Linux Audio

Check our new training course

Loading...
v6.13.7
  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 */
v4.10.11
 
  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/moduleparam.h>
 50#include <linux/kthread.h>
 51#include <linux/tick.h>
 
 
 
 
 
 
 52
 53#define CREATE_TRACE_POINTS
 54
 55#include "rcu.h"
 56
 57#ifdef MODULE_PARAM_PREFIX
 58#undef MODULE_PARAM_PREFIX
 59#endif
 60#define MODULE_PARAM_PREFIX "rcupdate."
 61
 62#ifndef CONFIG_TINY_RCU
 63module_param(rcu_expedited, int, 0);
 64module_param(rcu_normal, int, 0);
 65static int rcu_normal_after_boot;
 66module_param(rcu_normal_after_boot, int, 0);
 
 
 67#endif /* #ifndef CONFIG_TINY_RCU */
 68
 69#ifdef CONFIG_DEBUG_LOCK_ALLOC
 70/**
 71 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
 
 
 
 
 
 
 72 *
 73 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
 74 * RCU-sched read-side critical section.  In absence of
 75 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
 76 * critical section unless it can prove otherwise.  Note that disabling
 77 * of preemption (including disabling irqs) counts as an RCU-sched
 78 * read-side critical section.  This is useful for debug checks in functions
 79 * that required that they be called within an RCU-sched read-side
 80 * critical section.
 81 *
 82 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
 83 * and while lockdep is disabled.
 84 *
 85 * Note that if the CPU is in the idle loop from an RCU point of
 86 * view (ie: that we are in the section between rcu_idle_enter() and
 87 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
 88 * did an rcu_read_lock().  The reason for this is that RCU ignores CPUs
 89 * that are in such a section, considering these as in extended quiescent
 90 * state, so such a CPU is effectively never in an RCU read-side critical
 91 * section regardless of what RCU primitives it invokes.  This state of
 92 * affairs is required --- we need to keep an RCU-free window in idle
 93 * where the CPU may possibly enter into low power mode. This way we can
 94 * notice an extended quiescent state to other CPUs that started a grace
 95 * period. Otherwise we would delay any grace period as long as we run in
 96 * the idle task.
 97 *
 98 * Similarly, we avoid claiming an SRCU read lock held if the current
 99 * CPU is offline.
100 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
101int rcu_read_lock_sched_held(void)
102{
103	int lockdep_opinion = 0;
104
105	if (!debug_lockdep_rcu_enabled())
106		return 1;
107	if (!rcu_is_watching())
108		return 0;
109	if (!rcu_lockdep_current_cpu_online())
110		return 0;
111	if (debug_locks)
112		lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
113	return lockdep_opinion || !preemptible();
114}
115EXPORT_SYMBOL(rcu_read_lock_sched_held);
116#endif
117
118#ifndef CONFIG_TINY_RCU
119
120/*
121 * Should expedited grace-period primitives always fall back to their
122 * non-expedited counterparts?  Intended for use within RCU.  Note
123 * that if the user specifies both rcu_expedited and rcu_normal, then
124 * rcu_normal wins.  (Except during the time period during boot from
125 * when the first task is spawned until the rcu_exp_runtime_mode()
126 * core_initcall() is invoked, at which point everything is expedited.)
127 */
128bool rcu_gp_is_normal(void)
129{
130	return READ_ONCE(rcu_normal) &&
131	       rcu_scheduler_active != RCU_SCHEDULER_INIT;
132}
133EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
134
135static atomic_t rcu_expedited_nesting =
136	ATOMIC_INIT(IS_ENABLED(CONFIG_RCU_EXPEDITE_BOOT) ? 1 : 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
137
 
