Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * Read-Copy Update mechanism for mutual exclusion
4 *
5 * Copyright IBM Corporation, 2001
6 *
7 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
8 * Manfred Spraul <manfred@colorfullife.com>
9 *
10 * Based on the original work by Paul McKenney <paulmck@linux.ibm.com>
11 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
12 * Papers:
13 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
14 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
15 *
16 * For detailed explanation of Read-Copy Update mechanism see -
17 * http://lse.sourceforge.net/locking/rcupdate.html
18 *
19 */
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/torture.h>
29#include <linux/atomic.h>
30#include <linux/bitops.h>
31#include <linux/percpu.h>
32#include <linux/notifier.h>
33#include <linux/cpu.h>
34#include <linux/mutex.h>
35#include <linux/export.h>
36#include <linux/hardirq.h>
37#include <linux/delay.h>
38#include <linux/moduleparam.h>
39#include <linux/kthread.h>
40#include <linux/tick.h>
41#include <linux/rcupdate_wait.h>
42#include <linux/sched/isolation.h>
43#include <linux/kprobes.h>
44#include <linux/slab.h>
45#include <linux/irq_work.h>
46#include <linux/rcupdate_trace.h>
47
48#define CREATE_TRACE_POINTS
49
50#include "rcu.h"
51
52#ifdef MODULE_PARAM_PREFIX
53#undef MODULE_PARAM_PREFIX
54#endif
55#define MODULE_PARAM_PREFIX "rcupdate."
56
57#ifndef CONFIG_TINY_RCU
58module_param(rcu_expedited, int, 0444);
59module_param(rcu_normal, int, 0444);
60static int rcu_normal_after_boot = IS_ENABLED(CONFIG_PREEMPT_RT);
61#if !defined(CONFIG_PREEMPT_RT) || defined(CONFIG_NO_HZ_FULL)
62module_param(rcu_normal_after_boot, int, 0444);
63#endif
64#endif /* #ifndef CONFIG_TINY_RCU */
65
66#ifdef CONFIG_DEBUG_LOCK_ALLOC
67/**
68 * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
69 * @ret: Best guess answer if lockdep cannot be relied on
70 *
71 * Returns true if lockdep must be ignored, in which case ``*ret`` contains
72 * the best guess described below. Otherwise returns false, in which
73 * case ``*ret`` tells the caller nothing and the caller should instead
74 * consult lockdep.
75 *
76 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
77 * RCU-sched read-side critical section. In absence of
78 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
79 * critical section unless it can prove otherwise. Note that disabling
80 * of preemption (including disabling irqs) counts as an RCU-sched
81 * read-side critical section. This is useful for debug checks in functions
82 * that required that they be called within an RCU-sched read-side
83 * critical section.
84 *
85 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
86 * and while lockdep is disabled.
87 *
88 * Note that if the CPU is in the idle loop from an RCU point of view (ie:
89 * that we are in the section between ct_idle_enter() and ct_idle_exit())
90 * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
91 * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
92 * in such a section, considering these as in extended quiescent state,
93 * so such a CPU is effectively never in an RCU read-side critical section
94 * regardless of what RCU primitives it invokes. This state of affairs is
95 * required --- we need to keep an RCU-free window in idle where the CPU may
96 * possibly enter into low power mode. This way we can notice an extended
97 * quiescent state to other CPUs that started a grace period. Otherwise
98 * we would delay any grace period as long as we run in the idle task.
99 *
100 * Similarly, we avoid claiming an RCU read lock held if the current
101 * CPU is offline.
