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#include <linux/context_tracking_irq.h>
33
34#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
35#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
36#define ulong2long(a) (*(long *)(&(a)))
37#define USHORT_CMP_GE(a, b) (USHRT_MAX / 2 >= (unsigned short)((a) - (b)))
38#define USHORT_CMP_LT(a, b) (USHRT_MAX / 2 < (unsigned short)((a) - (b)))
39
40/* Exported common interfaces */
41void call_rcu(struct rcu_head *head, rcu_callback_t func);
42void rcu_barrier_tasks(void);
43void rcu_barrier_tasks_rude(void);
44void synchronize_rcu(void);
45unsigned long get_completed_synchronize_rcu(void);
46
47#ifdef CONFIG_PREEMPT_RCU
48
49void __rcu_read_lock(void);
50void __rcu_read_unlock(void);
51
52/*
53 * Defined as a macro as it is a very low level header included from
54 * areas that don't even know about current. This gives the rcu_read_lock()
55 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
56 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
57 */
58#define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
59
60#else /* #ifdef CONFIG_PREEMPT_RCU */
61
62#ifdef CONFIG_TINY_RCU
63#define rcu_read_unlock_strict() do { } while (0)
64#else
65void rcu_read_unlock_strict(void);
66#endif
67
68static inline void __rcu_read_lock(void)
69{
70 preempt_disable();
71}
72
73static inline void __rcu_read_unlock(void)
74{
75 preempt_enable();
76 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
77 rcu_read_unlock_strict();
78}
79
80static inline int rcu_preempt_depth(void)
81{
82 return 0;
83}
84
85#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
86
87/* Internal to kernel */
88void rcu_init(void);
89extern int rcu_scheduler_active;
90void rcu_sched_clock_irq(int user);
91void rcu_report_dead(unsigned int cpu);
92void rcutree_migrate_callbacks(int cpu);
93
94#ifdef CONFIG_TASKS_RCU_GENERIC
95void rcu_init_tasks_generic(void);
96#else
97static inline void rcu_init_tasks_generic(void) { }
98#endif
99
100#ifdef CONFIG_RCU_STALL_COMMON
101void rcu_sysrq_start(void);
102void rcu_sysrq_end(void);
103#else /* #ifdef CONFIG_RCU_STALL_COMMON */
104static inline void rcu_sysrq_start(void) { }
105static inline void rcu_sysrq_end(void) { }
106#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
107
108#if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK))
109void rcu_irq_work_resched(void);
110#else
111static inline void rcu_irq_work_resched(void) { }
112#endif
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 ct_idle_enter() and the ct_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 ct_irq_enter_irqson(); \
147 do { a; } while (0); \
148 ct_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// Bits for ->trc_reader_special.b.need_qs field.
173#define TRC_NEED_QS 0x1 // Task needs a quiescent state.
174#define TRC_NEED_QS_CHECKED 0x2 // Task has been checked for needing quiescent state.
175
176u8 rcu_trc_cmpxchg_need_qs(struct task_struct *t, u8 old, u8 new);
177void rcu_tasks_trace_qs_blkd(struct task_struct *t);
178
179# define rcu_tasks_trace_qs(t) \
180 do { \
181 int ___rttq_nesting = READ_ONCE((t)->trc_reader_nesting); \
182 \
183 if (likely(!READ_ONCE((t)->trc_reader_special.b.need_qs)) && \
184 likely(!___rttq_nesting)) { \
185 rcu_trc_cmpxchg_need_qs((t), 0, TRC_NEED_QS_CHECKED); \
186 } else if (___rttq_nesting && ___rttq_nesting != INT_MIN && \
187 !READ_ONCE((t)->trc_reader_special.b.blocked)) { \
188 rcu_tasks_trace_qs_blkd(t); \
189 } \
190 } while (0)
191# else
192# define rcu_tasks_trace_qs(t) do { } while (0)
193# endif
194
195#define rcu_tasks_qs(t, preempt) \
196do { \
197 rcu_tasks_classic_qs((t), (preempt)); \
198 rcu_tasks_trace_qs(t); \
199} while (0)
200
201# ifdef CONFIG_TASKS_RUDE_RCU
202void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
203void synchronize_rcu_tasks_rude(void);
204# endif
205
206#define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
207void exit_tasks_rcu_start(void);
208void exit_tasks_rcu_finish(void);
209#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
210#define rcu_tasks_classic_qs(t, preempt) do { } while (0)
211#define rcu_tasks_qs(t, preempt) do { } while (0)
212#define rcu_note_voluntary_context_switch(t) do { } while (0)
213#define call_rcu_tasks call_rcu
214#define synchronize_rcu_tasks synchronize_rcu
215static inline void exit_tasks_rcu_start(void) { }
216static inline void exit_tasks_rcu_finish(void) { }
217#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
218
219/**
220 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
221 *
222 * This macro resembles cond_resched(), except that it is defined to
223 * report potential quiescent states to RCU-tasks even if the cond_resched()
224 * machinery were to be shut off, as some advocate for PREEMPTION kernels.
