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