102 */
103static bool rcu_read_lock_held_common(bool *ret)
104{
105 if (!debug_lockdep_rcu_enabled()) {
106 *ret = true;
107 return true;
108 }
109 if (!rcu_is_watching()) {
110 *ret = false;
111 return true;
112 }
113 if (!rcu_lockdep_current_cpu_online()) {
114 *ret = false;
115 return true;
116 }
117 return false;
118}
119
120int rcu_read_lock_sched_held(void)
121{
122 bool ret;
123
124 if (rcu_read_lock_held_common(&ret))
125 return ret;
126 return lock_is_held(&rcu_sched_lock_map) || !preemptible();
127}
128EXPORT_SYMBOL(rcu_read_lock_sched_held);
129#endif
130
131#ifndef CONFIG_TINY_RCU
132
133/*
134 * Should expedited grace-period primitives always fall back to their
135 * non-expedited counterparts? Intended for use within RCU. Note
136 * that if the user specifies both rcu_expedited and rcu_normal, then
137 * rcu_normal wins. (Except during the time period during boot from
138 * when the first task is spawned until the rcu_set_runtime_mode()
139 * core_initcall() is invoked, at which point everything is expedited.)
140 */
141bool rcu_gp_is_normal(void)
142{
143 return READ_ONCE(rcu_normal) &&
144 rcu_scheduler_active != RCU_SCHEDULER_INIT;
145}
146EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
147
148static atomic_t rcu_async_hurry_nesting = ATOMIC_INIT(1);
149/*
150 * Should call_rcu() callbacks be processed with urgency or are
151 * they OK being executed with arbitrary delays?
152 */
153bool rcu_async_should_hurry(void)
154{
155 return !IS_ENABLED(CONFIG_RCU_LAZY) ||
156 atomic_read(&rcu_async_hurry_nesting);
157}
158EXPORT_SYMBOL_GPL(rcu_async_should_hurry);
159
160/**
161 * rcu_async_hurry - Make future async RCU callbacks not lazy.
162 *
163 * After a call to this function, future calls to call_rcu()
164 * will be processed in a timely fashion.
165 */
166void rcu_async_hurry(void)
167{
168 if (IS_ENABLED(CONFIG_RCU_LAZY))
169 atomic_inc(&rcu_async_hurry_nesting);
170}
171EXPORT_SYMBOL_GPL(rcu_async_hurry);
172
173/**
174 * rcu_async_relax - Make future async RCU callbacks lazy.
175 *
176 * After a call to this function, future calls to call_rcu()
177 * will be processed in a lazy fashion.
178 */
179void rcu_async_relax(void)
180{
181 if (IS_ENABLED(CONFIG_RCU_LAZY))
182 atomic_dec(&rcu_async_hurry_nesting);
183}
184EXPORT_SYMBOL_GPL(rcu_async_relax);
185
186static atomic_t rcu_expedited_nesting = ATOMIC_INIT(1);
187/*
188 * Should normal grace-period primitives be expedited? Intended for
189 * use within RCU. Note that this function takes the rcu_expedited
190 * sysfs/boot variable and rcu_scheduler_active into account as well
191 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
192 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
193 */
194bool rcu_gp_is_expedited(void)
195{
196 return rcu_expedited || atomic_read(&rcu_expedited_nesting);
197}
198EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
199
200/**
201 * rcu_expedite_gp - Expedite future RCU grace periods
202 *
203 * After a call to this function, future calls to synchronize_rcu() and
204 * friends act as the corresponding synchronize_rcu_expedited() function
205 * had instead been called.
206 */
207void rcu_expedite_gp(void)
208{
209 atomic_inc(&rcu_expedited_nesting);
210}
211EXPORT_SYMBOL_GPL(rcu_expedite_gp);
212
213/**
214 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
215 *
216 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
217 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
218 * and if the rcu_expedited sysfs/boot parameter is not set, then all
219 * subsequent calls to synchronize_rcu() and friends will return to
220 * their normal non-expedited behavior.
221 */
222void rcu_unexpedite_gp(void)
223{
224 atomic_dec(&rcu_expedited_nesting);
225}
226EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
227
228static bool rcu_boot_ended __read_mostly;
229
230/*
231 * Inform RCU of the end of the in-kernel boot sequence.
