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>
46 #define CREATE_TRACE_POINTS
50 #ifdef MODULE_PARAM_PREFIX
51 #undef MODULE_PARAM_PREFIX
53 #define MODULE_PARAM_PREFIX "rcupdate."
56 #define data_race(expr) \
61 #ifndef ASSERT_EXCLUSIVE_WRITER
62 #define ASSERT_EXCLUSIVE_WRITER(var) do { } while (0)
64 #ifndef ASSERT_EXCLUSIVE_ACCESS
65 #define ASSERT_EXCLUSIVE_ACCESS(var) do { } while (0)
68 #ifndef CONFIG_TINY_RCU
69 module_param(rcu_expedited
, int, 0);
70 module_param(rcu_normal
, int, 0);
71 static int rcu_normal_after_boot
;
72 module_param(rcu_normal_after_boot
, int, 0);
73 #endif /* #ifndef CONFIG_TINY_RCU */
75 #ifdef CONFIG_DEBUG_LOCK_ALLOC
77 * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
78 * @ret: Best guess answer if lockdep cannot be relied on
80 * Returns true if lockdep must be ignored, in which case ``*ret`` contains
81 * the best guess described below. Otherwise returns false, in which
82 * case ``*ret`` tells the caller nothing and the caller should instead
85 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
86 * RCU-sched read-side critical section. In absence of
87 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
88 * critical section unless it can prove otherwise. Note that disabling
89 * of preemption (including disabling irqs) counts as an RCU-sched
90 * read-side critical section. This is useful for debug checks in functions
91 * that required that they be called within an RCU-sched read-side
94 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
95 * and while lockdep is disabled.
97 * Note that if the CPU is in the idle loop from an RCU point of view (ie:
98 * that we are in the section between rcu_idle_enter() and rcu_idle_exit())
99 * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
100 * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
101 * in such a section, considering these as in extended quiescent state,
102 * so such a CPU is effectively never in an RCU read-side critical section
103 * regardless of what RCU primitives it invokes. This state of affairs is
104 * required --- we need to keep an RCU-free window in idle where the CPU may
105 * possibly enter into low power mode. This way we can notice an extended
106 * quiescent state to other CPUs that started a grace period. Otherwise
107 * we would delay any grace period as long as we run in the idle task.
109 * Similarly, we avoid claiming an RCU read lock held if the current
112 static bool rcu_read_lock_held_common(bool *ret
)
114 if (!debug_lockdep_rcu_enabled()) {
118 if (!rcu_is_watching()) {
122 if (!rcu_lockdep_current_cpu_online()) {
129 int rcu_read_lock_sched_held(void)
133 if (rcu_read_lock_held_common(&ret
))
135 return lock_is_held(&rcu_sched_lock_map
) || !preemptible();
137 EXPORT_SYMBOL(rcu_read_lock_sched_held
);
140 #ifndef CONFIG_TINY_RCU
143 * Should expedited grace-period primitives always fall back to their
144 * non-expedited counterparts? Intended for use within RCU. Note
145 * that if the user specifies both rcu_expedited and rcu_normal, then
146 * rcu_normal wins. (Except during the time period during boot from
147 * when the first task is spawned until the rcu_set_runtime_mode()
148 * core_initcall() is invoked, at which point everything is expedited.)
150 bool rcu_gp_is_normal(void)
152 return READ_ONCE(rcu_normal
) &&
153 rcu_scheduler_active
!= RCU_SCHEDULER_INIT
;
155 EXPORT_SYMBOL_GPL(rcu_gp_is_normal
);
157 static atomic_t rcu_expedited_nesting
= ATOMIC_INIT(1);
160 * Should normal grace-period primitives be expedited? Intended for
161 * use within RCU. Note that this function takes the rcu_expedited
162 * sysfs/boot variable and rcu_scheduler_active into account as well
163 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
164 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
166 bool rcu_gp_is_expedited(void)
168 return rcu_expedited
|| atomic_read(&rcu_expedited_nesting
);
170 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited
);
173 * rcu_expedite_gp - Expedite future RCU grace periods
175 * After a call to this function, future calls to synchronize_rcu() and
176 * friends act as the corresponding synchronize_rcu_expedited() function
177 * had instead been called.
179 void rcu_expedite_gp(void)
181 atomic_inc(&rcu_expedited_nesting
);
183 EXPORT_SYMBOL_GPL(rcu_expedite_gp
);
186 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
188 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
189 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
190 * and if the rcu_expedited sysfs/boot parameter is not set, then all
191 * subsequent calls to synchronize_rcu() and friends will return to
192 * their normal non-expedited behavior.
