1 // SPDX-License-Identifier: GPL-2.0+
3 * Read-Copy Update mechanism for mutual exclusion
5 * Copyright IBM Corporation, 2001
7 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
8 * Manfred Spraul <manfred@colorfullife.com>
10 * Based on the original work by Paul McKenney <paulmck@linux.ibm.com>
11 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
13 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
14 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
16 * For detailed explanation of Read-Copy Update mechanism see -
17 * http://lse.sourceforge.net/locking/rcupdate.html
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/init.h>
23 #include <linux/spinlock.h>
24 #include <linux/smp.h>
25 #include <linux/interrupt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/debug.h>
28 #include <linux/atomic.h>
29 #include <linux/bitops.h>
30 #include <linux/percpu.h>
31 #include <linux/notifier.h>
32 #include <linux/cpu.h>
33 #include <linux/mutex.h>
34 #include <linux/export.h>
35 #include <linux/hardirq.h>
36 #include <linux/delay.h>
37 #include <linux/moduleparam.h>
38 #include <linux/kthread.h>
39 #include <linux/tick.h>
40 #include <linux/rcupdate_wait.h>
41 #include <linux/sched/isolation.h>
42 #include <linux/kprobes.h>
43 #include <linux/slab.h>
44 #include <linux/irq_work.h>
45 #include <linux/rcupdate_trace.h>
47 #define CREATE_TRACE_POINTS
51 #ifdef MODULE_PARAM_PREFIX
52 #undef MODULE_PARAM_PREFIX
54 #define MODULE_PARAM_PREFIX "rcupdate."
57 #define data_race(expr) \
62 #ifndef ASSERT_EXCLUSIVE_WRITER
63 #define ASSERT_EXCLUSIVE_WRITER(var) do { } while (0)
65 #ifndef ASSERT_EXCLUSIVE_ACCESS
66 #define ASSERT_EXCLUSIVE_ACCESS(var) do { } while (0)
69 #ifndef CONFIG_TINY_RCU
70 module_param(rcu_expedited
, int, 0);
71 module_param(rcu_normal
, int, 0);
72 static int rcu_normal_after_boot
;
73 module_param(rcu_normal_after_boot
, int, 0);
74 #endif /* #ifndef CONFIG_TINY_RCU */
76 #ifdef CONFIG_DEBUG_LOCK_ALLOC
78 * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
79 * @ret: Best guess answer if lockdep cannot be relied on
81 * Returns true if lockdep must be ignored, in which case ``*ret`` contains
82 * the best guess described below. Otherwise returns false, in which
83 * case ``*ret`` tells the caller nothing and the caller should instead
86 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
87 * RCU-sched read-side critical section. In absence of
88 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
89 * critical section unless it can prove otherwise. Note that disabling
90 * of preemption (including disabling irqs) counts as an RCU-sched
91 * read-side critical section. This is useful for debug checks in functions
92 * that required that they be called within an RCU-sched read-side
95 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
96 * and while lockdep is disabled.
98 * Note that if the CPU is in the idle loop from an RCU point of view (ie:
99 * that we are in the section between rcu_idle_enter() and rcu_idle_exit())
100 * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
101 * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
102 * in such a section, considering these as in extended quiescent state,
103 * so such a CPU is effectively never in an RCU read-side critical section
104 * regardless of what RCU primitives it invokes. This state of affairs is
105 * required --- we need to keep an RCU-free window in idle where the CPU may
106 * possibly enter into low power mode. This way we can notice an extended
107 * quiescent state to other CPUs that started a grace period. Otherwise
108 * we would delay any grace period as long as we run in the idle task.
110 * Similarly, we avoid claiming an RCU read lock held if the current
113 static bool rcu_read_lock_held_common(bool *ret
)
115 if (!debug_lockdep_rcu_enabled()) {
119 if (!rcu_is_watching()) {
123 if (!rcu_lockdep_current_cpu_online()) {
130 int rcu_read_lock_sched_held(void)
134 if (rcu_read_lock_held_common(&ret
))
136 return lock_is_held(&rcu_sched_lock_map
) || !preemptible();
138 EXPORT_SYMBOL(rcu_read_lock_sched_held
);
141 #ifndef CONFIG_TINY_RCU
144 * Should expedited grace-period primitives always fall back to their
145 * non-expedited counterparts? Intended for use within RCU. Note
146 * that if the user specifies both rcu_expedited and rcu_normal, then
147 * rcu_normal wins. (Except during the time period during boot from
148 * when the first task is spawned until the rcu_set_runtime_mode()
149 * core_initcall() is invoked, at which point everything is expedited.)