225 */
226#define cond_resched_tasks_rcu_qs() \
227do { \
228 rcu_tasks_qs(current, false); \
229 cond_resched(); \
230} while (0)
231
232/*
233 * Infrastructure to implement the synchronize_() primitives in
234 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
235 */
236
237#if defined(CONFIG_TREE_RCU)
238#include <linux/rcutree.h>
239#elif defined(CONFIG_TINY_RCU)
240#include <linux/rcutiny.h>
241#else
242#error "Unknown RCU implementation specified to kernel configuration"
243#endif
244
245/*
246 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
247 * are needed for dynamic initialization and destruction of rcu_head
248 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
249 * dynamic initialization and destruction of statically allocated rcu_head
250 * structures. However, rcu_head structures allocated dynamically in the
251 * heap don't need any initialization.
252 */
253#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
254void init_rcu_head(struct rcu_head *head);
255void destroy_rcu_head(struct rcu_head *head);
256void init_rcu_head_on_stack(struct rcu_head *head);
257void destroy_rcu_head_on_stack(struct rcu_head *head);
258#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
259static inline void init_rcu_head(struct rcu_head *head) { }
260static inline void destroy_rcu_head(struct rcu_head *head) { }
261static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
262static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
263#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
264
265#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
266bool rcu_lockdep_current_cpu_online(void);
267#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
268static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
269#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
270
271extern struct lockdep_map rcu_lock_map;
272extern struct lockdep_map rcu_bh_lock_map;
273extern struct lockdep_map rcu_sched_lock_map;
274extern struct lockdep_map rcu_callback_map;
275
276#ifdef CONFIG_DEBUG_LOCK_ALLOC
277
278static inline void rcu_lock_acquire(struct lockdep_map *map)
279{
280 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
281}
282
283static inline void rcu_lock_release(struct lockdep_map *map)
284{
285 lock_release(map, _THIS_IP_);
286}
287
288int debug_lockdep_rcu_enabled(void);
289int rcu_read_lock_held(void);
290int rcu_read_lock_bh_held(void);
291int rcu_read_lock_sched_held(void);
292int rcu_read_lock_any_held(void);
293
294#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
295
296# define rcu_lock_acquire(a) do { } while (0)
297# define rcu_lock_release(a) do { } while (0)
298
299static inline int rcu_read_lock_held(void)
300{
301 return 1;
302}
303
304static inline int rcu_read_lock_bh_held(void)
305{
306 return 1;
307}
308
309static inline int rcu_read_lock_sched_held(void)
310{
311 return !preemptible();
312}
313
314static inline int rcu_read_lock_any_held(void)
315{
316 return !preemptible();
317}
318
319#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
320
321#ifdef CONFIG_PROVE_RCU
322
323/**
324 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
325 * @c: condition to check
326 * @s: informative message
327 */
328#define RCU_LOCKDEP_WARN(c, s) \
329 do { \
330 static bool __section(".data.unlikely") __warned; \
331 if ((c) && debug_lockdep_rcu_enabled() && !__warned) { \
332 __warned = true; \
333 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
334 } \
335 } while (0)
336
337#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
338static inline void rcu_preempt_sleep_check(void)
339{
340 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
341 "Illegal context switch in RCU read-side critical section");
342}
343#else /* #ifdef CONFIG_PROVE_RCU */
344static inline void rcu_preempt_sleep_check(void) { }
345#endif /* #else #ifdef CONFIG_PROVE_RCU */
346
347#define rcu_sleep_check() \
348 do { \
349 rcu_preempt_sleep_check(); \
350 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \
351 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
352 "Illegal context switch in RCU-bh read-side critical section"); \
353 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
354 "Illegal context switch in RCU-sched read-side critical section"); \
355 } while (0)
356
357#else /* #ifdef CONFIG_PROVE_RCU */
358
359#define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
360#define rcu_sleep_check() do { } while (0)
361
362#endif /* #else #ifdef CONFIG_PROVE_RCU */
363
364/*
365 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
366 * and rcu_assign_pointer(). Some of these could be folded into their
367 * callers, but they are left separate in order to ease introduction of
368 * multiple pointers markings to match different RCU implementations
369 * (e.g., __srcu), should this make sense in the future.
