xref: /linux/kernel/rcu/tree_stall.h (revision 96f30c8f0aa9923aa39b30bcaefeacf88b490231)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * RCU CPU stall warnings for normal RCU grace periods
4  *
5  * Copyright IBM Corporation, 2019
6  *
7  * Author: Paul E. McKenney <paulmck@linux.ibm.com>
8  */
9 
10 #include <linux/kvm_para.h>
11 #include <linux/rcu_notifier.h>
12 
13 //////////////////////////////////////////////////////////////////////////////
14 //
15 // Controlling CPU stall warnings, including delay calculation.
16 
17 /* panic() on RCU Stall sysctl. */
18 int sysctl_panic_on_rcu_stall __read_mostly;
19 int sysctl_max_rcu_stall_to_panic __read_mostly;
20 
21 #ifdef CONFIG_PROVE_RCU
22 #define RCU_STALL_DELAY_DELTA		(5 * HZ)
23 #else
24 #define RCU_STALL_DELAY_DELTA		0
25 #endif
26 #define RCU_STALL_MIGHT_DIV		8
27 #define RCU_STALL_MIGHT_MIN		(2 * HZ)
28 
29 int rcu_exp_jiffies_till_stall_check(void)
30 {
31 	int cpu_stall_timeout = READ_ONCE(rcu_exp_cpu_stall_timeout);
32 	int exp_stall_delay_delta = 0;
33 	int till_stall_check;
34 
35 	// Zero says to use rcu_cpu_stall_timeout, but in milliseconds.
36 	if (!cpu_stall_timeout)
37 		cpu_stall_timeout = jiffies_to_msecs(rcu_jiffies_till_stall_check());
38 
39 	// Limit check must be consistent with the Kconfig limits for
40 	// CONFIG_RCU_EXP_CPU_STALL_TIMEOUT, so check the allowed range.
41 	// The minimum clamped value is "2UL", because at least one full
42 	// tick has to be guaranteed.
43 	till_stall_check = clamp(msecs_to_jiffies(cpu_stall_timeout), 2UL, 300UL * HZ);
44 
45 	if (cpu_stall_timeout && jiffies_to_msecs(till_stall_check) != cpu_stall_timeout)
46 		WRITE_ONCE(rcu_exp_cpu_stall_timeout, jiffies_to_msecs(till_stall_check));
47 
48 #ifdef CONFIG_PROVE_RCU
49 	/* Add extra ~25% out of till_stall_check. */
50 	exp_stall_delay_delta = ((till_stall_check * 25) / 100) + 1;
51 #endif
52 
53 	return till_stall_check + exp_stall_delay_delta;
54 }
55 EXPORT_SYMBOL_GPL(rcu_exp_jiffies_till_stall_check);
56 
57 /* Limit-check stall timeouts specified at boottime and runtime. */
58 int rcu_jiffies_till_stall_check(void)
59 {
60 	int till_stall_check = READ_ONCE(rcu_cpu_stall_timeout);
61 
62 	/*
63 	 * Limit check must be consistent with the Kconfig limits
64 	 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
65 	 */
66 	if (till_stall_check < 3) {
67 		WRITE_ONCE(rcu_cpu_stall_timeout, 3);
68 		till_stall_check = 3;
69 	} else if (till_stall_check > 300) {
70 		WRITE_ONCE(rcu_cpu_stall_timeout, 300);
71 		till_stall_check = 300;
72 	}
73 	return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
74 }
75 EXPORT_SYMBOL_GPL(rcu_jiffies_till_stall_check);
76 
77 /**
78  * rcu_gp_might_be_stalled - Is it likely that the grace period is stalled?
79  *
80  * Returns @true if the current grace period is sufficiently old that
81  * it is reasonable to assume that it might be stalled.  This can be
82  * useful when deciding whether to allocate memory to enable RCU-mediated
83  * freeing on the one hand or just invoking synchronize_rcu() on the other.
84  * The latter is preferable when the grace period is stalled.
85  *
86  * Note that sampling of the .gp_start and .gp_seq fields must be done
87  * carefully to avoid false positives at the beginnings and ends of
88  * grace periods.
89  */
90 bool rcu_gp_might_be_stalled(void)
91 {
92 	unsigned long d = rcu_jiffies_till_stall_check() / RCU_STALL_MIGHT_DIV;
93 	unsigned long j = jiffies;
94 
95 	if (d < RCU_STALL_MIGHT_MIN)
96 		d = RCU_STALL_MIGHT_MIN;
97 	smp_mb(); // jiffies before .gp_seq to avoid false positives.
98 	if (!rcu_gp_in_progress())
99 		return false;
100 	// Long delays at this point avoids false positive, but a delay
101 	// of ULONG_MAX/4 jiffies voids your no-false-positive warranty.
102 	smp_mb(); // .gp_seq before second .gp_start
103 	// And ditto here.
104 	return !time_before(j, READ_ONCE(rcu_state.gp_start) + d);
105 }
106 
107 /* Don't do RCU CPU stall warnings during long sysrq printouts. */
108 void rcu_sysrq_start(void)
109 {
110 	if (!rcu_cpu_stall_suppress)
111 		rcu_cpu_stall_suppress = 2;
112 }
113 
114 void rcu_sysrq_end(void)
115 {
116 	if (rcu_cpu_stall_suppress == 2)
117 		rcu_cpu_stall_suppress = 0;
118 }
119 
120 /* Don't print RCU CPU stall warnings during a kernel panic. */
121 static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
122 {
123 	rcu_cpu_stall_suppress = 1;
124 	return NOTIFY_DONE;
125 }
126 
127 static struct notifier_block rcu_panic_block = {
128 	.notifier_call = rcu_panic,
129 };
130 
131 static int __init check_cpu_stall_init(void)
132 {
133 	atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
134 	return 0;
135 }
136 early_initcall(check_cpu_stall_init);
137 
138 /* If so specified via sysctl, panic, yielding cleaner stall-warning output. */
139 static void panic_on_rcu_stall(void)
140 {
141 	static int cpu_stall;
142 
143 	if (++cpu_stall < sysctl_max_rcu_stall_to_panic)
144 		return;
145 
146 	if (sysctl_panic_on_rcu_stall)
147 		panic("RCU Stall\n");
148 }
149 
150 /**
151  * rcu_cpu_stall_reset - restart stall-warning timeout for current grace period
152  *
153  * To perform the reset request from the caller, disable stall detection until
154  * 3 fqs loops have passed. This is required to ensure a fresh jiffies is
155  * loaded.  It should be safe to do from the fqs loop as enough timer
156  * interrupts and context switches should have passed.