151 bool rcu_gp_is_normal(void)
153 return READ_ONCE(rcu_normal
) &&
154 rcu_scheduler_active
!= RCU_SCHEDULER_INIT
;
156 EXPORT_SYMBOL_GPL(rcu_gp_is_normal
);
158 static atomic_t rcu_expedited_nesting
= ATOMIC_INIT(1);
161 * Should normal grace-period primitives be expedited? Intended for
162 * use within RCU. Note that this function takes the rcu_expedited
163 * sysfs/boot variable and rcu_scheduler_active into account as well
164 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
165 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
167 bool rcu_gp_is_expedited(void)
169 return rcu_expedited
|| atomic_read(&rcu_expedited_nesting
);
171 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited
);
174 * rcu_expedite_gp - Expedite future RCU grace periods
176 * After a call to this function, future calls to synchronize_rcu() and
177 * friends act as the corresponding synchronize_rcu_expedited() function
178 * had instead been called.
180 void rcu_expedite_gp(void)
182 atomic_inc(&rcu_expedited_nesting
);
184 EXPORT_SYMBOL_GPL(rcu_expedite_gp
);
187 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
189 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
190 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
191 * and if the rcu_expedited sysfs/boot parameter is not set, then all
192 * subsequent calls to synchronize_rcu() and friends will return to
193 * their normal non-expedited behavior.
195 void rcu_unexpedite_gp(void)
197 atomic_dec(&rcu_expedited_nesting
);
199 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp
);
201 static bool rcu_boot_ended __read_mostly
;
204 * Inform RCU of the end of the in-kernel boot sequence.
206 void rcu_end_inkernel_boot(void)
209 if (rcu_normal_after_boot
)
210 WRITE_ONCE(rcu_normal
, 1);
211 rcu_boot_ended
= true;
215 * Let rcutorture know when it is OK to turn it up to eleven.
217 bool rcu_inkernel_boot_has_ended(void)
219 return rcu_boot_ended
;
221 EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended
);
223 #endif /* #ifndef CONFIG_TINY_RCU */
226 * Test each non-SRCU synchronous grace-period wait API. This is
227 * useful just after a change in mode for these primitives, and
230 void rcu_test_sync_prims(void)
232 if (!IS_ENABLED(CONFIG_PROVE_RCU
))
235 synchronize_rcu_expedited();
238 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU)
241 * Switch to run-time mode once RCU has fully initialized.
243 static int __init
rcu_set_runtime_mode(void)
245 rcu_test_sync_prims();
246 rcu_scheduler_active
= RCU_SCHEDULER_RUNNING
;
247 kfree_rcu_scheduler_running();
248 rcu_test_sync_prims();
251 core_initcall(rcu_set_runtime_mode
);
253 #endif /* #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU) */
255 #ifdef CONFIG_DEBUG_LOCK_ALLOC
256 static struct lock_class_key rcu_lock_key
;
257 struct lockdep_map rcu_lock_map
= {
258 .name
= "rcu_read_lock",
259 .key
= &rcu_lock_key
,
260 .wait_type_outer
= LD_WAIT_FREE
,
261 .wait_type_inner
= LD_WAIT_CONFIG
, /* XXX PREEMPT_RCU ? */
263 EXPORT_SYMBOL_GPL(rcu_lock_map
);
265 static struct lock_class_key rcu_bh_lock_key
;
266 struct lockdep_map rcu_bh_lock_map
= {
267 .name
= "rcu_read_lock_bh",
268 .key
= &rcu_bh_lock_key
,
269 .wait_type_outer
= LD_WAIT_FREE
,
270 .wait_type_inner
= LD_WAIT_CONFIG
, /* PREEMPT_LOCK also makes BH preemptible */
272 EXPORT_SYMBOL_GPL(rcu_bh_lock_map
);
274 static struct lock_class_key rcu_sched_lock_key
;
275 struct lockdep_map rcu_sched_lock_map
= {
276 .name
= "rcu_read_lock_sched",
277 .key
= &rcu_sched_lock_key
,
278 .wait_type_outer
= LD_WAIT_FREE
,
279 .wait_type_inner
= LD_WAIT_SPIN
,
281 EXPORT_SYMBOL_GPL(rcu_sched_lock_map
);
283 // Tell lockdep when RCU callbacks are being invoked.
284 static struct lock_class_key rcu_callback_key
;
285 struct lockdep_map rcu_callback_map
=
286 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key
);
287 EXPORT_SYMBOL_GPL(rcu_callback_map
);
289 noinstr
int notrace
debug_lockdep_rcu_enabled(void)
291 return rcu_scheduler_active
!= RCU_SCHEDULER_INACTIVE
&& debug_locks
&&
292 current
->lockdep_recursion
== 0;
294 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled
);
297 * rcu_read_lock_held() - might we be in RCU read-side critical section?