370 */
371
372#ifdef __CHECKER__
373#define rcu_check_sparse(p, space) \
374 ((void)(((typeof(*p) space *)p) == p))
375#else /* #ifdef __CHECKER__ */
376#define rcu_check_sparse(p, space)
377#endif /* #else #ifdef __CHECKER__ */
378
379#define __unrcu_pointer(p, local) \
380({ \
381 typeof(*p) *local = (typeof(*p) *__force)(p); \
382 rcu_check_sparse(p, __rcu); \
383 ((typeof(*p) __force __kernel *)(local)); \
384})
385/**
386 * unrcu_pointer - mark a pointer as not being RCU protected
387 * @p: pointer needing to lose its __rcu property
388 *
389 * Converts @p from an __rcu pointer to a __kernel pointer.
390 * This allows an __rcu pointer to be used with xchg() and friends.
391 */
392#define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu))
393
394#define __rcu_access_pointer(p, local, space) \
395({ \
396 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
397 rcu_check_sparse(p, space); \
398 ((typeof(*p) __force __kernel *)(local)); \
399})
400#define __rcu_dereference_check(p, local, c, space) \
401({ \
402 /* Dependency order vs. p above. */ \
403 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
404 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
405 rcu_check_sparse(p, space); \
406 ((typeof(*p) __force __kernel *)(local)); \
407})
408#define __rcu_dereference_protected(p, local, c, space) \
409({ \
410 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
411 rcu_check_sparse(p, space); \
412 ((typeof(*p) __force __kernel *)(p)); \
413})
414#define __rcu_dereference_raw(p, local) \
415({ \
416 /* Dependency order vs. p above. */ \
417 typeof(p) local = READ_ONCE(p); \
418 ((typeof(*p) __force __kernel *)(local)); \
419})
420#define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu))
421
422/**
423 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
424 * @v: The value to statically initialize with.
425 */
426#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
427
428/**
429 * rcu_assign_pointer() - assign to RCU-protected pointer
430 * @p: pointer to assign to
431 * @v: value to assign (publish)
432 *
433 * Assigns the specified value to the specified RCU-protected
434 * pointer, ensuring that any concurrent RCU readers will see
435 * any prior initialization.
436 *
437 * Inserts memory barriers on architectures that require them
438 * (which is most of them), and also prevents the compiler from
439 * reordering the code that initializes the structure after the pointer
440 * assignment. More importantly, this call documents which pointers
441 * will be dereferenced by RCU read-side code.
442 *
443 * In some special cases, you may use RCU_INIT_POINTER() instead
444 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
445 * to the fact that it does not constrain either the CPU or the compiler.
446 * That said, using RCU_INIT_POINTER() when you should have used
447 * rcu_assign_pointer() is a very bad thing that results in
448 * impossible-to-diagnose memory corruption. So please be careful.
449 * See the RCU_INIT_POINTER() comment header for details.