232 */
233void rcu_end_inkernel_boot(void)
234{
235 rcu_unexpedite_gp();
236 rcu_async_relax();
237 if (rcu_normal_after_boot)
238 WRITE_ONCE(rcu_normal, 1);
239 rcu_boot_ended = true;
240}
241
242/*
243 * Let rcutorture know when it is OK to turn it up to eleven.
244 */
245bool rcu_inkernel_boot_has_ended(void)
246{
247 return rcu_boot_ended;
248}
249EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended);
250
251#endif /* #ifndef CONFIG_TINY_RCU */
252
253/*
254 * Test each non-SRCU synchronous grace-period wait API. This is
255 * useful just after a change in mode for these primitives, and
256 * during early boot.
257 */
258void rcu_test_sync_prims(void)
259{
260 if (!IS_ENABLED(CONFIG_PROVE_RCU))
261 return;
262 pr_info("Running RCU synchronous self tests\n");
263 synchronize_rcu();
264 synchronize_rcu_expedited();
265}
266
267#if !defined(CONFIG_TINY_RCU)
268
269/*
270 * Switch to run-time mode once RCU has fully initialized.
271 */
272static int __init rcu_set_runtime_mode(void)
273{
274 rcu_test_sync_prims();
275 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
276 kfree_rcu_scheduler_running();
277 rcu_test_sync_prims();
278 return 0;
279}
280core_initcall(rcu_set_runtime_mode);
281
282#endif /* #if !defined(CONFIG_TINY_RCU) */
283
284#ifdef CONFIG_DEBUG_LOCK_ALLOC
285static struct lock_class_key rcu_lock_key;
286struct lockdep_map rcu_lock_map = {
287 .name = "rcu_read_lock",
288 .key = &rcu_lock_key,
289 .wait_type_outer = LD_WAIT_FREE,
290 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT implies PREEMPT_RCU */
291};
292EXPORT_SYMBOL_GPL(rcu_lock_map);
293
294static struct lock_class_key rcu_bh_lock_key;
295struct lockdep_map rcu_bh_lock_map = {
296 .name = "rcu_read_lock_bh",
297 .key = &rcu_bh_lock_key,
298 .wait_type_outer = LD_WAIT_FREE,
299 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT makes BH preemptible. */
300};
301EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
302
303static struct lock_class_key rcu_sched_lock_key;
304struct lockdep_map rcu_sched_lock_map = {
305 .name = "rcu_read_lock_sched",
306 .key = &rcu_sched_lock_key,
307 .wait_type_outer = LD_WAIT_FREE,
308 .wait_type_inner = LD_WAIT_SPIN,
309};
310EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
311
312// Tell lockdep when RCU callbacks are being invoked.
313static struct lock_class_key rcu_callback_key;
314struct lockdep_map rcu_callback_map =
315 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
316EXPORT_SYMBOL_GPL(rcu_callback_map);
317
318noinstr int notrace debug_lockdep_rcu_enabled(void)
319{
320 return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) &&
321 current->lockdep_recursion == 0;
322}
323EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
324
325/**
326 * rcu_read_lock_held() - might we be in RCU read-side critical section?
327 *
328 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
329 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
330 * this assumes we are in an RCU read-side critical section unless it can
331 * prove otherwise. This is useful for debug checks in functions that
332 * require that they be called within an RCU read-side critical section.
333 *
334 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
335 * and while lockdep is disabled.
336 *
337 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
338 * occur in the same context, for example, it is illegal to invoke
339 * rcu_read_unlock() in process context if the matching rcu_read_lock()
340 * was invoked from within an irq handler.
341 *
342 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
343 * offline from an RCU perspective, so check for those as well.
344 */
345int rcu_read_lock_held(void)
346{
347 bool ret;
348
349 if (rcu_read_lock_held_common(&ret))
350 return ret;
351 return lock_is_held(&rcu_lock_map);
352}
353EXPORT_SYMBOL_GPL(rcu_read_lock_held);
354
355/**
356 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
357 *
358 * Check for bottom half being disabled, which covers both the
359 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
360 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
361 * will show the situation. This is useful for debug checks in functions
362 * that require that they be called within an RCU read-side critical
363 * section.