299 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
300 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
301 * this assumes we are in an RCU read-side critical section unless it can
302 * prove otherwise. This is useful for debug checks in functions that
303 * require that they be called within an RCU read-side critical section.
305 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
306 * and while lockdep is disabled.
308 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
309 * occur in the same context, for example, it is illegal to invoke
310 * rcu_read_unlock() in process context if the matching rcu_read_lock()
311 * was invoked from within an irq handler.
313 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
314 * offline from an RCU perspective, so check for those as well.
316 int rcu_read_lock_held(void)
320 if (rcu_read_lock_held_common(&ret
))
322 return lock_is_held(&rcu_lock_map
);
324 EXPORT_SYMBOL_GPL(rcu_read_lock_held
);
327 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
329 * Check for bottom half being disabled, which covers both the
330 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
331 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
332 * will show the situation. This is useful for debug checks in functions
333 * that require that they be called within an RCU read-side critical
336 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
338 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
339 * offline from an RCU perspective, so check for those as well.
341 int rcu_read_lock_bh_held(void)
345 if (rcu_read_lock_held_common(&ret
))
347 return in_softirq() || irqs_disabled();
349 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held
);
351 int rcu_read_lock_any_held(void)
355 if (rcu_read_lock_held_common(&ret
))
357 if (lock_is_held(&rcu_lock_map
) ||
358 lock_is_held(&rcu_bh_lock_map
) ||
359 lock_is_held(&rcu_sched_lock_map
))
361 return !preemptible();
363 EXPORT_SYMBOL_GPL(rcu_read_lock_any_held
);
365 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
368 * wakeme_after_rcu() - Callback function to awaken a task after grace period
369 * @head: Pointer to rcu_head member within rcu_synchronize structure
371 * Awaken the corresponding task now that a grace period has elapsed.
373 void wakeme_after_rcu(struct rcu_head
*head
)
375 struct rcu_synchronize
*rcu
;
377 rcu
= container_of(head
, struct rcu_synchronize
, head
);
378 complete(&rcu
->completion
);
380 EXPORT_SYMBOL_GPL(wakeme_after_rcu
);
382 void __wait_rcu_gp(bool checktiny
, int n
, call_rcu_func_t
*crcu_array
,
383 struct rcu_synchronize
*rs_array
)
388 /* Initialize and register callbacks for each crcu_array element. */
389 for (i
= 0; i
< n
; i
++) {
391 (crcu_array
[i
] == call_rcu
)) {
395 for (j
= 0; j
< i
; j
++)
396 if (crcu_array
[j
] == crcu_array
[i
])
399 init_rcu_head_on_stack(&rs_array
[i
].head
);
400 init_completion(&rs_array
[i
].completion
);
401 (crcu_array
[i
])(&rs_array
[i
].head
, wakeme_after_rcu
);
405 /* Wait for all callbacks to be invoked. */
406 for (i
= 0; i
< n
; i
++) {
408 (crcu_array
[i
] == call_rcu
))
410 for (j
= 0; j
< i
; j
++)
411 if (crcu_array
[j
] == crcu_array
[i
])
414 wait_for_completion(&rs_array
[i
].completion
);
415 destroy_rcu_head_on_stack(&rs_array
[i
].head
);
419 EXPORT_SYMBOL_GPL(__wait_rcu_gp
);
421 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
422 void init_rcu_head(struct rcu_head
*head
)
424 debug_object_init(head
, &rcuhead_debug_descr
);
426 EXPORT_SYMBOL_GPL(init_rcu_head
);
428 void destroy_rcu_head(struct rcu_head
*head
)
430 debug_object_free(head
, &rcuhead_debug_descr
);
432 EXPORT_SYMBOL_GPL(destroy_rcu_head
);
434 static bool rcuhead_is_static_object(void *addr
)
440 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
441 * @head: pointer to rcu_head structure to be initialized
443 * This function informs debugobjects of a new rcu_head structure that
444 * has been allocated as an auto variable on the stack. This function
445 * is not required for rcu_head structures that are statically defined or
446 * that are dynamically allocated on the heap. This function has no
447 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
449 void init_rcu_head_on_stack(struct rcu_head
*head
)
451 debug_object_init_on_stack(head
, &rcuhead_debug_descr
);
453 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack
);
456 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