450 *
451 * Note that rcu_assign_pointer() evaluates each of its arguments only
452 * once, appearances notwithstanding. One of the "extra" evaluations
453 * is in typeof() and the other visible only to sparse (__CHECKER__),
454 * neither of which actually execute the argument. As with most cpp
455 * macros, this execute-arguments-only-once property is important, so
456 * please be careful when making changes to rcu_assign_pointer() and the
457 * other macros that it invokes.
458 */
459#define rcu_assign_pointer(p, v) \
460do { \
461 uintptr_t _r_a_p__v = (uintptr_t)(v); \
462 rcu_check_sparse(p, __rcu); \
463 \
464 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
465 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
466 else \
467 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
468} while (0)
469
470/**
471 * rcu_replace_pointer() - replace an RCU pointer, returning its old value
472 * @rcu_ptr: RCU pointer, whose old value is returned
473 * @ptr: regular pointer
474 * @c: the lockdep conditions under which the dereference will take place
475 *
476 * Perform a replacement, where @rcu_ptr is an RCU-annotated
477 * pointer and @c is the lockdep argument that is passed to the
478 * rcu_dereference_protected() call used to read that pointer. The old
479 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
480 */
481#define rcu_replace_pointer(rcu_ptr, ptr, c) \
482({ \
483 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
484 rcu_assign_pointer((rcu_ptr), (ptr)); \
485 __tmp; \
486})
487
488/**
489 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
490 * @p: The pointer to read
491 *
492 * Return the value of the specified RCU-protected pointer, but omit the
493 * lockdep checks for being in an RCU read-side critical section. This is
494 * useful when the value of this pointer is accessed, but the pointer is
495 * not dereferenced, for example, when testing an RCU-protected pointer
496 * against NULL. Although rcu_access_pointer() may also be used in cases
497 * where update-side locks prevent the value of the pointer from changing,
498 * you should instead use rcu_dereference_protected() for this use case.
499 *
500 * It is also permissible to use rcu_access_pointer() when read-side
501 * access to the pointer was removed at least one grace period ago, as
502 * is the case in the context of the RCU callback that is freeing up
503 * the data, or after a synchronize_rcu() returns. This can be useful
504 * when tearing down multi-linked structures after a grace period
505 * has elapsed.
506 */
507#define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu)
508
509/**
510 * rcu_dereference_check() - rcu_dereference with debug checking
511 * @p: The pointer to read, prior to dereferencing
512 * @c: The conditions under which the dereference will take place
513 *
514 * Do an rcu_dereference(), but check that the conditions under which the
515 * dereference will take place are correct. Typically the conditions
516 * indicate the various locking conditions that should be held at that
517 * point. The check should return true if the conditions are satisfied.
518 * An implicit check for being in an RCU read-side critical section
519 * (rcu_read_lock()) is included.
520 *
521 * For example:
522 *
523 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
524 *
525 * could be used to indicate to lockdep that foo->bar may only be dereferenced
526 * if either rcu_read_lock() is held, or that the lock required to replace
527 * the bar struct at foo->bar is held.
528 *
529 * Note that the list of conditions may also include indications of when a lock
530 * need not be held, for example during initialisation or destruction of the
531 * target struct:
532 *
533 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
534 * atomic_read(&foo->usage) == 0);
535 *
536 * Inserts memory barriers on architectures that require them
537 * (currently only the Alpha), prevents the compiler from refetching
538 * (and from merging fetches), and, more importantly, documents exactly
539 * which pointers are protected by RCU and checks that the pointer is
540 * annotated as __rcu.
541 */
542#define rcu_dereference_check(p, c) \
543 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
544 (c) || rcu_read_lock_held(), __rcu)
545
546/**
547 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
548 * @p: The pointer to read, prior to dereferencing
549 * @c: The conditions under which the dereference will take place
550 *
551 * This is the RCU-bh counterpart to rcu_dereference_check(). However,
552 * please note that starting in v5.0 kernels, vanilla RCU grace periods
553 * wait for local_bh_disable() regions of code in addition to regions of
554 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means
555 * that synchronize_rcu(), call_rcu, and friends all take not only
556 * rcu_read_lock() but also rcu_read_lock_bh() into account.
