct lockdep_map rcu_lock_map;
261 extern struct lockdep_map rcu_bh_lock_map;
262 extern struct lockdep_map rcu_sched_lock_map;
263 extern struct lockdep_map rcu_callback_map;
264 
265 #ifdef CONFIG_DEBUG_LOCK_ALLOC
266 
rcu_lock_acquire(struct lockdep_map * map)267 static inline void rcu_lock_acquire(struct lockdep_map *map)
268 {
269 	lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
270 }
271 
rcu_lock_release(struct lockdep_map * map)272 static inline void rcu_lock_release(struct lockdep_map *map)
273 {
274 	lock_release(map, _THIS_IP_);
275 }
276 
277 int debug_lockdep_rcu_enabled(void);
278 int rcu_read_lock_held(void);
279 int rcu_read_lock_bh_held(void);
280 int rcu_read_lock_sched_held(void);
281 int rcu_read_lock_any_held(void);
282 
283 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
284 
285 # define rcu_lock_acquire(a)		do { } while (0)
286 # define rcu_lock_release(a)		do { } while (0)
287 
rcu_read_lock_held(void)288 static inline int rcu_read_lock_held(void)
289 {
290 	return 1;
291 }
292 
rcu_read_lock_bh_held(void)293 static inline int rcu_read_lock_bh_held(void)
294 {
295 	return 1;
296 }
297 
rcu_read_lock_sched_held(void)298 static inline int rcu_read_lock_sched_held(void)
299 {
300 	return !preemptible();
301 }
302 
rcu_read_lock_any_held(void)303 static inline int rcu_read_lock_any_held(void)
304 {
305 	return !preemptible();
306 }
307 
308 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
309 
310 #ifdef CONFIG_PROVE_RCU
311 
312 /**
313  * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
314  * @c: condition to check
315  * @s: informative message
316  *
317  * This checks debug_lockdep_rcu_enabled() before checking (c) to
318  * prevent early boot splats due to lockdep not yet being initialized,
319  * and rechecks it after checking (c) to prevent false-positive splats
320  * due to races with lockdep being disabled.  See commit 3066820034b5dd
321  * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail.
322  */
323 #define RCU_LOCKDEP_WARN(c, s)						\
324 	do {								\
325 		static bool __section(".data.unlikely") __warned;	\
326 		if (debug_lockdep_rcu_enabled() && (c) &&		\
327 		    debug_lockdep_rcu_enabled() && !__warned) {		\
328 			__warned = true;				\
329 			lockdep_rcu_suspicious(__FILE__, __LINE__, s);	\
330 		}							\
331 	} while (0)
332 
333 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
rcu_preempt_sleep_check(void)334 static inline void rcu_preempt_sleep_check(void)
335 {
336 	RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
337 			 "Illegal context switch in RCU read-side critical section");
338 }
339 #else /* #ifdef CONFIG_PROVE_RCU */
rcu_preempt_sleep_check(void)340 static inline void rcu_preempt_sleep_check(void) { }
341 #endif /* #else #ifdef CONFIG_PROVE_RCU */
342 
343 #define rcu_sleep_check()						\
344 	do {								\
345 		rcu_preempt_sleep_check();				\
346 		if (!IS_ENABLED(CONFIG_PREEMPT_RT))			\
347 		    RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map),	\
348 				 "Illegal context switch in RCU-bh read-side critical section"); \
349 		RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map),	\
350 				 "Illegal context switch in RCU-sched read-side critical section"); \
351 	} while (0)
352 
353 #else /* #ifdef CONFIG_PROVE_RCU */
354 
355 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
356 #define rcu_sleep_check() do { } while (0)
357 
358 #endif /* #else #ifdef CONFIG_PROVE_RCU */
359 
360 /*
361  * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
362  * and rcu_assign_pointer().  Some of these could be folded into their
363  * callers, but they are left separate in order to ease introduction of
364  * multiple pointers markings to match different RCU implementations
365  * (e.g., __srcu), should this make sense in the future.
