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 <linux/cleanup.h>
31#include <asm/processor.h>
32#include <linux/cpumask.h>
33#include <linux/context_tracking_irq.h>
34
35#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
36#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
37#define ulong2long(a) (*(long *)(&(a)))
38#define USHORT_CMP_GE(a, b) (USHRT_MAX / 2 >= (unsigned short)((a) - (b)))
39#define USHORT_CMP_LT(a, b) (USHRT_MAX / 2 < (unsigned short)((a) - (b)))
40
41/* Exported common interfaces */
42void call_rcu(struct rcu_head *head, rcu_callback_t func);
43void rcu_barrier_tasks(void);
44void rcu_barrier_tasks_rude(void);
45void synchronize_rcu(void);
46
47struct rcu_gp_oldstate;
48unsigned long get_completed_synchronize_rcu(void);
49void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp);
50
51// Maximum number of unsigned long values corresponding to
52// not-yet-completed RCU grace periods.
53#define NUM_ACTIVE_RCU_POLL_OLDSTATE 2
54
55/**
56 * same_state_synchronize_rcu - Are two old-state values identical?
57 * @oldstate1: First old-state value.
58 * @oldstate2: Second old-state value.
59 *
60 * The two old-state values must have been obtained from either
61 * get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or
62 * get_completed_synchronize_rcu(). Returns @true if the two values are
63 * identical and @false otherwise. This allows structures whose lifetimes
64 * are tracked by old-state values to push these values to a list header,
65 * allowing those structures to be slightly smaller.
66 */
67static inline bool same_state_synchronize_rcu(unsigned long oldstate1, unsigned long oldstate2)
68{
69 return oldstate1 == oldstate2;
70}
71
72#ifdef CONFIG_PREEMPT_RCU
73
74void __rcu_read_lock(void);
75void __rcu_read_unlock(void);
76
77/*
78 * Defined as a macro as it is a very low level header included from
79 * areas that don't even know about current. This gives the rcu_read_lock()
80 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
81 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
82 */
83#define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
84
85#else /* #ifdef CONFIG_PREEMPT_RCU */
86
87#ifdef CONFIG_TINY_RCU
88#define rcu_read_unlock_strict() do { } while (0)
89#else
90void rcu_read_unlock_strict(void);
91#endif
92
93static inline void __rcu_read_lock(void)
94{
95 preempt_disable();
96}
97
98static inline void __rcu_read_unlock(void)
99{
100 preempt_enable();
101 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
102 rcu_read_unlock_strict();
103}
104
105static inline int rcu_preempt_depth(void)
106{
107 return 0;
108}
109
110#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
111
112#ifdef CONFIG_RCU_LAZY
113void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func);
114#else
115static inline void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func)
116{
117 call_rcu(head, func);
118}
119#endif
120
121/* Internal to kernel */
122void rcu_init(void);
123extern int rcu_scheduler_active;
124void rcu_sched_clock_irq(int user);
125
126#ifdef CONFIG_TASKS_RCU_GENERIC
127void rcu_init_tasks_generic(void);
128#else
129static inline void rcu_init_tasks_generic(void) { }
130#endif
131
132#ifdef CONFIG_RCU_STALL_COMMON
133void rcu_sysrq_start(void);
134void rcu_sysrq_end(void);
135#else /* #ifdef CONFIG_RCU_STALL_COMMON */
136static inline void rcu_sysrq_start(void) { }
137static inline void rcu_sysrq_end(void) { }
138#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
139
140#if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK))
141void rcu_irq_work_resched(void);
142#else
143static inline void rcu_irq_work_resched(void) { }
144#endif
145
146#ifdef CONFIG_RCU_NOCB_CPU
147void rcu_init_nohz(void);
148int rcu_nocb_cpu_offload(int cpu);
149int rcu_nocb_cpu_deoffload(int cpu);
150void rcu_nocb_flush_deferred_wakeup(void);
151#else /* #ifdef CONFIG_RCU_NOCB_CPU */
152static inline void rcu_init_nohz(void) { }
153static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
154static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
155static inline void rcu_nocb_flush_deferred_wakeup(void) { }
156#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
157
158/*
159 * Note a quasi-voluntary context switch for RCU-tasks's benefit.
160 * This is a macro rather than an inline function to avoid #include hell.