138/*
139 * Should normal grace-period primitives be expedited?  Intended for
140 * use within RCU.  Note that this function takes the rcu_expedited
141 * sysfs/boot variable and rcu_scheduler_active into account as well
142 * as the rcu_expedite_gp() nesting.  So looping on rcu_unexpedite_gp()
143 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
144 */
145bool rcu_gp_is_expedited(void)
146{
147	return rcu_expedited || atomic_read(&rcu_expedited_nesting) ||
148	       rcu_scheduler_active == RCU_SCHEDULER_INIT;
149}
150EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
151
152/**
153 * rcu_expedite_gp - Expedite future RCU grace periods
154 *
155 * After a call to this function, future calls to synchronize_rcu() and
156 * friends act as the corresponding synchronize_rcu_expedited() function
157 * had instead been called.
158 */
159void rcu_expedite_gp(void)
160{
161	atomic_inc(&rcu_expedited_nesting);
162}
163EXPORT_SYMBOL_GPL(rcu_expedite_gp);
164
165/**
166 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
167 *
168 * Undo a prior call to rcu_expedite_gp().  If all prior calls to
169 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
170 * and if the rcu_expedited sysfs/boot parameter is not set, then all
171 * subsequent calls to synchronize_rcu() and friends will return to
172 * their normal non-expedited behavior.
173 */
174void rcu_unexpedite_gp(void)
175{
176	atomic_dec(&rcu_expedited_nesting);
177}
178EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
179
 
 
180/*
181 * Inform RCU of the end of the in-kernel boot sequence.
182 */
183void rcu_end_inkernel_boot(void)
184{
185	if (IS_ENABLED(CONFIG_RCU_EXPEDITE_BOOT))
186		rcu_unexpedite_gp();
187	if (rcu_normal_after_boot)
188		WRITE_ONCE(rcu_normal, 1);
 
 
 
 
 
 
 
 
 
189}
 
190
191#endif /* #ifndef CONFIG_TINY_RCU */
192
193#ifdef CONFIG_PREEMPT_RCU
194
195/*
196 * Preemptible RCU implementation for rcu_read_lock().
197 * Just increment ->rcu_read_lock_nesting, shared state will be updated
198 * if we block.
199 */
200void __rcu_read_lock(void)
201{
202	current->rcu_read_lock_nesting++;
203	barrier();  /* critical section after entry code. */
 
 
 
204}
205EXPORT_SYMBOL_GPL(__rcu_read_lock);
 
206
207/*
208 * Preemptible RCU implementation for rcu_read_unlock().
209 * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
210 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
211 * invoke rcu_read_unlock_special() to clean up after a context switch
212 * in an RCU read-side critical section and other special cases.
213 */
214void __rcu_read_unlock(void)
215{
216	struct task_struct *t = current;
217
218	if (t->rcu_read_lock_nesting != 1) {
219		--t->rcu_read_lock_nesting;
220	} else {
221		barrier();  /* critical section before exit code. */
222		t->rcu_read_lock_nesting = INT_MIN;
223		barrier();  /* assign before ->rcu_read_unlock_special load */
224		if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s)))
225			rcu_read_unlock_special(t);
226		barrier();  /* ->rcu_read_unlock_special load before assign */
227		t->rcu_read_lock_nesting = 0;
228	}
229#ifdef CONFIG_PROVE_LOCKING
230	{
231		int rrln = READ_ONCE(t->rcu_read_lock_nesting);
232
233		WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
234	}
235#endif /* #ifdef CONFIG_PROVE_LOCKING */
236}
237EXPORT_SYMBOL_GPL(__rcu_read_unlock);
238
239#endif /* #ifdef CONFIG_PREEMPT_RCU */
240
241#ifdef CONFIG_DEBUG_LOCK_ALLOC
242static struct lock_class_key rcu_lock_key;
243struct lockdep_map rcu_lock_map =
244	STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
 
 
 
 
245EXPORT_SYMBOL_GPL(rcu_lock_map);
246
247static struct lock_class_key rcu_bh_lock_key;
248struct lockdep_map rcu_bh_lock_map =
249	STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
 
 
 
 
250EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
251
252static struct lock_class_key rcu_sched_lock_key;
253struct lockdep_map rcu_sched_lock_map =
254	STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
 
 
 