366  */
367 
368 #ifdef __CHECKER__
369 #define rcu_check_sparse(p, space) \
370 	((void)(((typeof(*p) space *)p) == p))
371 #else /* #ifdef __CHECKER__ */
372 #define rcu_check_sparse(p, space)
373 #endif /* #else #ifdef __CHECKER__ */
374 
375 /**
376  * unrcu_pointer - mark a pointer as not being RCU protected
377  * @p: pointer needing to lose its __rcu property
378  *
379  * Converts @p from an __rcu pointer to a __kernel pointer.
380  * This allows an __rcu pointer to be used with xchg() and friends.
381  */
382 #define unrcu_pointer(p)						\
383 ({									\
384 	typeof(*p) *_________p1 = (typeof(*p) *__force)(p);		\
385 	rcu_check_sparse(p, __rcu);					\
386 	((typeof(*p) __force __kernel *)(_________p1)); 		\
387 })
388 
389 #define __rcu_access_pointer(p, space) \
390 ({ \
391 	typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
392 	rcu_check_sparse(p, space); \
393 	((typeof(*p) __force __kernel *)(_________p1)); \
394 })
395 #define __rcu_dereference_check(p, c, space) \
396 ({ \
397 	/* Dependency order vs. p above. */ \
398 	typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \
399 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
400 	rcu_check_sparse(p, space); \
401 	((typeof(*p) __force __kernel *)(________p1)); \
402 })
403 #define __rcu_dereference_protected(p, c, space) \
404 ({ \
405 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
406 	rcu_check_sparse(p, space); \
407 	((typeof(*p) __force __kernel *)(p)); \
408 })
409 #define rcu_dereference_raw(p) \
410 ({ \
411 	/* Dependency order vs. p above. */ \
412 	typeof(p) ________p1 = READ_ONCE(p); \
413 	((typeof(*p) __force __kernel *)(________p1)); \
414 })
415 
416 /**
417  * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
418  * @v: The value to statically initialize with.
419  */
420 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
421 
422 /**
423  * rcu_assign_pointer() - assign to RCU-protected pointer
424  * @p: pointer to assign to
425  * @v: value to assign (publish)
426  *
427  * Assigns the specified value to the specified RCU-protected
428  * pointer, ensuring that any concurrent RCU readers will see
429  * any prior initialization.
430  *
431  * Inserts memory barriers on architectures that require them
432  * (which is most of them), and also prevents the compiler from
433  * reordering the code that initializes the structure after the pointer
434  * assignment.  More importantly, this call documents which pointers
435  * will be dereferenced by RCU read-side code.
436  *
437  * In some special cases, you may use RCU_INIT_POINTER() instead
438  * of rcu_assign_pointer().  RCU_INIT_POINTER() is a bit faster due
439  * to the fact that it does not constrain either the CPU or the compiler.
440  * That said, using RCU_INIT_POINTER() when you should have used
441  * rcu_assign_pointer() is a very bad thing that results in
442  * impossible-to-diagnose memory corruption.  So please be careful.
443  * See the RCU_INIT_POINTER() comment header for details.
444  *
445  * Note that rcu_assign_pointer() evaluates each of its arguments only
446  * once, appearances notwithstanding.  One of the "extra" evaluations
447  * is in typeof() and the other visible only to sparse (__CHECKER__),
448  * neither of which actually execute the argument.  As with most cpp
449  * macros, this execute-arguments-only-once property is important, so
450  * please be careful when making changes to rcu_assign_pointer() and the
451  * other macros that it invokes.
452  */
453 #define rcu_assign_pointer(p, v)					      \
454 do {									      \
455 	uintptr_t _r_a_p__v = (uintptr_t)(v);				      \
456 	rcu_check_sparse(p, __rcu);					      \
457 									      \
458 	if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL)	      \
459 		WRITE_ONCE((p), (typeof(p))(_r_a_p__v));		      \
460 	else								      \
461 		smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
462 } while (0)
463 
464 /**
465  * rcu_replace_pointer() - replace an RCU pointer, returning its old value
466  * @rcu_ptr: RCU pointer, whose old value is returned
467  * @ptr: regular pointer
468  * @c: the lockdep conditions under which the dereference will take place
469  *
470  * Perform a replacement, where @rcu_ptr is an RCU-annotated
471  * pointer and @c is the lockdep argument that is passed to the
472  * rcu_dereference_protected() call used to read that pointer.  The old
473  * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
474  */
475 #define rcu_replace_pointer(rcu_ptr, ptr, c)				\
476 ({									\
477 	typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c));	\
478 	rcu_assign_pointer((rcu_ptr), (ptr));				\
479 	__tmp;								\
480 })
481 
482 /**
483  * rcu_access_pointer() - fetch RCU pointer with no dereferencing
484  * @p: The pointer to read
485  *
486  * Return the value of the specified RCU-protected pointer, but omit the
487  * lockdep checks for being in an RCU read-side critical section.  This is
488  * useful when the value of this pointer is accessed, but the pointer is
489  * not dereferenced, for example, when testing an RCU-protected pointer
490  * against NULL.  Although rcu_access_pointer() may also be used in cases
491  * where update-side locks prevent the value of the pointer from changing,
492  * you should instead use rcu_dereference_protected() for this use case.
