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_expedited_nesting = ATOMIC_INIT(1);
150 * Should normal grace-period primitives be expedited? Intended for
151 * use within RCU. Note that this function takes the rcu_expedited
152 * sysfs/boot variable and rcu_scheduler_active into account as well
153 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
154 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
156 bool rcu_gp_is_expedited(void)
158 return rcu_expedited || atomic_read(&rcu_expedited_nesting);
160 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
163 * rcu_expedite_gp - Expedite future RCU grace periods
165 * After a call to this function, future calls to synchronize_rcu() and
166 * friends act as the corresponding synchronize_rcu_expedited() function
167 * had instead been called.
169 void rcu_expedite_gp(void)
171 atomic_inc(&rcu_expedited_nesting);
173 EXPORT_SYMBOL_GPL(rcu_expedite_gp);
176 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
178 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
179 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
180 * and if the rcu_expedited sysfs/boot parameter is not set, then all
181 * subsequent calls to synchronize_rcu() and friends will return to
182 * their normal non-expedited behavior.
184 void rcu_unexpedite_gp(void)
186 atomic_dec(&rcu_expedited_nesting);
188 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
190 static bool rcu_boot_ended __read_mostly;
193 * Inform RCU of the end of the in-kernel boot sequence.
195 void rcu_end_inkernel_boot(void)
198 if (rcu_normal_after_boot)
199 WRITE_ONCE(rcu_normal, 1);
200 rcu_boot_ended = true;
204 * Let rcutorture know when it is OK to turn it up to eleven.
206 bool rcu_inkernel_boot_has_ended(void)
208 return rcu_boot_ended;
210 EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended);
212 #endif /* #ifndef CONFIG_TINY_RCU */
215 * Test each non-SRCU synchronous grace-period wait API. This is
216 * useful just after a change in mode for these primitives, and
219 void rcu_test_sync_prims(void)
221 if (!IS_ENABLED(CONFIG_PROVE_RCU))
224 synchronize_rcu_expedited();
227 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU)
230 * Switch to run-time mode once RCU has fully initialized.
232 static int __init rcu_set_runtime_mode(void)
234 rcu_test_sync_prims();
235 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
236 kfree_rcu_scheduler_running();
237 rcu_test_sync_prims();
240 core_initcall(rcu_set_runtime_mode);
242 #endif /* #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU) */
244 #ifdef CONFIG_DEBUG_LOCK_ALLOC
245 static struct lock_class_key rcu_lock_key;
246 struct lockdep_map rcu_lock_map = {
247 .name = "rcu_read_lock",
248 .key = &rcu_lock_key,
249 .wait_type_outer = LD_WAIT_FREE,
250 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT implies PREEMPT_RCU */
252 EXPORT_SYMBOL_GPL(rcu_lock_map);
254 static struct lock_class_key rcu_bh_lock_key;
255 struct lockdep_map rcu_bh_lock_map = {
256 .name = "rcu_read_lock_bh",
257 .key = &rcu_bh_lock_key,
258 .wait_type_outer = LD_WAIT_FREE,
259 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT makes BH preemptible. */
261 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
263 static struct lock_class_key rcu_sched_lock_key;
264 struct lockdep_map rcu_sched_lock_map = {
265 .name = "rcu_read_lock_sched",
266 .key = &rcu_sched_lock_key,
267 .wait_type_outer = LD_WAIT_FREE,
268 .wait_type_inner = LD_WAIT_SPIN,
270 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
272 // Tell lockdep when RCU callbacks are being invoked.
273 static struct lock_class_key rcu_callback_key;
274 struct lockdep_map rcu_callback_map =
275 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
276 EXPORT_SYMBOL_GPL(rcu_callback_map);
278 noinstr int notrace debug_lockdep_rcu_enabled(void)
280 return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) &&
281 current->lockdep_recursion == 0;
283 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
286 * rcu_read_lock_held() - might we be in RCU read-side critical section?
288 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
289 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
290 * this assumes we are in an RCU read-side critical section unless it can
291 * prove otherwise. This is useful for debug checks in functions that
292 * require that they be called within an RCU read-side critical section.
294 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
295 * and while lockdep is disabled.