557 */
558#define rcu_dereference_bh_check(p, c) \
559 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
560 (c) || rcu_read_lock_bh_held(), __rcu)
561
562/**
563 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
564 * @p: The pointer to read, prior to dereferencing
565 * @c: The conditions under which the dereference will take place
566 *
567 * This is the RCU-sched counterpart to rcu_dereference_check().
568 * However, please note that starting in v5.0 kernels, vanilla RCU grace
569 * periods wait for preempt_disable() regions of code in addition to
570 * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
571 * This means that synchronize_rcu(), call_rcu, and friends all take not
572 * only rcu_read_lock() but also rcu_read_lock_sched() into account.
573 */
574#define rcu_dereference_sched_check(p, c) \
575 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
576 (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) \
587 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu)
588
589/**
590 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
591 * @p: The pointer to read, prior to dereferencing
592 * @c: The conditions under which the dereference will take place
593 *
594 * Return the value of the specified RCU-protected pointer, but omit
595 * the READ_ONCE(). This is useful in cases where update-side locks
596 * prevent the value of the pointer from changing. Please note that this
597 * primitive does *not* prevent the compiler from repeating this reference
598 * or combining it with other references, so it should not be used without
599 * protection of appropriate locks.
600 *
601 * This function is only for update-side use. Using this function
602 * when protected only by rcu_read_lock() will result in infrequent
603 * but very ugly failures.
604 */
605#define rcu_dereference_protected(p, c) \
606 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu)
607
608
609/**
610 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
611 * @p: The pointer to read, prior to dereferencing
612 *
613 * This is a simple wrapper around rcu_dereference_check().
614 */
615#define rcu_dereference(p) rcu_dereference_check(p, 0)
616
617/**
618 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
619 * @p: The pointer to read, prior to dereferencing
620 *
621 * Makes rcu_dereference_check() do the dirty work.
622 */
623#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
624
625/**
626 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
627 * @p: The pointer to read, prior to dereferencing
628 *
629 * Makes rcu_dereference_check() do the dirty work.
630 */
631#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
632
633/**
634 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
635 * @p: The pointer to hand off
636 *
637 * This is simply an identity function, but it documents where a pointer
638 * is handed off from RCU to some other synchronization mechanism, for
639 * example, reference counting or locking. In C11, it would map to
640 * kill_dependency(). It could be used as follows::
641 *
642 * rcu_read_lock();
643 * p = rcu_dereference(gp);
644 * long_lived = is_long_lived(p);
645 * if (long_lived) {
646 * if (!atomic_inc_not_zero(p->refcnt))
647 * long_lived = false;
648 * else
649 * p = rcu_pointer_handoff(p);
650 * }
651 * rcu_read_unlock();
652 */
653#define rcu_pointer_handoff(p) (p)
654
655/**
656 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
657 *
658 * When synchronize_rcu() is invoked on one CPU while other CPUs
659 * are within RCU read-side critical sections, then the
660 * synchronize_rcu() is guaranteed to block until after all the other
661 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
662 * on one CPU while other CPUs are within RCU read-side critical
663 * sections, invocation of the corresponding RCU callback is deferred
664 * until after the all the other CPUs exit their critical sections.
665 *
666 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
667 * wait for regions of code with preemption disabled, including regions of
668 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which
669 * define synchronize_sched(), only code enclosed within rcu_read_lock()
670 * and rcu_read_unlock() are guaranteed to be waited for.
671 *
672 * Note, however, that RCU callbacks are permitted to run concurrently
673 * with new RCU read-side critical sections. One way that this can happen
674 * is via the following sequence of events: (1) CPU 0 enters an RCU
675 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
676 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
677 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
678 * callback is invoked. This is legal, because the RCU read-side critical
679 * section that was running concurrently with the call_rcu() (and which
680 * therefore might be referencing something that the corresponding RCU
681 * callback would free up) has completed before the corresponding
682 * RCU callback is invoked.