493  *
494  * It is also permissible to use rcu_access_pointer() when read-side
495  * access to the pointer was removed at least one grace period ago, as
496  * is the case in the context of the RCU callback that is freeing up
497  * the data, or after a synchronize_rcu() returns.  This can be useful
498  * when tearing down multi-linked structures after a grace period
499  * has elapsed.
500  */
501 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
502 
503 /**
504  * rcu_dereference_check() - rcu_dereference with debug checking
505  * @p: The pointer to read, prior to dereferencing
506  * @c: The conditions under which the dereference will take place
507  *
508  * Do an rcu_dereference(), but check that the conditions under which the
509  * dereference will take place are correct.  Typically the conditions
510  * indicate the various locking conditions that should be held at that
511  * point.  The check should return true if the conditions are satisfied.
512  * An implicit check for being in an RCU read-side critical section
513  * (rcu_read_lock()) is included.
514  *
515  * For example:
516  *
517  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
518  *
519  * could be used to indicate to lockdep that foo->bar may only be dereferenced
520  * if either rcu_read_lock() is held, or that the lock required to replace
521  * the bar struct at foo->bar is held.
522  *
523  * Note that the list of conditions may also include indications of when a lock
524  * need not be held, for example during initialisation or destruction of the
525  * target struct:
526  *
527  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
528  *					      atomic_read(&foo->usage) == 0);
529  *
530  * Inserts memory barriers on architectures that require them
531  * (currently only the Alpha), prevents the compiler from refetching
532  * (and from merging fetches), and, more importantly, documents exactly
533  * which pointers are protected by RCU and checks that the pointer is
534  * annotated as __rcu.
535  */
536 #define rcu_dereference_check(p, c) \
537 	__rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
538 
539 /**
540  * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
541  * @p: The pointer to read, prior to dereferencing
542  * @c: The conditions under which the dereference will take place
543  *
544  * This is the RCU-bh counterpart to rcu_dereference_check().  However,
545  * please note that starting in v5.0 kernels, vanilla RCU grace periods
546  * wait for local_bh_disable() regions of code in addition to regions of
547  * code demarked by rcu_read_lock() and rcu_read_unlock().  This means
548  * that synchronize_rcu(), call_rcu, and friends all take not only
549  * rcu_read_lock() but also rcu_read_lock_bh() into account.
550  */
551 #define rcu_dereference_bh_check(p, c) \
552 	__rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
553 
554 /**
555  * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
556  * @p: The pointer to read, prior to dereferencing
557  * @c: The conditions under which the dereference will take place
558  *
559  * This is the RCU-sched counterpart to rcu_dereference_check().
560  * However, please note that starting in v5.0 kernels, vanilla RCU grace
561  * periods wait for preempt_disable() regions of code in addition to
562  * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
563  * This means that synchronize_rcu(), call_rcu, and friends all take not
564  * only rcu_read_lock() but also rcu_read_lock_sched() into account.
565  */
566 #define rcu_dereference_sched_check(p, c) \
567 	__rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
568 				__rcu)
569 
570 /*
571  * The tracing infrastructure traces RCU (we want that), but unfortunately
572  * some of the RCU checks causes tracing to lock up the system.
573  *
574  * The no-tracing version of rcu_dereference_raw() must not call
575  * rcu_read_lock_held().