297 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
298 * occur in the same context, for example, it is illegal to invoke
299 * rcu_read_unlock() in process context if the matching rcu_read_lock()
300 * was invoked from within an irq handler.
302 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
303 * offline from an RCU perspective, so check for those as well.
305 int rcu_read_lock_held(void)
309 if (rcu_read_lock_held_common(&ret))
311 return lock_is_held(&rcu_lock_map);
313 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
316 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
318 * Check for bottom half being disabled, which covers both the
319 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
320 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
321 * will show the situation. This is useful for debug checks in functions
322 * that require that they be called within an RCU read-side critical
325 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
327 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
328 * offline from an RCU perspective, so check for those as well.
330 int rcu_read_lock_bh_held(void)
334 if (rcu_read_lock_held_common(&ret))
336 return in_softirq() || irqs_disabled();
338 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
340 int rcu_read_lock_any_held(void)
344 if (rcu_read_lock_held_common(&ret))
346 if (lock_is_held(&rcu_lock_map) ||
347 lock_is_held(&rcu_bh_lock_map) ||
348 lock_is_held(&rcu_sched_lock_map))
350 return !preemptible();
352 EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
354 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
357 * wakeme_after_rcu() - Callback function to awaken a task after grace period
358 * @head: Pointer to rcu_head member within rcu_synchronize structure
360 * Awaken the corresponding task now that a grace period has elapsed.
362 void wakeme_after_rcu(struct rcu_head *head)
364 struct rcu_synchronize *rcu;
366 rcu = container_of(head, struct rcu_synchronize, head);
367 complete(&rcu->completion);
369 EXPORT_SYMBOL_GPL(wakeme_after_rcu);
371 void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
372 struct rcu_synchronize *rs_array)
377 /* Initialize and register callbacks for each crcu_array element. */
378 for (i = 0; i < n; i++) {
380 (crcu_array[i] == call_rcu)) {
384 for (j = 0; j < i; j++)
385 if (crcu_array[j] == crcu_array[i])
388 init_rcu_head_on_stack(&rs_array[i].head);
389 init_completion(&rs_array[i].completion);
390 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
394 /* Wait for all callbacks to be invoked. */
395 for (i = 0; i < n; i++) {
397 (crcu_array[i] == call_rcu))
399 for (j = 0; j < i; j++)
400 if (crcu_array[j] == crcu_array[i])
403 wait_for_completion(&rs_array[i].completion);
404 destroy_rcu_head_on_stack(&rs_array[i].head);
408 EXPORT_SYMBOL_GPL(__wait_rcu_gp);
410 void finish_rcuwait(struct rcuwait *w)
412 rcu_assign_pointer(w->task, NULL);
413 __set_current_state(TASK_RUNNING);
415 EXPORT_SYMBOL_GPL(finish_rcuwait);
417 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
418 void init_rcu_head(struct rcu_head *head)
420 debug_object_init(head, &rcuhead_debug_descr);
422 EXPORT_SYMBOL_GPL(init_rcu_head);
424 void destroy_rcu_head(struct rcu_head *head)
426 debug_object_free(head, &rcuhead_debug_descr);
428 EXPORT_SYMBOL_GPL(destroy_rcu_head);
430 static bool rcuhead_is_static_object(void *addr)
436 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
437 * @head: pointer to rcu_head structure to be initialized
439 * This function informs debugobjects of a new rcu_head structure that
440 * has been allocated as an auto variable on the stack. This function
441 * is not required for rcu_head structures that are statically defined or
442 * that are dynamically allocated on the heap. This function has no
443 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
445 void init_rcu_head_on_stack(struct rcu_head *head)
447 debug_object_init_on_stack(head, &rcuhead_debug_descr);
449 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
452 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
453 * @head: pointer to rcu_head structure to be initialized
455 * This function informs debugobjects that an on-stack rcu_head structure
456 * is about to go out of scope. As with init_rcu_head_on_stack(), this
457 * function is not required for rcu_head structures that are statically
458 * defined or that are dynamically allocated on the heap. Also as with
459 * init_rcu_head_on_stack(), this function has no effect for
460 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
462 void destroy_rcu_head_on_stack(struct rcu_head *head)
464 debug_object_free(head, &rcuhead_debug_descr);
466 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
468 const struct debug_obj_descr rcuhead_debug_descr = {
470 .is_static_object = rcuhead_is_static_object,
472 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
473 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
475 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
476 void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
478 unsigned long c_old, unsigned long c)
480 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
482 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
484 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
488 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
489 /* Get rcutorture access to sched_setaffinity(). */
490 long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
494 ret = sched_setaffinity(pid, in_mask);
495 WARN_ONCE(ret, "%s: sched_setaffinity() returned %d\n", __func__, ret);
498 EXPORT_SYMBOL_GPL(rcutorture_sched_setaffinity);
501 #ifdef CONFIG_RCU_STALL_COMMON
502 int rcu_cpu_stall_ftrace_dump __read_mostly;
503 module_param(rcu_cpu_stall_ftrace_dump, int, 0644);
504 int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings.