364 *
365 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
366 *
367 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
368 * offline from an RCU perspective, so check for those as well.
369 */
370int rcu_read_lock_bh_held(void)
371{
372 bool ret;
373
374 if (rcu_read_lock_held_common(&ret))
375 return ret;
376 return in_softirq() || irqs_disabled();
377}
378EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
379
380int rcu_read_lock_any_held(void)
381{
382 bool ret;
383
384 if (rcu_read_lock_held_common(&ret))
385 return ret;
386 if (lock_is_held(&rcu_lock_map) ||
387 lock_is_held(&rcu_bh_lock_map) ||
388 lock_is_held(&rcu_sched_lock_map))
389 return 1;
390 return !preemptible();
391}
392EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
393
394#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
395
396/**
397 * wakeme_after_rcu() - Callback function to awaken a task after grace period
398 * @head: Pointer to rcu_head member within rcu_synchronize structure
399 *
400 * Awaken the corresponding task now that a grace period has elapsed.
401 */
402void wakeme_after_rcu(struct rcu_head *head)
403{
404 struct rcu_synchronize *rcu;
405
406 rcu = container_of(head, struct rcu_synchronize, head);
407 complete(&rcu->completion);
408}
409EXPORT_SYMBOL_GPL(wakeme_after_rcu);
410
411void __wait_rcu_gp(bool checktiny, unsigned int state, int n, call_rcu_func_t *crcu_array,
412 struct rcu_synchronize *rs_array)
413{
414 int i;
415 int j;
416
417 /* Initialize and register callbacks for each crcu_array element. */
418 for (i = 0; i < n; i++) {
419 if (checktiny &&
420 (crcu_array[i] == call_rcu)) {
421 might_sleep();
422 continue;
423 }
424 for (j = 0; j < i; j++)
425 if (crcu_array[j] == crcu_array[i])
426 break;
427 if (j == i) {
428 init_rcu_head_on_stack(&rs_array[i].head);
429 init_completion(&rs_array[i].completion);
430 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
431 }
432 }
433
434 /* Wait for all callbacks to be invoked. */
435 for (i = 0; i < n; i++) {
436 if (checktiny &&
437 (crcu_array[i] == call_rcu))
438 continue;
439 for (j = 0; j < i; j++)
440 if (crcu_array[j] == crcu_array[i])
441 break;
442 if (j == i) {
443 wait_for_completion_state(&rs_array[i].completion, state);
444 destroy_rcu_head_on_stack(&rs_array[i].head);
445 }
446 }
447}
448EXPORT_SYMBOL_GPL(__wait_rcu_gp);
449
450void finish_rcuwait(struct rcuwait *w)
451{
452 rcu_assign_pointer(w->task, NULL);
453 __set_current_state(TASK_RUNNING);
454}
455EXPORT_SYMBOL_GPL(finish_rcuwait);
456
457#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
458void init_rcu_head(struct rcu_head *head)
459{
460 debug_object_init(head, &rcuhead_debug_descr);
461}
462EXPORT_SYMBOL_GPL(init_rcu_head);
463
464void destroy_rcu_head(struct rcu_head *head)
465{
466 debug_object_free(head, &rcuhead_debug_descr);
467}
468EXPORT_SYMBOL_GPL(destroy_rcu_head);
469
470static bool rcuhead_is_static_object(void *addr)
471{
472 return true;
473}
474
475/**
476 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
477 * @head: pointer to rcu_head structure to be initialized
478 *
479 * This function informs debugobjects of a new rcu_head structure that
480 * has been allocated as an auto variable on the stack. This function
481 * is not required for rcu_head structures that are statically defined or
482 * that are dynamically allocated on the heap. This function has no
483 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
484 */
485void init_rcu_head_on_stack(struct rcu_head *head)
486{
487 debug_object_init_on_stack(head, &rcuhead_debug_descr);
488}
489EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
490
491/**
492 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
493 * @head: pointer to rcu_head structure to be initialized
494 *