157  *
158  * The caller must disable hard irqs.
159  */
160 void rcu_cpu_stall_reset(void)
161 {
162 	WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 3);
163 	WRITE_ONCE(rcu_state.jiffies_stall, ULONG_MAX);
164 }
165 
166 //////////////////////////////////////////////////////////////////////////////
167 //
168 // Interaction with RCU grace periods
169 
170 /* Start of new grace period, so record stall time (and forcing times). */
171 static void record_gp_stall_check_time(void)
172 {
173 	unsigned long j = jiffies;
174 	unsigned long j1;
175 
176 	WRITE_ONCE(rcu_state.gp_start, j);
177 	j1 = rcu_jiffies_till_stall_check();
178 	smp_mb(); // ->gp_start before ->jiffies_stall and caller's ->gp_seq.
179 	WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 0);
180 	WRITE_ONCE(rcu_state.jiffies_stall, j + j1);
181 	rcu_state.jiffies_resched = j + j1 / 2;
182 	rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs);
183 }
184 
185 /* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */
186 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
187 {
188 	rdp->ticks_this_gp = 0;
189 	rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id());
190 	WRITE_ONCE(rdp->last_fqs_resched, jiffies);
191 }
192 
193 /*
194  * If too much time has passed in the current grace period, and if
195  * so configured, go kick the relevant kthreads.
196  */
197 static void rcu_stall_kick_kthreads(void)
198 {
199 	unsigned long j;
200 
201 	if (!READ_ONCE(rcu_kick_kthreads))
202 		return;
203 	j = READ_ONCE(rcu_state.jiffies_kick_kthreads);
204 	if (time_after(jiffies, j) && rcu_state.gp_kthread &&
205 	    (rcu_gp_in_progress() || READ_ONCE(rcu_state.gp_flags))) {
206 		WARN_ONCE(1, "Kicking %s grace-period kthread\n",
207 			  rcu_state.name);
208 		rcu_ftrace_dump(DUMP_ALL);
209 		wake_up_process(rcu_state.gp_kthread);
210 		WRITE_ONCE(rcu_state.jiffies_kick_kthreads, j + HZ);
211 	}
212 }
213 
214 /*
215  * Handler for the irq_work request posted about halfway into the RCU CPU
216  * stall timeout, and used to detect excessive irq disabling.  Set state
217  * appropriately, but just complain if there is unexpected state on entry.
218  */
219 static void rcu_iw_handler(struct irq_work *iwp)
220 {
221 	struct rcu_data *rdp;
222 	struct rcu_node *rnp;
223 
224 	rdp = container_of(iwp, struct rcu_data, rcu_iw);
225 	rnp = rdp->mynode;
226 	raw_spin_lock_rcu_node(rnp);
227 	if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) {
228 		rdp->rcu_iw_gp_seq = rnp->gp_seq;
229 		rdp->rcu_iw_pending = false;
230 	}
231 	raw_spin_unlock_rcu_node(rnp);
232 }
233 
234 //////////////////////////////////////////////////////////////////////////////
235 //
236 // Printing RCU CPU stall warnings
237 
238 #ifdef CONFIG_PREEMPT_RCU
239 
240 /*
241  * Dump detailed information for all tasks blocking the current RCU
242  * grace period on the specified rcu_node structure.
243  */
244 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
245 {
246 	unsigned long flags;
247 	struct task_struct *t;
248 
249 	raw_spin_lock_irqsave_rcu_node(rnp, flags);
250 	if (!rcu_preempt_blocked_readers_cgp(rnp)) {
251 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
252 		return;
253 	}
254 	t = list_entry(rnp->gp_tasks->prev,
255 		       struct task_struct, rcu_node_entry);
256 	list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
257 		/*
258 		 * We could be printing a lot while holding a spinlock.
259 		 * Avoid triggering hard lockup.
260 		 */
261 		touch_nmi_watchdog();
262 		sched_show_task(t);
263 	}
264 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
265 }
266 
267 // Communicate task state back to the RCU CPU stall warning request.
268 struct rcu_stall_chk_rdr {
269 	int nesting;
270 	union rcu_special rs;
271 	bool on_blkd_list;
272 };
273 
274 /*
275  * Report out the state of a not-running task that is stalling the
276  * current RCU grace period.
277  */
278 static int check_slow_task(struct task_struct *t, void *arg)
279 {
280 	struct rcu_stall_chk_rdr *rscrp = arg;
281 
282 	if (task_curr(t))
283 		return -EBUSY; // It is running, so decline to inspect it.
284 	rscrp->nesting = t->rcu_read_lock_nesting;
285 	rscrp->rs = t->rcu_read_unlock_special;
286 	rscrp->on_blkd_list = !list_empty(&t->rcu_node_entry);
287 	return 0;
288 }
289 
290 /*
291  * Scan the current list of tasks blocked within RCU read-side critical
292  * sections, printing out the tid of each of the first few of them.
