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/sched.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/context_tracking_irq.h>
33
34 token_context_lock(RCU, __reentrant_ctx_lock);
35 token_context_lock_instance(RCU, RCU_SCHED);
36 token_context_lock_instance(RCU, RCU_BH);
37
38 /*
39 * A convenience macro that can be used for RCU-protected globals or struct
40 * members; adds type qualifier __rcu, and also enforces __guarded_by(RCU).
41 */
42 #define __rcu_guarded __rcu __guarded_by(RCU)
43
44 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
45 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
46
47 #define RCU_SEQ_CTR_SHIFT 2
48 #define RCU_SEQ_STATE_MASK ((1 << RCU_SEQ_CTR_SHIFT) - 1)
49
50 /* Exported common interfaces */
51 void call_rcu(struct rcu_head *head, rcu_callback_t func);
52 void rcu_barrier_tasks(void);
53 void synchronize_rcu(void);
54
55 /*
56 * Grace-period sequence snapshot for the polled RCU APIs: ->norm for the
57 * normal grace period and ->exp for the expedited one. ->exp is unused by
58 * Tiny RCU, but is present unconditionally so that a single definition
59 * serves both Tiny RCU and Tree RCU.
60 */
61 struct rcu_gp_seq {
62 unsigned long norm;
63 unsigned long exp;
64 };
65 unsigned long get_completed_synchronize_rcu(void);
66 void get_completed_synchronize_rcu_full(struct rcu_gp_seq *gsp);
67
68 // Maximum number of unsigned long values corresponding to
69 // not-yet-completed RCU grace periods.
70 #define NUM_ACTIVE_RCU_POLL_OLDSTATE 2
71
72 /**
73 * same_state_synchronize_rcu - Are two old-state values identical?
74 * @oldstate1: First old-state value.
75 * @oldstate2: Second old-state value.
76 *
77 * The two old-state values must have been obtained from either
78 * get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or
79 * get_completed_synchronize_rcu(). Returns @true if the two values are
80 * identical and @false otherwise. This allows structures whose lifetimes
81 * are tracked by old-state values to push these values to a list header,
82 * allowing those structures to be slightly smaller.
83 */
same_state_synchronize_rcu(unsigned long oldstate1,unsigned long oldstate2)84 static inline bool same_state_synchronize_rcu(unsigned long oldstate1, unsigned long oldstate2)
85 {
86 return oldstate1 == oldstate2;
87 }
88
89 #ifdef CONFIG_PREEMPT_RCU
90
91 void __rcu_read_lock(void);
92 void __rcu_read_unlock(void);
93
94 /*
95 * Defined as a macro as it is a very low level header included from
96 * areas that don't even know about current. This gives the rcu_read_lock()
97 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
98 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
99 */
100 #define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
101
102 #else /* #ifdef CONFIG_PREEMPT_RCU */
103
104 #ifdef CONFIG_TINY_RCU
105 #define rcu_read_unlock_strict() do { } while (0)
106 #else
107 void rcu_read_unlock_strict(void);
108 #endif
109
__rcu_read_lock(void)110 static inline void __rcu_read_lock(void)
111 {
112 preempt_disable();
113 }
114
__rcu_read_unlock(void)115 static inline void __rcu_read_unlock(void)
116 {
117 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
118 rcu_read_unlock_strict();
119 preempt_enable();
120 }
121
rcu_preempt_depth(void)122 static inline int rcu_preempt_depth(void)
123 {
124 return 0;
125 }
126
127 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
128
129 #ifdef CONFIG_RCU_LAZY
130 void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func);
131 #else
call_rcu_hurry(struct rcu_head * head,rcu_callback_t func)132 static inline void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func)
133 {
134 call_rcu(head, func);
135 }
136 #endif
137
138 /* Internal to kernel */
139 void rcu_init(void);
140 extern int rcu_scheduler_active;
141 void rcu_sched_clock_irq(int user);
142
143 #ifdef CONFIG_RCU_STALL_COMMON
144 void rcu_sysrq_start(void);
145 void rcu_sysrq_end(void);
146 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
rcu_sysrq_start(void)147 static inline void rcu_sysrq_start(void) { }
rcu_sysrq_end(void)148 static inline void rcu_sysrq_end(void) { }
149 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
150
151 #if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_VIRT_XFER_TO_GUEST_WORK))
152 void rcu_irq_work_resched(void);
153 #else
rcu_irq_work_resched(void)154 static __always_inline void rcu_irq_work_resched(void) { }
155 #endif
156
157 #ifdef CONFIG_RCU_NOCB_CPU
158 void rcu_init_nohz(void);
159 int rcu_nocb_cpu_offload(int cpu);
160 int rcu_nocb_cpu_deoffload(int cpu);
161 void rcu_nocb_flush_deferred_wakeup(void);
162
163 #define RCU_NOCB_LOCKDEP_WARN(c, s) RCU_LOCKDEP_WARN(c, s)
164
165 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
166
rcu_init_nohz(void)167 static inline void rcu_init_nohz(void) { }
rcu_nocb_cpu_offload(int cpu)168 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
rcu_nocb_cpu_deoffload(int cpu)169 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
rcu_nocb_flush_deferred_wakeup(void)170 static inline void rcu_nocb_flush_deferred_wakeup(void) { }
171
172 #define RCU_NOCB_LOCKDEP_WARN(c, s)
173
174 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
175
176 /*
177 * Note a quasi-voluntary context switch for RCU-tasks's benefit.
178 * This is a macro rather than an inline function to avoid #include hell.