495 * This function informs debugobjects that an on-stack rcu_head structure
496 * is about to go out of scope. As with init_rcu_head_on_stack(), this
497 * function is not required for rcu_head structures that are statically
498 * defined or that are dynamically allocated on the heap. Also as with
499 * init_rcu_head_on_stack(), this function has no effect for
500 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
501 */
502void destroy_rcu_head_on_stack(struct rcu_head *head)
503{
504 debug_object_free(head, &rcuhead_debug_descr);
505}
506EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
507
508const struct debug_obj_descr rcuhead_debug_descr = {
509 .name = "rcu_head",
510 .is_static_object = rcuhead_is_static_object,
511};
512EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
513#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
514
515#if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
516void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
517 unsigned long secs,
518 unsigned long c_old, unsigned long c)
519{
520 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
521}
522EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
523#else
524#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
525 do { } while (0)
526#endif
527
528#if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST) || IS_ENABLED(CONFIG_LOCK_TORTURE_TEST) || IS_MODULE(CONFIG_LOCK_TORTURE_TEST)
529/* Get rcutorture access to sched_setaffinity(). */
530long torture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
531{
532 int ret;
533
534 ret = sched_setaffinity(pid, in_mask);
535 WARN_ONCE(ret, "%s: sched_setaffinity(%d) returned %d\n", __func__, pid, ret);
536 return ret;
537}
538EXPORT_SYMBOL_GPL(torture_sched_setaffinity);
539#endif
540
541int rcu_cpu_stall_notifiers __read_mostly; // !0 = provide stall notifiers (rarely useful)
542EXPORT_SYMBOL_GPL(rcu_cpu_stall_notifiers);
543
544#ifdef CONFIG_RCU_STALL_COMMON
545int rcu_cpu_stall_ftrace_dump __read_mostly;
546module_param(rcu_cpu_stall_ftrace_dump, int, 0644);
547#ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
548module_param(rcu_cpu_stall_notifiers, int, 0444);
549#endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
550int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings.
551EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
552module_param(rcu_cpu_stall_suppress, int, 0644);
553int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
554module_param(rcu_cpu_stall_timeout, int, 0644);
555int rcu_exp_cpu_stall_timeout __read_mostly = CONFIG_RCU_EXP_CPU_STALL_TIMEOUT;
556module_param(rcu_exp_cpu_stall_timeout, int, 0644);
557int rcu_cpu_stall_cputime __read_mostly = IS_ENABLED(CONFIG_RCU_CPU_STALL_CPUTIME);
558module_param(rcu_cpu_stall_cputime, int, 0644);
559bool rcu_exp_stall_task_details __read_mostly;
560module_param(rcu_exp_stall_task_details, bool, 0644);
561#endif /* #ifdef CONFIG_RCU_STALL_COMMON */
562
563// Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
564// warnings. Also used by rcutorture even if stall warnings are excluded.
565int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls.
566EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot);
567module_param(rcu_cpu_stall_suppress_at_boot, int, 0444);
568
569/**
570 * get_completed_synchronize_rcu - Return a pre-completed polled state cookie
571 *
572 * Returns a value that will always be treated by functions like
573 * poll_state_synchronize_rcu() as a cookie whose grace period has already
574 * completed.
575 */
576unsigned long get_completed_synchronize_rcu(void)
577{
578 return RCU_GET_STATE_COMPLETED;
579}
580EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu);
581
582#ifdef CONFIG_PROVE_RCU
583
584/*
585 * Early boot self test parameters.
