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."
56 #ifndef CONFIG_TINY_RCU
57 module_param(rcu_expedited, int, 0444);
58 module_param(rcu_normal, int, 0444);
59 static int rcu_normal_after_boot = IS_ENABLED(CONFIG_PREEMPT_RT);
60 #if !defined(CONFIG_PREEMPT_RT) || defined(CONFIG_NO_HZ_FULL)
61 module_param(rcu_normal_after_boot, int, 0444);
63 #endif /* #ifndef CONFIG_TINY_RCU */
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
67 * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
68 * @ret: Best guess answer if lockdep cannot be relied on
70 * Returns true if lockdep must be ignored, in which case ``*ret`` contains
71 * the best guess described below. Otherwise returns false, in which
72 * case ``*ret`` tells the caller nothing and the caller should instead
75 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
76 * RCU-sched read-side critical section. In absence of
77 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
78 * critical section unless it can prove otherwise. Note that disabling
79 * of preemption (including disabling irqs) counts as an RCU-sched
80 * read-side critical section. This is useful for debug checks in functions
81 * that required that they be called within an RCU-sched read-side
84 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
85 * and while lockdep is disabled.
87 * Note that if the CPU is in the idle loop from an RCU point of view (ie:
88 * that we are in the section between ct_idle_enter() and ct_idle_exit())
89 * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
90 * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
91 * in such a section, considering these as in extended quiescent state,
92 * so such a CPU is effectively never in an RCU read-side critical section
93 * regardless of what RCU primitives it invokes. This state of affairs is
94 * required --- we need to keep an RCU-free window in idle where the CPU may
95 * possibly enter into low power mode. This way we can notice an extended
96 * quiescent state to other CPUs that started a grace period. Otherwise
97 * we would delay any grace period as long as we run in the idle task.
99 * Similarly, we avoid claiming an RCU read lock held if the current
102 static bool rcu_read_lock_held_common(bool *ret)
104 if (!debug_lockdep_rcu_enabled()) {
108 if (!rcu_is_watching()) {
112 if (!rcu_lockdep_current_cpu_online()) {
119 int rcu_read_lock_sched_held(void)
123 if (rcu_read_lock_held_common(&ret))
125 return lock_is_held(&rcu_sched_lock_map) || !preemptible();
127 EXPORT_SYMBOL(rcu_read_lock_sched_held);
130 #ifndef CONFIG_TINY_RCU
133 * Should expedited grace-period primitives always fall back to their
134 * non-expedited counterparts? Intended for use within RCU. Note
135 * that if the user specifies both rcu_expedited and rcu_normal, then
136 * rcu_normal wins. (Except during the time period during boot from
137 * when the first task is spawned until the rcu_set_runtime_mode()
138 * core_initcall() is invoked, at which point everything is expedited.)
140 bool rcu_gp_is_normal(void)
142 return READ_ONCE(rcu_normal) &&
143 rcu_scheduler_active != RCU_SCHEDULER_INIT;
145 EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
147 static atomic_t rcu_async_hurry_nesting = ATOMIC_INIT(1);
149 * Should call_rcu() callbacks be processed with urgency or are
150 * they OK being executed with arbitrary delays?
152 bool rcu_async_should_hurry(void)
154 return !IS_ENABLED(CONFIG_RCU_LAZY) ||
155 atomic_read(&rcu_async_hurry_nesting);
157 EXPORT_SYMBOL_GPL(rcu_async_should_hurry);
160 * rcu_async_hurry - Make future async RCU callbacks not lazy.
162 * After a call to this function, future calls to call_rcu()
163 * will be processed in a timely fashion.
165 void rcu_async_hurry(void)
167 if (IS_ENABLED(CONFIG_RCU_LAZY))
168 atomic_inc(&rcu_async_hurry_nesting);
170 EXPORT_SYMBOL_GPL(rcu_async_hurry);
173 * rcu_async_relax - Make future async RCU callbacks lazy.
175 * After a call to this function, future calls to call_rcu()
176 * will be processed in a lazy fashion.