576  */
577 #define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu)
578 
579 /**
580  * rcu_dereference_protected() - fetch RCU pointer when updates prevented
581  * @p: The pointer to read, prior to dereferencing
582  * @c: The conditions under which the dereference will take place
583  *
584  * Return the value of the specified RCU-protected pointer, but omit
585  * the READ_ONCE().  This is useful in cases where update-side locks
586  * prevent the value of the pointer from changing.  Please note that this
587  * primitive does *not* prevent the compiler from repeating this reference
588  * or combining it with other references, so it should not be used without
589  * protection of appropriate locks.
590  *
591  * This function is only for update-side use.  Using this function
592  * when protected only by rcu_read_lock() will result in infrequent
593  * but very ugly failures.
594  */
595 #define rcu_dereference_protected(p, c) \
596 	__rcu_dereference_protected((p), (c), __rcu)
597 
598 
599 /**
600  * rcu_dereference() - fetch RCU-protected pointer for dereferencing
601  * @p: The pointer to read, prior to dereferencing
602  *
603  * This is a simple wrapper around rcu_dereference_check().
604  */
605 #define rcu_dereference(p) rcu_dereference_check(p, 0)
606 
607 /**
608  * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
609  * @p: The pointer to read, prior to dereferencing
610  *
611  * Makes rcu_dereference_check() do the dirty work.
612  */
613 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
614 
615 /**
616  * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
617  * @p: The pointer to read, prior to dereferencing
618  *
619  * Makes rcu_dereference_check() do the dirty work.
620  */
621 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
622 
623 /**
624  * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
625  * @p: The pointer to hand off
626  *
627  * This is simply an identity function, but it documents where a pointer
628  * is handed off from RCU to some other synchronization mechanism, for
629  * example, reference counting or locking.  In C11, it would map to
630  * kill_dependency().  It could be used as follows::
631  *
632  *	rcu_read_lock();
633  *	p = rcu_dereference(gp);
634  *	long_lived = is_long_lived(p);
635  *	if (long_lived) {
636  *		if (!atomic_inc_not_zero(p->refcnt))
637  *			long_lived = false;
638  *		else
639  *			p = rcu_pointer_handoff(p);
640  *	}
641  *	rcu_read_unlock();
642  */
643 #define rcu_pointer_handoff(p) (p)
644 
645 /**
646  * rcu_read_lock() - mark the beginning of an RCU read-side critical section
647  *
648  * When synchronize_rcu() is invoked on one CPU while other CPUs
649  * are within RCU read-side critical sections, then the
650  * synchronize_rcu() is guaranteed to block until after all the other
651  * CPUs exit their critical sections.  Similarly, if call_rcu() is invoked
652  * on one CPU while other CPUs are within RCU read-side critical
653  * sections, invocation of the corresponding RCU callback is deferred
654  * until after the all the other CPUs exit their critical sections.
655  *
656  * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
657  * wait for regions of code with preemption disabled, including regions of
658  * code with interrupts or softirqs disabled.  In pre-v5.0 kernels, which
659  * define synchronize_sched(), only code enclosed within rcu_read_lock()
660  * and rcu_read_unlock() are guaranteed to be waited for.
661  *
662  * Note, however, that RCU callbacks are permitted to run concurrently
663  * with new RCU read-side critical sections.  One way that this can happen
664  * is via the following sequence of events: (1) CPU 0 enters an RCU
665  * read-side critical section, (2) CPU 1 invokes call_rcu() to register
666  * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
667  * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
668  * callback is invoked.  This is legal, because the RCU read-side critical
669  * section that was running concurrently with the call_rcu() (and which
670  * therefore might be referencing something that the corresponding RCU
671  * callback would free up) has completed before the corresponding
672  * RCU callback is invoked.
673  *
674  * RCU read-side critical sections may be nested.  Any deferred actions
675  * will be deferred until the outermost RCU read-side critical section
676  * completes.
677  *
678  * You can avoid reading and understanding the next paragraph by
679  * following this rule: don't put anything in an rcu_read_lock() RCU
680  * read-side critical section that would block in a !PREEMPTION kernel.
681  * But if you want the full story, read on!
682  *
683  * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
684  * it is illegal to block while in an RCU read-side critical section.
