• 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 #ifdef CONFIG_RCU_LAZY
85 void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func);
86 #else
call_rcu_hurry(struct rcu_head * head,rcu_callback_t func)87 static inline void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func)
88 {
89 	call_rcu(head, func);
90 }
91 #endif
92 
93 /* Internal to kernel */
94 void rcu_init(void);
95 extern int rcu_scheduler_active __read_mostly;
96 void rcu_sched_clock_irq(int user);
97 void rcu_report_dead(unsigned int cpu);
98 void rcutree_migrate_callbacks(int cpu);
99 
100 #ifdef CONFIG_TASKS_RCU_GENERIC
101 void rcu_init_tasks_generic(void);
102 #else
rcu_init_tasks_generic(void)103 static inline void rcu_init_tasks_generic(void) { }
104 #endif
105 
106 #ifdef CONFIG_RCU_STALL_COMMON
107 void rcu_sysrq_start(void);
108 void rcu_sysrq_end(void);
109 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
rcu_sysrq_start(void)110 static inline void rcu_sysrq_start(void) { }
rcu_sysrq_end(void)111 static inline void rcu_sysrq_end(void) { }
112 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
113 
114 #ifdef CONFIG_NO_HZ_FULL
115 void rcu_user_enter(void);
116 void rcu_user_exit(void);
117 #else
rcu_user_enter(void)118 static inline void rcu_user_enter(void) { }
rcu_user_exit(void)119 static inline void rcu_user_exit(void) { }
120 #endif /* CONFIG_NO_HZ_FULL */
121 
122 #ifdef CONFIG_RCU_NOCB_CPU
123 void rcu_init_nohz(void);
124 int rcu_nocb_cpu_offload(int cpu);
125 int rcu_nocb_cpu_deoffload(int cpu);
126 void rcu_nocb_flush_deferred_wakeup(void);
127 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
rcu_init_nohz(void)128 static inline void rcu_init_nohz(void) { }
rcu_nocb_cpu_offload(int cpu)129 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
rcu_nocb_cpu_deoffload(int cpu)130 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
rcu_nocb_flush_deferred_wakeup(void)131 static inline void rcu_nocb_flush_deferred_wakeup(void) { }
132 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
133 
134 /**
135  * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
136  * @a: Code that RCU needs to pay attention to.
137  *
138  * RCU read-side critical sections are forbidden in the inner idle loop,
139  * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU
140  * will happily ignore any such read-side critical sections.  However,
141  * things like powertop need tracepoints in the inner idle loop.
142  *
143  * This macro provides the way out:  RCU_NONIDLE(do_something_with_RCU())
144  * will tell RCU that it needs to pay attention, invoke its argument
145  * (in this example, calling the do_something_with_RCU() function),
146  * and then tell RCU to go back to ignoring this CPU.  It is permissible
147  * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
148  * on the order of a million or so, even on 32-bit systems).  It is
149  * not legal to block within RCU_NONIDLE(), nor is it permissible to
150  * transfer control either into or out of RCU_NONIDLE()'s statement.
151  */
152 #define RCU_NONIDLE(a) \
153 	do { \
154 		rcu_irq_enter_irqson(); \
155 		do { a; } while (0); \
156 		rcu_irq_exit_irqson(); \
157 	} while (0)
158 
159 /*
160  * Note a quasi-voluntary context switch for RCU-tasks's benefit.
161  * This is a macro rather than an inline function to avoid #include hell.