457 * @head: pointer to rcu_head structure to be initialized
459 * This function informs debugobjects that an on-stack rcu_head structure
460 * is about to go out of scope. As with init_rcu_head_on_stack(), this
461 * function is not required for rcu_head structures that are statically
462 * defined or that are dynamically allocated on the heap. Also as with
463 * init_rcu_head_on_stack(), this function has no effect for
464 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
466 void destroy_rcu_head_on_stack(struct rcu_head
*head
)
468 debug_object_free(head
, &rcuhead_debug_descr
);
470 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack
);
472 struct debug_obj_descr rcuhead_debug_descr
= {
474 .is_static_object
= rcuhead_is_static_object
,
476 EXPORT_SYMBOL_GPL(rcuhead_debug_descr
);
477 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
479 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
480 void do_trace_rcu_torture_read(const char *rcutorturename
, struct rcu_head
*rhp
,
482 unsigned long c_old
, unsigned long c
)
484 trace_rcu_torture_read(rcutorturename
, rhp
, secs
, c_old
, c
);
486 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read
);
488 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
492 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
493 /* Get rcutorture access to sched_setaffinity(). */
494 long rcutorture_sched_setaffinity(pid_t pid
, const struct cpumask
*in_mask
)
498 ret
= sched_setaffinity(pid
, in_mask
);
499 WARN_ONCE(ret
, "%s: sched_setaffinity() returned %d\n", __func__
, ret
);
502 EXPORT_SYMBOL_GPL(rcutorture_sched_setaffinity
);
505 #ifdef CONFIG_RCU_STALL_COMMON
506 int rcu_cpu_stall_ftrace_dump __read_mostly
;
507 module_param(rcu_cpu_stall_ftrace_dump
, int, 0644);
508 int rcu_cpu_stall_suppress __read_mostly
; // !0 = suppress stall warnings.
509 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress
);
510 module_param(rcu_cpu_stall_suppress
, int, 0644);
511 int rcu_cpu_stall_timeout __read_mostly
= CONFIG_RCU_CPU_STALL_TIMEOUT
;
512 module_param(rcu_cpu_stall_timeout
, int, 0644);
513 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
515 // Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
516 // warnings. Also used by rcutorture even if stall warnings are excluded.
517 int rcu_cpu_stall_suppress_at_boot __read_mostly
; // !0 = suppress boot stalls.
518 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot
);
519 module_param(rcu_cpu_stall_suppress_at_boot
, int, 0444);
521 #ifdef CONFIG_PROVE_RCU
524 * Early boot self test parameters.
526 static bool rcu_self_test
;
527 module_param(rcu_self_test
, bool, 0444);
529 static int rcu_self_test_counter
;
531 static void test_callback(struct rcu_head
*r
)
533 rcu_self_test_counter
++;
534 pr_info("RCU test callback executed %d\n", rcu_self_test_counter
);
537 DEFINE_STATIC_SRCU(early_srcu
);
539 struct early_boot_kfree_rcu
{
543 static void early_boot_test_call_rcu(void)
545 static struct rcu_head head
;
546 static struct rcu_head shead
;
547 struct early_boot_kfree_rcu
*rhp
;
549 call_rcu(&head
, test_callback
);
550 if (IS_ENABLED(CONFIG_SRCU
))
551 call_srcu(&early_srcu
, &shead
, test_callback
);
552 rhp
= kmalloc(sizeof(*rhp
), GFP_KERNEL
);
553 if (!WARN_ON_ONCE(!rhp
))
557 void rcu_early_boot_tests(void)
559 pr_info("Running RCU self tests\n");
562 early_boot_test_call_rcu();
563 rcu_test_sync_prims();
566 static int rcu_verify_early_boot_tests(void)
569 int early_boot_test_counter
= 0;
572 early_boot_test_counter
++;
574 if (IS_ENABLED(CONFIG_SRCU
)) {
575 early_boot_test_counter
++;
576 srcu_barrier(&early_srcu
);
579 if (rcu_self_test_counter
!= early_boot_test_counter
) {
586 late_initcall(rcu_verify_early_boot_tests
);
588 void rcu_early_boot_tests(void) {}
589 #endif /* CONFIG_PROVE_RCU */
593 #ifndef CONFIG_TINY_RCU
596 * Print any significant non-default boot-time settings.
598 void __init
rcupdate_announce_bootup_oddness(void)
601 pr_info("\tNo expedited grace period (rcu_normal).\n");
602 else if (rcu_normal_after_boot
)
603 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
604 else if (rcu_expedited
)
605 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
606 if (rcu_cpu_stall_suppress
)
607 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
608 if (rcu_cpu_stall_timeout
!= CONFIG_RCU_CPU_STALL_TIMEOUT
)
609 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout
);
610 rcu_tasks_bootup_oddness();
613 #endif /* #ifndef CONFIG_TINY_RCU */