293  */
294 static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
295 	__releases(rnp->lock)
296 {
297 	int i = 0;
298 	int ndetected = 0;
299 	struct rcu_stall_chk_rdr rscr;
300 	struct task_struct *t;
301 	struct task_struct *ts[8];
302 
303 	lockdep_assert_irqs_disabled();
304 	if (!rcu_preempt_blocked_readers_cgp(rnp)) {
305 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
306 		return 0;
307 	}
308 	pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
309 	       rnp->level, rnp->grplo, rnp->grphi);
310 	t = list_entry(rnp->gp_tasks->prev,
311 		       struct task_struct, rcu_node_entry);
312 	list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
313 		get_task_struct(t);
314 		ts[i++] = t;
315 		if (i >= ARRAY_SIZE(ts))
316 			break;
317 	}
318 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
319 	while (i) {
320 		t = ts[--i];
321 		if (task_call_func(t, check_slow_task, &rscr))
322 			pr_cont(" P%d", t->pid);
323 		else
324 			pr_cont(" P%d/%d:%c%c%c%c",
325 				t->pid, rscr.nesting,
326 				".b"[rscr.rs.b.blocked],
327 				".q"[rscr.rs.b.need_qs],
328 				".e"[rscr.rs.b.exp_hint],
329 				".l"[rscr.on_blkd_list]);
330 		lockdep_assert_irqs_disabled();
331 		put_task_struct(t);
332 		ndetected++;
333 	}
334 	pr_cont("\n");
335 	return ndetected;
336 }
337 
338 #else /* #ifdef CONFIG_PREEMPT_RCU */
339 
340 /*
341  * Because preemptible RCU does not exist, we never have to check for
342  * tasks blocked within RCU read-side critical sections.
343  */
344 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
345 {
346 }
347 
348 /*
349  * Because preemptible RCU does not exist, we never have to check for
350  * tasks blocked within RCU read-side critical sections.
351  */
352 static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
353 	__releases(rnp->lock)
354 {
355 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
356 	return 0;
357 }
358 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
359 
360 /*
361  * Dump stacks of all tasks running on stalled CPUs.  First try using
362  * NMIs, but fall back to manual remote stack tracing on architectures
363  * that don't support NMI-based stack dumps.  The NMI-triggered stack
364  * traces are more accurate because they are printed by the target CPU.
365  */
366 static void rcu_dump_cpu_stacks(void)
367 {
368 	int cpu;
369 	unsigned long flags;
370 	struct rcu_node *rnp;
371 
372 	rcu_for_each_leaf_node(rnp) {
373 		raw_spin_lock_irqsave_rcu_node(rnp, flags);
374 		for_each_leaf_node_possible_cpu(rnp, cpu)
375 			if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
376 				if (cpu_is_offline(cpu))
377 					pr_err("Offline CPU %d blocking current GP.\n", cpu);
378 				else
379 					dump_cpu_task(cpu);
380 			}
381 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
382 	}
383 }
384 
385 static const char * const gp_state_names[] = {
386 	[RCU_GP_IDLE] = "RCU_GP_IDLE",
387 	[RCU_GP_WAIT_GPS] = "RCU_GP_WAIT_GPS",
388 	[RCU_GP_DONE_GPS] = "RCU_GP_DONE_GPS",
389 	[RCU_GP_ONOFF] = "RCU_GP_ONOFF",
390 	[RCU_GP_INIT] = "RCU_GP_INIT",
391 	[RCU_GP_WAIT_FQS] = "RCU_GP_WAIT_FQS",
392 	[RCU_GP_DOING_FQS] = "RCU_GP_DOING_FQS",
393 	[RCU_GP_CLEANUP] = "RCU_GP_CLEANUP",
394 	[RCU_GP_CLEANED] = "RCU_GP_CLEANED",
395 };
396 
397 /*
398  * Convert a ->gp_state value to a character string.
399  */
400 static const char *gp_state_getname(short gs)
401 {
402 	if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names))
403 		return "???";
404 	return gp_state_names[gs];
405 }
406 
407 /* Is the RCU grace-period kthread being starved of CPU time? */
408 static bool rcu_is_gp_kthread_starving(unsigned long *jp)
409 {
410 	unsigned long j = jiffies - READ_ONCE(rcu_state.gp_activity);
411 
412 	if (jp)
413 		*jp = j;
414 	return j > 2 * HZ;
415 }
416 
417 static bool rcu_is_rcuc_kthread_starving(struct rcu_data *rdp, unsigned long *jp)
418 {
419 	int cpu;
420 	struct task_struct *rcuc;
421 	unsigned long j;
422 
423 	rcuc = rdp->rcu_cpu_kthread_task;
424 	if (!rcuc)
425 		return false;
426 
427 	cpu = task_cpu(rcuc);
428 	if (cpu_is_offline(cpu) || idle_cpu(cpu))
429 		return false;
430 
431 	j = jiffies - READ_ONCE(rdp->rcuc_activity);
432 
433 	if (jp)
434 		*jp = j;
435 	return j > 2 * HZ;
436 }
437 
438 static void print_cpu_stat_info(int cpu)
439 {
440 	struct rcu_snap_record rsr, *rsrp;
441 	struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
442 	struct kernel_cpustat *kcsp = &kcpustat_cpu(cpu);
443 
444 	if (!rcu_cpu_stall_cputime)
445 		return;
446 
447 	rsrp = &rdp->snap_record;
448 	if (rsrp->gp_seq != rdp->gp_seq)
449 		return;
450 
451 	rsr.cputime_irq     = kcpustat_field(kcsp, CPUTIME_IRQ, cpu);
452 	rsr.cputime_softirq = kcpustat_field(kcsp, CPUTIME_SOFTIRQ, cpu);
453 	rsr.cputime_system  = kcpustat_field(kcsp, CPUTIME_SYSTEM, cpu);
454 
455 	pr_err("\t         hardirqs   softirqs   csw/system\n");
456 	pr_err("\t number: %8ld %10d %12lld\n",
457 		kstat_cpu_irqs_sum(cpu) - rsrp->nr_hardirqs,
458 		kstat_cpu_softirqs_sum(cpu) - rsrp->nr_softirqs,
459 		nr_context_switches_cpu(cpu) - rsrp->nr_csw);
460 	pr_err("\tcputime: %8lld %10lld %12lld   ==> %d(ms)\n",
461 		div_u64(rsr.cputime_irq - rsrp->cputime_irq, NSEC_PER_MSEC),
462 		div_u64(rsr.cputime_softirq - rsrp->cputime_softirq, NSEC_PER_MSEC),
463 		div_u64(rsr.cputime_system - rsrp->cputime_system, NSEC_PER_MSEC),
464 		jiffies_to_msecs(jiffies - rsrp->jiffies));
465 }
466 
467 /*
468  * Print out diagnostic information for the specified stalled CPU.