162  */
163 #ifdef CONFIG_TASKS_RCU_GENERIC
164 
165 # ifdef CONFIG_TASKS_RCU
166 # define rcu_tasks_classic_qs(t, preempt)				\
167 	do {								\
168 		if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout))	\
169 			WRITE_ONCE((t)->rcu_tasks_holdout, false);	\
170 	} while (0)
171 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
172 void synchronize_rcu_tasks(void);
173 # else
174 # define rcu_tasks_classic_qs(t, preempt) do { } while (0)
175 # define call_rcu_tasks call_rcu
176 # define synchronize_rcu_tasks synchronize_rcu
177 # endif
178 
179 # ifdef CONFIG_TASKS_TRACE_RCU
180 # define rcu_tasks_trace_qs(t)						\
181 	do {								\
182 		if (!likely(READ_ONCE((t)->trc_reader_checked)) &&	\
183 		    !unlikely(READ_ONCE((t)->trc_reader_nesting))) {	\
184 			smp_store_release(&(t)->trc_reader_checked, true); \
185 			smp_mb(); /* Readers partitioned by store. */	\
186 		}							\
187 	} while (0)
188 # else
189 # define rcu_tasks_trace_qs(t) do { } while (0)
190 # endif
191 
192 #define rcu_tasks_qs(t, preempt)					\
193 do {									\
194 	rcu_tasks_classic_qs((t), (preempt));				\
195 	rcu_tasks_trace_qs((t));					\
196 } while (0)
197 
198 # ifdef CONFIG_TASKS_RUDE_RCU
199 void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
200 void synchronize_rcu_tasks_rude(void);
201 # endif
202 
203 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
204 void exit_tasks_rcu_start(void);
205 void exit_tasks_rcu_stop(void);
206 void exit_tasks_rcu_finish(void);
207 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
208 #define rcu_tasks_qs(t, preempt) do { } while (0)
209 #define rcu_note_voluntary_context_switch(t) do { } while (0)
210 #define call_rcu_tasks call_rcu
211 #define synchronize_rcu_tasks synchronize_rcu
exit_tasks_rcu_start(void)212 static inline void exit_tasks_rcu_start(void) { }
exit_tasks_rcu_stop(void)213 static inline void exit_tasks_rcu_stop(void) { }
exit_tasks_rcu_finish(void)214 static inline void exit_tasks_rcu_finish(void) { }
215 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
216 
217 /**
218  * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
219  *
220  * This macro resembles cond_resched(), except that it is defined to
221  * report potential quiescent states to RCU-tasks even if the cond_resched()
222  * machinery were to be shut off, as some advocate for PREEMPTION kernels.
223  */
224 #define cond_resched_tasks_rcu_qs() \
225 do { \
226 	rcu_tasks_qs(current, false); \
227 	cond_resched(); \
228 } while (0)
229 
230 /*
231  * Infrastructure to implement the synchronize_() primitives in
232  * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
233  */
234 
235 #if defined(CONFIG_TREE_RCU)
236 #include <linux/rcutree.h>
237 #elif defined(CONFIG_TINY_RCU)
238 #include <linux/rcutiny.h>
239 #else
240 #error "Unknown RCU implementation specified to kernel configuration"
241 #endif
242 
243 /*
244  * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
245  * are needed for dynamic initialization and destruction of rcu_head
246  * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
247  * dynamic initialization and destruction of statically allocated rcu_head
248  * structures.  However, rcu_head structures allocated dynamically in the
249  * heap don't need any initialization.
250  */
251 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
252 void init_rcu_head(struct rcu_head *head);
253 void destroy_rcu_head(struct rcu_head *head);
254 void init_rcu_head_on_stack(struct rcu_head *head);
255 void destroy_rcu_head_on_stack(struct rcu_head *head);
256 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
init_rcu_head(struct rcu_head * head)257 static inline void init_rcu_head(struct rcu_head *head) { }
destroy_rcu_head(struct rcu_head * head)258 static inline void destroy_rcu_head(struct rcu_head *head) { }
init_rcu_head_on_stack(struct rcu_head * head)259 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
destroy_rcu_head_on_stack(struct rcu_head * head)260 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
261 #endif	/* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
262 
263 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
264 bool rcu_lockdep_current_cpu_online(void);
265 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
rcu_lockdep_current_cpu_online(void)266 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
267 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
268 
269 extern struct lockdep_map rcu_lock_map;
270 extern struct lockdep_map rcu_bh_lock_map;
271 extern struct lockdep_map rcu_sched_lock_map;
272 extern struct lockdep_map rcu_callback_map;
273 
274 #ifdef CONFIG_DEBUG_LOCK_ALLOC
275 
rcu_lock_acquire(struct lockdep_map * map)276 static inline void rcu_lock_acquire(struct lockdep_map *map)
277 {
278 	lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
279 }
280 
rcu_lock_release(struct lockdep_map * map)281 static inline void rcu_lock_release(struct lockdep_map *map)
282 {
283 	lock_release(map, _THIS_IP_);
284 }
285 
286 int debug_lockdep_rcu_enabled(void);
287 int rcu_read_lock_held(void);
288 int rcu_read_lock_bh_held(void);
289 int rcu_read_lock_sched_held(void);
290 int rcu_read_lock_any_held(void);
291 
292 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
293 
294 # define rcu_lock_acquire(a)		do { } while (0)
295 # define rcu_lock_release(a)		do { } while (0)
296 
rcu_read_lock_held(void)297 static inline int rcu_read_lock_held(void)
298 {
299 	return 1;
300 }
301 
rcu_read_lock_bh_held(void)302 static inline int rcu_read_lock_bh_held(void)
303 {
304 	return 1;
305 }
306 
rcu_read_lock_sched_held(void)307 static inline int rcu_read_lock_sched_held(void)
308 {
309 	return !preemptible();
310 }
311 
rcu_read_lock_any_held(void)312 static inline int rcu_read_lock_any_held(void)
313 {
314 	return !preemptible();
315 }
316 
317 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
318 
319 #ifdef CONFIG_PROVE_RCU
320 
321 /**
322  * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
323  * @c: condition to check
324  * @s: informative message
325  *
326  * This checks debug_lockdep_rcu_enabled() before checking (c) to
327  * prevent early boot splats due to lockdep not yet being initialized,
328  * and rechecks it after checking (c) to prevent false-positive splats
329  * due to races with lockdep being disabled.  See commit 3066820034b5dd
330  * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail.