178 void rcu_async_relax(void)
180 if (IS_ENABLED(CONFIG_RCU_LAZY))
181 atomic_dec(&rcu_async_hurry_nesting);
183 EXPORT_SYMBOL_GPL(rcu_async_relax);
185 static atomic_t rcu_expedited_nesting = ATOMIC_INIT(1);
187 * Should normal grace-period primitives be expedited? Intended for
188 * use within RCU. Note that this function takes the rcu_expedited
189 * sysfs/boot variable and rcu_scheduler_active into account as well
190 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
191 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
193 bool rcu_gp_is_expedited(void)
195 return rcu_expedited || atomic_read(&rcu_expedited_nesting);
197 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
200 * rcu_expedite_gp - Expedite future RCU grace periods
202 * After a call to this function, future calls to synchronize_rcu() and
203 * friends act as the corresponding synchronize_rcu_expedited() function
204 * had instead been called.
206 void rcu_expedite_gp(void)
208 atomic_inc(&rcu_expedited_nesting);
210 EXPORT_SYMBOL_GPL(rcu_expedite_gp);
213 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
215 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
216 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
217 * and if the rcu_expedited sysfs/boot parameter is not set, then all
218 * subsequent calls to synchronize_rcu() and friends will return to
219 * their normal non-expedited behavior.
221 void rcu_unexpedite_gp(void)
223 atomic_dec(&rcu_expedited_nesting);
225 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
227 static bool rcu_boot_ended __read_mostly;
230 * Inform RCU of the end of the in-kernel boot sequence.
232 void rcu_end_inkernel_boot(void)
236 if (rcu_normal_after_boot)
237 WRITE_ONCE(rcu_normal, 1);
238 rcu_boot_ended = true;
242 * Let rcutorture know when it is OK to turn it up to eleven.
244 bool rcu_inkernel_boot_has_ended(void)
246 return rcu_boot_ended;
248 EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended);
250 #endif /* #ifndef CONFIG_TINY_RCU */
253 * Test each non-SRCU synchronous grace-period wait API. This is
254 * useful just after a change in mode for these primitives, and
257 void rcu_test_sync_prims(void)
259 if (!IS_ENABLED(CONFIG_PROVE_RCU))
261 pr_info("Running RCU synchronous self tests\n");
263 synchronize_rcu_expedited();
266 #if !defined(CONFIG_TINY_RCU)
269 * Switch to run-time mode once RCU has fully initialized.
271 static int __init rcu_set_runtime_mode(void)
273 rcu_test_sync_prims();
274 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
275 kfree_rcu_scheduler_running();
276 rcu_test_sync_prims();
279 core_initcall(rcu_set_runtime_mode);
281 #endif /* #if !defined(CONFIG_TINY_RCU) */
283 #ifdef CONFIG_DEBUG_LOCK_ALLOC
284 static struct lock_class_key rcu_lock_key;
285 struct lockdep_map rcu_lock_map = {
286 .name = "rcu_read_lock",
287 .key = &rcu_lock_key,
288 .wait_type_outer = LD_WAIT_FREE,
289 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT implies PREEMPT_RCU */
291 EXPORT_SYMBOL_GPL(rcu_lock_map);
293 static struct lock_class_key rcu_bh_lock_key;
294 struct lockdep_map rcu_bh_lock_map = {
295 .name = "rcu_read_lock_bh",
296 .key = &rcu_bh_lock_key,
297 .wait_type_outer = LD_WAIT_FREE,
298 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT makes BH preemptible. */
300 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
302 static struct lock_class_key rcu_sched_lock_key;
303 struct lockdep_map rcu_sched_lock_map = {
304 .name = "rcu_read_lock_sched",
305 .key = &rcu_sched_lock_key,
306 .wait_type_outer = LD_WAIT_FREE,
307 .wait_type_inner = LD_WAIT_SPIN,
309 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
311 // Tell lockdep when RCU callbacks are being invoked.
312 static struct lock_class_key rcu_callback_key;
313 struct lockdep_map rcu_callback_map =
314 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
315 EXPORT_SYMBOL_GPL(rcu_callback_map);
317 noinstr int notrace debug_lockdep_rcu_enabled(void)
319 return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) &&
320 current->lockdep_recursion == 0;
322 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
325 * rcu_read_lock_held() - might we be in RCU read-side critical section?
327 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
328 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
329 * this assumes we are in an RCU read-side critical section unless it can
330 * prove otherwise. This is useful for debug checks in functions that
331 * require that they be called within an RCU read-side critical section.
333 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
334 * and while lockdep is disabled.