179 */
180 #ifdef CONFIG_TASKS_RCU_GENERIC
181
182 # ifdef CONFIG_TASKS_RCU
183 # define rcu_tasks_classic_qs(t, preempt) \
184 do { \
185 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \
186 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
187 } while (0)
188 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
189 void synchronize_rcu_tasks(void);
190 void rcu_tasks_torture_stats_print(char *tt, char *tf);
191 # else
192 # define rcu_tasks_classic_qs(t, preempt) do { } while (0)
193 # define call_rcu_tasks call_rcu
194 # define synchronize_rcu_tasks synchronize_rcu
195 # endif
196
197 #define rcu_tasks_qs(t, preempt) rcu_tasks_classic_qs((t), (preempt))
198
199 # ifdef CONFIG_TASKS_RUDE_RCU
200 void synchronize_rcu_tasks_rude(void);
201 void rcu_tasks_rude_torture_stats_print(char *tt, char *tf);
202 # endif
203
204 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
205 void exit_tasks_rcu_start(void);
206 void exit_tasks_rcu_finish(void);
207 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
208 #define rcu_tasks_classic_qs(t, preempt) do { } while (0)
209 #define rcu_tasks_qs(t, preempt) do { } while (0)
210 #define rcu_note_voluntary_context_switch(t) do { } while (0)
211 #define call_rcu_tasks call_rcu
212 #define synchronize_rcu_tasks synchronize_rcu
exit_tasks_rcu_start(void)213 static inline void exit_tasks_rcu_start(void) { }
exit_tasks_rcu_finish(void)214 static inline void exit_tasks_rcu_finish(void) { }
215 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
216
217 /**
218 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
219 *
220 * This function resembles cond_resched(), except that it is defined to
221 * report potential quiescent states to RCU-tasks even if the cond_resched()
222 * machinery were to be shut off, as some advocate for PREEMPTION kernels.
223 */
cond_resched_tasks_rcu_qs(void)224 static inline void cond_resched_tasks_rcu_qs(void)
225 {
226 rcu_tasks_qs(current, false);
227 cond_resched();
228 }
229
230 /**
231 * rcu_softirq_qs_periodic - Report RCU and RCU-Tasks quiescent states
232 * @old_ts: jiffies at start of processing.
233 *
234 * This helper is for long-running softirq handlers, such as NAPI threads in
235 * networking. The caller should initialize the variable passed in as @old_ts
236 * at the beginning of the softirq handler. When invoked frequently, this macro
237 * will invoke rcu_softirq_qs() every 100 milliseconds thereafter, which will
238 * provide both RCU and RCU-Tasks quiescent states. Note that this macro
239 * modifies its old_ts argument.
240 *
241 * Because regions of code that have disabled softirq act as RCU read-side
242 * critical sections, this macro should be invoked with softirq (and
243 * preemption) enabled.
244 *
245 * The macro is not needed when CONFIG_PREEMPT_RT is defined. RT kernels would
246 * have more chance to invoke schedule() calls and provide necessary quiescent
247 * states. As a contrast, calling cond_resched() only won't achieve the same
248 * effect because cond_resched() does not provide RCU-Tasks quiescent states.
249 */
250 #define rcu_softirq_qs_periodic(old_ts) \
251 do { \
252 if (!IS_ENABLED(CONFIG_PREEMPT_RT) && \
253 time_after(jiffies, (old_ts) + HZ / 10)) { \
254 preempt_disable(); \
255 rcu_softirq_qs(); \
256 preempt_enable(); \
257 (old_ts) = jiffies; \
258 } \
259 } while (0)
260
261 /*
262 * Infrastructure to implement the synchronize_() primitives in
263 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
264 */
265
266 #if defined(CONFIG_TREE_RCU)
267 #include <linux/rcutree.h>
268 #elif defined(CONFIG_TINY_RCU)
269 #include <linux/rcutiny.h>
270 #else
271 #error "Unknown RCU implementation specified to kernel configuration"
272 #endif
273
274 /*
275 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
276 * are needed for dynamic initialization and destruction of rcu_head
277 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
278 * dynamic initialization and destruction of statically allocated rcu_head
279 * structures. However, rcu_head structures allocated dynamically in the
280 * heap don't need any initialization.
281 */
282 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
283 void init_rcu_head(struct rcu_head *head);
284 void destroy_rcu_head(struct rcu_head *head);
285 void init_rcu_head_on_stack(struct rcu_head *head);
286 void destroy_rcu_head_on_stack(struct rcu_head *head);
287 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
init_rcu_head(struct rcu_head * head)288 static inline void init_rcu_head(struct rcu_head *head) { }
destroy_rcu_head(struct rcu_head * head)289 static inline void destroy_rcu_head(struct rcu_head *head) { }
init_rcu_head_on_stack(struct rcu_head * head)290 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
destroy_rcu_head_on_stack(struct rcu_head * head)291 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
292 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
293
294 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
295 bool rcu_lockdep_current_cpu_online(void);
296 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
rcu_lockdep_current_cpu_online(void)297 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
298 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
299
300 extern struct lockdep_map rcu_lock_map;
301 extern struct lockdep_map rcu_bh_lock_map;
302 extern struct lockdep_map rcu_sched_lock_map;
303 extern struct lockdep_map rcu_callback_map;
304
305 #ifdef CONFIG_DEBUG_LOCK_ALLOC
306
rcu_lock_acquire(struct lockdep_map * map)307 static inline void rcu_lock_acquire(struct lockdep_map *map)
308 {
309 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
310 }
311
rcu_try_lock_acquire(struct lockdep_map * map)312 static inline void rcu_try_lock_acquire(struct lockdep_map *map)
313 {
314 lock_acquire(map, 0, 1, 2, 0, NULL, _THIS_IP_);
315 }
316
rcu_lock_release(struct lockdep_map * map)317 static inline void rcu_lock_release(struct lockdep_map *map)
318 {
319 lock_release(map, _THIS_IP_);
320 }
321
322 int debug_lockdep_rcu_enabled(void);
323 int rcu_read_lock_held(void);
324 int rcu_read_lock_bh_held(void);
325 int rcu_read_lock_sched_held(void);
326 int rcu_read_lock_any_held(void);
327
328 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
329
330 # define rcu_lock_acquire(a) do { } while (0)
331 # define rcu_try_lock_acquire(a) do { } while (0)
332 # define rcu_lock_release(a) do { } while (0)
333
rcu_read_lock_held(void)334 static inline int rcu_read_lock_held(void)
335 {
336 return 1;
337 }
338
rcu_read_lock_bh_held(void)339 static inline int rcu_read_lock_bh_held(void)
340 {
341 return 1;
342 }
343
rcu_read_lock_sched_held(void)344 static inline int rcu_read_lock_sched_held(void)
345 {
346 return !preemptible();
347 }
348
rcu_read_lock_any_held(void)349 static inline int rcu_read_lock_any_held(void)
350 {
351 return !preemptible();
352 }
353
debug_lockdep_rcu_enabled(void)354 static inline int debug_lockdep_rcu_enabled(void)
355 {
356 return 0;
357 }
358
359 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
360
361 #ifdef CONFIG_PROVE_RCU
362
363 /**
364 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
365 * @c: condition to check
366 * @s: informative message
367 *
368 * This checks debug_lockdep_rcu_enabled() before checking (c) to
369 * prevent early boot splats due to lockdep not yet being initialized,
370 * and rechecks it after checking (c) to prevent false-positive splats
371 * due to races with lockdep being disabled. See commit 3066820034b5dd
372 * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail.