331  */
332 #define RCU_LOCKDEP_WARN(c, s)						\
333 	do {								\
334 		static bool __section(".data.unlikely") __warned;	\
335 		if (debug_lockdep_rcu_enabled() && (c) &&		\
336 		    debug_lockdep_rcu_enabled() && !__warned) {		\
337 			__warned = true;				\
338 			lockdep_rcu_suspicious(__FILE__, __LINE__, s);	\
339 		}							\
340 	} while (0)
341 
342 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
rcu_preempt_sleep_check(void)343 static inline void rcu_preempt_sleep_check(void)
344 {
345 	RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
346 			 "Illegal context switch in RCU read-side critical section");
347 }
348 #else /* #ifdef CONFIG_PROVE_RCU */
rcu_preempt_sleep_check(void)349 static inline void rcu_preempt_sleep_check(void) { }
350 #endif /* #else #ifdef CONFIG_PROVE_RCU */
351 
352 #define rcu_sleep_check()						\
353 	do {								\
354 		rcu_preempt_sleep_check();				\
355 		if (!IS_ENABLED(CONFIG_PREEMPT_RT))			\
356 		    RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map),	\
357 				 "Illegal context switch in RCU-bh read-side critical section"); \
358 		RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map),	\
359 				 "Illegal context switch in RCU-sched read-side critical section"); \
360 	} while (0)
361 
362 #else /* #ifdef CONFIG_PROVE_RCU */
363 
364 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
365 #define rcu_sleep_check() do { } while (0)
366 
367 #endif /* #else #ifdef CONFIG_PROVE_RCU */
368 
369 /*
370  * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
371  * and rcu_assign_pointer().  Some of these could be folded into their
372  * callers, but they are left separate in order to ease introduction of
373  * multiple pointers markings to match different RCU implementations
374  * (e.g., __srcu), should this make sense in the future.
375  */
376 
377 #ifdef __CHECKER__
378 #define rcu_check_sparse(p, space) \
379 	((void)(((typeof(*p) space *)p) == p))
380 #else /* #ifdef __CHECKER__ */
381 #define rcu_check_sparse(p, space)
382 #endif /* #else #ifdef __CHECKER__ */
383 
384 /**
385  * unrcu_pointer - mark a pointer as not being RCU protected
386  * @p: pointer needing to lose its __rcu property
387  *
388  * Converts @p from an __rcu pointer to a __kernel pointer.
389  * This allows an __rcu pointer to be used with xchg() and friends.
390  */
391 #define unrcu_pointer(p)						\
392 ({									\
393 	typeof(*p) *_________p1 = (typeof(*p) *__force)(p);		\
394 	rcu_check_sparse(p, __rcu);					\
395 	((typeof(*p) __force __kernel *)(_________p1)); 		\
396 })
397 
398 #define __rcu_access_pointer(p, space) \
399 ({ \
400 	typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
401 	rcu_check_sparse(p, space); \
402 	((typeof(*p) __force __kernel *)(_________p1)); \
403 })
404 #define __rcu_dereference_check(p, c, space) \
405 ({ \
406 	/* Dependency order vs. p above. */ \
407 	typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \
408 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
409 	rcu_check_sparse(p, space); \
410 	((typeof(*p) __force __kernel *)(________p1)); \
411 })
412 #define __rcu_dereference_protected(p, c, space) \
413 ({ \
414 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
415 	rcu_check_sparse(p, space); \
416 	((typeof(*p) __force __kernel *)(p)); \
417 })
418 #define rcu_dereference_raw(p) \
419 ({ \
420 	/* Dependency order vs. p above. */ \
421 	typeof(p) ________p1 = READ_ONCE(p); \
422 	((typeof(*p) __force __kernel *)(________p1)); \
423 })
424 
425 /**
426  * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
427  * @v: The value to statically initialize with.
428  */
429 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
430 
431 /**
432  * rcu_assign_pointer() - assign to RCU-protected pointer
433  * @p: pointer to assign to
434  * @v: value to assign (publish)
435  *
436  * Assigns the specified value to the specified RCU-protected
437  * pointer, ensuring that any concurrent RCU readers will see
438  * any prior initialization.