161 */
162#ifdef CONFIG_TASKS_RCU_GENERIC
163
164# ifdef CONFIG_TASKS_RCU
165# define rcu_tasks_classic_qs(t, preempt) \
166 do { \
167 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \
168 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
169 } while (0)
170void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
171void synchronize_rcu_tasks(void);
172# else
173# define rcu_tasks_classic_qs(t, preempt) do { } while (0)
174# define call_rcu_tasks call_rcu
175# define synchronize_rcu_tasks synchronize_rcu
176# endif
177
178# ifdef CONFIG_TASKS_TRACE_RCU
179// Bits for ->trc_reader_special.b.need_qs field.
180#define TRC_NEED_QS 0x1 // Task needs a quiescent state.
181#define TRC_NEED_QS_CHECKED 0x2 // Task has been checked for needing quiescent state.
182
183u8 rcu_trc_cmpxchg_need_qs(struct task_struct *t, u8 old, u8 new);
184void rcu_tasks_trace_qs_blkd(struct task_struct *t);
185
186# define rcu_tasks_trace_qs(t) \
187 do { \
188 int ___rttq_nesting = READ_ONCE((t)->trc_reader_nesting); \
189 \
190 if (likely(!READ_ONCE((t)->trc_reader_special.b.need_qs)) && \
191 likely(!___rttq_nesting)) { \
192 rcu_trc_cmpxchg_need_qs((t), 0, TRC_NEED_QS_CHECKED); \
193 } else if (___rttq_nesting && ___rttq_nesting != INT_MIN && \
194 !READ_ONCE((t)->trc_reader_special.b.blocked)) { \
195 rcu_tasks_trace_qs_blkd(t); \
196 } \
197 } while (0)
198# else
199# define rcu_tasks_trace_qs(t) do { } while (0)
200# endif
201
202#define rcu_tasks_qs(t, preempt) \
203do { \
204 rcu_tasks_classic_qs((t), (preempt)); \
205 rcu_tasks_trace_qs(t); \
206} while (0)
207
208# ifdef CONFIG_TASKS_RUDE_RCU
209void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
210void synchronize_rcu_tasks_rude(void);
211# endif
212
213#define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
214void exit_tasks_rcu_start(void);
215void exit_tasks_rcu_stop(void);
216void exit_tasks_rcu_finish(void);
217#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
218#define rcu_tasks_classic_qs(t, preempt) do { } while (0)
219#define rcu_tasks_qs(t, preempt) do { } while (0)
220#define rcu_note_voluntary_context_switch(t) do { } while (0)
221#define call_rcu_tasks call_rcu
222#define synchronize_rcu_tasks synchronize_rcu
223static inline void exit_tasks_rcu_start(void) { }
224static inline void exit_tasks_rcu_stop(void) { }
225static inline void exit_tasks_rcu_finish(void) { }
226#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
227
228/**
229 * rcu_trace_implies_rcu_gp - does an RCU Tasks Trace grace period imply an RCU grace period?
230 *
231 * As an accident of implementation, an RCU Tasks Trace grace period also
232 * acts as an RCU grace period. However, this could change at any time.
233 * Code relying on this accident must call this function to verify that
234 * this accident is still happening.
235 *
236 * You have been warned!
237 */
238static inline bool rcu_trace_implies_rcu_gp(void) { return true; }
239
240/**
241 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
242 *
243 * This macro resembles cond_resched(), except that it is defined to
244 * report potential quiescent states to RCU-tasks even if the cond_resched()
245 * machinery were to be shut off, as some advocate for PREEMPTION kernels.
246 */
247#define cond_resched_tasks_rcu_qs() \
248do { \
249 rcu_tasks_qs(current, false); \
250 cond_resched(); \
251} while (0)
252
253/*
254 * Infrastructure to implement the synchronize_() primitives in
255 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
256 */
257
258#if defined(CONFIG_TREE_RCU)
259#include <linux/rcutree.h>
260#elif defined(CONFIG_TINY_RCU)
261#include <linux/rcutiny.h>
262#else
263#error "Unknown RCU implementation specified to kernel configuration"
264#endif
265
266/*
267 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
268 * are needed for dynamic initialization and destruction of rcu_head
269 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
270 * dynamic initialization and destruction of statically allocated rcu_head
271 * structures. However, rcu_head structures allocated dynamically in the
272 * heap don't need any initialization.