 
255EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
256
 
257static struct lock_class_key rcu_callback_key;
258struct lockdep_map rcu_callback_map =
259	STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
260EXPORT_SYMBOL_GPL(rcu_callback_map);
261
262int notrace debug_lockdep_rcu_enabled(void)
263{
264	return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && debug_locks &&
265	       current->lockdep_recursion == 0;
266}
267EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
268
269/**
270 * rcu_read_lock_held() - might we be in RCU read-side critical section?
271 *
272 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
273 * read-side critical section.  In absence of CONFIG_DEBUG_LOCK_ALLOC,
274 * this assumes we are in an RCU read-side critical section unless it can
275 * prove otherwise.  This is useful for debug checks in functions that
276 * require that they be called within an RCU read-side critical section.
277 *
278 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
279 * and while lockdep is disabled.
280 *
281 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
282 * occur in the same context, for example, it is illegal to invoke
283 * rcu_read_unlock() in process context if the matching rcu_read_lock()
284 * was invoked from within an irq handler.
285 *
286 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
287 * offline from an RCU perspective, so check for those as well.
288 */
289int rcu_read_lock_held(void)
290{
291	if (!debug_lockdep_rcu_enabled())
292		return 1;
293	if (!rcu_is_watching())
294		return 0;
295	if (!rcu_lockdep_current_cpu_online())
296		return 0;
297	return lock_is_held(&rcu_lock_map);
298}
299EXPORT_SYMBOL_GPL(rcu_read_lock_held);
300
301/**
302 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
303 *
304 * Check for bottom half being disabled, which covers both the
305 * CONFIG_PROVE_RCU and not cases.  Note that if someone uses
306 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
307 * will show the situation.  This is useful for debug checks in functions
308 * that require that they be called within an RCU read-side critical
309 * section.
310 *
311 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
312 *
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.
315 */
316int rcu_read_lock_bh_held(void)
317{
318	if (!debug_lockdep_rcu_enabled())
319		return 1;
320	if (!rcu_is_watching())
321		return 0;
322	if (!rcu_lockdep_current_cpu_online())
323		return 0;
324	return in_softirq() || irqs_disabled();
325}
326EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
328#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
329
330/**
331 * wakeme_after_rcu() - Callback function to awaken a task after grace period
332 * @head: Pointer to rcu_head member within rcu_synchronize structure
333 *
334 * Awaken the corresponding task now that a grace period has elapsed.
335 */
336void wakeme_after_rcu(struct rcu_head *head)
337{
338	struct rcu_synchronize *rcu;
339
340	rcu = container_of(head, struct rcu_synchronize, head);
341	complete(&rcu->completion);
342}
343EXPORT_SYMBOL_GPL(wakeme_after_rcu);
344
345void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
346		   struct rcu_synchronize *rs_array)
347{
348	int i;
 
349
350	/* Initialize and register callbacks for each flavor specified. */
351	for (i = 0; i < n; i++) {
352		if (checktiny &&
353		    (crcu_array[i] == call_rcu ||
354		     crcu_array[i] == call_rcu_bh)) {
355			might_sleep();
356			continue;
357		}
358		init_rcu_head_on_stack(&rs_array[i].head);
359		init_completion(&rs_array[i].completion);
360		(crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
 
 
 
 
 
361	}
362
363	/* Wait for all callbacks to be invoked. */
364	for (i = 0; i < n; i++) {
365		if (checktiny &&
366		    (crcu_array[i] == call_rcu ||
367		     crcu_array[i] == call_rcu_bh))
368			continue;
369		wait_for_completion(&rs_array[i].completion);
370		destroy_rcu_head_on_stack(&rs_array[i].head);
 
 
 
 
 
371	}
372}
373EXPORT_SYMBOL_GPL(__wait_rcu_gp);
374
 
 
 
 
 
 
 
375#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
376void init_rcu_head(struct rcu_head *head)
377{
378	debug_object_init(head, &rcuhead_debug_descr);
379}
 
380
381void destroy_rcu_head(struct rcu_head *head)
382{
383	debug_object_free(head, &rcuhead_debug_descr);
384}
 