439  *
440  * Inserts memory barriers on architectures that require them
441  * (which is most of them), and also prevents the compiler from
442  * reordering the code that initializes the structure after the pointer
443  * assignment.  More importantly, this call documents which pointers
444  * will be dereferenced by RCU read-side code.
445  *
446  * In some special cases, you may use RCU_INIT_POINTER() instead
447  * of rcu_assign_pointer().  RCU_INIT_POINTER() is a bit faster due
448  * to the fact that it does not constrain either the CPU or the compiler.
449  * That said, using RCU_INIT_POINTER() when you should have used
450  * rcu_assign_pointer() is a very bad thing that results in
451  * impossible-to-diagnose memory corruption.  So please be careful.
452  * See the RCU_INIT_POINTER() comment header for details.
453  *
454  * Note that rcu_assign_pointer() evaluates each of its arguments only
455  * once, appearances notwithstanding.  One of the "extra" evaluations
456  * is in typeof() and the other visible only to sparse (__CHECKER__),
457  * neither of which actually execute the argument.  As with most cpp
458  * macros, this execute-arguments-only-once property is important, so
459  * please be careful when making changes to rcu_assign_pointer() and the
460  * other macros that it invokes.
461  */
462 #define rcu_assign_pointer(p, v)					      \
463 do {									      \
464 	uintptr_t _r_a_p__v = (uintptr_t)(v);				      \
465 	rcu_check_sparse(p, __rcu);					      \
466 									      \
467 	if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL)	      \
468 		WRITE_ONCE((p), (typeof(p))(_r_a_p__v));		      \
469 	else								      \
470 		smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
471 } while (0)
472 
473 /**
474  * rcu_replace_pointer() - replace an RCU pointer, returning its old value
475  * @rcu_ptr: RCU pointer, whose old value is returned
476  * @ptr: regular pointer
477  * @c: the lockdep conditions under which the dereference will take place
478  *
479  * Perform a replacement, where @rcu_ptr is an RCU-annotated
480  * pointer and @c is the lockdep argument that is passed to the
481  * rcu_dereference_protected() call used to read that pointer.  The old
482  * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
483  */
484 #define rcu_replace_pointer(rcu_ptr, ptr, c)				\
485 ({									\
486 	typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c));	\
487 	rcu_assign_pointer((rcu_ptr), (ptr));				\
488 	__tmp;								\
489 })
490 
491 /**
492  * rcu_access_pointer() - fetch RCU pointer with no dereferencing
493  * @p: The pointer to read
494  *
495  * Return the value of the specified RCU-protected pointer, but omit the
496  * lockdep checks for being in an RCU read-side critical section.  This is
497  * useful when the value of this pointer is accessed, but the pointer is
498  * not dereferenced, for example, when testing an RCU-protected pointer
499  * against NULL.  Although rcu_access_pointer() may also be used in cases
500  * where update-side locks prevent the value of the pointer from changing,
501  * you should instead use rcu_dereference_protected() for this use case.
502  *
503  * It is also permissible to use rcu_access_pointer() when read-side
504  * access to the pointer was removed at least one grace period ago, as
505  * is the case in the context of the RCU callback that is freeing up
506  * the data, or after a synchronize_rcu() returns.  This can be useful
507  * when tearing down multi-linked structures after a grace period
508  * has elapsed.
509  */
510 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
511 
512 /**
513  * rcu_dereference_check() - rcu_dereference with debug checking
514  * @p: The pointer to read, prior to dereferencing
515  * @c: The conditions under which the dereference will take place
516  *
517  * Do an rcu_dereference(), but check that the conditions under which the
518  * dereference will take place are correct.  Typically the conditions
519  * indicate the various locking conditions that should be held at that
520  * point.  The check should return true if the conditions are satisfied.
521  * An implicit check for being in an RCU read-side critical section
522  * (rcu_read_lock()) is included.
523  *
524  * For example:
525  *
526  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
527  *
528  * could be used to indicate to lockdep that foo->bar may only be dereferenced
529  * if either rcu_read_lock() is held, or that the lock required to replace
530  * the bar struct at foo->bar is held.
531  *
532  * Note that the list of conditions may also include indications of when a lock
533  * need not be held, for example during initialisation or destruction of the
534  * target struct:
535  *
536  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
537  *					      atomic_read(&foo->usage) == 0);
538  *
539  * Inserts memory barriers on architectures that require them
540  * (currently only the Alpha), prevents the compiler from refetching
541  * (and from merging fetches), and, more importantly, documents exactly
542  * which pointers are protected by RCU and checks that the pointer is
543  * annotated as __rcu.