505 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
506 module_param(rcu_cpu_stall_suppress, int, 0644);
507 int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
508 module_param(rcu_cpu_stall_timeout, int, 0644);
509 int rcu_exp_cpu_stall_timeout __read_mostly = CONFIG_RCU_EXP_CPU_STALL_TIMEOUT;
510 module_param(rcu_exp_cpu_stall_timeout, int, 0644);
511 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
513 // Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
514 // warnings. Also used by rcutorture even if stall warnings are excluded.
515 int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls.
516 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot);
517 module_param(rcu_cpu_stall_suppress_at_boot, int, 0444);
520 * get_completed_synchronize_rcu - Return a pre-completed polled state cookie
522 * Returns a value that will always be treated by functions like
523 * poll_state_synchronize_rcu() as a cookie whose grace period has already
526 unsigned long get_completed_synchronize_rcu(void)
528 return RCU_GET_STATE_COMPLETED;
530 EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu);
532 #ifdef CONFIG_PROVE_RCU
535 * Early boot self test parameters.
537 static bool rcu_self_test;
538 module_param(rcu_self_test, bool, 0444);
540 static int rcu_self_test_counter;
542 static void test_callback(struct rcu_head *r)
544 rcu_self_test_counter++;
545 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
548 DEFINE_STATIC_SRCU(early_srcu);
549 static unsigned long early_srcu_cookie;
551 struct early_boot_kfree_rcu {
555 static void early_boot_test_call_rcu(void)
557 static struct rcu_head head;
558 static struct rcu_head shead;
559 struct early_boot_kfree_rcu *rhp;
561 call_rcu(&head, test_callback);
562 if (IS_ENABLED(CONFIG_SRCU)) {
563 early_srcu_cookie = start_poll_synchronize_srcu(&early_srcu);
564 call_srcu(&early_srcu, &shead, test_callback);
566 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
567 if (!WARN_ON_ONCE(!rhp))
571 void rcu_early_boot_tests(void)
573 pr_info("Running RCU self tests\n");
576 early_boot_test_call_rcu();
577 rcu_test_sync_prims();
580 static int rcu_verify_early_boot_tests(void)
583 int early_boot_test_counter = 0;
586 early_boot_test_counter++;
588 if (IS_ENABLED(CONFIG_SRCU)) {
589 early_boot_test_counter++;
590 srcu_barrier(&early_srcu);
591 WARN_ON_ONCE(!poll_state_synchronize_srcu(&early_srcu, early_srcu_cookie));
594 if (rcu_self_test_counter != early_boot_test_counter) {
601 late_initcall(rcu_verify_early_boot_tests);
603 void rcu_early_boot_tests(void) {}
604 #endif /* CONFIG_PROVE_RCU */
608 #ifndef CONFIG_TINY_RCU
611 * Print any significant non-default boot-time settings.
613 void __init rcupdate_announce_bootup_oddness(void)
616 pr_info("\tNo expedited grace period (rcu_normal).\n");
617 else if (rcu_normal_after_boot)
618 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
619 else if (rcu_expedited)
620 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
621 if (rcu_cpu_stall_suppress)
622 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
623 if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT)
624 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout);
625 rcu_tasks_bootup_oddness();
628 #endif /* #ifndef CONFIG_TINY_RCU */