385
386static bool rcuhead_is_static_object(void *addr)
387{
388	return true;
389}
390
391/**
392 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
393 * @head: pointer to rcu_head structure to be initialized
394 *
395 * This function informs debugobjects of a new rcu_head structure that
396 * has been allocated as an auto variable on the stack.  This function
397 * is not required for rcu_head structures that are statically defined or
398 * that are dynamically allocated on the heap.  This function has no
399 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
400 */
401void init_rcu_head_on_stack(struct rcu_head *head)
402{
403	debug_object_init_on_stack(head, &rcuhead_debug_descr);
404}
405EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
406
407/**
408 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
409 * @head: pointer to rcu_head structure to be initialized
410 *
411 * This function informs debugobjects that an on-stack rcu_head structure
412 * is about to go out of scope.  As with init_rcu_head_on_stack(), this
413 * function is not required for rcu_head structures that are statically
414 * defined or that are dynamically allocated on the heap.  Also as with
415 * init_rcu_head_on_stack(), this function has no effect for
416 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
417 */
418void destroy_rcu_head_on_stack(struct rcu_head *head)
419{
420	debug_object_free(head, &rcuhead_debug_descr);
421}
422EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
423
424struct debug_obj_descr rcuhead_debug_descr = {
425	.name = "rcu_head",
426	.is_static_object = rcuhead_is_static_object,
427};
428EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
429#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
430
431#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
432void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
433			       unsigned long secs,
434			       unsigned long c_old, unsigned long c)
435{
436	trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
437}
438EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
439#else
440#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
441	do { } while (0)
442#endif
443
444#ifdef CONFIG_RCU_STALL_COMMON
 
 
 
 
445
446#ifdef CONFIG_PROVE_RCU
447#define RCU_STALL_DELAY_DELTA	       (5 * HZ)
448#else
449#define RCU_STALL_DELAY_DELTA	       0
 
450#endif
451
452int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
453static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
454
 
 
 
 
 
 
 
 
455module_param(rcu_cpu_stall_suppress, int, 0644);
 