544  */
545 #define rcu_dereference_check(p, c) \
546 	__rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
547 
548 /**
549  * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
550  * @p: The pointer to read, prior to dereferencing
551  * @c: The conditions under which the dereference will take place
552  *
553  * This is the RCU-bh counterpart to rcu_dereference_check().  However,
554  * please note that starting in v5.0 kernels, vanilla RCU grace periods
555  * wait for local_bh_disable() regions of code in addition to regions of
556  * code demarked by rcu_read_lock() and rcu_read_unlock().  This means
557  * that synchronize_rcu(), call_rcu, and friends all take not only
558  * rcu_read_lock() but also rcu_read_lock_bh() into account.
559  */
560 #define rcu_dereference_bh_check(p, c) \
561 	__rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
562 
563 /**
564  * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
565  * @p: The pointer to read, prior to dereferencing
566  * @c: The conditions under which the dereference will take place
567  *
568  * This is the RCU-sched counterpart to rcu_dereference_check().
569  * However, please note that starting in v5.0 kernels, vanilla RCU grace
570  * periods wait for preempt_disable() regions of code in addition to
571  * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
572  * This means that synchronize_rcu(), call_rcu, and friends all take not
573  * only rcu_read_lock() but also rcu_read_lock_sched() into account.
574  */
575 #define rcu_dereference_sched_check(p, c) \
576 	__rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
577 				__rcu)
578 
579 /*
580  * The tracing infrastructure traces RCU (we want that), but unfortunately
581  * some of the RCU checks causes tracing to lock up the system.
582  *
583  * The no-tracing version of rcu_dereference_raw() must not call
584  * rcu_read_lock_held().
585  */
586 #define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu)
587 
588 /**
589  * rcu_dereference_protected() - fetch RCU pointer when updates prevented
590  * @p: The pointer to read, prior to dereferencing
591  * @c: The conditions under which the dereference will take place
592  *
593  * Return the value of the specified RCU-protected pointer, but omit
594  * the READ_ONCE().  This is useful in cases where update-side locks
595  * prevent the value of the pointer from changing.  Please note that this
596  * primitive does *not* prevent the compiler from repeating this reference
597  * or combining it with other references, so it should not be used without
598  * protection of appropriate locks.
599  *
600  * This function is only for update-side use.  Using this function
601  * when protected only by rcu_read_lock() will result in infrequent
602  * but very ugly failures.
603  */
604 #define rcu_dereference_protected(p, c) \
605 	__rcu_dereference_protected((p), (c), __rcu)
606 
607 
608 /**
609  * rcu_dereference() - fetch RCU-protected pointer for dereferencing
610  * @p: The pointer to read, prior to dereferencing
611  *
612  * This is a simple wrapper around rcu_dereference_check().
613  */
614 #define rcu_dereference(p) rcu_dereference_check(p, 0)
615 
616 /**
617  * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
618  * @p: The pointer to read, prior to dereferencing
619  *
620  * Makes rcu_dereference_check() do the dirty work.
621  */
622 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
623 
624 /**
625  * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
626  * @p: The pointer to read, prior to dereferencing
627  *
628  * Makes rcu_dereference_check() do the dirty work.
629  */
630 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
631 
632 /**
633  * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
634  * @p: The pointer to hand off
635  *
636  * This is simply an identity function, but it documents where a pointer
637  * is handed off from RCU to some other synchronization mechanism, for
638  * example, reference counting or locking.  In C11, it would map to
639  * kill_dependency().  It could be used as follows::
640  *
641  *	rcu_read_lock();
642  *	p = rcu_dereference(gp);
643  *	long_lived = is_long_lived(p);
644  *	if (long_lived) {
645  *		if (!atomic_inc_not_zero(p->refcnt))
646  *			long_lived = false;
647  *		else
648  *			p = rcu_pointer_handoff(p);
649  *	}
650  *	rcu_read_unlock();
651  */
652 #define rcu_pointer_handoff(p) (p)
653 
654 /**
655  * rcu_read_lock() - mark the beginning of an RCU read-side critical section
656  *
657  * When synchronize_rcu() is invoked on one CPU while other CPUs
658  * are within RCU read-side critical sections, then the
659  * synchronize_rcu() is guaranteed to block until after all the other
660  * CPUs exit their critical sections.  Similarly, if call_rcu() is invoked
661  * on one CPU while other CPUs are within RCU read-side critical
662  * sections, invocation of the corresponding RCU callback is deferred
663  * until after the all the other CPUs exit their critical sections.