194 void rcu_unexpedite_gp(void)
196 atomic_dec(&rcu_expedited_nesting
);
198 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp
);
200 static bool rcu_boot_ended __read_mostly
;
203 * Inform RCU of the end of the in-kernel boot sequence.
205 void rcu_end_inkernel_boot(void)
208 if (rcu_normal_after_boot
)
209 WRITE_ONCE(rcu_normal
, 1);
214 * Let rcutorture know when it is OK to turn it up to eleven.
216 bool rcu_inkernel_boot_has_ended(void)
218 return rcu_boot_ended
;
220 EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended
);
222 #endif /* #ifndef CONFIG_TINY_RCU */
225 * Test each non-SRCU synchronous grace-period wait API. This is
226 * useful just after a change in mode for these primitives, and
229 void rcu_test_sync_prims(void)
231 if (!IS_ENABLED(CONFIG_PROVE_RCU
))
234 synchronize_rcu_expedited();
237 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU)
240 * Switch to run-time mode once RCU has fully initialized.
242 static int __init
rcu_set_runtime_mode(void)
244 rcu_test_sync_prims();
245 rcu_scheduler_active
= RCU_SCHEDULER_RUNNING
;
246 kfree_rcu_scheduler_running();
247 rcu_test_sync_prims();
250 core_initcall(rcu_set_runtime_mode
);
252 #endif /* #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU) */
254 #ifdef CONFIG_DEBUG_LOCK_ALLOC
255 static struct lock_class_key rcu_lock_key
;
256 struct lockdep_map rcu_lock_map
= {
257 .name
= "rcu_read_lock",
258 .key
= &rcu_lock_key
,
259 .wait_type_outer
= LD_WAIT_FREE
,
260 .wait_type_inner
= LD_WAIT_CONFIG
, /* XXX PREEMPT_RCU ? */
262 EXPORT_SYMBOL_GPL(rcu_lock_map
);
264 static struct lock_class_key rcu_bh_lock_key
;
265 struct lockdep_map rcu_bh_lock_map
= {
266 .name
= "rcu_read_lock_bh",
267 .key
= &rcu_bh_lock_key
,
268 .wait_type_outer
= LD_WAIT_FREE
,
269 .wait_type_inner
= LD_WAIT_CONFIG
, /* PREEMPT_LOCK also makes BH preemptible */
271 EXPORT_SYMBOL_GPL(rcu_bh_lock_map
);
273 static struct lock_class_key rcu_sched_lock_key
;
274 struct lockdep_map rcu_sched_lock_map
= {
275 .name
= "rcu_read_lock_sched",
276 .key
= &rcu_sched_lock_key
,
277 .wait_type_outer
= LD_WAIT_FREE
,
278 .wait_type_inner
= LD_WAIT_SPIN
,
280 EXPORT_SYMBOL_GPL(rcu_sched_lock_map
);
282 static struct lock_class_key rcu_callback_key
;
283 struct lockdep_map rcu_callback_map
=
284 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key
);
285 EXPORT_SYMBOL_GPL(rcu_callback_map
);
287 noinstr
int notrace
debug_lockdep_rcu_enabled(void)
289 return rcu_scheduler_active
!= RCU_SCHEDULER_INACTIVE
&& debug_locks
&&
290 current
->lockdep_recursion
== 0;
292 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled
);
295 * rcu_read_lock_held() - might we be in RCU read-side critical section?
297 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
298 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
299 * this assumes we are in an RCU read-side critical section unless it can
300 * prove otherwise. This is useful for debug checks in functions that
301 * require that they be called within an RCU read-side critical section.
303 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
304 * and while lockdep is disabled.
306 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
307 * occur in the same context, for example, it is illegal to invoke
308 * rcu_read_unlock() in process context if the matching rcu_read_lock()
309 * was invoked from within an irq handler.
311 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
312 * offline from an RCU perspective, so check for those as well.
314 int rcu_read_lock_held(void)
318 if (rcu_read_lock_held_common(&ret
))
320 return lock_is_held(&rcu_lock_map
);
322 EXPORT_SYMBOL_GPL(rcu_read_lock_held
);
325 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
327 * Check for bottom half being disabled, which covers both the
328 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
329 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
330 * will show the situation. This is useful for debug checks in functions
331 * that require that they be called within an RCU read-side critical
334 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
336 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
337 * offline from an RCU perspective, so check for those as well.