373 */
374 #define RCU_LOCKDEP_WARN(c, s) \
375 do { \
376 static bool __section(".data..unlikely") __warned; \
377 if (debug_lockdep_rcu_enabled() && (c) && \
378 debug_lockdep_rcu_enabled() && !__warned) { \
379 __warned = true; \
380 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
381 } \
382 } while (0)
383
384 #ifndef CONFIG_PREEMPT_RCU
rcu_preempt_sleep_check(void)385 static inline void rcu_preempt_sleep_check(void)
386 {
387 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
388 "Illegal context switch in RCU read-side critical section");
389 }
390 #else // #ifndef CONFIG_PREEMPT_RCU
rcu_preempt_sleep_check(void)391 static inline void rcu_preempt_sleep_check(void) { }
392 #endif // #else // #ifndef CONFIG_PREEMPT_RCU
393
394 #define rcu_sleep_check() \
395 do { \
396 rcu_preempt_sleep_check(); \
397 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \
398 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
399 "Illegal context switch in RCU-bh read-side critical section"); \
400 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
401 "Illegal context switch in RCU-sched read-side critical section"); \
402 } while (0)
403
404 // See RCU_LOCKDEP_WARN() for an explanation of the double call to
405 // debug_lockdep_rcu_enabled().
lockdep_assert_rcu_helper(bool c,const struct __ctx_lock_RCU * ctx)406 static __always_inline bool lockdep_assert_rcu_helper(bool c, const struct __ctx_lock_RCU *ctx)
407 __assumes_shared_ctx_lock(RCU) __assumes_shared_ctx_lock(ctx)
408 {
409 return debug_lockdep_rcu_enabled() &&
410 (c || !rcu_is_watching() || !rcu_lockdep_current_cpu_online()) &&
411 debug_lockdep_rcu_enabled();
412 }
413
414 /**
415 * lockdep_assert_in_rcu_read_lock - WARN if not protected by rcu_read_lock()
416 *
417 * Splats if lockdep is enabled and there is no rcu_read_lock() in effect.
418 */
419 #define lockdep_assert_in_rcu_read_lock() \
420 WARN_ON_ONCE(lockdep_assert_rcu_helper(!lock_is_held(&rcu_lock_map), RCU))
421
422 /**
423 * lockdep_assert_in_rcu_read_lock_bh - WARN if not protected by rcu_read_lock_bh()
424 *
425 * Splats if lockdep is enabled and there is no rcu_read_lock_bh() in effect.
426 * Note that local_bh_disable() and friends do not suffice here, instead an
427 * actual rcu_read_lock_bh() is required.
428 */
429 #define lockdep_assert_in_rcu_read_lock_bh() \
430 WARN_ON_ONCE(lockdep_assert_rcu_helper(!lock_is_held(&rcu_bh_lock_map), RCU_BH))
431
432 /**
433 * lockdep_assert_in_rcu_read_lock_sched - WARN if not protected by rcu_read_lock_sched()
434 *
435 * Splats if lockdep is enabled and there is no rcu_read_lock_sched()
436 * in effect. Note that preempt_disable() and friends do not suffice here,
437 * instead an actual rcu_read_lock_sched() is required.
438 */
439 #define lockdep_assert_in_rcu_read_lock_sched() \
440 WARN_ON_ONCE(lockdep_assert_rcu_helper(!lock_is_held(&rcu_sched_lock_map), RCU_SCHED))
441
442 /**
443 * lockdep_assert_in_rcu_reader - WARN if not within some type of RCU reader
444 *
445 * Splats if lockdep is enabled and there is no RCU reader of any
446 * type in effect. Note that regions of code protected by things like
447 * preempt_disable, local_bh_disable(), and local_irq_disable() all qualify
448 * as RCU readers.
449 *
450 * Note that this will never trigger in PREEMPT_NONE or PREEMPT_VOLUNTARY
451 * kernels that are not also built with PREEMPT_COUNT. But if you have
452 * lockdep enabled, you might as well also enable PREEMPT_COUNT.
453 */
454 #define lockdep_assert_in_rcu_reader() \
455 WARN_ON_ONCE(lockdep_assert_rcu_helper(!lock_is_held(&rcu_lock_map) && \
456 !lock_is_held(&rcu_bh_lock_map) && \
457 !lock_is_held(&rcu_sched_lock_map) && \
458 preemptible(), RCU))
459
460 #else /* #ifdef CONFIG_PROVE_RCU */
461
462 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
463 #define rcu_sleep_check() do { } while (0)
464
465 #define lockdep_assert_in_rcu_read_lock() __assume_shared_ctx_lock(RCU)
466 #define lockdep_assert_in_rcu_read_lock_bh() __assume_shared_ctx_lock(RCU_BH)
467 #define lockdep_assert_in_rcu_read_lock_sched() __assume_shared_ctx_lock(RCU_SCHED)
468 #define lockdep_assert_in_rcu_reader() __assume_shared_ctx_lock(RCU)
469
470 #endif /* #else #ifdef CONFIG_PROVE_RCU */
471
472 /*
473 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
474 * and rcu_assign_pointer(). Some of these could be folded into their
475 * callers, but they are left separate in order to ease introduction of
476 * multiple pointers markings to match different RCU implementations
477 * (e.g., __srcu), should this make sense in the future.