273 */
274#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
275void init_rcu_head(struct rcu_head *head);
276void destroy_rcu_head(struct rcu_head *head);
277void init_rcu_head_on_stack(struct rcu_head *head);
278void destroy_rcu_head_on_stack(struct rcu_head *head);
279#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
280static inline void init_rcu_head(struct rcu_head *head) { }
281static inline void destroy_rcu_head(struct rcu_head *head) { }
282static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
283static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
284#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
285
286#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
287bool rcu_lockdep_current_cpu_online(void);
288#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
289static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
290#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
291
292extern struct lockdep_map rcu_lock_map;
293extern struct lockdep_map rcu_bh_lock_map;
294extern struct lockdep_map rcu_sched_lock_map;
295extern struct lockdep_map rcu_callback_map;
296
297#ifdef CONFIG_DEBUG_LOCK_ALLOC
298
299static inline void rcu_lock_acquire(struct lockdep_map *map)
300{
301 lock_acquire(lock: map, subclass: 0, trylock: 0, read: 2, check: 0, NULL, _THIS_IP_);
302}
303
304static inline void rcu_lock_release(struct lockdep_map *map)
305{
306 lock_release(lock: map, _THIS_IP_);
307}
308
309int debug_lockdep_rcu_enabled(void);
310int rcu_read_lock_held(void);
311int rcu_read_lock_bh_held(void);
312int rcu_read_lock_sched_held(void);
313int rcu_read_lock_any_held(void);
314
315#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
316
317# define rcu_lock_acquire(a) do { } while (0)
318# define rcu_lock_release(a) do { } while (0)
319
320static inline int rcu_read_lock_held(void)
321{
322 return 1;
323}
324
325static inline int rcu_read_lock_bh_held(void)
326{
327 return 1;
328}
329
330static inline int rcu_read_lock_sched_held(void)
331{
332 return !preemptible();
333}
334
335static inline int rcu_read_lock_any_held(void)
336{
337 return !preemptible();
338}
339
340static inline int debug_lockdep_rcu_enabled(void)
341{
342 return 0;
343}
344
345#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
346
347#ifdef CONFIG_PROVE_RCU
348
349/**
350 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
351 * @c: condition to check
352 * @s: informative message
353 *
354 * This checks debug_lockdep_rcu_enabled() before checking (c) to
355 * prevent early boot splats due to lockdep not yet being initialized,
356 * and rechecks it after checking (c) to prevent false-positive splats
357 * due to races with lockdep being disabled. See commit 3066820034b5dd
358 * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail.
359 */
360#define RCU_LOCKDEP_WARN(c, s) \
361 do { \
362 static bool __section(".data.unlikely") __warned; \
363 if (debug_lockdep_rcu_enabled() && (c) && \
364 debug_lockdep_rcu_enabled() && !__warned) { \
365 __warned = true; \
366 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
367 } \
368 } while (0)
369
370#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
371static inline void rcu_preempt_sleep_check(void)
372{
373 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
374 "Illegal context switch in RCU read-side critical section");
375}
376#else /* #ifdef CONFIG_PROVE_RCU */
377static inline void rcu_preempt_sleep_check(void) { }
378#endif /* #else #ifdef CONFIG_PROVE_RCU */
379
380#define rcu_sleep_check() \
381 do { \
382 rcu_preempt_sleep_check(); \
383 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \
384 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
385 "Illegal context switch in RCU-bh read-side critical section"); \
386 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
387 "Illegal context switch in RCU-sched read-side critical section"); \
388 } while (0)
389
390#else /* #ifdef CONFIG_PROVE_RCU */
391
392#define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
393#define rcu_sleep_check() do { } while (0)
394
395#endif /* #else #ifdef CONFIG_PROVE_RCU */
396
397/*
398 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
399 * and rcu_assign_pointer(). Some of these could be folded into their
400 * callers, but they are left separate in order to ease introduction of
401 * multiple pointers markings to match different RCU implementations
402 * (e.g., __srcu), should this make sense in the future.
403 */
404
405#ifdef __CHECKER__
406#define rcu_check_sparse(p, space) \
407 ((void)(((typeof(*p) space *)p) == p))
408#else /* #ifdef __CHECKER__ */
409#define rcu_check_sparse(p, space)
410#endif /* #else #ifdef __CHECKER__ */
411
412#define __unrcu_pointer(p, local) \
413({ \
414 typeof(*p) *local = (typeof(*p) *__force)(p); \
415 rcu_check_sparse(p, __rcu); \
416 ((typeof(*p) __force __kernel *)(local)); \
417})
418/**
419 * unrcu_pointer - mark a pointer as not being RCU protected
420 * @p: pointer needing to lose its __rcu property
421 *
422 * Converts @p from an __rcu pointer to a __kernel pointer.
423 * This allows an __rcu pointer to be used with xchg() and friends.