469  *
470  * If the specified CPU is aware of the current RCU grace period, then
471  * print the number of scheduling clock interrupts the CPU has taken
472  * during the time that it has been aware.  Otherwise, print the number
473  * of RCU grace periods that this CPU is ignorant of, for example, "1"
474  * if the CPU was aware of the previous grace period.
475  *
476  * Also print out idle info.
477  */
478 static void print_cpu_stall_info(int cpu)
479 {
480 	unsigned long delta;
481 	bool falsepositive;
482 	struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
483 	char *ticks_title;
484 	unsigned long ticks_value;
485 	bool rcuc_starved;
486 	unsigned long j;
487 	char buf[32];
488 
489 	/*
490 	 * We could be printing a lot while holding a spinlock.  Avoid
491 	 * triggering hard lockup.
492 	 */
493 	touch_nmi_watchdog();
494 
495 	ticks_value = rcu_seq_ctr(rcu_state.gp_seq - rdp->gp_seq);
496 	if (ticks_value) {
497 		ticks_title = "GPs behind";
498 	} else {
499 		ticks_title = "ticks this GP";
500 		ticks_value = rdp->ticks_this_gp;
501 	}
502 	delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq);
503 	falsepositive = rcu_is_gp_kthread_starving(NULL) &&
504 			rcu_dynticks_in_eqs(ct_dynticks_cpu(cpu));
505 	rcuc_starved = rcu_is_rcuc_kthread_starving(rdp, &j);
506 	if (rcuc_starved)
507 		// Print signed value, as negative values indicate a probable bug.
508 		snprintf(buf, sizeof(buf), " rcuc=%ld jiffies(starved)", j);
509 	pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%04x/%ld/%#lx softirq=%u/%u fqs=%ld%s%s\n",
510 	       cpu,
511 	       "O."[!!cpu_online(cpu)],
512 	       "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
513 	       "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)],
514 	       !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' :
515 			rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' :
516 				"!."[!delta],
517 	       ticks_value, ticks_title,
518 	       ct_dynticks_cpu(cpu) & 0xffff,
519 	       ct_dynticks_nesting_cpu(cpu), ct_dynticks_nmi_nesting_cpu(cpu),
520 	       rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
521 	       data_race(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
522 	       rcuc_starved ? buf : "",
523 	       falsepositive ? " (false positive?)" : "");
524 
525 	print_cpu_stat_info(cpu);
526 }
527 
528 /* Complain about starvation of grace-period kthread.  */
529 static void rcu_check_gp_kthread_starvation(void)
530 {
531 	int cpu;
532 	struct task_struct *gpk = rcu_state.gp_kthread;
533 	unsigned long j;
534 
535 	if (rcu_is_gp_kthread_starving(&j)) {
536 		cpu = gpk ? task_cpu(gpk) : -1;
537 		pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x ->cpu=%d\n",
538 		       rcu_state.name, j,
539 		       (long)rcu_seq_current(&rcu_state.gp_seq),
540 		       data_race(READ_ONCE(rcu_state.gp_flags)),
541 		       gp_state_getname(rcu_state.gp_state),
542 		       data_race(READ_ONCE(rcu_state.gp_state)),
543 		       gpk ? data_race(READ_ONCE(gpk->__state)) : ~0, cpu);
544 		if (gpk) {
545 			struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
546 
547 			pr_err("\tUnless %s kthread gets sufficient CPU time, OOM is now expected behavior.\n", rcu_state.name);
548 			pr_err("RCU grace-period kthread stack dump:\n");
549 			sched_show_task(gpk);
550 			if (cpu_is_offline(cpu)) {
551 				pr_err("RCU GP kthread last ran on offline CPU %d.\n", cpu);
552 			} else if (!(data_race(READ_ONCE(rdp->mynode->qsmask)) & rdp->grpmask)) {
553 				pr_err("Stack dump where RCU GP kthread last ran:\n");
554 				dump_cpu_task(cpu);
555 			}
556 			wake_up_process(gpk);
557 		}
558 	}
559 }
560 
561 /* Complain about missing wakeups from expired fqs wait timer */
562 static void rcu_check_gp_kthread_expired_fqs_timer(void)
563 {
564 	struct task_struct *gpk = rcu_state.gp_kthread;
565 	short gp_state;
566 	unsigned long jiffies_fqs;
567 	int cpu;
568 
569 	/*
570 	 * Order reads of .gp_state and .jiffies_force_qs.
571 	 * Matching smp_wmb() is present in rcu_gp_fqs_loop().
572 	 */
573 	gp_state = smp_load_acquire(&rcu_state.gp_state);
574 	jiffies_fqs = READ_ONCE(rcu_state.jiffies_force_qs);
575 
576 	if (gp_state == RCU_GP_WAIT_FQS &&
577 	    time_after(jiffies, jiffies_fqs + RCU_STALL_MIGHT_MIN) &&
578 	    gpk && !READ_ONCE(gpk->on_rq)) {
579 		cpu = task_cpu(gpk);
580 		pr_err("%s kthread timer wakeup didn't happen for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x\n",
581 		       rcu_state.name, (jiffies - jiffies_fqs),
582 		       (long)rcu_seq_current(&rcu_state.gp_seq),
583 		       data_race(READ_ONCE(rcu_state.gp_flags)), // Diagnostic read
584 		       gp_state_getname(RCU_GP_WAIT_FQS), RCU_GP_WAIT_FQS,
585 		       data_race(READ_ONCE(gpk->__state)));
586 		pr_err("\tPossible timer handling issue on cpu=%d timer-softirq=%u\n",
587 		       cpu, kstat_softirqs_cpu(TIMER_SOFTIRQ, cpu));
588 	}
589 }
590 
591 static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
592 {
593 	int cpu;
594 	unsigned long flags;
595 	unsigned long gpa;
596 	unsigned long j;
597 	int ndetected = 0;
598 	struct rcu_node *rnp;
599 	long totqlen = 0;
600 
601 	lockdep_assert_irqs_disabled();
602 
603 	/* Kick and suppress, if so configured. */
604 	rcu_stall_kick_kthreads();
605 	if (rcu_stall_is_suppressed())
606 		return;
607 
608 	/*
609 	 * OK, time to rat on our buddy...