478 */
479
480 #ifdef __CHECKER__
481 #define rcu_check_sparse(p, space) \
482 ((void)(((typeof(*p) space *)p) == p))
483 #else /* #ifdef __CHECKER__ */
484 #define rcu_check_sparse(p, space)
485 #endif /* #else #ifdef __CHECKER__ */
486
487 #define __unrcu_pointer(p, local) \
488 context_unsafe( \
489 typeof(*p) *local = (typeof(*p) *__force)(p); \
490 rcu_check_sparse(p, __rcu); \
491 ((typeof(*p) __force __kernel *)(local)) \
492 )
493 /**
494 * unrcu_pointer - mark a pointer as not being RCU protected
495 * @p: pointer needing to lose its __rcu property
496 *
497 * Converts @p from an __rcu pointer to a __kernel pointer.
498 * This allows an __rcu pointer to be used with xchg() and friends.
499 */
500 #define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu))
501
502 #define __rcu_access_pointer(p, local, space) context_unsafe( \
503 ({ \
504 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
505 rcu_check_sparse(p, space); \
506 ((typeof(*p) __force __kernel *)(local)); \
507 }) )
508 #define __rcu_dereference_check(p, local, c, space) \
509 ({ \
510 /* Dependency order vs. p above. */ \
511 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
512 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
513 rcu_check_sparse(p, space); \
514 ((typeof(*p) __force __kernel *)(local)); \
515 })
516 #define __rcu_dereference_protected(p, local, c, space) \
517 ({ \
518 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
519 rcu_check_sparse(p, space); \
520 ((typeof(*p) __force __kernel *)(p)); \
521 })
522 #define __rcu_dereference_raw(p, local) \
523 ({ \
524 /* Dependency order vs. p above. */ \
525 typeof(p) local = READ_ONCE(p); \
526 ((typeof(*p) __force __kernel *)(local)); \
527 })
528 #define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu))
529
530 /**
531 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
532 * @v: The value to statically initialize with.
533 */
534 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
535
536 /**
537 * rcu_assign_pointer() - assign to RCU-protected pointer
538 * @p: pointer to assign to
539 * @v: value to assign (publish)
540 *
541 * Assigns the specified value to the specified RCU-protected
542 * pointer, ensuring that any concurrent RCU readers will see
543 * any prior initialization.
544 *
545 * Inserts memory barriers on architectures that require them
546 * (which is most of them), and also prevents the compiler from
547 * reordering the code that initializes the structure after the pointer
548 * assignment. More importantly, this call documents which pointers
549 * will be dereferenced by RCU read-side code.
550 *
551 * In some special cases, you may use RCU_INIT_POINTER() instead
552 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
553 * to the fact that it does not constrain either the CPU or the compiler.
554 * That said, using RCU_INIT_POINTER() when you should have used
555 * rcu_assign_pointer() is a very bad thing that results in
556 * impossible-to-diagnose memory corruption. So please be careful.
557 * See the RCU_INIT_POINTER() comment header for details.
558 *
559 * Note that rcu_assign_pointer() evaluates each of its arguments only
560 * once, appearances notwithstanding. One of the "extra" evaluations
561 * is in typeof() and the other visible only to sparse (__CHECKER__),
562 * neither of which actually execute the argument. As with most cpp
563 * macros, this execute-arguments-only-once property is important, so
564 * please be careful when making changes to rcu_assign_pointer() and the
565 * other macros that it invokes.
566 */
567 #define rcu_assign_pointer(p, v) \
568 context_unsafe( \
569 uintptr_t _r_a_p__v = (uintptr_t)(v); \
570 rcu_check_sparse(p, __rcu); \
571 \
572 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
573 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
574 else \
575 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
576 )
577
578 /**
579 * rcu_replace_pointer() - replace an RCU pointer, returning its old value
580 * @rcu_ptr: RCU pointer, whose old value is returned
581 * @ptr: regular pointer
582 * @c: the lockdep conditions under which the dereference will take place
583 *
584 * Perform a replacement, where @rcu_ptr is an RCU-annotated
585 * pointer and @c is the lockdep argument that is passed to the
586 * rcu_dereference_protected() call used to read that pointer. The old
587 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
588 */
589 #define rcu_replace_pointer(rcu_ptr, ptr, c) \
590 ({ \
591 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
592 rcu_assign_pointer((rcu_ptr), (ptr)); \
593 __tmp; \
594 })
595
596 /**
597 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
598 * @p: The pointer to read
599 *
600 * Return the value of the specified RCU-protected pointer, but omit the
601 * lockdep checks for being in an RCU read-side critical section. This is
602 * useful when the value of this pointer is accessed, but the pointer is
603 * not dereferenced, for example, when testing an RCU-protected pointer
604 * against NULL. Within an RCU read-side critical section, there is little
605 * reason to use rcu_access_pointer(). Although rcu_access_pointer() may
606 * also be used in cases where update-side locks prevent the value of the
607 * pointer from changing, you should instead use rcu_dereference_protected()
608 * for this use case. It is also permissible to use rcu_access_pointer()
609 * within lockless updaters to obtain the old value for an atomic operation,
610 * for example, for cmpxchg().
611 *
612 * It is usually best to test the rcu_access_pointer() return value
613 * directly in order to avoid accidental dereferences being introduced
614 * by later inattentive changes. In other words, assigning the
615 * rcu_access_pointer() return value to a local variable results in an
616 * accident waiting to happen.