683 *
684 * RCU read-side critical sections may be nested. Any deferred actions
685 * will be deferred until the outermost RCU read-side critical section
686 * completes.
687 *
688 * You can avoid reading and understanding the next paragraph by
689 * following this rule: don't put anything in an rcu_read_lock() RCU
690 * read-side critical section that would block in a !PREEMPTION kernel.
691 * But if you want the full story, read on!
692 *
693 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
694 * it is illegal to block while in an RCU read-side critical section.
695 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
696 * kernel builds, RCU read-side critical sections may be preempted,
697 * but explicit blocking is illegal. Finally, in preemptible RCU
698 * implementations in real-time (with -rt patchset) kernel builds, RCU
699 * read-side critical sections may be preempted and they may also block, but
700 * only when acquiring spinlocks that are subject to priority inheritance.
701 */
702static __always_inline void rcu_read_lock(void)
703{
704 __rcu_read_lock();
705 __acquire(RCU);
706 rcu_lock_acquire(&rcu_lock_map);
707 RCU_LOCKDEP_WARN(!rcu_is_watching(),
708 "rcu_read_lock() used illegally while idle");
709}
710
711/*
712 * So where is rcu_write_lock()? It does not exist, as there is no
713 * way for writers to lock out RCU readers. This is a feature, not
714 * a bug -- this property is what provides RCU's performance benefits.
715 * Of course, writers must coordinate with each other. The normal
716 * spinlock primitives work well for this, but any other technique may be
717 * used as well. RCU does not care how the writers keep out of each
718 * others' way, as long as they do so.
719 */
720
721/**
722 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
723 *
724 * In almost all situations, rcu_read_unlock() is immune from deadlock.
725 * In recent kernels that have consolidated synchronize_sched() and
726 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
727 * also extends to the scheduler's runqueue and priority-inheritance
728 * spinlocks, courtesy of the quiescent-state deferral that is carried
729 * out when rcu_read_unlock() is invoked with interrupts disabled.
730 *
731 * See rcu_read_lock() for more information.
732 */
733static inline void rcu_read_unlock(void)
734{
735 RCU_LOCKDEP_WARN(!rcu_is_watching(),
736 "rcu_read_unlock() used illegally while idle");
737 __release(RCU);
738 __rcu_read_unlock();
739 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
740}
741
742/**
743 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
744 *
745 * This is equivalent to rcu_read_lock(), but also disables softirqs.
746 * Note that anything else that disables softirqs can also serve as an RCU
747 * read-side critical section. However, please note that this equivalence
748 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and
749 * rcu_read_lock_bh() were unrelated.
750 *
751 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
752 * must occur in the same context, for example, it is illegal to invoke
753 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
754 * was invoked from some other task.
755 */
756static inline void rcu_read_lock_bh(void)
757{
758 local_bh_disable();
759 __acquire(RCU_BH);
760 rcu_lock_acquire(&rcu_bh_lock_map);
761 RCU_LOCKDEP_WARN(!rcu_is_watching(),
762 "rcu_read_lock_bh() used illegally while idle");
763}
764
765/**
766 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
767 *
768 * See rcu_read_lock_bh() for more information.
769 */
770static inline void rcu_read_unlock_bh(void)
771{
772 RCU_LOCKDEP_WARN(!rcu_is_watching(),
773 "rcu_read_unlock_bh() used illegally while idle");
774 rcu_lock_release(&rcu_bh_lock_map);
775 __release(RCU_BH);
776 local_bh_enable();
777}
778
779/**
780 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
781 *
782 * This is equivalent to rcu_read_lock(), but also disables preemption.
783 * Read-side critical sections can also be introduced by anything else that
784 * disables preemption, including local_irq_disable() and friends. However,
785 * please note that the equivalence to rcu_read_lock() applies only to
786 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
787 * were unrelated.
788 *
789 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
790 * must occur in the same context, for example, it is illegal to invoke
791 * rcu_read_unlock_sched() from process context if the matching
792 * rcu_read_lock_sched() was invoked from an NMI handler.