685  * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
686  * kernel builds, RCU read-side critical sections may be preempted,
687  * but explicit blocking is illegal.  Finally, in preemptible RCU
688  * implementations in real-time (with -rt patchset) kernel builds, RCU
689  * read-side critical sections may be preempted and they may also block, but
690  * only when acquiring spinlocks that are subject to priority inheritance.
691  */
rcu_read_lock(void)692 static __always_inline void rcu_read_lock(void)
693 {
694 	__rcu_read_lock();
695 	__acquire(RCU);
696 	rcu_lock_acquire(&rcu_lock_map);
697 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
698 			 "rcu_read_lock() used illegally while idle");
699 }
700 
701 /*
702  * So where is rcu_write_lock()?  It does not exist, as there is no
703  * way for writers to lock out RCU readers.  This is a feature, not
704  * a bug -- this property is what provides RCU's performance benefits.
705  * Of course, writers must coordinate with each other.  The normal
706  * spinlock primitives work well for this, but any other technique may be
707  * used as well.  RCU does not care how the writers keep out of each
708  * others' way, as long as they do so.
709  */
710 
711 /**
712  * rcu_read_unlock() - marks the end of an RCU read-side critical section.
713  *
714  * In almost all situations, rcu_read_unlock() is immune from deadlock.
715  * In recent kernels that have consolidated synchronize_sched() and
716  * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
717  * also extends to the scheduler's runqueue and priority-inheritance
718  * spinlocks, courtesy of the quiescent-state deferral that is carried
719  * out when rcu_read_unlock() is invoked with interrupts disabled.
720  *
721  * See rcu_read_lock() for more information.
722  */
rcu_read_unlock(void)723 static inline void rcu_read_unlock(void)
724 {
725 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
726 			 "rcu_read_unlock() used illegally while idle");
727 	__release(RCU);
728 	__rcu_read_unlock();
729 	rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
730 }
731 
732 /**
733  * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
734  *
735  * This is equivalent to rcu_read_lock(), but also disables softirqs.
736  * Note that anything else that disables softirqs can also serve as an RCU
737  * read-side critical section.  However, please note that this equivalence
738  * applies only to v5.0 and later.  Before v5.0, rcu_read_lock() and
739  * rcu_read_lock_bh() were unrelated.
740  *
741  * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
742  * must occur in the same context, for example, it is illegal to invoke
743  * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
744  * was invoked from some other task.
745  */
rcu_read_lock_bh(void)746 static inline void rcu_read_lock_bh(void)
747 {
748 	local_bh_disable();
749 	__acquire(RCU_BH);
750 	rcu_lock_acquire(&rcu_bh_lock_map);
751 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
752 			 "rcu_read_lock_bh() used illegally while idle");
753 }
754 
755 /**
756  * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
757  *
758  * See rcu_read_lock_bh() for more information.
759  */
rcu_read_unlock_bh(void)760 static inline void rcu_read_unlock_bh(void)
761 {
762 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
763 			 "rcu_read_unlock_bh() used illegally while idle");
764 	rcu_lock_release(&rcu_bh_lock_map);
765 	__release(RCU_BH);
766 	local_bh_enable();
767 }
768 
769 /**
770  * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
771  *
772  * This is equivalent to rcu_read_lock(), but also disables preemption.
773  * Read-side critical sections can also be introduced by anything else that
774  * disables preemption, including local_irq_disable() and friends.  However,
775  * please note that the equivalence to rcu_read_lock() applies only to
776  * v5.0 and later.  Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
777  * were unrelated.
778  *
779  * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
780  * must occur in the same context, for example, it is illegal to invoke
781  * rcu_read_unlock_sched() from process context if the matching
782  * rcu_read_lock_sched() was invoked from an NMI handler.
783  */
rcu_read_lock_sched(void)784 static inline void rcu_read_lock_sched(void)
785 {
786 	preempt_disable();
787 	__acquire(RCU_SCHED);
788 	rcu_lock_acquire(&rcu_sched_lock_map);
789 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
790 			 "rcu_read_lock_sched() used illegally while idle");
791 }
792 
793 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_lock_sched_notrace(void)794 static inline notrace void rcu_read_lock_sched_notrace(void)
795 {
796 	preempt_disable_notrace();
797 	__acquire(RCU_SCHED);
798 }
799 
800 /**
801  * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
802  *
803  * See rcu_read_lock_sched() for more information.