456module_param(rcu_cpu_stall_timeout, int, 0644);
457
458int rcu_jiffies_till_stall_check(void)
459{
460	int till_stall_check = READ_ONCE(rcu_cpu_stall_timeout);
461
462	/*
463	 * Limit check must be consistent with the Kconfig limits
464	 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
465	 */
466	if (till_stall_check < 3) {
467		WRITE_ONCE(rcu_cpu_stall_timeout, 3);
468		till_stall_check = 3;
469	} else if (till_stall_check > 300) {
470		WRITE_ONCE(rcu_cpu_stall_timeout, 300);
471		till_stall_check = 300;
472	}
473	return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
474}
475
476void rcu_sysrq_start(void)
477{
478	if (!rcu_cpu_stall_suppress)
479		rcu_cpu_stall_suppress = 2;
480}
481
482void rcu_sysrq_end(void)
483{
484	if (rcu_cpu_stall_suppress == 2)
485		rcu_cpu_stall_suppress = 0;
486}
487
488static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
489{
490	rcu_cpu_stall_suppress = 1;
491	return NOTIFY_DONE;
492}
493
494static struct notifier_block rcu_panic_block = {
495	.notifier_call = rcu_panic,
496};
497
498static int __init check_cpu_stall_init(void)
499{
500	atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
501	return 0;
502}
503early_initcall(check_cpu_stall_init);
504
505#endif /* #ifdef CONFIG_RCU_STALL_COMMON */
506
507#ifdef CONFIG_TASKS_RCU
508
509/*
510 * Simple variant of RCU whose quiescent states are voluntary context switch,
511 * user-space execution, and idle.  As such, grace periods can take one good
512 * long time.  There are no read-side primitives similar to rcu_read_lock()
513 * and rcu_read_unlock() because this implementation is intended to get
514 * the system into a safe state for some of the manipulations involved in
515 * tracing and the like.  Finally, this implementation does not support
516 * high call_rcu_tasks() rates from multiple CPUs.  If this is required,
517 * per-CPU callback lists will be needed.
518 */
519
520/* Global list of callbacks and associated lock. */
521static struct rcu_head *rcu_tasks_cbs_head;
522static struct rcu_head **rcu_tasks_cbs_tail = &rcu_tasks_cbs_head;
523static DECLARE_WAIT_QUEUE_HEAD(rcu_tasks_cbs_wq);
524static DEFINE_RAW_SPINLOCK(rcu_tasks_cbs_lock);
525
526/* Track exiting tasks in order to allow them to be waited for. */
527DEFINE_SRCU(tasks_rcu_exit_srcu);
528
529/* Control stall timeouts.  Disable with <= 0, otherwise jiffies till stall. */
530static int rcu_task_stall_timeout __read_mostly = HZ * 60 * 10;
531module_param(rcu_task_stall_timeout, int, 0644);
532
533static void rcu_spawn_tasks_kthread(void);
534static struct task_struct *rcu_tasks_kthread_ptr;
535
536/*
537 * Post an RCU-tasks callback.  First call must be from process context
538 * after the scheduler if fully operational.
539 */
540void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func)
541{
542	unsigned long flags;
543	bool needwake;
544	bool havetask = READ_ONCE(rcu_tasks_kthread_ptr);
545
546	rhp->next = NULL;
547	rhp->func = func;
548	raw_spin_lock_irqsave(&rcu_tasks_cbs_lock, flags);
549	needwake = !rcu_tasks_cbs_head;
550	*rcu_tasks_cbs_tail = rhp;
551	rcu_tasks_cbs_tail = &rhp->next;
552	raw_spin_unlock_irqrestore(&rcu_tasks_cbs_lock, flags);
553	/* We can't create the thread unless interrupts are enabled. */
554	if ((needwake && havetask) ||
555	    (!havetask && !irqs_disabled_flags(flags))) {
556		rcu_spawn_tasks_kthread();
557		wake_up(&rcu_tasks_cbs_wq);
558	}
559}
560EXPORT_SYMBOL_GPL(call_rcu_tasks);
561
562/**
563 * synchronize_rcu_tasks - wait until an rcu-tasks grace period has elapsed.
564 *
565 * Control will return to the caller some time after a full rcu-tasks
566 * grace period has elapsed, in other words after all currently
567 * executing rcu-tasks read-side critical sections have elapsed.  These
568 * read-side critical sections are delimited by calls to schedule(),
569 * cond_resched_rcu_qs(), idle execution, userspace execution, calls
570 * to synchronize_rcu_tasks(), and (in theory, anyway) cond_resched().
571 *
572 * This is a very specialized primitive, intended only for a few uses in
573 * tracing and other situations requiring manipulation of function
574 * preambles and profiling hooks.  The synchronize_rcu_tasks() function
575 * is not (yet) intended for heavy use from multiple CPUs.
576 *
577 * Note that this guarantee implies further memory-ordering guarantees.
578 * On systems with more than one CPU, when synchronize_rcu_tasks() returns,
579 * each CPU is guaranteed to have executed a full memory barrier since the
580 * end of its last RCU-tasks read-side critical section whose beginning
581 * preceded the call to synchronize_rcu_tasks().  In addition, each CPU
582 * having an RCU-tasks read-side critical section that extends beyond
583 * the return from synchronize_rcu_tasks() is guaranteed to have executed
584 * a full memory barrier after the beginning of synchronize_rcu_tasks()
585 * and before the beginning of that RCU-tasks read-side critical section.
586 * Note that these guarantees include CPUs that are offline, idle, or
587 * executing in user mode, as well as CPUs that are executing in the kernel.
588 *
589 * Furthermore, if CPU A invoked synchronize_rcu_tasks(), which returned
590 * to its caller on CPU B, then both CPU A and CPU B are guaranteed
591 * to have executed a full memory barrier during the execution of
592 * synchronize_rcu_tasks() -- even if CPU A and CPU B are the same CPU
593 * (but again only if the system has more than one CPU).
594 */
595void synchronize_rcu_tasks(void)
596{
597	/* Complain if the scheduler has not started.  */
598	RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
599			 "synchronize_rcu_tasks called too soon");
600
601	/* Wait for the grace period. */
602	wait_rcu_gp(call_rcu_tasks);
603}
604EXPORT_SYMBOL_GPL(synchronize_rcu_tasks);
605
606/**
607 * rcu_barrier_tasks - Wait for in-flight call_rcu_tasks() callbacks.
608 *
609 * Although the current implementation is guaranteed to wait, it is not
610 * obligated to, for example, if there are no pending callbacks.
 