424 */
425#define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu))
426
427#define __rcu_access_pointer(p, local, space) \
428({ \
429 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
430 rcu_check_sparse(p, space); \
431 ((typeof(*p) __force __kernel *)(local)); \
432})
433#define __rcu_dereference_check(p, local, c, space) \
434({ \
435 /* Dependency order vs. p above. */ \
436 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
437 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
438 rcu_check_sparse(p, space); \
439 ((typeof(*p) __force __kernel *)(local)); \
440})
441#define __rcu_dereference_protected(p, local, c, space) \
442({ \
443 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
444 rcu_check_sparse(p, space); \
445 ((typeof(*p) __force __kernel *)(p)); \
446})
447#define __rcu_dereference_raw(p, local) \
448({ \
449 /* Dependency order vs. p above. */ \
450 typeof(p) local = READ_ONCE(p); \
451 ((typeof(*p) __force __kernel *)(local)); \
452})
453#define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu))
454
455/**
456 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
457 * @v: The value to statically initialize with.
458 */
459#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
460
461/**
462 * rcu_assign_pointer() - assign to RCU-protected pointer
463 * @p: pointer to assign to
464 * @v: value to assign (publish)
465 *
466 * Assigns the specified value to the specified RCU-protected
467 * pointer, ensuring that any concurrent RCU readers will see
468 * any prior initialization.
469 *
470 * Inserts memory barriers on architectures that require them
471 * (which is most of them), and also prevents the compiler from
472 * reordering the code that initializes the structure after the pointer
473 * assignment. More importantly, this call documents which pointers
474 * will be dereferenced by RCU read-side code.
475 *
476 * In some special cases, you may use RCU_INIT_POINTER() instead
477 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
478 * to the fact that it does not constrain either the CPU or the compiler.
479 * That said, using RCU_INIT_POINTER() when you should have used
480 * rcu_assign_pointer() is a very bad thing that results in
481 * impossible-to-diagnose memory corruption. So please be careful.
482 * See the RCU_INIT_POINTER() comment header for details.
483 *
484 * Note that rcu_assign_pointer() evaluates each of its arguments only
485 * once, appearances notwithstanding. One of the "extra" evaluations
486 * is in typeof() and the other visible only to sparse (__CHECKER__),
487 * neither of which actually execute the argument. As with most cpp
488 * macros, this execute-arguments-only-once property is important, so
489 * please be careful when making changes to rcu_assign_pointer() and the
490 * other macros that it invokes.
491 */
492#define rcu_assign_pointer(p, v) \
493do { \
494 uintptr_t _r_a_p__v = (uintptr_t)(v); \
495 rcu_check_sparse(p, __rcu); \
496 \
497 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
498 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
499 else \
500 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
501} while (0)
502
503/**
504 * rcu_replace_pointer() - replace an RCU pointer, returning its old value
505 * @rcu_ptr: RCU pointer, whose old value is returned
506 * @ptr: regular pointer
507 * @c: the lockdep conditions under which the dereference will take place
508 *
509 * Perform a replacement, where @rcu_ptr is an RCU-annotated
510 * pointer and @c is the lockdep argument that is passed to the
511 * rcu_dereference_protected() call used to read that pointer. The old
512 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
513 */
514#define rcu_replace_pointer(rcu_ptr, ptr, c) \
515({ \
516 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
517 rcu_assign_pointer((rcu_ptr), (ptr)); \
518 __tmp; \
519})
520
521/**
522 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
523 * @p: The pointer to read
524 *
525 * Return the value of the specified RCU-protected pointer, but omit the
526 * lockdep checks for being in an RCU read-side critical section. This is
527 * useful when the value of this pointer is accessed, but the pointer is
528 * not dereferenced, for example, when testing an RCU-protected pointer
529 * against NULL. Although rcu_access_pointer() may also be used in cases
530 * where update-side locks prevent the value of the pointer from changing,
531 * you should instead use rcu_dereference_protected() for this use case.
532 * Within an RCU read-side critical section, there is little reason to
533 * use rcu_access_pointer().
534 *
535 * It is usually best to test the rcu_access_pointer() return value
536 * directly in order to avoid accidental dereferences being introduced
537 * by later inattentive changes. In other words, assigning the
538 * rcu_access_pointer() return value to a local variable results in an
539 * accident waiting to happen.
540 *
541 * It is also permissible to use rcu_access_pointer() when read-side
542 * access to the pointer was removed at least one grace period ago, as is
543 * the case in the context of the RCU callback that is freeing up the data,
544 * or after a synchronize_rcu() returns. This can be useful when tearing
545 * down multi-linked structures after a grace period has elapsed. However,
546 * rcu_dereference_protected() is normally preferred for this use case.