336 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
337 * occur in the same context, for example, it is illegal to invoke
338 * rcu_read_unlock() in process context if the matching rcu_read_lock()
339 * was invoked from within an irq handler.
341 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
342 * offline from an RCU perspective, so check for those as well.
344 int rcu_read_lock_held(void)
348 if (rcu_read_lock_held_common(&ret))
350 return lock_is_held(&rcu_lock_map);
352 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
355 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
357 * Check for bottom half being disabled, which covers both the
358 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
359 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
360 * will show the situation. This is useful for debug checks in functions
361 * that require that they be called within an RCU read-side critical
364 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
366 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
367 * offline from an RCU perspective, so check for those as well.
369 int rcu_read_lock_bh_held(void)
373 if (rcu_read_lock_held_common(&ret))
375 return in_softirq() || irqs_disabled();
377 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
379 int rcu_read_lock_any_held(void)
383 if (rcu_read_lock_held_common(&ret))
385 if (lock_is_held(&rcu_lock_map) ||
386 lock_is_held(&rcu_bh_lock_map) ||
387 lock_is_held(&rcu_sched_lock_map))
389 return !preemptible();
391 EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
393 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
396 * wakeme_after_rcu() - Callback function to awaken a task after grace period
397 * @head: Pointer to rcu_head member within rcu_synchronize structure
399 * Awaken the corresponding task now that a grace period has elapsed.
401 void wakeme_after_rcu(struct rcu_head *head)
403 struct rcu_synchronize *rcu;
405 rcu = container_of(head, struct rcu_synchronize, head);
406 complete(&rcu->completion);
408 EXPORT_SYMBOL_GPL(wakeme_after_rcu);
410 void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
411 struct rcu_synchronize *rs_array)
416 /* Initialize and register callbacks for each crcu_array element. */
417 for (i = 0; i < n; i++) {
419 (crcu_array[i] == call_rcu)) {
423 for (j = 0; j < i; j++)
424 if (crcu_array[j] == crcu_array[i])
427 init_rcu_head_on_stack(&rs_array[i].head);
428 init_completion(&rs_array[i].completion);
429 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
433 /* Wait for all callbacks to be invoked. */
434 for (i = 0; i < n; i++) {
436 (crcu_array[i] == call_rcu))
438 for (j = 0; j < i; j++)
439 if (crcu_array[j] == crcu_array[i])
442 wait_for_completion(&rs_array[i].completion);
443 destroy_rcu_head_on_stack(&rs_array[i].head);
447 EXPORT_SYMBOL_GPL(__wait_rcu_gp);
449 void finish_rcuwait(struct rcuwait *w)
451 rcu_assign_pointer(w->task, NULL);
452 __set_current_state(TASK_RUNNING);
454 EXPORT_SYMBOL_GPL(finish_rcuwait);
456 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
457 void init_rcu_head(struct rcu_head *head)
459 debug_object_init(head, &rcuhead_debug_descr);
461 EXPORT_SYMBOL_GPL(init_rcu_head);
463 void destroy_rcu_head(struct rcu_head *head)
465 debug_object_free(head, &rcuhead_debug_descr);
467 EXPORT_SYMBOL_GPL(destroy_rcu_head);
469 static bool rcuhead_is_static_object(void *addr)
475 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
476 * @head: pointer to rcu_head structure to be initialized
478 * This function informs debugobjects of a new rcu_head structure that
479 * has been allocated as an auto variable on the stack. This function
480 * is not required for rcu_head structures that are statically defined or
481 * that are dynamically allocated on the heap. This function has no
482 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
484 void init_rcu_head_on_stack(struct rcu_head *head)
486 debug_object_init_on_stack(head, &rcuhead_debug_descr);
488 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
491 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
492 * @head: pointer to rcu_head structure to be initialized
494 * This function informs debugobjects that an on-stack rcu_head structure
495 * is about to go out of scope. As with init_rcu_head_on_stack(), this
496 * function is not required for rcu_head structures that are statically
497 * defined or that are dynamically allocated on the heap. Also as with
498 * init_rcu_head_on_stack(), this function has no effect for
499 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
501 void destroy_rcu_head_on_stack(struct rcu_head *head)
503 debug_object_free(head, &rcuhead_debug_descr);
505 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
507 const struct debug_obj_descr rcuhead_debug_descr = {
509 .is_static_object = rcuhead_is_static_object,
511 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
512 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
514 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
515 void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
517 unsigned long c_old, unsigned long c)
519 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
521 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
523 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
527 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
528 /* Get rcutorture access to sched_setaffinity(). */
529 long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
533 ret = sched_setaffinity(pid, in_mask);
534 WARN_ONCE(ret, "%s: sched_setaffinity() returned %d\n", __func__, ret);
537 EXPORT_SYMBOL_GPL(rcutorture_sched_setaffinity);
540 #ifdef CONFIG_RCU_STALL_COMMON
541 int rcu_cpu_stall_ftrace_dump __read_mostly;
542 module_param(rcu_cpu_stall_ftrace_dump, int, 0644);
543 int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings.