610 	 * See Documentation/RCU/stallwarn.rst for info on how to debug
611 	 * RCU CPU stall warnings.
612 	 */
613 	trace_rcu_stall_warning(rcu_state.name, TPS("StallDetected"));
614 	pr_err("INFO: %s detected stalls on CPUs/tasks:\n", rcu_state.name);
615 	rcu_for_each_leaf_node(rnp) {
616 		raw_spin_lock_irqsave_rcu_node(rnp, flags);
617 		if (rnp->qsmask != 0) {
618 			for_each_leaf_node_possible_cpu(rnp, cpu)
619 				if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
620 					print_cpu_stall_info(cpu);
621 					ndetected++;
622 				}
623 		}
624 		ndetected += rcu_print_task_stall(rnp, flags); // Releases rnp->lock.
625 		lockdep_assert_irqs_disabled();
626 	}
627 
628 	for_each_possible_cpu(cpu)
629 		totqlen += rcu_get_n_cbs_cpu(cpu);
630 	pr_err("\t(detected by %d, t=%ld jiffies, g=%ld, q=%lu ncpus=%d)\n",
631 	       smp_processor_id(), (long)(jiffies - gps),
632 	       (long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
633 	       data_race(rcu_state.n_online_cpus)); // Diagnostic read
634 	if (ndetected) {
635 		rcu_dump_cpu_stacks();
636 
637 		/* Complain about tasks blocking the grace period. */
638 		rcu_for_each_leaf_node(rnp)
639 			rcu_print_detail_task_stall_rnp(rnp);
640 	} else {
641 		if (rcu_seq_current(&rcu_state.gp_seq) != gp_seq) {
642 			pr_err("INFO: Stall ended before state dump start\n");
643 		} else {
644 			j = jiffies;
645 			gpa = data_race(READ_ONCE(rcu_state.gp_activity));
646 			pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
647 			       rcu_state.name, j - gpa, j, gpa,
648 			       data_race(READ_ONCE(jiffies_till_next_fqs)),
649 			       data_race(READ_ONCE(rcu_get_root()->qsmask)));
650 		}
651 	}
652 	/* Rewrite if needed in case of slow consoles. */
653 	if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
654 		WRITE_ONCE(rcu_state.jiffies_stall,
655 			   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
656 
657 	rcu_check_gp_kthread_expired_fqs_timer();
658 	rcu_check_gp_kthread_starvation();
659 
660 	panic_on_rcu_stall();
661 
662 	rcu_force_quiescent_state();  /* Kick them all. */
663 }
664 
665 static void print_cpu_stall(unsigned long gps)
666 {
667 	int cpu;
668 	unsigned long flags;
669 	struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
670 	struct rcu_node *rnp = rcu_get_root();
671 	long totqlen = 0;
672 
673 	lockdep_assert_irqs_disabled();
674 
675 	/* Kick and suppress, if so configured. */
676 	rcu_stall_kick_kthreads();
677 	if (rcu_stall_is_suppressed())
678 		return;
679 
680 	/*
681 	 * OK, time to rat on ourselves...
682 	 * See Documentation/RCU/stallwarn.rst for info on how to debug
683 	 * RCU CPU stall warnings.
684 	 */
685 	trace_rcu_stall_warning(rcu_state.name, TPS("SelfDetected"));
686 	pr_err("INFO: %s self-detected stall on CPU\n", rcu_state.name);
687 	raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags);
688 	print_cpu_stall_info(smp_processor_id());
689 	raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags);
690 	for_each_possible_cpu(cpu)
691 		totqlen += rcu_get_n_cbs_cpu(cpu);
692 	pr_err("\t(t=%lu jiffies g=%ld q=%lu ncpus=%d)\n",
693 		jiffies - gps,
694 		(long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
695 		data_race(rcu_state.n_online_cpus)); // Diagnostic read
696 
697 	rcu_check_gp_kthread_expired_fqs_timer();
698 	rcu_check_gp_kthread_starvation();
699 
700 	rcu_dump_cpu_stacks();
701 
702 	raw_spin_lock_irqsave_rcu_node(rnp, flags);
703 	/* Rewrite if needed in case of slow consoles. */
704 	if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
705 		WRITE_ONCE(rcu_state.jiffies_stall,
706 			   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
707 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
708 
709 	panic_on_rcu_stall();
710 
711 	/*
712 	 * Attempt to revive the RCU machinery by forcing a context switch.
713 	 *
714 	 * A context switch would normally allow the RCU state machine to make
715 	 * progress and it could be we're stuck in kernel space without context
716 	 * switches for an entirely unreasonable amount of time.
717 	 */
718 	set_tsk_need_resched(current);
719 	set_preempt_need_resched();
720 }
721 
722 static void check_cpu_stall(struct rcu_data *rdp)
723 {
724 	bool self_detected;
725 	unsigned long gs1;
726 	unsigned long gs2;
727 	unsigned long gps;
728 	unsigned long j;
729 	unsigned long jn;
730 	unsigned long js;
731 	struct rcu_node *rnp;
732 
733 	lockdep_assert_irqs_disabled();
734 	if ((rcu_stall_is_suppressed() && !READ_ONCE(rcu_kick_kthreads)) ||
735 	    !rcu_gp_in_progress())
736 		return;
737 	rcu_stall_kick_kthreads();
738 
739 	/*
740 	 * Check if it was requested (via rcu_cpu_stall_reset()) that the FQS
741 	 * loop has to set jiffies to ensure a non-stale jiffies value. This
742 	 * is required to have good jiffies value after coming out of long
743 	 * breaks of jiffies updates. Not doing so can cause false positives.