586 */
587static bool rcu_self_test;
588module_param(rcu_self_test, bool, 0444);
589
590static int rcu_self_test_counter;
591
592static void test_callback(struct rcu_head *r)
593{
594 rcu_self_test_counter++;
595 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
596}
597
598DEFINE_STATIC_SRCU(early_srcu);
599static unsigned long early_srcu_cookie;
600
601struct early_boot_kfree_rcu {
602 struct rcu_head rh;
603};
604
605static void early_boot_test_call_rcu(void)
606{
607 static struct rcu_head head;
608 int idx;
609 static struct rcu_head shead;
610 struct early_boot_kfree_rcu *rhp;
611
612 idx = srcu_down_read(&early_srcu);
613 srcu_up_read(&early_srcu, idx);
614 call_rcu(&head, test_callback);
615 early_srcu_cookie = start_poll_synchronize_srcu(&early_srcu);
616 call_srcu(&early_srcu, &shead, test_callback);
617 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
618 if (!WARN_ON_ONCE(!rhp))
619 kfree_rcu(rhp, rh);
620}
621
622void rcu_early_boot_tests(void)
623{
624 pr_info("Running RCU self tests\n");
625
626 if (rcu_self_test)
627 early_boot_test_call_rcu();
628 rcu_test_sync_prims();
629}
630
631static int rcu_verify_early_boot_tests(void)
632{
633 int ret = 0;
634 int early_boot_test_counter = 0;
635
636 if (rcu_self_test) {
637 early_boot_test_counter++;
638 rcu_barrier();
639 early_boot_test_counter++;
640 srcu_barrier(&early_srcu);
641 WARN_ON_ONCE(!poll_state_synchronize_srcu(&early_srcu, early_srcu_cookie));
642 cleanup_srcu_struct(&early_srcu);
643 }
644 if (rcu_self_test_counter != early_boot_test_counter) {
645 WARN_ON(1);
646 ret = -1;
647 }
648
649 return ret;
650}
651late_initcall(rcu_verify_early_boot_tests);
652#else
653void rcu_early_boot_tests(void) {}
654#endif /* CONFIG_PROVE_RCU */
655
656#include "tasks.h"
657
658#ifndef CONFIG_TINY_RCU
659
660/*
661 * Print any significant non-default boot-time settings.
662 */
663void __init rcupdate_announce_bootup_oddness(void)
664{
665 if (rcu_normal)
666 pr_info("\tNo expedited grace period (rcu_normal).\n");
667 else if (rcu_normal_after_boot)
668 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
669 else if (rcu_expedited)
670 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
671 if (rcu_cpu_stall_suppress)
672 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
673 if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT)
674 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout);
675 rcu_tasks_bootup_oddness();
676}
677
678#endif /* #ifndef CONFIG_TINY_RCU */
1/*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
17 *
18 * Copyright IBM Corporation, 2001
19 *
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 *
23 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
24 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * Papers:
26 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
27 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 *
29 * For detailed explanation of Read-Copy Update mechanism see -
30 * http://lse.sourceforge.net/locking/rcupdate.html
31 *
32 */
33#include <linux/types.h>
34#include <linux/kernel.h>
35#include <linux/init.h>
36#include <linux/spinlock.h>
37#include <linux/smp.h>
38#include <linux/interrupt.h>
39#include <linux/sched.h>
40#include <linux/atomic.h>
41#include <linux/bitops.h>
42#include <linux/percpu.h>
43#include <linux/notifier.h>
44#include <linux/cpu.h>
45#include <linux/mutex.h>
46#include <linux/export.h>
47#include <linux/hardirq.h>
48#include <linux/delay.h>
49#include <linux/module.h>
50
51#define CREATE_TRACE_POINTS
52
53#include "rcu.h"
54
55MODULE_ALIAS("rcupdate");
56#ifdef MODULE_PARAM_PREFIX
57#undef MODULE_PARAM_PREFIX
58#endif
59#define MODULE_PARAM_PREFIX "rcupdate."
60
61module_param(rcu_expedited, int, 0);
62
63#ifdef CONFIG_PREEMPT_RCU
64
65/*
66 * Preemptible RCU implementation for rcu_read_lock().
67 * Just increment ->rcu_read_lock_nesting, shared state will be updated
68 * if we block.