664  *
665  * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
666  * wait for regions of code with preemption disabled, including regions of
667  * code with interrupts or softirqs disabled.  In pre-v5.0 kernels, which
668  * define synchronize_sched(), only code enclosed within rcu_read_lock()
669  * and rcu_read_unlock() are guaranteed to be waited for.
670  *
671  * Note, however, that RCU callbacks are permitted to run concurrently
672  * with new RCU read-side critical sections.  One way that this can happen
673  * is via the following sequence of events: (1) CPU 0 enters an RCU
674  * read-side critical section, (2) CPU 1 invokes call_rcu() to register
675  * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
676  * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
677  * callback is invoked.  This is legal, because the RCU read-side critical
678  * section that was running concurrently with the call_rcu() (and which
679  * therefore might be referencing something that the corresponding RCU
680  * callback would free up) has completed before the corresponding
681  * RCU callback is invoked.
682  *
683  * RCU read-side critical sections may be nested.  Any deferred actions
684  * will be deferred until the outermost RCU read-side critical section
685  * completes.
686  *
687  * You can avoid reading and understanding the next paragraph by
688  * following this rule: don't put anything in an rcu_read_lock() RCU
689  * read-side critical section that would block in a !PREEMPTION kernel.
690  * But if you want the full story, read on!
691  *
692  * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
693  * it is illegal to block while in an RCU read-side critical section.
694  * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
695  * kernel builds, RCU read-side critical sections may be preempted,
696  * but explicit blocking is illegal.  Finally, in preemptible RCU
697  * implementations in real-time (with -rt patchset) kernel builds, RCU
698  * read-side critical sections may be preempted and they may also block, but
699  * only when acquiring spinlocks that are subject to priority inheritance.
700  */
rcu_read_lock(void)701 static __always_inline void rcu_read_lock(void)
702 {
703 	__rcu_read_lock();
704 	__acquire(RCU);
705 	rcu_lock_acquire(&rcu_lock_map);
706 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
707 			 "rcu_read_lock() used illegally while idle");
708 }
709 
710 /*
711  * So where is rcu_write_lock()?  It does not exist, as there is no
712  * way for writers to lock out RCU readers.  This is a feature, not
713  * a bug -- this property is what provides RCU's performance benefits.
714  * Of course, writers must coordinate with each other.  The normal
715  * spinlock primitives work well for this, but any other technique may be
716  * used as well.  RCU does not care how the writers keep out of each
717  * others' way, as long as they do so.
718  */
719 
720 /**
721  * rcu_read_unlock() - marks the end of an RCU read-side critical section.
722  *
723  * In almost all situations, rcu_read_unlock() is immune from deadlock.
724  * In recent kernels that have consolidated synchronize_sched() and
725  * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
726  * also extends to the scheduler's runqueue and priority-inheritance
727  * spinlocks, courtesy of the quiescent-state deferral that is carried
728  * out when rcu_read_unlock() is invoked with interrupts disabled.
729  *
730  * See rcu_read_lock() for more information.
731  */
rcu_read_unlock(void)732 static inline void rcu_read_unlock(void)
733 {
734 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
735 			 "rcu_read_unlock() used illegally while idle");
736 	__release(RCU);
737 	__rcu_read_unlock();
738 	rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
739 }
740 
741 /**
742  * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
743  *
744  * This is equivalent to rcu_read_lock(), but also disables softirqs.
745  * Note that anything else that disables softirqs can also serve as an RCU
746  * read-side critical section.  However, please note that this equivalence
747  * applies only to v5.0 and later.  Before v5.0, rcu_read_lock() and
748  * rcu_read_lock_bh() were unrelated.
749  *
750  * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
751  * must occur in the same context, for example, it is illegal to invoke
752  * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
753  * was invoked from some other task.
754  */
rcu_read_lock_bh(void)755 static inline void rcu_read_lock_bh(void)
756 {
757 	local_bh_disable();
758 	__acquire(RCU_BH);
759 	rcu_lock_acquire(&rcu_bh_lock_map);
760 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
761 			 "rcu_read_lock_bh() used illegally while idle");
762 }
763 
764 /**
765  * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
766  *
767  * See rcu_read_lock_bh() for more information.
768  */
rcu_read_unlock_bh(void)769 static inline void rcu_read_unlock_bh(void)
770 {
771 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
772 			 "rcu_read_unlock_bh() used illegally while idle");
773 	rcu_lock_release(&rcu_bh_lock_map);
774 	__release(RCU_BH);
775 	local_bh_enable();
776 }
777 
778 /**
779  * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
780  *
781  * This is equivalent to rcu_read_lock(), but also disables preemption.