617 *
618 * It is also permissible to use rcu_access_pointer() when read-side
619 * access to the pointer was removed at least one grace period ago, as is
620 * the case in the context of the RCU callback that is freeing up the data,
621 * or after a synchronize_rcu() returns. This can be useful when tearing
622 * down multi-linked structures after a grace period has elapsed. However,
623 * rcu_dereference_protected() is normally preferred for this use case.
624 */
625 #define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu)
626
627 /**
628 * rcu_dereference_check() - rcu_dereference with debug checking
629 * @p: The pointer to read, prior to dereferencing
630 * @c: The conditions under which the dereference will take place
631 *
632 * Do an rcu_dereference(), but check that the conditions under which the
633 * dereference will take place are correct. Typically the conditions
634 * indicate the various locking conditions that should be held at that
635 * point. The check should return true if the conditions are satisfied.
636 * An implicit check for being in an RCU read-side critical section
637 * (rcu_read_lock()) is included.
638 *
639 * For example:
640 *
641 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
642 *
643 * could be used to indicate to lockdep that foo->bar may only be dereferenced
644 * if either rcu_read_lock() is held, or that the lock required to replace
645 * the bar struct at foo->bar is held.
646 *
647 * Note that the list of conditions may also include indications of when a lock
648 * need not be held, for example during initialisation or destruction of the
649 * target struct:
650 *
651 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
652 * atomic_read(&foo->usage) == 0);
653 *
654 * Inserts memory barriers on architectures that require them
655 * (currently only the Alpha), prevents the compiler from refetching
656 * (and from merging fetches), and, more importantly, documents exactly
657 * which pointers are protected by RCU and checks that the pointer is
658 * annotated as __rcu.
659 */
660 #define rcu_dereference_check(p, c) \
661 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
662 (c) || rcu_read_lock_held(), __rcu)
663
664 /**
665 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
666 * @p: The pointer to read, prior to dereferencing
667 * @c: The conditions under which the dereference will take place
668 *
669 * This is the RCU-bh counterpart to rcu_dereference_check(). However,
670 * please note that starting in v5.0 kernels, vanilla RCU grace periods
671 * wait for local_bh_disable() regions of code in addition to regions of
672 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means
673 * that synchronize_rcu(), call_rcu, and friends all take not only
674 * rcu_read_lock() but also rcu_read_lock_bh() into account.
675 */
676 #define rcu_dereference_bh_check(p, c) \
677 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
678 (c) || rcu_read_lock_bh_held(), __rcu)
679
680 /**
681 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
682 * @p: The pointer to read, prior to dereferencing
683 * @c: The conditions under which the dereference will take place
684 *
685 * This is the RCU-sched counterpart to rcu_dereference_check().
686 * However, please note that starting in v5.0 kernels, vanilla RCU grace
687 * periods wait for preempt_disable() regions of code in addition to
688 * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
689 * This means that synchronize_rcu(), call_rcu, and friends all take not
690 * only rcu_read_lock() but also rcu_read_lock_sched() into account.
691 */
692 #define rcu_dereference_sched_check(p, c) \
693 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
694 (c) || rcu_read_lock_sched_held(), \
695 __rcu)
696
697 /**
698 * rcu_dereference_all_check() - rcu_dereference_all with debug checking
699 * @p: The pointer to read, prior to dereferencing
700 * @c: The conditions under which the dereference will take place
701 *
702 * This is similar to rcu_dereference_check(), but allows protection
703 * by all forms of vanilla RCU readers, including preemption disabled,
704 * bh-disabled, and interrupt-disabled regions of code. Note that "vanilla
705 * RCU" excludes SRCU and the various Tasks RCU flavors. Please note
706 * that this macro should not be backported to any Linux-kernel version
707 * preceding v5.0 due to changes in synchronize_rcu() semantics prior
708 * to that version.
709 */
710 #define rcu_dereference_all_check(p, c) \
711 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
712 (c) || rcu_read_lock_any_held(), \
713 __rcu)
714
715 /*
716 * The tracing infrastructure traces RCU (we want that), but unfortunately
717 * some of the RCU checks causes tracing to lock up the system.
718 *
719 * The no-tracing version of rcu_dereference_raw() must not call
720 * rcu_read_lock_held().
721 */
722 #define rcu_dereference_raw_check(p) \
723 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu)
724
725 /**
726 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
727 * @p: The pointer to read, prior to dereferencing
728 * @c: The conditions under which the dereference will take place
729 *
730 * Return the value of the specified RCU-protected pointer, but omit
731 * the READ_ONCE(). This is useful in cases where update-side locks
732 * prevent the value of the pointer from changing. Please note that this
733 * primitive does *not* prevent the compiler from repeating this reference
734 * or combining it with other references, so it should not be used without
735 * protection of appropriate locks.
736 *
737 * This function is only for update-side use. Using this function
738 * when protected only by rcu_read_lock() will result in infrequent
739 * but very ugly failures.
740 */
741 #define rcu_dereference_protected(p, c) \
742 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu)
743
744
745 /**
746 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
747 * @p: The pointer to read, prior to dereferencing
748 *
749 * This is a simple wrapper around rcu_dereference_check().
750 */
751 #define rcu_dereference(p) rcu_dereference_check(p, 0)
752
753 /**
754 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
755 * @p: The pointer to read, prior to dereferencing
756 *
757 * Makes rcu_dereference_check() do the dirty work.
758 */
759 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
760
761 /**
762 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
763 * @p: The pointer to read, prior to dereferencing
764 *
765 * Makes rcu_dereference_check() do the dirty work.
766 */
767 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
768
769 /**
770 * rcu_dereference_all() - fetch RCU-all-protected pointer for dereferencing
771 * @p: The pointer to read, prior to dereferencing
772 *
773 * Makes rcu_dereference_check() do the dirty work.