804  */
rcu_read_unlock_sched(void)805 static inline void rcu_read_unlock_sched(void)
806 {
807 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
808 			 "rcu_read_unlock_sched() used illegally while idle");
809 	rcu_lock_release(&rcu_sched_lock_map);
810 	__release(RCU_SCHED);
811 	preempt_enable();
812 }
813 
814 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_unlock_sched_notrace(void)815 static inline notrace void rcu_read_unlock_sched_notrace(void)
816 {
817 	__release(RCU_SCHED);
818 	preempt_enable_notrace();
819 }
820 
821 /**
822  * RCU_INIT_POINTER() - initialize an RCU protected pointer
823  * @p: The pointer to be initialized.
824  * @v: The value to initialized the pointer to.
825  *
826  * Initialize an RCU-protected pointer in special cases where readers
827  * do not need ordering constraints on the CPU or the compiler.  These
828  * special cases are:
829  *
830  * 1.	This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
831  * 2.	The caller has taken whatever steps are required to prevent
832  *	RCU readers from concurrently accessing this pointer *or*
833  * 3.	The referenced data structure has already been exposed to
834  *	readers either at compile time or via rcu_assign_pointer() *and*
835  *
836  *	a.	You have not made *any* reader-visible changes to
837  *		this structure since then *or*
838  *	b.	It is OK for readers accessing this structure from its
839  *		new location to see the old state of the structure.  (For
840  *		example, the changes were to statistical counters or to
841  *		other state where exact synchronization is not required.)
842  *
843  * Failure to follow these rules governing use of RCU_INIT_POINTER() will
844  * result in impossible-to-diagnose memory corruption.  As in the structures
845  * will look OK in crash dumps, but any concurrent RCU readers might
846  * see pre-initialized values of the referenced data structure.  So
847  * please be very careful how you use RCU_INIT_POINTER()!!!
848  *
849  * If you are creating an RCU-protected linked structure that is accessed
850  * by a single external-to-structure RCU-protected pointer, then you may
851  * use RCU_INIT_POINTER() to initialize the internal RCU-protected
852  * pointers, but you must use rcu_assign_pointer() to initialize the
853  * external-to-structure pointer *after* you have completely initialized
854  * the reader-accessible portions of the linked structure.
855  *
856  * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
857  * ordering guarantees for either the CPU or the compiler.
858  */
859 #define RCU_INIT_POINTER(p, v) \
860 	do { \
861 		rcu_check_sparse(p, __rcu); \
862 		WRITE_ONCE(p, RCU_INITIALIZER(v)); \
863 	} while (0)
864 
865 /**
866  * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
867  * @p: The pointer to be initialized.
868  * @v: The value to initialized the pointer to.
869  *
870  * GCC-style initialization for an RCU-protected pointer in a structure field.
871  */
872 #define RCU_POINTER_INITIALIZER(p, v) \
873 		.p = RCU_INITIALIZER(v)
874 
875 /*
876  * Does the specified offset indicate that the corresponding rcu_head
877  * structure can be handled by kvfree_rcu()?
878  */
879 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
880 
881 /**
882  * kfree_rcu() - kfree an object after a grace period.
883  * @ptr: pointer to kfree for both single- and double-argument invocations.
884  * @rhf: the name of the struct rcu_head within the type of @ptr,
885  *       but only for double-argument invocations.
886  *
887  * Many rcu callbacks functions just call kfree() on the base structure.
888  * These functions are trivial, but their size adds up, and furthermore
889  * when they are used in a kernel module, that module must invoke the
890  * high-latency rcu_barrier() function at module-unload time.
891  *
892  * The kfree_rcu() function handles this issue.  Rather than encoding a
893  * function address in the embedded rcu_head structure, kfree_rcu() instead
894  * encodes the offset of the rcu_head structure within the base structure.
895  * Because the functions are not allowed in the low-order 4096 bytes of
896  * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
897  * If the offset is larger than 4095 bytes, a compile-time error will
898  * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
899  * either fall back to use of call_rcu() or rearrange the structure to
900  * position the rcu_head structure into the first 4096 bytes.