547 */
548#define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu)
549
550/**
551 * rcu_dereference_check() - rcu_dereference 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 * Do an rcu_dereference(), but check that the conditions under which the
556 * dereference will take place are correct. Typically the conditions
557 * indicate the various locking conditions that should be held at that
558 * point. The check should return true if the conditions are satisfied.
559 * An implicit check for being in an RCU read-side critical section
560 * (rcu_read_lock()) is included.
561 *
562 * For example:
563 *
564 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
565 *
566 * could be used to indicate to lockdep that foo->bar may only be dereferenced
567 * if either rcu_read_lock() is held, or that the lock required to replace
568 * the bar struct at foo->bar is held.
569 *
570 * Note that the list of conditions may also include indications of when a lock
571 * need not be held, for example during initialisation or destruction of the
572 * target struct:
573 *
574 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
575 * atomic_read(&foo->usage) == 0);
576 *
577 * Inserts memory barriers on architectures that require them
578 * (currently only the Alpha), prevents the compiler from refetching
579 * (and from merging fetches), and, more importantly, documents exactly
580 * which pointers are protected by RCU and checks that the pointer is
581 * annotated as __rcu.
582 */
583#define rcu_dereference_check(p, c) \
584 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
585 (c) || rcu_read_lock_held(), __rcu)
586
587/**
588 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
589 * @p: The pointer to read, prior to dereferencing
590 * @c: The conditions under which the dereference will take place
591 *
592 * This is the RCU-bh counterpart to rcu_dereference_check(). However,
593 * please note that starting in v5.0 kernels, vanilla RCU grace periods
594 * wait for local_bh_disable() regions of code in addition to regions of
595 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means
596 * that synchronize_rcu(), call_rcu, and friends all take not only
597 * rcu_read_lock() but also rcu_read_lock_bh() into account.
598 */
599#define rcu_dereference_bh_check(p, c) \
600 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
601 (c) || rcu_read_lock_bh_held(), __rcu)
602
603/**
604 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
605 * @p: The pointer to read, prior to dereferencing
606 * @c: The conditions under which the dereference will take place
607 *
608 * This is the RCU-sched counterpart to rcu_dereference_check().
609 * However, please note that starting in v5.0 kernels, vanilla RCU grace
610 * periods wait for preempt_disable() regions of code in addition to
611 * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
612 * This means that synchronize_rcu(), call_rcu, and friends all take not
613 * only rcu_read_lock() but also rcu_read_lock_sched() into account.
614 */
615#define rcu_dereference_sched_check(p, c) \
616 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
617 (c) || rcu_read_lock_sched_held(), \
618 __rcu)
619
620/*
621 * The tracing infrastructure traces RCU (we want that), but unfortunately
622 * some of the RCU checks causes tracing to lock up the system.
623 *
624 * The no-tracing version of rcu_dereference_raw() must not call
625 * rcu_read_lock_held().
626 */
627#define rcu_dereference_raw_check(p) \
628 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu)
629
630/**
631 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
632 * @p: The pointer to read, prior to dereferencing
633 * @c: The conditions under which the dereference will take place
634 *
635 * Return the value of the specified RCU-protected pointer, but omit
636 * the READ_ONCE(). This is useful in cases where update-side locks
637 * prevent the value of the pointer from changing. Please note that this
638 * primitive does *not* prevent the compiler from repeating this reference
639 * or combining it with other references, so it should not be used without
640 * protection of appropriate locks.
641 *
642 * This function is only for update-side use. Using this function
643 * when protected only by rcu_read_lock() will result in infrequent
644 * but very ugly failures.
645 */
646#define rcu_dereference_protected(p, c) \
647 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu)
648
649
650/**
651 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
652 * @p: The pointer to read, prior to dereferencing
653 *
654 * This is a simple wrapper around rcu_dereference_check().
655 */
656#define rcu_dereference(p) rcu_dereference_check(p, 0)
657
658/**
659 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
660 * @p: The pointer to read, prior to dereferencing
661 *
662 * Makes rcu_dereference_check() do the dirty work.
663 */
664#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
665
666/**
667 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
668 * @p: The pointer to read, prior to dereferencing
669 *
670 * Makes rcu_dereference_check() do the dirty work.