774 */
775 #define rcu_dereference_all(p) rcu_dereference_all_check(p, 0)
776
777 /**
778 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
779 * @p: The pointer to hand off
780 *
781 * This is simply an identity function, but it documents where a pointer
782 * is handed off from RCU to some other synchronization mechanism, for
783 * example, reference counting or locking. In C11, it would map to
784 * kill_dependency(). It could be used as follows::
785 *
786 * rcu_read_lock();
787 * p = rcu_dereference(gp);
788 * long_lived = is_long_lived(p);
789 * if (long_lived) {
790 * if (!atomic_inc_not_zero(p->refcnt))
791 * long_lived = false;
792 * else
793 * p = rcu_pointer_handoff(p);
794 * }
795 * rcu_read_unlock();
796 */
797 #define rcu_pointer_handoff(p) (p)
798
799 /**
800 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
801 *
802 * When synchronize_rcu() is invoked on one CPU while other CPUs
803 * are within RCU read-side critical sections, then the
804 * synchronize_rcu() is guaranteed to block until after all the other
805 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
806 * on one CPU while other CPUs are within RCU read-side critical
807 * sections, invocation of the corresponding RCU callback is deferred
808 * until after the all the other CPUs exit their critical sections.
809 *
810 * Both synchronize_rcu() and call_rcu() also wait for regions of code
811 * with preemption disabled, including regions of code with interrupts or
812 * softirqs disabled.
813 *
814 * Note, however, that RCU callbacks are permitted to run concurrently
815 * with new RCU read-side critical sections. One way that this can happen
816 * is via the following sequence of events: (1) CPU 0 enters an RCU
817 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
818 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
819 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
820 * callback is invoked. This is legal, because the RCU read-side critical
821 * section that was running concurrently with the call_rcu() (and which
822 * therefore might be referencing something that the corresponding RCU
823 * callback would free up) has completed before the corresponding
824 * RCU callback is invoked.
825 *
826 * RCU read-side critical sections may be nested. Any deferred actions
827 * will be deferred until the outermost RCU read-side critical section
828 * completes.
829 *
830 * You can avoid reading and understanding the next paragraph by
831 * following this rule: don't put anything in an rcu_read_lock() RCU
832 * read-side critical section that would block in a !PREEMPTION kernel.
833 * But if you want the full story, read on!
834 *
835 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
836 * it is illegal to block while in an RCU read-side critical section.
837 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
838 * kernel builds, RCU read-side critical sections may be preempted,
839 * but explicit blocking is illegal. Finally, in preemptible RCU
840 * implementations in real-time (with -rt patchset) kernel builds, RCU
841 * read-side critical sections may be preempted and they may also block, but
842 * only when acquiring spinlocks that are subject to priority inheritance.
843 */
rcu_read_lock(void)844 static __always_inline void rcu_read_lock(void)
845 __acquires_shared(RCU)
846 {
847 __rcu_read_lock();
848 __acquire_shared(RCU);
849 rcu_lock_acquire(&rcu_lock_map);
850 RCU_LOCKDEP_WARN(!rcu_is_watching(),
851 "rcu_read_lock() used illegally while idle");
852 }
853
854 /*
855 * So where is rcu_write_lock()? It does not exist, as there is no
856 * way for writers to lock out RCU readers. This is a feature, not
857 * a bug -- this property is what provides RCU's performance benefits.
858 * Of course, writers must coordinate with each other. The normal
859 * spinlock primitives work well for this, but any other technique may be
860 * used as well. RCU does not care how the writers keep out of each
861 * others' way, as long as they do so.
862 */
863
864 /**
865 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
866 *
867 * In almost all situations, rcu_read_unlock() is immune from deadlock.
868 * This deadlock immunity also extends to the scheduler's runqueue
869 * and priority-inheritance spinlocks, courtesy of the quiescent-state
870 * deferral that is carried out when rcu_read_unlock() is invoked with
871 * interrupts disabled.
872 *
873 * See rcu_read_lock() for more information.
874 */
rcu_read_unlock(void)875 static inline void rcu_read_unlock(void)
876 __releases_shared(RCU)
877 {
878 RCU_LOCKDEP_WARN(!rcu_is_watching(),
879 "rcu_read_unlock() used illegally while idle");
880 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
881 __release_shared(RCU);
882 __rcu_read_unlock();
883 }
884
885 /**
886 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
887 *
888 * This is equivalent to rcu_read_lock(), but also disables softirqs.
889 * Note that anything else that disables softirqs can also serve as an RCU
890 * read-side critical section. However, please note that this equivalence
891 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and
892 * rcu_read_lock_bh() were unrelated.
893 *
894 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
895 * must occur in the same context, for example, it is illegal to invoke
896 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
897 * was invoked from some other task.
898 */
rcu_read_lock_bh(void)899 static inline void rcu_read_lock_bh(void)
900 __acquires_shared(RCU) __acquires_shared(RCU_BH)
901 {
902 local_bh_disable();
903 __acquire_shared(RCU);
904 __acquire_shared(RCU_BH);
905 rcu_lock_acquire(&rcu_bh_lock_map);
906 RCU_LOCKDEP_WARN(!rcu_is_watching(),
907 "rcu_read_lock_bh() used illegally while idle");
908 }
909
910 /**
911 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
912 *
913 * See rcu_read_lock_bh() for more information.
914 */
rcu_read_unlock_bh(void)915 static inline void rcu_read_unlock_bh(void)
916 __releases_shared(RCU) __releases_shared(RCU_BH)
917 {
918 RCU_LOCKDEP_WARN(!rcu_is_watching(),
919 "rcu_read_unlock_bh() used illegally while idle");
920 rcu_lock_release(&rcu_bh_lock_map);
921 __release_shared(RCU_BH);
922 __release_shared(RCU);
923 local_bh_enable();
924 }
925
926 /**
927 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
928 *
929 * This is equivalent to rcu_read_lock(), but also disables preemption.