611 */
612void rcu_barrier_tasks(void)
613{
614	/* There is only one callback queue, so this is easy.  ;-) */
615	synchronize_rcu_tasks();
616}
617EXPORT_SYMBOL_GPL(rcu_barrier_tasks);
618
619/* See if tasks are still holding out, complain if so. */
620static void check_holdout_task(struct task_struct *t,
621			       bool needreport, bool *firstreport)
622{
623	int cpu;
624
625	if (!READ_ONCE(t->rcu_tasks_holdout) ||
626	    t->rcu_tasks_nvcsw != READ_ONCE(t->nvcsw) ||
627	    !READ_ONCE(t->on_rq) ||
628	    (IS_ENABLED(CONFIG_NO_HZ_FULL) &&
629	     !is_idle_task(t) && t->rcu_tasks_idle_cpu >= 0)) {
630		WRITE_ONCE(t->rcu_tasks_holdout, false);
631		list_del_init(&t->rcu_tasks_holdout_list);
632		put_task_struct(t);
633		return;
634	}
635	if (!needreport)
636		return;
637	if (*firstreport) {
638		pr_err("INFO: rcu_tasks detected stalls on tasks:\n");
639		*firstreport = false;
640	}
641	cpu = task_cpu(t);
642	pr_alert("%p: %c%c nvcsw: %lu/%lu holdout: %d idle_cpu: %d/%d\n",
643		 t, ".I"[is_idle_task(t)],
644		 "N."[cpu < 0 || !tick_nohz_full_cpu(cpu)],
645		 t->rcu_tasks_nvcsw, t->nvcsw, t->rcu_tasks_holdout,
646		 t->rcu_tasks_idle_cpu, cpu);
647	sched_show_task(t);
648}
649
650/* RCU-tasks kthread that detects grace periods and invokes callbacks. */
651static int __noreturn rcu_tasks_kthread(void *arg)
652{
653	unsigned long flags;
654	struct task_struct *g, *t;
655	unsigned long lastreport;
656	struct rcu_head *list;
657	struct rcu_head *next;
658	LIST_HEAD(rcu_tasks_holdouts);
659
660	/* Run on housekeeping CPUs by default.  Sysadm can move if desired. */
661	housekeeping_affine(current);
662
663	/*
664	 * Each pass through the following loop makes one check for
665	 * newly arrived callbacks, and, if there are some, waits for
666	 * one RCU-tasks grace period and then invokes the callbacks.
667	 * This loop is terminated by the system going down.  ;-)
668	 */
669	for (;;) {
670
671		/* Pick up any new callbacks. */
672		raw_spin_lock_irqsave(&rcu_tasks_cbs_lock, flags);
673		list = rcu_tasks_cbs_head;
674		rcu_tasks_cbs_head = NULL;
675		rcu_tasks_cbs_tail = &rcu_tasks_cbs_head;
676		raw_spin_unlock_irqrestore(&rcu_tasks_cbs_lock, flags);
677
678		/* If there were none, wait a bit and start over. */
679		if (!list) {
680			wait_event_interruptible(rcu_tasks_cbs_wq,
681						 rcu_tasks_cbs_head);
682			if (!rcu_tasks_cbs_head) {
683				WARN_ON(signal_pending(current));
684				schedule_timeout_interruptible(HZ/10);
685			}
686			continue;
687		}
688
689		/*
690		 * Wait for all pre-existing t->on_rq and t->nvcsw
691		 * transitions to complete.  Invoking synchronize_sched()
692		 * suffices because all these transitions occur with
693		 * interrupts disabled.  Without this synchronize_sched(),
694		 * a read-side critical section that started before the
695		 * grace period might be incorrectly seen as having started
696		 * after the grace period.
697		 *
698		 * This synchronize_sched() also dispenses with the
699		 * need for a memory barrier on the first store to
700		 * ->rcu_tasks_holdout, as it forces the store to happen
701		 * after the beginning of the grace period.
702		 */
703		synchronize_sched();
704
705		/*
706		 * There were callbacks, so we need to wait for an
707		 * RCU-tasks grace period.  Start off by scanning
708		 * the task list for tasks that are not already
709		 * voluntarily blocked.  Mark these tasks and make
710		 * a list of them in rcu_tasks_holdouts.
711		 */
712		rcu_read_lock();
713		for_each_process_thread(g, t) {
714			if (t != current && READ_ONCE(t->on_rq) &&
715			    !is_idle_task(t)) {
716				get_task_struct(t);
717				t->rcu_tasks_nvcsw = READ_ONCE(t->nvcsw);
718				WRITE_ONCE(t->rcu_tasks_holdout, true);