339 int rcu_read_lock_bh_held(void)
343 if (rcu_read_lock_held_common(&ret
))
345 return in_softirq() || irqs_disabled();
347 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held
);
349 int rcu_read_lock_any_held(void)
353 if (rcu_read_lock_held_common(&ret
))
355 if (lock_is_held(&rcu_lock_map
) ||
356 lock_is_held(&rcu_bh_lock_map
) ||
357 lock_is_held(&rcu_sched_lock_map
))
359 return !preemptible();
361 EXPORT_SYMBOL_GPL(rcu_read_lock_any_held
);
363 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
366 * wakeme_after_rcu() - Callback function to awaken a task after grace period
367 * @head: Pointer to rcu_head member within rcu_synchronize structure
369 * Awaken the corresponding task now that a grace period has elapsed.
371 void wakeme_after_rcu(struct rcu_head
*head
)
373 struct rcu_synchronize
*rcu
;
375 rcu
= container_of(head
, struct rcu_synchronize
, head
);
376 complete(&rcu
->completion
);
378 EXPORT_SYMBOL_GPL(wakeme_after_rcu
);
380 void __wait_rcu_gp(bool checktiny
, int n
, call_rcu_func_t
*crcu_array
,
381 struct rcu_synchronize
*rs_array
)
386 /* Initialize and register callbacks for each crcu_array element. */
387 for (i
= 0; i
< n
; i
++) {
389 (crcu_array
[i
] == call_rcu
)) {
393 init_rcu_head_on_stack(&rs_array
[i
].head
);
394 init_completion(&rs_array
[i
].completion
);
395 for (j
= 0; j
< i
; j
++)
396 if (crcu_array
[j
] == crcu_array
[i
])
399 (crcu_array
[i
])(&rs_array
[i
].head
, wakeme_after_rcu
);
402 /* Wait for all callbacks to be invoked. */
403 for (i
= 0; i
< n
; i
++) {
405 (crcu_array
[i
] == call_rcu
))
407 for (j
= 0; j
< i
; j
++)
408 if (crcu_array
[j
] == crcu_array
[i
])
411 wait_for_completion(&rs_array
[i
].completion
);
412 destroy_rcu_head_on_stack(&rs_array
[i
].head
);
415 EXPORT_SYMBOL_GPL(__wait_rcu_gp
);
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 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 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
511 // Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
512 // warnings. Also used by rcutorture even if stall warnings are excluded.
513 int rcu_cpu_stall_suppress_at_boot __read_mostly
; // !0 = suppress boot stalls.
514 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot
);
515 module_param(rcu_cpu_stall_suppress_at_boot
, int, 0444);
517 #ifdef CONFIG_PROVE_RCU
520 * Early boot self test parameters.
522 static bool rcu_self_test
;
523 module_param(rcu_self_test
, bool, 0444);
525 static int rcu_self_test_counter
;
527 static void test_callback(struct rcu_head
*r
)
529 rcu_self_test_counter
++;
530 pr_info("RCU test callback executed %d\n", rcu_self_test_counter
);
533 DEFINE_STATIC_SRCU(early_srcu
);
535 struct early_boot_kfree_rcu
{
539 static void early_boot_test_call_rcu(void)
541 static struct rcu_head head
;
542 static struct rcu_head shead
;
543 struct early_boot_kfree_rcu
*rhp
;
545 call_rcu(&head
, test_callback
);
546 if (IS_ENABLED(CONFIG_SRCU
))
547 call_srcu(&early_srcu
, &shead
, test_callback
);
548 rhp
= kmalloc(sizeof(*rhp
), GFP_KERNEL
);
549 if (!WARN_ON_ONCE(!rhp
))
553 void rcu_early_boot_tests(void)
555 pr_info("Running RCU self tests\n");
558 early_boot_test_call_rcu();
559 rcu_test_sync_prims();
562 static int rcu_verify_early_boot_tests(void)
565 int early_boot_test_counter
= 0;
568 early_boot_test_counter
++;
570 if (IS_ENABLED(CONFIG_SRCU
)) {
571 early_boot_test_counter
++;
572 srcu_barrier(&early_srcu
);
575 if (rcu_self_test_counter
!= early_boot_test_counter
) {
582 late_initcall(rcu_verify_early_boot_tests
);
584 void rcu_early_boot_tests(void) {}
585 #endif /* CONFIG_PROVE_RCU */
589 #ifndef CONFIG_TINY_RCU
592 * Print any significant non-default boot-time settings.
594 void __init
rcupdate_announce_bootup_oddness(void)
597 pr_info("\tNo expedited grace period (rcu_normal).\n");
598 else if (rcu_normal_after_boot
)
599 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
600 else if (rcu_expedited
)
601 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
602 if (rcu_cpu_stall_suppress
)
603 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
604 if (rcu_cpu_stall_timeout
!= CONFIG_RCU_CPU_STALL_TIMEOUT
)
605 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout
);
606 rcu_tasks_bootup_oddness();
609 #endif /* #ifndef CONFIG_TINY_RCU */