69 */
70void __rcu_read_lock(void)
71{
72 current->rcu_read_lock_nesting++;
73 barrier(); /* critical section after entry code. */
74}
75EXPORT_SYMBOL_GPL(__rcu_read_lock);
76
77/*
78 * Preemptible RCU implementation for rcu_read_unlock().
79 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
80 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
81 * invoke rcu_read_unlock_special() to clean up after a context switch
82 * in an RCU read-side critical section and other special cases.
83 */
84void __rcu_read_unlock(void)
85{
86 struct task_struct *t = current;
87
88 if (t->rcu_read_lock_nesting != 1) {
89 --t->rcu_read_lock_nesting;
90 } else {
91 barrier(); /* critical section before exit code. */
92 t->rcu_read_lock_nesting = INT_MIN;
93#ifdef CONFIG_PROVE_RCU_DELAY
94 udelay(10); /* Make preemption more probable. */
95#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
96 barrier(); /* assign before ->rcu_read_unlock_special load */
97 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
98 rcu_read_unlock_special(t);
99 barrier(); /* ->rcu_read_unlock_special load before assign */
100 t->rcu_read_lock_nesting = 0;
101 }
102#ifdef CONFIG_PROVE_LOCKING
103 {
104 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
105
106 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
107 }
108#endif /* #ifdef CONFIG_PROVE_LOCKING */
109}
110EXPORT_SYMBOL_GPL(__rcu_read_unlock);
111
112#endif /* #ifdef CONFIG_PREEMPT_RCU */
113
114#ifdef CONFIG_DEBUG_LOCK_ALLOC
115static struct lock_class_key rcu_lock_key;
116struct lockdep_map rcu_lock_map =
117 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
118EXPORT_SYMBOL_GPL(rcu_lock_map);
119
120static struct lock_class_key rcu_bh_lock_key;
121struct lockdep_map rcu_bh_lock_map =
122 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
123EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
124
125static struct lock_class_key rcu_sched_lock_key;
126struct lockdep_map rcu_sched_lock_map =
127 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
128EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
129
130static struct lock_class_key rcu_callback_key;
131struct lockdep_map rcu_callback_map =
132 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
133EXPORT_SYMBOL_GPL(rcu_callback_map);
134
135int notrace debug_lockdep_rcu_enabled(void)
136{
137 return rcu_scheduler_active && debug_locks &&
138 current->lockdep_recursion == 0;
139}
140EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
141
142/**
143 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
144 *
145 * Check for bottom half being disabled, which covers both the
146 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
147 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
148 * will show the situation. This is useful for debug checks in functions
149 * that require that they be called within an RCU read-side critical
150 * section.
151 *
152 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
153 *
154 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
155 * offline from an RCU perspective, so check for those as well.
156 */
157int rcu_read_lock_bh_held(void)
158{
159 if (!debug_lockdep_rcu_enabled())
160 return 1;
161 if (!rcu_is_watching())
162 return 0;
163 if (!rcu_lockdep_current_cpu_online())
164 return 0;
165 return in_softirq() || irqs_disabled();
166}
167EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
168
169#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
170
171struct rcu_synchronize {
172 struct rcu_head head;
173 struct completion completion;
174};
175
176/*
177 * Awaken the corresponding synchronize_rcu() instance now that a
178 * grace period has elapsed.
179 */
180static void wakeme_after_rcu(struct rcu_head *head)
181{
182 struct rcu_synchronize *rcu;
183
184 rcu = container_of(head, struct rcu_synchronize, head);
185 complete(&rcu->completion);
186}
187
188void wait_rcu_gp(call_rcu_func_t crf)
189{
190 struct rcu_synchronize rcu;
191
192 init_rcu_head_on_stack(&rcu.head);
193 init_completion(&rcu.completion);
194 /* Will wake me after RCU finished. */
195 crf(&rcu.head, wakeme_after_rcu);
196 /* Wait for it. */
197 wait_for_completion(&rcu.completion);
198 destroy_rcu_head_on_stack(&rcu.head);
199}
200EXPORT_SYMBOL_GPL(wait_rcu_gp);
201
202#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
203static inline void debug_init_rcu_head(struct rcu_head *head)
204{
205 debug_object_init(head, &rcuhead_debug_descr);
206}
207
208static inline void debug_rcu_head_free(struct rcu_head *head)
209{
210 debug_object_free(head, &rcuhead_debug_descr);
211}
212
213/*
214 * fixup_activate is called when:
215 * - an active object is activated
216 * - an unknown object is activated (might be a statically initialized object)
217 * Activation is performed internally by call_rcu().