719				list_add(&t->rcu_tasks_holdout_list,
720					 &rcu_tasks_holdouts);
721			}
722		}
723		rcu_read_unlock();
724
725		/*
726		 * Wait for tasks that are in the process of exiting.
727		 * This does only part of the job, ensuring that all
728		 * tasks that were previously exiting reach the point
729		 * where they have disabled preemption, allowing the
730		 * later synchronize_sched() to finish the job.
731		 */
732		synchronize_srcu(&tasks_rcu_exit_srcu);
733
734		/*
735		 * Each pass through the following loop scans the list
736		 * of holdout tasks, removing any that are no longer
737		 * holdouts.  When the list is empty, we are done.
738		 */
739		lastreport = jiffies;
740		while (!list_empty(&rcu_tasks_holdouts)) {
741			bool firstreport;
742			bool needreport;
743			int rtst;
744			struct task_struct *t1;
745
746			schedule_timeout_interruptible(HZ);
747			rtst = READ_ONCE(rcu_task_stall_timeout);
748			needreport = rtst > 0 &&
749				     time_after(jiffies, lastreport + rtst);
750			if (needreport)
751				lastreport = jiffies;
752			firstreport = true;
753			WARN_ON(signal_pending(current));
754			list_for_each_entry_safe(t, t1, &rcu_tasks_holdouts,
755						rcu_tasks_holdout_list) {
756				check_holdout_task(t, needreport, &firstreport);
757				cond_resched();
758			}
759		}
760
761		/*
762		 * Because ->on_rq and ->nvcsw are not guaranteed
763		 * to have a full memory barriers prior to them in the
764		 * schedule() path, memory reordering on other CPUs could
765		 * cause their RCU-tasks read-side critical sections to
766		 * extend past the end of the grace period.  However,
767		 * because these ->nvcsw updates are carried out with
768		 * interrupts disabled, we can use synchronize_sched()
769		 * to force the needed ordering on all such CPUs.
770		 *
771		 * This synchronize_sched() also confines all
772		 * ->rcu_tasks_holdout accesses to be within the grace
773		 * period, avoiding the need for memory barriers for
774		 * ->rcu_tasks_holdout accesses.
775		 *
776		 * In addition, this synchronize_sched() waits for exiting
777		 * tasks to complete their final preempt_disable() region
778		 * of execution, cleaning up after the synchronize_srcu()
779		 * above.
780		 */
781		synchronize_sched();
782
783		/* Invoke the callbacks. */
784		while (list) {
785			next = list->next;
786			local_bh_disable();
787			list->func(list);
788			local_bh_enable();
789			list = next;
790			cond_resched();
791		}
792		schedule_timeout_uninterruptible(HZ/10);
793	}
794}
795
796/* Spawn rcu_tasks_kthread() at first call to call_rcu_tasks(). */
797static void rcu_spawn_tasks_kthread(void)
798{
799	static DEFINE_MUTEX(rcu_tasks_kthread_mutex);
800	struct task_struct *t;
801
802	if (READ_ONCE(rcu_tasks_kthread_ptr)) {
803		smp_mb(); /* Ensure caller sees full kthread. */
804		return;
805	}
806	mutex_lock(&rcu_tasks_kthread_mutex);
807	if (rcu_tasks_kthread_ptr) {
808		mutex_unlock(&rcu_tasks_kthread_mutex);
809		return;
810	}
811	t = kthread_run(rcu_tasks_kthread, NULL, "rcu_tasks_kthread");
812	BUG_ON(IS_ERR(t));
813	smp_mb(); /* Ensure others see full kthread. */
814	WRITE_ONCE(rcu_tasks_kthread_ptr, t);
815	mutex_unlock(&rcu_tasks_kthread_mutex);
816}
817
818#endif /* #ifdef CONFIG_TASKS_RCU */
819
820/*
821 * Test each non-SRCU synchronous grace-period wait API.  This is
822 * useful just after a change in mode for these primitives, and
823 * during early boot.
824 */
825void rcu_test_sync_prims(void)
826{
827	if (!IS_ENABLED(CONFIG_PROVE_RCU))
828		return;
829	synchronize_rcu();
830	synchronize_rcu_bh();
831	synchronize_sched();
832	synchronize_rcu_expedited();
833	synchronize_rcu_bh_expedited();
834	synchronize_sched_expedited();
835}
 