671 */
672#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
673
674/**
675 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
676 * @p: The pointer to hand off
677 *
678 * This is simply an identity function, but it documents where a pointer
679 * is handed off from RCU to some other synchronization mechanism, for
680 * example, reference counting or locking. In C11, it would map to
681 * kill_dependency(). It could be used as follows::
682 *
683 * rcu_read_lock();
684 * p = rcu_dereference(gp);
685 * long_lived = is_long_lived(p);
686 * if (long_lived) {
687 * if (!atomic_inc_not_zero(p->refcnt))
688 * long_lived = false;
689 * else
690 * p = rcu_pointer_handoff(p);
691 * }
692 * rcu_read_unlock();
693 */
694#define rcu_pointer_handoff(p) (p)
695
696/**
697 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
698 *
699 * When synchronize_rcu() is invoked on one CPU while other CPUs
700 * are within RCU read-side critical sections, then the
701 * synchronize_rcu() is guaranteed to block until after all the other
702 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
703 * on one CPU while other CPUs are within RCU read-side critical
704 * sections, invocation of the corresponding RCU callback is deferred
705 * until after the all the other CPUs exit their critical sections.
706 *
707 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
708 * wait for regions of code with preemption disabled, including regions of
709 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which
710 * define synchronize_sched(), only code enclosed within rcu_read_lock()
711 * and rcu_read_unlock() are guaranteed to be waited for.
712 *
713 * Note, however, that RCU callbacks are permitted to run concurrently
714 * with new RCU read-side critical sections. One way that this can happen
715 * is via the following sequence of events: (1) CPU 0 enters an RCU
716 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
717 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
718 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
719 * callback is invoked. This is legal, because the RCU read-side critical
720 * section that was running concurrently with the call_rcu() (and which
721 * therefore might be referencing something that the corresponding RCU
722 * callback would free up) has completed before the corresponding
723 * RCU callback is invoked.
724 *
725 * RCU read-side critical sections may be nested. Any deferred actions
726 * will be deferred until the outermost RCU read-side critical section
727 * completes.
728 *
729 * You can avoid reading and understanding the next paragraph by
730 * following this rule: don't put anything in an rcu_read_lock() RCU
731 * read-side critical section that would block in a !PREEMPTION kernel.
732 * But if you want the full story, read on!
733 *
734 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
735 * it is illegal to block while in an RCU read-side critical section.
736 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
737 * kernel builds, RCU read-side critical sections may be preempted,
738 * but explicit blocking is illegal. Finally, in preemptible RCU
739 * implementations in real-time (with -rt patchset) kernel builds, RCU
740 * read-side critical sections may be preempted and they may also block, but
741 * only when acquiring spinlocks that are subject to priority inheritance.
742 */
743static __always_inline void rcu_read_lock(void)
744{
745 __rcu_read_lock();
746 __acquire(RCU);
747 rcu_lock_acquire(map: &rcu_lock_map);
748 RCU_LOCKDEP_WARN(!rcu_is_watching(),
749 "rcu_read_lock() used illegally while idle");
750}
751
752/*
753 * So where is rcu_write_lock()? It does not exist, as there is no
754 * way for writers to lock out RCU readers. This is a feature, not
755 * a bug -- this property is what provides RCU's performance benefits.
756 * Of course, writers must coordinate with each other. The normal
757 * spinlock primitives work well for this, but any other technique may be
758 * used as well. RCU does not care how the writers keep out of each
759 * others' way, as long as they do so.
760 */
761
762/**
763 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
764 *
765 * In almost all situations, rcu_read_unlock() is immune from deadlock.
766 * In recent kernels that have consolidated synchronize_sched() and
767 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
768 * also extends to the scheduler's runqueue and priority-inheritance
769 * spinlocks, courtesy of the quiescent-state deferral that is carried
770 * out when rcu_read_unlock() is invoked with interrupts disabled.
771 *
772 * See rcu_read_lock() for more information.
773 */
774static inline void rcu_read_unlock(void)
775{
776 RCU_LOCKDEP_WARN(!rcu_is_watching(),
777 "rcu_read_unlock() used illegally while idle");
778 __release(RCU);
779 __rcu_read_unlock();
780 rcu_lock_release(map: &rcu_lock_map); /* Keep acq info for rls diags. */
781}
782
783/**
784 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
785 *
786 * This is equivalent to rcu_read_lock(), but also disables softirqs.
787 * Note that anything else that disables softirqs can also serve as an RCU
788 * read-side critical section. However, please note that this equivalence
789 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and
790 * rcu_read_lock_bh() were unrelated.
791 *
792 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
793 * must occur in the same context, for example, it is illegal to invoke
794 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
795 * was invoked from some other task.
796 */
797static inline void rcu_read_lock_bh(void)
798{
799 local_bh_disable();
800 __acquire(RCU_BH);
801 rcu_lock_acquire(map: &rcu_bh_lock_map);
802 RCU_LOCKDEP_WARN(!rcu_is_watching(),
803 "rcu_read_lock_bh() used illegally while idle");
804}
805
806/**
807 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
808 *
809 * See rcu_read_lock_bh() for more information.