744 	 */
745 	if (READ_ONCE(rcu_state.nr_fqs_jiffies_stall) > 0)
746 		return;
747 
748 	j = jiffies;
749 
750 	/*
751 	 * Lots of memory barriers to reject false positives.
752 	 *
753 	 * The idea is to pick up rcu_state.gp_seq, then
754 	 * rcu_state.jiffies_stall, then rcu_state.gp_start, and finally
755 	 * another copy of rcu_state.gp_seq.  These values are updated in
756 	 * the opposite order with memory barriers (or equivalent) during
757 	 * grace-period initialization and cleanup.  Now, a false positive
758 	 * can occur if we get an new value of rcu_state.gp_start and a old
759 	 * value of rcu_state.jiffies_stall.  But given the memory barriers,
760 	 * the only way that this can happen is if one grace period ends
761 	 * and another starts between these two fetches.  This is detected
762 	 * by comparing the second fetch of rcu_state.gp_seq with the
763 	 * previous fetch from rcu_state.gp_seq.
764 	 *
765 	 * Given this check, comparisons of jiffies, rcu_state.jiffies_stall,
766 	 * and rcu_state.gp_start suffice to forestall false positives.
767 	 */
768 	gs1 = READ_ONCE(rcu_state.gp_seq);
769 	smp_rmb(); /* Pick up ->gp_seq first... */
770 	js = READ_ONCE(rcu_state.jiffies_stall);
771 	smp_rmb(); /* ...then ->jiffies_stall before the rest... */
772 	gps = READ_ONCE(rcu_state.gp_start);
773 	smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */
774 	gs2 = READ_ONCE(rcu_state.gp_seq);
775 	if (gs1 != gs2 ||
776 	    ULONG_CMP_LT(j, js) ||
777 	    ULONG_CMP_GE(gps, js))
778 		return; /* No stall or GP completed since entering function. */
779 	rnp = rdp->mynode;
780 	jn = jiffies + ULONG_MAX / 2;
781 	self_detected = READ_ONCE(rnp->qsmask) & rdp->grpmask;
782 	if (rcu_gp_in_progress() &&
783 	    (self_detected || ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) &&
784 	    cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) {
785 		/*
786 		 * If a virtual machine is stopped by the host it can look to
787 		 * the watchdog like an RCU stall. Check to see if the host
788 		 * stopped the vm.
789 		 */
790 		if (kvm_check_and_clear_guest_paused())
791 			return;
792 
793 		rcu_stall_notifier_call_chain(RCU_STALL_NOTIFY_NORM, (void *)j - gps);
794 		if (self_detected) {
795 			/* We haven't checked in, so go dump stack. */
796 			print_cpu_stall(gps);
797 		} else {
798 			/* They had a few time units to dump stack, so complain. */
799 			print_other_cpu_stall(gs2, gps);
800 		}
801 
802 		if (READ_ONCE(rcu_cpu_stall_ftrace_dump))
803 			rcu_ftrace_dump(DUMP_ALL);
804 
805 		if (READ_ONCE(rcu_state.jiffies_stall) == jn) {
806 			jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
807 			WRITE_ONCE(rcu_state.jiffies_stall, jn);
808 		}
809 	}
810 }
811 
812 //////////////////////////////////////////////////////////////////////////////
813 //
814 // RCU forward-progress mechanisms, including for callback invocation.
815 
816 
817 /*
818  * Check to see if a failure to end RCU priority inversion was due to
819  * a CPU not passing through a quiescent state.  When this happens, there
820  * is nothing that RCU priority boosting can do to help, so we shouldn't
821  * count this as an RCU priority boosting failure.  A return of true says
822  * RCU priority boosting is to blame, and false says otherwise.  If false
823  * is returned, the first of the CPUs to blame is stored through cpup.
824  * If there was no CPU blocking the current grace period, but also nothing
825  * in need of being boosted, *cpup is set to -1.  This can happen in case
826  * of vCPU preemption while the last CPU is reporting its quiscent state,
827  * for example.
828  *
829  * If cpup is NULL, then a lockless quick check is carried out, suitable
830  * for high-rate usage.  On the other hand, if cpup is non-NULL, each
831  * rcu_node structure's ->lock is acquired, ruling out high-rate usage.
832  */
833 bool rcu_check_boost_fail(unsigned long gp_state, int *cpup)
834 {
835 	bool atb = false;
836 	int cpu;
837 	unsigned long flags;
838 	struct rcu_node *rnp;
839 
840 	rcu_for_each_leaf_node(rnp) {
841 		if (!cpup) {
842 			if (data_race(READ_ONCE(rnp->qsmask))) {
843 				return false;
844 			} else {
845 				if (READ_ONCE(rnp->gp_tasks))
846 					atb = true;
847 				continue;
848 			}
849 		}
850 		*cpup = -1;
851 		raw_spin_lock_irqsave_rcu_node(rnp, flags);
852 		if (rnp->gp_tasks)
853 			atb = true;
854 		if (!rnp->qsmask) {
855 			// No CPUs without quiescent states for this rnp.
856 			raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
857 			continue;
858 		}
859 		// Find the first holdout CPU.
860 		for_each_leaf_node_possible_cpu(rnp, cpu) {
861 			if (rnp->qsmask & (1UL << (cpu - rnp->grplo))) {
862 				raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
863 				*cpup = cpu;
864 				return false;
865 			}
866 		}
867 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
868 	}
869 	// Can't blame CPUs, so must blame RCU priority boosting.