218 */
219static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state)
220{
221 struct rcu_head *head = addr;
222
223 switch (state) {
224
225 case ODEBUG_STATE_NOTAVAILABLE:
226 /*
227 * This is not really a fixup. We just make sure that it is
228 * tracked in the object tracker.
229 */
230 debug_object_init(head, &rcuhead_debug_descr);
231 debug_object_activate(head, &rcuhead_debug_descr);
232 return 0;
233 default:
234 return 1;
235 }
236}
237
238/**
239 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
240 * @head: pointer to rcu_head structure to be initialized
241 *
242 * This function informs debugobjects of a new rcu_head structure that
243 * has been allocated as an auto variable on the stack. This function
244 * is not required for rcu_head structures that are statically defined or
245 * that are dynamically allocated on the heap. This function has no
246 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
247 */
248void init_rcu_head_on_stack(struct rcu_head *head)
249{
250 debug_object_init_on_stack(head, &rcuhead_debug_descr);
251}
252EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
253
254/**
255 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
256 * @head: pointer to rcu_head structure to be initialized
257 *
258 * This function informs debugobjects that an on-stack rcu_head structure
259 * is about to go out of scope. As with init_rcu_head_on_stack(), this
260 * function is not required for rcu_head structures that are statically
261 * defined or that are dynamically allocated on the heap. Also as with
262 * init_rcu_head_on_stack(), this function has no effect for
263 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
264 */
265void destroy_rcu_head_on_stack(struct rcu_head *head)
266{
267 debug_object_free(head, &rcuhead_debug_descr);
268}
269EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
270
271struct debug_obj_descr rcuhead_debug_descr = {
272 .name = "rcu_head",
273 .fixup_activate = rcuhead_fixup_activate,
274};
275EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
276#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
277
278#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
279void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
280 unsigned long secs,
281 unsigned long c_old, unsigned long c)
282{
283 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
284}
285EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
286#else
287#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
288 do { } while (0)
289#endif
290
291#ifdef CONFIG_RCU_STALL_COMMON
292
293#ifdef CONFIG_PROVE_RCU
294#define RCU_STALL_DELAY_DELTA (5 * HZ)
295#else
296#define RCU_STALL_DELAY_DELTA 0
297#endif
298
299int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
300static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
301
302module_param(rcu_cpu_stall_suppress, int, 0644);
303module_param(rcu_cpu_stall_timeout, int, 0644);
304
305int rcu_jiffies_till_stall_check(void)
306{
307 int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout);
308
309 /*
310 * Limit check must be consistent with the Kconfig limits
311 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
312 */
313 if (till_stall_check < 3) {
314 ACCESS_ONCE(rcu_cpu_stall_timeout) = 3;
315 till_stall_check = 3;
316 } else if (till_stall_check > 300) {
317 ACCESS_ONCE(rcu_cpu_stall_timeout) = 300;
318 till_stall_check = 300;
319 }
320 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
321}
322
323static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
324{
325 rcu_cpu_stall_suppress = 1;
326 return NOTIFY_DONE;
327}
328
329static struct notifier_block rcu_panic_block = {
330 .notifier_call = rcu_panic,
331};
332
333static int __init check_cpu_stall_init(void)
334{
335 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
336 return 0;
337}
338early_initcall(check_cpu_stall_init);
339
340#endif /* #ifdef CONFIG_RCU_STALL_COMMON */