930 * Read-side critical sections can also be introduced by anything else that
931 * disables preemption, including local_irq_disable() and friends. However,
932 * please note that the equivalence to rcu_read_lock() applies only to
933 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
934 * were unrelated.
935 *
936 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
937 * must occur in the same context, for example, it is illegal to invoke
938 * rcu_read_unlock_sched() from process context if the matching
939 * rcu_read_lock_sched() was invoked from an NMI handler.
940 */
rcu_read_lock_sched(void)941 static inline void rcu_read_lock_sched(void)
942 __acquires_shared(RCU) __acquires_shared(RCU_SCHED)
943 {
944 preempt_disable();
945 __acquire_shared(RCU);
946 __acquire_shared(RCU_SCHED);
947 rcu_lock_acquire(&rcu_sched_lock_map);
948 RCU_LOCKDEP_WARN(!rcu_is_watching(),
949 "rcu_read_lock_sched() used illegally while idle");
950 }
951
952 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_lock_sched_notrace(void)953 static inline notrace void rcu_read_lock_sched_notrace(void)
954 __acquires_shared(RCU) __acquires_shared(RCU_SCHED)
955 {
956 preempt_disable_notrace();
957 __acquire_shared(RCU);
958 __acquire_shared(RCU_SCHED);
959 }
960
961 /**
962 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
963 *
964 * See rcu_read_lock_sched() for more information.
965 */
rcu_read_unlock_sched(void)966 static inline void rcu_read_unlock_sched(void)
967 __releases_shared(RCU) __releases_shared(RCU_SCHED)
968 {
969 RCU_LOCKDEP_WARN(!rcu_is_watching(),
970 "rcu_read_unlock_sched() used illegally while idle");
971 rcu_lock_release(&rcu_sched_lock_map);
972 __release_shared(RCU_SCHED);
973 __release_shared(RCU);
974 preempt_enable();
975 }
976
977 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
rcu_read_unlock_sched_notrace(void)978 static inline notrace void rcu_read_unlock_sched_notrace(void)
979 __releases_shared(RCU) __releases_shared(RCU_SCHED)
980 {
981 __release_shared(RCU_SCHED);
982 __release_shared(RCU);
983 preempt_enable_notrace();
984 }
985
rcu_read_lock_dont_migrate(void)986 static __always_inline void rcu_read_lock_dont_migrate(void)
987 __acquires_shared(RCU)
988 {
989 if (IS_ENABLED(CONFIG_PREEMPT_RCU))
990 migrate_disable();
991 rcu_read_lock();
992 }
993
rcu_read_unlock_migrate(void)994 static inline void rcu_read_unlock_migrate(void)
995 __releases_shared(RCU)
996 {
997 rcu_read_unlock();
998 if (IS_ENABLED(CONFIG_PREEMPT_RCU))
999 migrate_enable();
1000 }
1001
1002 /**
1003 * RCU_INIT_POINTER() - initialize an RCU protected pointer
1004 * @p: The pointer to be initialized.
1005 * @v: The value to initialized the pointer to.
1006 *
1007 * Initialize an RCU-protected pointer in special cases where readers
1008 * do not need ordering constraints on the CPU or the compiler. These
1009 * special cases are:
1010 *
1011 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
1012 * 2. The caller has taken whatever steps are required to prevent
1013 * RCU readers from concurrently accessing this pointer *or*
1014 * 3. The referenced data structure has already been exposed to
1015 * readers either at compile time or via rcu_assign_pointer() *and*
1016 *
1017 * a. You have not made *any* reader-visible changes to
1018 * this structure since then *or*
1019 * b. It is OK for readers accessing this structure from its
1020 * new location to see the old state of the structure. (For
1021 * example, the changes were to statistical counters or to
1022 * other state where exact synchronization is not required.)
1023 *
1024 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1025 * result in impossible-to-diagnose memory corruption. As in the structures
1026 * will look OK in crash dumps, but any concurrent RCU readers might
1027 * see pre-initialized values of the referenced data structure. So
1028 * please be very careful how you use RCU_INIT_POINTER()!!!
1029 *
1030 * If you are creating an RCU-protected linked structure that is accessed
1031 * by a single external-to-structure RCU-protected pointer, then you may
1032 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1033 * pointers, but you must use rcu_assign_pointer() to initialize the
1034 * external-to-structure pointer *after* you have completely initialized
1035 * the reader-accessible portions of the linked structure.
1036 *
1037 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1038 * ordering guarantees for either the CPU or the compiler.
1039 */
1040 #define RCU_INIT_POINTER(p, v) \
1041 context_unsafe( \
1042 rcu_check_sparse(p, __rcu); \
1043 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
1044 )
1045
1046 /**
1047 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1048 * @p: The pointer to be initialized.
1049 * @v: The value to initialized the pointer to.
1050 *
1051 * GCC-style initialization for an RCU-protected pointer in a structure field.
1052 */
1053 #define RCU_POINTER_INITIALIZER(p, v) \
1054 .p = RCU_INITIALIZER(v)
1055
1056 /**
1057 * kfree_rcu() - kfree an object after a grace period.
1058 * @ptr: pointer to kfree for double-argument invocations.
1059 * @rhf: the name of the struct rcu_head within the type of @ptr.
1060 *
1061 * Many rcu callbacks functions just call kfree() on the base structure.
1062 * These functions are trivial, but their size adds up, and furthermore
1063 * when they are used in a kernel module, that module must invoke the
1064 * high-latency rcu_barrier() function at module-unload time.
1065 *
1066 * The kfree_rcu() function handles this issue. In order to have a universal
1067 * callback function handling different offsets of rcu_head, the callback needs
1068 * to determine the starting address of the freed object, which can be a large
1069 * kmalloc or vmalloc allocation. To allow simply aligning the pointer down to
1070 * page boundary for those, only offsets up to 4095 bytes can be accommodated.
1071 * If the offset is larger than 4095 bytes, a compile-time error will
1072 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
1073 * either fall back to use of call_rcu() or rearrange the structure to
1074 * position the rcu_head structure into the first 4096 bytes.