810 */
811static inline void rcu_read_unlock_bh(void)
812{
813 RCU_LOCKDEP_WARN(!rcu_is_watching(),
814 "rcu_read_unlock_bh() used illegally while idle");
815 rcu_lock_release(map: &rcu_bh_lock_map);
816 __release(RCU_BH);
817 local_bh_enable();
818}
819
820/**
821 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
822 *
823 * This is equivalent to rcu_read_lock(), but also disables preemption.
824 * Read-side critical sections can also be introduced by anything else that
825 * disables preemption, including local_irq_disable() and friends. However,
826 * please note that the equivalence to rcu_read_lock() applies only to
827 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
828 * were unrelated.
829 *
830 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
831 * must occur in the same context, for example, it is illegal to invoke
832 * rcu_read_unlock_sched() from process context if the matching
833 * rcu_read_lock_sched() was invoked from an NMI handler.
834 */
835static inline void rcu_read_lock_sched(void)
836{
837 preempt_disable();
838 __acquire(RCU_SCHED);
839 rcu_lock_acquire(map: &rcu_sched_lock_map);
840 RCU_LOCKDEP_WARN(!rcu_is_watching(),
841 "rcu_read_lock_sched() used illegally while idle");
842}
843
844/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
845static inline notrace void rcu_read_lock_sched_notrace(void)
846{
847 preempt_disable_notrace();
848 __acquire(RCU_SCHED);
849}
850
851/**
852 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
853 *
854 * See rcu_read_lock_sched() for more information.
855 */
856static inline void rcu_read_unlock_sched(void)
857{
858 RCU_LOCKDEP_WARN(!rcu_is_watching(),
859 "rcu_read_unlock_sched() used illegally while idle");
860 rcu_lock_release(map: &rcu_sched_lock_map);
861 __release(RCU_SCHED);
862 preempt_enable();
863}
864
865/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
866static inline notrace void rcu_read_unlock_sched_notrace(void)
867{
868 __release(RCU_SCHED);
869 preempt_enable_notrace();
870}
871
872/**
873 * RCU_INIT_POINTER() - initialize an RCU protected pointer
874 * @p: The pointer to be initialized.
875 * @v: The value to initialized the pointer to.
876 *
877 * Initialize an RCU-protected pointer in special cases where readers
878 * do not need ordering constraints on the CPU or the compiler. These
879 * special cases are:
880 *
881 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
882 * 2. The caller has taken whatever steps are required to prevent
883 * RCU readers from concurrently accessing this pointer *or*
884 * 3. The referenced data structure has already been exposed to
885 * readers either at compile time or via rcu_assign_pointer() *and*
886 *
887 * a. You have not made *any* reader-visible changes to
888 * this structure since then *or*
889 * b. It is OK for readers accessing this structure from its
890 * new location to see the old state of the structure. (For
891 * example, the changes were to statistical counters or to
892 * other state where exact synchronization is not required.)
893 *
894 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
895 * result in impossible-to-diagnose memory corruption. As in the structures
896 * will look OK in crash dumps, but any concurrent RCU readers might
897 * see pre-initialized values of the referenced data structure. So
898 * please be very careful how you use RCU_INIT_POINTER()!!!
899 *
900 * If you are creating an RCU-protected linked structure that is accessed
901 * by a single external-to-structure RCU-protected pointer, then you may
902 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
903 * pointers, but you must use rcu_assign_pointer() to initialize the
904 * external-to-structure pointer *after* you have completely initialized
905 * the reader-accessible portions of the linked structure.
906 *
907 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
908 * ordering guarantees for either the CPU or the compiler.
909 */
910#define RCU_INIT_POINTER(p, v) \
911 do { \
912 rcu_check_sparse(p, __rcu); \
913 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
914 } while (0)
915
916/**
917 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
918 * @p: The pointer to be initialized.
919 * @v: The value to initialized the pointer to.
920 *
921 * GCC-style initialization for an RCU-protected pointer in a structure field.
922 */
923#define RCU_POINTER_INITIALIZER(p, v) \
924 .p = RCU_INITIALIZER(v)
925
926/*
927 * Does the specified offset indicate that the corresponding rcu_head
928 * structure can be handled by kvfree_rcu()?
929 */
930#define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
931
932/**
933 * kfree_rcu() - kfree an object after a grace period.
934 * @ptr: pointer to kfree for double-argument invocations.