782  * Read-side critical sections can also be introduced by anything else that
783  * disables preemption, including local_irq_disable() and friends.  However,
784  * please note that the equivalence to rcu_read_lock() applies only to
785  * v5.0 and later.  Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
786  * were unrelated.
787  *
788  * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
789  * must occur in the same context, for example, it is illegal to invoke
790  * rcu_read_unlock_sched() from process context if the matching
791  * rcu_read_lock_sched() was invoked from an NMI handler.
792  */
rcu_read_lock_sched(void)793 static inline void rcu_read_lock_sched(void)
794 {
795 	preempt_disable();
796 	__acquire(RCU_SCHED);
797 	rcu_lock_acquire(&rcu_sched_lock_map);
798 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
799 			 "rcu_read_lock_sched() used illegally while idle");
800 }
801 
802 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_lock_sched_notrace(void)803 static inline notrace void rcu_read_lock_sched_notrace(void)
804 {
805 	preempt_disable_notrace();
806 	__acquire(RCU_SCHED);
807 }
808 
809 /**
810  * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
811  *
812  * See rcu_read_lock_sched() for more information.
813  */
rcu_read_unlock_sched(void)814 static inline void rcu_read_unlock_sched(void)
815 {
816 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
817 			 "rcu_read_unlock_sched() used illegally while idle");
818 	rcu_lock_release(&rcu_sched_lock_map);
819 	__release(RCU_SCHED);
820 	preempt_enable();
821 }
822 
823 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_unlock_sched_notrace(void)824 static inline notrace void rcu_read_unlock_sched_notrace(void)
825 {
826 	__release(RCU_SCHED);
827 	preempt_enable_notrace();
828 }
829 
830 /**
831  * RCU_INIT_POINTER() - initialize an RCU protected pointer
832  * @p: The pointer to be initialized.
833  * @v: The value to initialized the pointer to.
834  *
835  * Initialize an RCU-protected pointer in special cases where readers
836  * do not need ordering constraints on the CPU or the compiler.  These
837  * special cases are:
838  *
839  * 1.	This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
840  * 2.	The caller has taken whatever steps are required to prevent
841  *	RCU readers from concurrently accessing this pointer *or*
842  * 3.	The referenced data structure has already been exposed to
843  *	readers either at compile time or via rcu_assign_pointer() *and*
844  *
845  *	a.	You have not made *any* reader-visible changes to
846  *		this structure since then *or*
847  *	b.	It is OK for readers accessing this structure from its
848  *		new location to see the old state of the structure.  (For
849  *		example, the changes were to statistical counters or to
850  *		other state where exact synchronization is not required.)
851  *
852  * Failure to follow these rules governing use of RCU_INIT_POINTER() will
853  * result in impossible-to-diagnose memory corruption.  As in the structures
854  * will look OK in crash dumps, but any concurrent RCU readers might
855  * see pre-initialized values of the referenced data structure.  So
856  * please be very careful how you use RCU_INIT_POINTER()!!!
857  *
858  * If you are creating an RCU-protected linked structure that is accessed
859  * by a single external-to-structure RCU-protected pointer, then you may
860  * use RCU_INIT_POINTER() to initialize the internal RCU-protected
861  * pointers, but you must use rcu_assign_pointer() to initialize the
862  * external-to-structure pointer *after* you have completely initialized
863  * the reader-accessible portions of the linked structure.
864  *
865  * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
866  * ordering guarantees for either the CPU or the compiler.
867  */
868 #define RCU_INIT_POINTER(p, v) \
869 	do { \
870 		rcu_check_sparse(p, __rcu); \
871 		WRITE_ONCE(p, RCU_INITIALIZER(v)); \
872 	} while (0)
873 
874 /**
875  * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
876  * @p: The pointer to be initialized.
877  * @v: The value to initialized the pointer to.
878  *
879  * GCC-style initialization for an RCU-protected pointer in a structure field.
880  */
881 #define RCU_POINTER_INITIALIZER(p, v) \
882 		.p = RCU_INITIALIZER(v)
883 
884 /*
885  * Does the specified offset indicate that the corresponding rcu_head
886  * structure can be handled by kvfree_rcu()?
887  */
888 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
889 
890 /**
891  * kfree_rcu() - kfree an object after a grace period.
892  * @ptr: pointer to kfree for both single- and double-argument invocations.
893  * @rhf: the name of the struct rcu_head within the type of @ptr,
894  *       but only for double-argument invocations.
895  *
896  * Many rcu callbacks functions just call kfree() on the base structure.
897  * These functions are trivial, but their size adds up, and furthermore
898  * when they are used in a kernel module, that module must invoke the
899  * high-latency rcu_barrier() function at module-unload time.