793 */
794static inline void rcu_read_lock_sched(void)
795{
796 preempt_disable();
797 __acquire(RCU_SCHED);
798 rcu_lock_acquire(&rcu_sched_lock_map);
799 RCU_LOCKDEP_WARN(!rcu_is_watching(),
800 "rcu_read_lock_sched() used illegally while idle");
801}
802
803/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
804static inline notrace void rcu_read_lock_sched_notrace(void)
805{
806 preempt_disable_notrace();
807 __acquire(RCU_SCHED);
808}
809
810/**
811 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
812 *
813 * See rcu_read_lock_sched() for more information.
814 */
815static inline void rcu_read_unlock_sched(void)
816{
817 RCU_LOCKDEP_WARN(!rcu_is_watching(),
818 "rcu_read_unlock_sched() used illegally while idle");
819 rcu_lock_release(&rcu_sched_lock_map);
820 __release(RCU_SCHED);
821 preempt_enable();
822}
823
824/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
825static inline notrace void rcu_read_unlock_sched_notrace(void)
826{
827 __release(RCU_SCHED);
828 preempt_enable_notrace();
829}
830
831/**
832 * RCU_INIT_POINTER() - initialize an RCU protected pointer
833 * @p: The pointer to be initialized.
834 * @v: The value to initialized the pointer to.
835 *
836 * Initialize an RCU-protected pointer in special cases where readers
837 * do not need ordering constraints on the CPU or the compiler. These
838 * special cases are:
839 *
840 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
841 * 2. The caller has taken whatever steps are required to prevent
842 * RCU readers from concurrently accessing this pointer *or*
843 * 3. The referenced data structure has already been exposed to
844 * readers either at compile time or via rcu_assign_pointer() *and*
845 *
846 * a. You have not made *any* reader-visible changes to
847 * this structure since then *or*
848 * b. It is OK for readers accessing this structure from its
849 * new location to see the old state of the structure. (For
850 * example, the changes were to statistical counters or to
851 * other state where exact synchronization is not required.)
852 *
853 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
854 * result in impossible-to-diagnose memory corruption. As in the structures
855 * will look OK in crash dumps, but any concurrent RCU readers might
856 * see pre-initialized values of the referenced data structure. So
857 * please be very careful how you use RCU_INIT_POINTER()!!!
858 *
859 * If you are creating an RCU-protected linked structure that is accessed
860 * by a single external-to-structure RCU-protected pointer, then you may
861 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
862 * pointers, but you must use rcu_assign_pointer() to initialize the
863 * external-to-structure pointer *after* you have completely initialized
864 * the reader-accessible portions of the linked structure.
865 *
866 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
867 * ordering guarantees for either the CPU or the compiler.
868 */
869#define RCU_INIT_POINTER(p, v) \
870 do { \
871 rcu_check_sparse(p, __rcu); \
872 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
873 } while (0)
874
875/**
876 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
877 * @p: The pointer to be initialized.
878 * @v: The value to initialized the pointer to.
879 *
880 * GCC-style initialization for an RCU-protected pointer in a structure field.
881 */
882#define RCU_POINTER_INITIALIZER(p, v) \
883 .p = RCU_INITIALIZER(v)
884
885/*
886 * Does the specified offset indicate that the corresponding rcu_head
887 * structure can be handled by kvfree_rcu()?
888 */
889#define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
890
891/**
892 * kfree_rcu() - kfree an object after a grace period.
893 * @ptr: pointer to kfree for both single- and double-argument invocations.
894 * @rhf: the name of the struct rcu_head within the type of @ptr,
895 * but only for double-argument invocations.
896 *
897 * Many rcu callbacks functions just call kfree() on the base structure.
898 * These functions are trivial, but their size adds up, and furthermore
899 * when they are used in a kernel module, that module must invoke the
900 * high-latency rcu_barrier() function at module-unload time.
901 *
902 * The kfree_rcu() function handles this issue. Rather than encoding a
903 * function address in the embedded rcu_head structure, kfree_rcu() instead
904 * encodes the offset of the rcu_head structure within the base structure.