935 * @rhf: the name of the struct rcu_head within the type of @ptr.
936 *
937 * Many rcu callbacks functions just call kfree() on the base structure.
938 * These functions are trivial, but their size adds up, and furthermore
939 * when they are used in a kernel module, that module must invoke the
940 * high-latency rcu_barrier() function at module-unload time.
941 *
942 * The kfree_rcu() function handles this issue. Rather than encoding a
943 * function address in the embedded rcu_head structure, kfree_rcu() instead
944 * encodes the offset of the rcu_head structure within the base structure.
945 * Because the functions are not allowed in the low-order 4096 bytes of
946 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
947 * If the offset is larger than 4095 bytes, a compile-time error will
948 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
949 * either fall back to use of call_rcu() or rearrange the structure to
950 * position the rcu_head structure into the first 4096 bytes.
951 *
952 * The object to be freed can be allocated either by kmalloc() or
953 * kmem_cache_alloc().
954 *
955 * Note that the allowable offset might decrease in the future.
956 *
957 * The BUILD_BUG_ON check must not involve any function calls, hence the
958 * checks are done in macros here.
959 */
960#define kfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf)
961#define kvfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf)
962
963/**
964 * kfree_rcu_mightsleep() - kfree an object after a grace period.
965 * @ptr: pointer to kfree for single-argument invocations.
966 *
967 * When it comes to head-less variant, only one argument
968 * is passed and that is just a pointer which has to be
969 * freed after a grace period. Therefore the semantic is
970 *
971 * kfree_rcu_mightsleep(ptr);
972 *
973 * where @ptr is the pointer to be freed by kvfree().
974 *
975 * Please note, head-less way of freeing is permitted to
976 * use from a context that has to follow might_sleep()
977 * annotation. Otherwise, please switch and embed the
978 * rcu_head structure within the type of @ptr.
979 */
980#define kfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr)
981#define kvfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr)
982
983#define kvfree_rcu_arg_2(ptr, rhf) \
984do { \
985 typeof (ptr) ___p = (ptr); \
986 \
987 if (___p) { \
988 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \
989 kvfree_call_rcu(&((___p)->rhf), (void *) (___p)); \
990 } \
991} while (0)
992
993#define kvfree_rcu_arg_1(ptr) \
994do { \
995 typeof(ptr) ___p = (ptr); \
996 \
997 if (___p) \
998 kvfree_call_rcu(NULL, (void *) (___p)); \
999} while (0)
1000
1001/*
1002 * Place this after a lock-acquisition primitive to guarantee that
1003 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
1004 * if the UNLOCK and LOCK are executed by the same CPU or if the
1005 * UNLOCK and LOCK operate on the same lock variable.
1006 */
1007#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
1008#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
1009#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1010#define smp_mb__after_unlock_lock() do { } while (0)
1011#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1012
1013
1014/* Has the specified rcu_head structure been handed to call_rcu()? */
1015
1016/**
1017 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
1018 * @rhp: The rcu_head structure to initialize.
1019 *
1020 * If you intend to invoke rcu_head_after_call_rcu() to test whether a
1021 * given rcu_head structure has already been passed to call_rcu(), then
1022 * you must also invoke this rcu_head_init() function on it just after
1023 * allocating that structure. Calls to this function must not race with
1024 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
1025 */
1026static inline void rcu_head_init(struct rcu_head *rhp)
1027{
1028 rhp->func = (rcu_callback_t)~0L;
1029}
1030
1031/**
1032 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
1033 * @rhp: The rcu_head structure to test.
1034 * @f: The function passed to call_rcu() along with @rhp.
1035 *
1036 * Returns @true if the @rhp has been passed to call_rcu() with @func,
1037 * and @false otherwise. Emits a warning in any other case, including
1038 * the case where @rhp has already been invoked after a grace period.
1039 * Calls to this function must not race with callback invocation. One way
1040 * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
1041 * in an RCU read-side critical section that includes a read-side fetch
1042 * of the pointer to the structure containing @rhp.
1043 */
1044static inline bool
1045rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1046{
1047 rcu_callback_t func = READ_ONCE(rhp->func);
1048
1049 if (func == f)
1050 return true;
1051 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1052 return false;
1053}
1054
1055/* kernel/ksysfs.c definitions */
1056extern int rcu_expedited;
1057extern int rcu_normal;
1058
1059DEFINE_LOCK_GUARD_0(rcu, rcu_read_lock(), rcu_read_unlock())
1060
1061#endif /* __LINUX_RCUPDATE_H */
1062

source code of linux/include/linux/rcupdate.h