544 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
545 module_param(rcu_cpu_stall_suppress, int, 0644);
546 int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
547 module_param(rcu_cpu_stall_timeout, int, 0644);
548 int rcu_exp_cpu_stall_timeout __read_mostly = CONFIG_RCU_EXP_CPU_STALL_TIMEOUT;
549 module_param(rcu_exp_cpu_stall_timeout, int, 0644);
550 int rcu_cpu_stall_cputime __read_mostly = IS_ENABLED(CONFIG_RCU_CPU_STALL_CPUTIME);
551 module_param(rcu_cpu_stall_cputime, int, 0644);
552 bool rcu_exp_stall_task_details __read_mostly;
553 module_param(rcu_exp_stall_task_details, bool, 0644);
554 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
556 // Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
557 // warnings. Also used by rcutorture even if stall warnings are excluded.
558 int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls.
559 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot);
560 module_param(rcu_cpu_stall_suppress_at_boot, int, 0444);
563 * get_completed_synchronize_rcu - Return a pre-completed polled state cookie
565 * Returns a value that will always be treated by functions like
566 * poll_state_synchronize_rcu() as a cookie whose grace period has already
569 unsigned long get_completed_synchronize_rcu(void)
571 return RCU_GET_STATE_COMPLETED;
573 EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu);
575 #ifdef CONFIG_PROVE_RCU
578 * Early boot self test parameters.
580 static bool rcu_self_test;
581 module_param(rcu_self_test, bool, 0444);
583 static int rcu_self_test_counter;
585 static void test_callback(struct rcu_head *r)
587 rcu_self_test_counter++;
588 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
591 DEFINE_STATIC_SRCU(early_srcu);
592 static unsigned long early_srcu_cookie;
594 struct early_boot_kfree_rcu {
598 static void early_boot_test_call_rcu(void)
600 static struct rcu_head head;
602 static struct rcu_head shead;
603 struct early_boot_kfree_rcu *rhp;
605 idx = srcu_down_read(&early_srcu);
606 srcu_up_read(&early_srcu, idx);
607 call_rcu(&head, test_callback);
608 early_srcu_cookie = start_poll_synchronize_srcu(&early_srcu);
609 call_srcu(&early_srcu, &shead, test_callback);
610 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
611 if (!WARN_ON_ONCE(!rhp))
615 void rcu_early_boot_tests(void)
617 pr_info("Running RCU self tests\n");
620 early_boot_test_call_rcu();
621 rcu_test_sync_prims();
624 static int rcu_verify_early_boot_tests(void)
627 int early_boot_test_counter = 0;
630 early_boot_test_counter++;
632 early_boot_test_counter++;
633 srcu_barrier(&early_srcu);
634 WARN_ON_ONCE(!poll_state_synchronize_srcu(&early_srcu, early_srcu_cookie));
635 cleanup_srcu_struct(&early_srcu);
637 if (rcu_self_test_counter != early_boot_test_counter) {
644 late_initcall(rcu_verify_early_boot_tests);
646 void rcu_early_boot_tests(void) {}
647 #endif /* CONFIG_PROVE_RCU */
651 #ifndef CONFIG_TINY_RCU
654 * Print any significant non-default boot-time settings.
656 void __init rcupdate_announce_bootup_oddness(void)
659 pr_info("\tNo expedited grace period (rcu_normal).\n");
660 else if (rcu_normal_after_boot)
661 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
662 else if (rcu_expedited)
663 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
664 if (rcu_cpu_stall_suppress)
665 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
666 if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT)
667 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout);
668 rcu_tasks_bootup_oddness();
671 #endif /* #ifndef CONFIG_TINY_RCU */