1075 *
1076 * The object to be freed can be allocated either by kmalloc(),
1077 * kmalloc_nolock(), or kmem_cache_alloc().
1078 *
1079 * Note that the allowable offset might decrease in the future.
1080 *
1081 * The BUILD_BUG_ON check must not involve any function calls, hence the
1082 * checks are done in macros here.
1083 */
1084 #define kfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf)
1085 #define kvfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf)
1086
1087 /**
1088 * kfree_rcu_mightsleep() - kfree an object after a grace period.
1089 * @ptr: pointer to kfree for single-argument invocations.
1090 *
1091 * When it comes to head-less variant, only one argument
1092 * is passed and that is just a pointer which has to be
1093 * freed after a grace period. Therefore the semantic is
1094 *
1095 * kfree_rcu_mightsleep(ptr);
1096 *
1097 * where @ptr is the pointer to be freed by kvfree().
1098 *
1099 * Please note, head-less way of freeing is permitted to
1100 * use from a context that has to follow might_sleep()
1101 * annotation. Otherwise, please switch and embed the
1102 * rcu_head structure within the type of @ptr.
1103 */
1104 #define kfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr)
1105 #define kvfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr)
1106
1107 /*
1108 * In mm/slab_common.c, no suitable header to include here.
1109 */
1110 void kvfree_call_rcu(struct kvfree_rcu_head *head, void *ptr);
1111 void kfree_call_rcu_nolock(struct kvfree_rcu_head *head, void *ptr);
1112
1113 /*
1114 * The BUILD_BUG_ON() makes sure the rcu_head offset can be handled. See the
1115 * comment of kfree_rcu() for details.
1116 */
1117 #define kvfree_rcu_arg_2(ptr, kvrhf) \
1118 do { \
1119 typeof (ptr) ___p = (ptr); \
1120 struct kvfree_rcu_head *___head; \
1121 \
1122 if (___p) { \
1123 BUILD_BUG_ON(offsetof(typeof(*(ptr)), kvrhf) >= 4096); \
1124 ___head = (struct kvfree_rcu_head *) &(___p)->kvrhf; \
1125 kvfree_call_rcu(___head, (void *) (___p)); \
1126 } \
1127 } while (0)
1128
1129 #define kvfree_rcu_arg_1(ptr) \
1130 do { \
1131 typeof(ptr) ___p = (ptr); \
1132 \
1133 if (___p) \
1134 kvfree_call_rcu(NULL, (void *) (___p)); \
1135 } while (0)
1136
1137 /**
1138 * kfree_rcu_nolock() - a version of kfree_rcu() that can be called in any context.
1139 * @ptr: pointer to kfree for double-argument invocations.
1140 * @kvrhf: the name of the struct kvfree_rcu_head within the type of @ptr.
1141 *
1142 * With KVFREE_RCU_BATCHED, kfree_rcu_nolock() tries hard to free objects
1143 * without any deferred processing, but may still defer freeing.
1144 * Large kmalloc and vmalloc objects are always deferred.
1145 *
1146 * kfree_rcu_nolock() supports 2-arg variant only.
1147 */
1148 #define kfree_rcu_nolock(ptr, kvrhf) \
1149 do { \
1150 typeof (ptr) ___p = (ptr); \
1151 \
1152 if (___p) { \
1153 BUILD_BUG_ON(offsetof(typeof(*(ptr)), kvrhf) >= 4096); \
1154 kfree_call_rcu_nolock(&((___p)->kvrhf), (void *) (___p)); \
1155 } \
1156 } while (0)
1157
1158 /*
1159 * Place this after a lock-acquisition primitive to guarantee that
1160 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
1161 * if the UNLOCK and LOCK are executed by the same CPU or if the
1162 * UNLOCK and LOCK operate on the same lock variable.
1163 */
1164 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
1165 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
1166 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1167 #define smp_mb__after_unlock_lock() do { } while (0)
1168 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1169
1170
1171 /* Has the specified rcu_head structure been handed to call_rcu()? */
1172
1173 /**
1174 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
1175 * @rhp: The rcu_head structure to initialize.
1176 *
1177 * If you intend to invoke rcu_head_after_call_rcu() to test whether a
1178 * given rcu_head structure has already been passed to call_rcu(), then
1179 * you must also invoke this rcu_head_init() function on it just after
1180 * allocating that structure. Calls to this function must not race with
1181 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
1182 */
rcu_head_init(struct rcu_head * rhp)1183 static inline void rcu_head_init(struct rcu_head *rhp)
1184 {
1185 rhp->func = (rcu_callback_t)~0L;
1186 }
1187
1188 /**
1189 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
1190 * @rhp: The rcu_head structure to test.
1191 * @f: The function passed to call_rcu() along with @rhp.
1192 *
1193 * Returns @true if the @rhp has been passed to call_rcu() with @func,
1194 * and @false otherwise. Emits a warning in any other case, including
1195 * the case where @rhp has already been invoked after a grace period.
1196 * Calls to this function must not race with callback invocation. One way
1197 * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
1198 * in an RCU read-side critical section that includes a read-side fetch
1199 * of the pointer to the structure containing @rhp.
1200 */
1201 static inline bool
rcu_head_after_call_rcu(struct rcu_head * rhp,rcu_callback_t f)1202 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1203 {
1204 rcu_callback_t func = READ_ONCE(rhp->func);
1205
1206 if (func == f)
1207 return true;
1208 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1209 return false;
1210 }
1211
1212 /* kernel/ksysfs.c definitions */
1213 extern int rcu_expedited;
1214 extern int rcu_normal;
1215
1216 DEFINE_LOCK_GUARD_0(rcu, rcu_read_lock(), rcu_read_unlock())
1217 DECLARE_LOCK_GUARD_0_ATTRS(rcu, __acquires_shared(RCU), __releases_shared(RCU))
1218
1219 #endif /* __LINUX_RCUPDATE_H */
1220