901  *
902  * Note that the allowable offset might decrease in the future, for example,
903  * to allow something like kmem_cache_free_rcu().
904  *
905  * The BUILD_BUG_ON check must not involve any function calls, hence the
906  * checks are done in macros here.
907  */
908 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf)
909 
910 /**
911  * kvfree_rcu() - kvfree an object after a grace period.
912  *
913  * This macro consists of one or two arguments and it is
914  * based on whether an object is head-less or not. If it
915  * has a head then a semantic stays the same as it used
916  * to be before:
917  *
918  *     kvfree_rcu(ptr, rhf);
919  *
920  * where @ptr is a pointer to kvfree(), @rhf is the name
921  * of the rcu_head structure within the type of @ptr.
922  *
923  * When it comes to head-less variant, only one argument
924  * is passed and that is just a pointer which has to be
925  * freed after a grace period. Therefore the semantic is
926  *
927  *     kvfree_rcu(ptr);
928  *
929  * where @ptr is a pointer to kvfree().
930  *
931  * Please note, head-less way of freeing is permitted to
932  * use from a context that has to follow might_sleep()
933  * annotation. Otherwise, please switch and embed the
934  * rcu_head structure within the type of @ptr.
935  */
936 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__,		\
937 	kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__)
938 
939 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME
940 #define kvfree_rcu_arg_2(ptr, rhf)					\
941 do {									\
942 	typeof (ptr) ___p = (ptr);					\
943 									\
944 	if (___p) {									\
945 		BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf)));	\
946 		kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long)		\
947 			(offsetof(typeof(*(ptr)), rhf)));				\
948 	}										\
949 } while (0)
950 
951 #define kvfree_rcu_arg_1(ptr)					\
952 do {								\
953 	typeof(ptr) ___p = (ptr);				\
954 								\
955 	if (___p)						\
956 		kvfree_call_rcu(NULL, (rcu_callback_t) (___p));	\
957 } while (0)
958 
959 /*
960  * Place this after a lock-acquisition primitive to guarantee that
961  * an UNLOCK+LOCK pair acts as a full barrier.  This guarantee applies
962  * if the UNLOCK and LOCK are executed by the same CPU or if the
963  * UNLOCK and LOCK operate on the same lock variable.
964  */
965 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
966 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
967 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
968 #define smp_mb__after_unlock_lock()	do { } while (0)
969 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
970 
971 
972 /* Has the specified rcu_head structure been handed to call_rcu()? */
973 
974 /**
975  * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
976  * @rhp: The rcu_head structure to initialize.
977  *
978  * If you intend to invoke rcu_head_after_call_rcu() to test whether a
979  * given rcu_head structure has already been passed to call_rcu(), then
980  * you must also invoke this rcu_head_init() function on it just after
981  * allocating that structure.  Calls to this function must not race with
982  * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
983  */
rcu_head_init(struct rcu_head * rhp)984 static inline void rcu_head_init(struct rcu_head *rhp)
985 {
986 	rhp->func = (rcu_callback_t)~0L;
987 }
988 
989 /**
990  * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
991  * @rhp: The rcu_head structure to test.
992  * @f: The function passed to call_rcu() along with @rhp.
993  *
994  * Returns @true if the @rhp has been passed to call_rcu() with @func,
995  * and @false otherwise.  Emits a warning in any other case, including
996  * the case where @rhp has already been invoked after a grace period.
997  * Calls to this function must not race with callback invocation.  One way
998  * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
999  * in an RCU read-side critical section that includes a read-side fetch
1000  * of the pointer to the structure containing @rhp.
1001  */
1002 static inline bool
rcu_head_after_call_rcu(struct rcu_head * rhp,rcu_callback_t f)1003 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1004 {
1005 	rcu_callback_t func = READ_ONCE(rhp->func);
1006 
1007 	if (func == f)
1008 		return true;
1009 	WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1010 	return false;
1011 }
1012 
1013 /* kernel/ksysfs.c definitions */
1014 extern int rcu_expedited;
1015 extern int rcu_normal;
1016 
1017 #endif /* __LINUX_RCUPDATE_H */
1018