870 	return atb;
871 }
872 EXPORT_SYMBOL_GPL(rcu_check_boost_fail);
873 
874 /*
875  * Show the state of the grace-period kthreads.
876  */
877 void show_rcu_gp_kthreads(void)
878 {
879 	unsigned long cbs = 0;
880 	int cpu;
881 	unsigned long j;
882 	unsigned long ja;
883 	unsigned long jr;
884 	unsigned long js;
885 	unsigned long jw;
886 	struct rcu_data *rdp;
887 	struct rcu_node *rnp;
888 	struct task_struct *t = READ_ONCE(rcu_state.gp_kthread);
889 
890 	j = jiffies;
891 	ja = j - data_race(READ_ONCE(rcu_state.gp_activity));
892 	jr = j - data_race(READ_ONCE(rcu_state.gp_req_activity));
893 	js = j - data_race(READ_ONCE(rcu_state.gp_start));
894 	jw = j - data_race(READ_ONCE(rcu_state.gp_wake_time));
895 	pr_info("%s: wait state: %s(%d) ->state: %#x ->rt_priority %u delta ->gp_start %lu ->gp_activity %lu ->gp_req_activity %lu ->gp_wake_time %lu ->gp_wake_seq %ld ->gp_seq %ld ->gp_seq_needed %ld ->gp_max %lu ->gp_flags %#x\n",
896 		rcu_state.name, gp_state_getname(rcu_state.gp_state),
897 		data_race(READ_ONCE(rcu_state.gp_state)),
898 		t ? data_race(READ_ONCE(t->__state)) : 0x1ffff, t ? t->rt_priority : 0xffU,
899 		js, ja, jr, jw, (long)data_race(READ_ONCE(rcu_state.gp_wake_seq)),
900 		(long)data_race(READ_ONCE(rcu_state.gp_seq)),
901 		(long)data_race(READ_ONCE(rcu_get_root()->gp_seq_needed)),
902 		data_race(READ_ONCE(rcu_state.gp_max)),
903 		data_race(READ_ONCE(rcu_state.gp_flags)));
904 	rcu_for_each_node_breadth_first(rnp) {
905 		if (ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq), READ_ONCE(rnp->gp_seq_needed)) &&
906 		    !data_race(READ_ONCE(rnp->qsmask)) && !data_race(READ_ONCE(rnp->boost_tasks)) &&
907 		    !data_race(READ_ONCE(rnp->exp_tasks)) && !data_race(READ_ONCE(rnp->gp_tasks)))
908 			continue;
909 		pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld ->qsmask %#lx %c%c%c%c ->n_boosts %ld\n",
910 			rnp->grplo, rnp->grphi,
911 			(long)data_race(READ_ONCE(rnp->gp_seq)),
912 			(long)data_race(READ_ONCE(rnp->gp_seq_needed)),
913 			data_race(READ_ONCE(rnp->qsmask)),
914 			".b"[!!data_race(READ_ONCE(rnp->boost_kthread_task))],
915 			".B"[!!data_race(READ_ONCE(rnp->boost_tasks))],
916 			".E"[!!data_race(READ_ONCE(rnp->exp_tasks))],
917 			".G"[!!data_race(READ_ONCE(rnp->gp_tasks))],
918 			data_race(READ_ONCE(rnp->n_boosts)));
919 		if (!rcu_is_leaf_node(rnp))
920 			continue;
921 		for_each_leaf_node_possible_cpu(rnp, cpu) {
922 			rdp = per_cpu_ptr(&rcu_data, cpu);
923 			if (READ_ONCE(rdp->gpwrap) ||
924 			    ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq),
925 					 READ_ONCE(rdp->gp_seq_needed)))
926 				continue;
927 			pr_info("\tcpu %d ->gp_seq_needed %ld\n",
928 				cpu, (long)data_race(READ_ONCE(rdp->gp_seq_needed)));
929 		}
930 	}
931 	for_each_possible_cpu(cpu) {
932 		rdp = per_cpu_ptr(&rcu_data, cpu);
933 		cbs += data_race(READ_ONCE(rdp->n_cbs_invoked));
934 		if (rcu_segcblist_is_offloaded(&rdp->cblist))
935 			show_rcu_nocb_state(rdp);
936 	}
937 	pr_info("RCU callbacks invoked since boot: %lu\n", cbs);
938 	show_rcu_tasks_gp_kthreads();
939 }
940 EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
941 
942 /*
943  * This function checks for grace-period requests that fail to motivate
944  * RCU to come out of its idle mode.
945  */
946 static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
947 				     const unsigned long gpssdelay)
948 {
949 	unsigned long flags;
950 	unsigned long j;
951 	struct rcu_node *rnp_root = rcu_get_root();
952 	static atomic_t warned = ATOMIC_INIT(0);
953 
954 	if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() ||
955 	    ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
956 			 READ_ONCE(rnp_root->gp_seq_needed)) ||
957 	    !smp_load_acquire(&rcu_state.gp_kthread)) // Get stable kthread.
958 		return;
959 	j = jiffies; /* Expensive access, and in common case don't get here. */
960 	if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
961 	    time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
962 	    atomic_read(&warned))
963 		return;
964 
965 	raw_spin_lock_irqsave_rcu_node(rnp, flags);
966 	j = jiffies;
967 	if (rcu_gp_in_progress() ||
968 	    ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
969 			 READ_ONCE(rnp_root->gp_seq_needed)) ||
970 	    time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
971 	    time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
972 	    atomic_read(&warned)) {
973 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
974 		return;
975 	}
976 	/* Hold onto the leaf lock to make others see warned==1. */
977 
978 	if (rnp_root != rnp)
979 		raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
980 	j = jiffies;
981 	if (rcu_gp_in_progress() ||
982 	    ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
983 			 READ_ONCE(rnp_root->gp_seq_needed)) ||
984 	    time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
985 	    time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
986 	    atomic_xchg(&warned, 1)) {
987 		if (rnp_root != rnp)
988 			/* irqs remain disabled. */
989 			raw_spin_unlock_rcu_node(rnp_root);
990 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
991 		return;
992 	}
993 	WARN_ON(1);
994 	if (rnp_root != rnp)
995 		raw_spin_unlock_rcu_node(rnp_root);
996 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
997 	show_rcu_gp_kthreads();
998 }
999 
1000 /*
1001  * Do a forward-progress check for rcutorture.  This is normally invoked
1002  * due to an OOM event.  The argument "j" gives the time period during
1003  * which rcutorture would like progress to have been made.