836
837#ifdef CONFIG_PROVE_RCU
838
839/*
840 * Early boot self test parameters, one for each flavor
841 */
842static bool rcu_self_test;
843static bool rcu_self_test_bh;
844static bool rcu_self_test_sched;
845
846module_param(rcu_self_test, bool, 0444);
847module_param(rcu_self_test_bh, bool, 0444);
848module_param(rcu_self_test_sched, bool, 0444);
849
850static int rcu_self_test_counter;
851
852static void test_callback(struct rcu_head *r)
853{
854	rcu_self_test_counter++;
855	pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
856}
857
 
 
 
 
 
 
 
858static void early_boot_test_call_rcu(void)
859{
860	static struct rcu_head head;
 
 
 
861
 
 
862	call_rcu(&head, test_callback);
863}
864
865static void early_boot_test_call_rcu_bh(void)
866{
867	static struct rcu_head head;
868
869	call_rcu_bh(&head, test_callback);
870}
871
872static void early_boot_test_call_rcu_sched(void)
873{
874	static struct rcu_head head;
875
876	call_rcu_sched(&head, test_callback);
877}
878
879void rcu_early_boot_tests(void)
880{
881	pr_info("Running RCU self tests\n");
882
883	if (rcu_self_test)
884		early_boot_test_call_rcu();
885	if (rcu_self_test_bh)
886		early_boot_test_call_rcu_bh();
887	if (rcu_self_test_sched)
888		early_boot_test_call_rcu_sched();
889	rcu_test_sync_prims();
890}
891
892static int rcu_verify_early_boot_tests(void)
893{
894	int ret = 0;
895	int early_boot_test_counter = 0;
896
897	if (rcu_self_test) {
898		early_boot_test_counter++;
899		rcu_barrier();
900	}
901	if (rcu_self_test_bh) {
902		early_boot_test_counter++;
903		rcu_barrier_bh();
904	}
905	if (rcu_self_test_sched) {
906		early_boot_test_counter++;
907		rcu_barrier_sched();
 
 
908	}
909
910	if (rcu_self_test_counter != early_boot_test_counter) {
911		WARN_ON(1);
912		ret = -1;
913	}
914
915	return ret;
916}
917late_initcall(rcu_verify_early_boot_tests);
918#else
919void rcu_early_boot_tests(void) {}
920#endif /* CONFIG_PROVE_RCU */