905 * Because the functions are not allowed in the low-order 4096 bytes of
906 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
907 * If the offset is larger than 4095 bytes, a compile-time error will
908 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
909 * either fall back to use of call_rcu() or rearrange the structure to
910 * position the rcu_head structure into the first 4096 bytes.
911 *
912 * Note that the allowable offset might decrease in the future, for example,
913 * to allow something like kmem_cache_free_rcu().
914 *
915 * The BUILD_BUG_ON check must not involve any function calls, hence the
916 * checks are done in macros here.
917 */
918#define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf)
919
920/**
921 * kvfree_rcu() - kvfree an object after a grace period.
922 *
923 * This macro consists of one or two arguments and it is
924 * based on whether an object is head-less or not. If it
925 * has a head then a semantic stays the same as it used
926 * to be before:
927 *
928 * kvfree_rcu(ptr, rhf);
929 *
930 * where @ptr is a pointer to kvfree(), @rhf is the name
931 * of the rcu_head structure within the type of @ptr.
932 *
933 * When it comes to head-less variant, only one argument
934 * is passed and that is just a pointer which has to be
935 * freed after a grace period. Therefore the semantic is
936 *
937 * kvfree_rcu(ptr);
938 *
939 * where @ptr is the pointer to be freed by kvfree().
940 *
941 * Please note, head-less way of freeing is permitted to
942 * use from a context that has to follow might_sleep()
943 * annotation. Otherwise, please switch and embed the
944 * rcu_head structure within the type of @ptr.
945 */
946#define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \
947 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__)
948
949#define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME
950#define kvfree_rcu_arg_2(ptr, rhf) \
951do { \
952 typeof (ptr) ___p = (ptr); \
953 \
954 if (___p) { \
955 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \
956 kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long) \
957 (offsetof(typeof(*(ptr)), rhf))); \
958 } \
959} while (0)
960
961#define kvfree_rcu_arg_1(ptr) \
962do { \
963 typeof(ptr) ___p = (ptr); \
964 \
965 if (___p) \
966 kvfree_call_rcu(NULL, (rcu_callback_t) (___p)); \
967} while (0)
968
969/*
970 * Place this after a lock-acquisition primitive to guarantee that
971 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
972 * if the UNLOCK and LOCK are executed by the same CPU or if the
973 * UNLOCK and LOCK operate on the same lock variable.
974 */
975#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
976#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
977#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
978#define smp_mb__after_unlock_lock() do { } while (0)
979#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
980
981
982/* Has the specified rcu_head structure been handed to call_rcu()? */
983
984/**
985 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
986 * @rhp: The rcu_head structure to initialize.
987 *
988 * If you intend to invoke rcu_head_after_call_rcu() to test whether a
989 * given rcu_head structure has already been passed to call_rcu(), then
990 * you must also invoke this rcu_head_init() function on it just after
991 * allocating that structure. Calls to this function must not race with
992 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
993 */
994static inline void rcu_head_init(struct rcu_head *rhp)
995{
996 rhp->func = (rcu_callback_t)~0L;
997}
998
999/**
1000 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
1001 * @rhp: The rcu_head structure to test.
1002 * @f: The function passed to call_rcu() along with @rhp.
1003 *
1004 * Returns @true if the @rhp has been passed to call_rcu() with @func,
1005 * and @false otherwise. Emits a warning in any other case, including
1006 * the case where @rhp has already been invoked after a grace period.
1007 * Calls to this function must not race with callback invocation. One way
1008 * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
1009 * in an RCU read-side critical section that includes a read-side fetch
1010 * of the pointer to the structure containing @rhp.
1011 */
1012static inline bool
1013rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1014{
1015 rcu_callback_t func = READ_ONCE(rhp->func);
1016
1017 if (func == f)
1018 return true;
1019 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1020 return false;
1021}
1022
1023/* kernel/ksysfs.c definitions */
1024extern int rcu_expedited;
1025extern int rcu_normal;
1026
1027#endif /* __LINUX_RCUPDATE_H */
1028

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