1004  */
1005 void rcu_fwd_progress_check(unsigned long j)
1006 {
1007 	unsigned long cbs;
1008 	int cpu;
1009 	unsigned long max_cbs = 0;
1010 	int max_cpu = -1;
1011 	struct rcu_data *rdp;
1012 
1013 	if (rcu_gp_in_progress()) {
1014 		pr_info("%s: GP age %lu jiffies\n",
1015 			__func__, jiffies - data_race(READ_ONCE(rcu_state.gp_start)));
1016 		show_rcu_gp_kthreads();
1017 	} else {
1018 		pr_info("%s: Last GP end %lu jiffies ago\n",
1019 			__func__, jiffies - data_race(READ_ONCE(rcu_state.gp_end)));
1020 		preempt_disable();
1021 		rdp = this_cpu_ptr(&rcu_data);
1022 		rcu_check_gp_start_stall(rdp->mynode, rdp, j);
1023 		preempt_enable();
1024 	}
1025 	for_each_possible_cpu(cpu) {
1026 		cbs = rcu_get_n_cbs_cpu(cpu);
1027 		if (!cbs)
1028 			continue;
1029 		if (max_cpu < 0)
1030 			pr_info("%s: callbacks", __func__);
1031 		pr_cont(" %d: %lu", cpu, cbs);
1032 		if (cbs <= max_cbs)
1033 			continue;
1034 		max_cbs = cbs;
1035 		max_cpu = cpu;
1036 	}
1037 	if (max_cpu >= 0)
1038 		pr_cont("\n");
1039 }
1040 EXPORT_SYMBOL_GPL(rcu_fwd_progress_check);
1041 
1042 /* Commandeer a sysrq key to dump RCU's tree. */
1043 static bool sysrq_rcu;
1044 module_param(sysrq_rcu, bool, 0444);
1045 
1046 /* Dump grace-period-request information due to commandeered sysrq. */
1047 static void sysrq_show_rcu(u8 key)
1048 {
1049 	show_rcu_gp_kthreads();
1050 }
1051 
1052 static const struct sysrq_key_op sysrq_rcudump_op = {
1053 	.handler = sysrq_show_rcu,
1054 	.help_msg = "show-rcu(y)",
1055 	.action_msg = "Show RCU tree",
1056 	.enable_mask = SYSRQ_ENABLE_DUMP,
1057 };
1058 
1059 static int __init rcu_sysrq_init(void)
1060 {
1061 	if (sysrq_rcu)
1062 		return register_sysrq_key('y', &sysrq_rcudump_op);
1063 	return 0;
1064 }
1065 early_initcall(rcu_sysrq_init);
1066 
1067 #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1068 
1069 //////////////////////////////////////////////////////////////////////////////
1070 //
1071 // RCU CPU stall-warning notifiers
1072 
1073 static ATOMIC_NOTIFIER_HEAD(rcu_cpu_stall_notifier_list);
1074 
1075 /**
1076  * rcu_stall_chain_notifier_register - Add an RCU CPU stall notifier
1077  * @n: Entry to add.
1078  *
1079  * Adds an RCU CPU stall notifier to an atomic notifier chain.
1080  * The @action passed to a notifier will be @RCU_STALL_NOTIFY_NORM or
1081  * friends.  The @data will be the duration of the stalled grace period,
1082  * in jiffies, coerced to a void* pointer.
1083  *
1084  * Returns 0 on success, %-EEXIST on error.
1085  */
1086 int rcu_stall_chain_notifier_register(struct notifier_block *n)
1087 {
1088 	int rcsn = rcu_cpu_stall_notifiers;
1089 
1090 	WARN(1, "Adding %pS() to RCU stall notifier list (%s).\n", n->notifier_call,
1091 	     rcsn ? "possibly suppressing RCU CPU stall warnings" : "failed, so all is well");
1092 	if (rcsn)
1093 		return atomic_notifier_chain_register(&rcu_cpu_stall_notifier_list, n);
1094 	return -EEXIST;
1095 }
1096 EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_register);
1097 
1098 /**
1099  * rcu_stall_chain_notifier_unregister - Remove an RCU CPU stall notifier
1100  * @n: Entry to add.
1101  *
1102  * Removes an RCU CPU stall notifier from an atomic notifier chain.
1103  *
1104  * Returns zero on success, %-ENOENT on failure.
1105  */
1106 int rcu_stall_chain_notifier_unregister(struct notifier_block *n)
1107 {
1108 	return atomic_notifier_chain_unregister(&rcu_cpu_stall_notifier_list, n);
1109 }
1110 EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_unregister);
1111 
1112 /*
1113  * rcu_stall_notifier_call_chain - Call functions in an RCU CPU stall notifier chain
1114  * @val: Value passed unmodified to notifier function
1115  * @v: Pointer passed unmodified to notifier function
1116  *
1117  * Calls each function in the RCU CPU stall notifier chain in turn, which
1118  * is an atomic call chain.  See atomic_notifier_call_chain() for more
1119  * information.
1120  *
1121  * This is for use within RCU, hence the omission of the extra asterisk
1122  * to indicate a non-kerneldoc format header comment.
1123  */
1124 int rcu_stall_notifier_call_chain(unsigned long val, void *v)
1125 {
1126 	return atomic_notifier_call_chain(&rcu_cpu_stall_notifier_list, val, v);
1127 }
1128 
1129 #endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1130