900  *
901  * The kfree_rcu() function handles this issue.  Rather than encoding a
902  * function address in the embedded rcu_head structure, kfree_rcu() instead
903  * encodes the offset of the rcu_head structure within the base structure.
904  * Because the functions are not allowed in the low-order 4096 bytes of
905  * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
906  * If the offset is larger than 4095 bytes, a compile-time error will
907  * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
908  * either fall back to use of call_rcu() or rearrange the structure to
909  * position the rcu_head structure into the first 4096 bytes.
910  *
911  * Note that the allowable offset might decrease in the future, for example,
912  * to allow something like kmem_cache_free_rcu().
913  *
914  * The BUILD_BUG_ON check must not involve any function calls, hence the
915  * checks are done in macros here.
916  */
917 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf)
918 
919 /**
920  * kvfree_rcu() - kvfree an object after a grace period.
921  *
922  * This macro consists of one or two arguments and it is
923  * based on whether an object is head-less or not. If it
924  * has a head then a semantic stays the same as it used
925  * to be before:
926  *
927  *     kvfree_rcu(ptr, rhf);
928  *
929  * where @ptr is a pointer to kvfree(), @rhf is the name
930  * of the rcu_head structure within the type of @ptr.
931  *
932  * When it comes to head-less variant, only one argument
933  * is passed and that is just a pointer which has to be
934  * freed after a grace period. Therefore the semantic is
935  *
936  *     kvfree_rcu(ptr);
937  *
938  * where @ptr is a pointer to kvfree().
939  *
940  * Please note, head-less way of freeing is permitted to
941  * use from a context that has to follow might_sleep()
942  * annotation. Otherwise, please switch and embed the
943  * rcu_head structure within the type of @ptr.
944  */
945 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__,		\
946 	kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__)
947 
948 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME
949 #define kvfree_rcu_arg_2(ptr, rhf)					\
950 do {									\
951 	typeof (ptr) ___p = (ptr);					\
952 									\
953 	if (___p) {									\
954 		BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf)));	\
955 		kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long)		\
956 			(offsetof(typeof(*(ptr)), rhf)));				\
957 	}										\
958 } while (0)
959 
960 #define kvfree_rcu_arg_1(ptr)					\
961 do {								\
962 	typeof(ptr) ___p = (ptr);				\
963 								\
964 	if (___p)						\
965 		kvfree_call_rcu(NULL, (rcu_callback_t) (___p));	\
966 } while (0)
967 
968 /*
969  * Place this after a lock-acquisition primitive to guarantee that
970  * an UNLOCK+LOCK pair acts as a full barrier.  This guarantee applies
971  * if the UNLOCK and LOCK are executed by the same CPU or if the
972  * UNLOCK and LOCK operate on the same lock variable.
973  */
974 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
975 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
976 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
977 #define smp_mb__after_unlock_lock()	do { } while (0)
978 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
979 
980 
981 /* Has the specified rcu_head structure been handed to call_rcu()? */
982 
983 /**
984  * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
985  * @rhp: The rcu_head structure to initialize.
986  *
987  * If you intend to invoke rcu_head_after_call_rcu() to test whether a
988  * given rcu_head structure has already been passed to call_rcu(), then
989  * you must also invoke this rcu_head_init() function on it just after
990  * allocating that structure.  Calls to this function must not race with
991  * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
992  */
rcu_head_init(struct rcu_head * rhp)993 static inline void rcu_head_init(struct rcu_head *rhp)
994 {
995 	rhp->func = (rcu_callback_t)~0L;
996 }
997 
998 /**
999  * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
1000  * @rhp: The rcu_head structure to test.
1001  * @f: The function passed to call_rcu() along with @rhp.
1002  *
1003  * Returns @true if the @rhp has been passed to call_rcu() with @func,
1004  * and @false otherwise.  Emits a warning in any other case, including
1005  * the case where @rhp has already been invoked after a grace period.
1006  * Calls to this function must not race with callback invocation.  One way
1007  * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
1008  * in an RCU read-side critical section that includes a read-side fetch
1009  * of the pointer to the structure containing @rhp.
1010  */
1011 static inline bool
rcu_head_after_call_rcu(struct rcu_head * rhp,rcu_callback_t f)1012 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1013 {
1014 	rcu_callback_t func = READ_ONCE(rhp->func);
1015 
1016 	if (func == f)
1017 		return true;
1018 	WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1019 	return false;
1020 }
1021 
1022 /* kernel/ksysfs.c definitions */
1023 extern int rcu_expedited;
1024 extern int rcu_normal;
1025 
1026 #endif /* __LINUX_RCUPDATE_H */
1027