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/console.h>
11#include <linux/kvm_para.h>
12#include <linux/rcu_notifier.h>
13#include <linux/smp.h>
14
15//////////////////////////////////////////////////////////////////////////////
16//
17// Controlling CPU stall warnings, including delay calculation.
18
19/* panic() on RCU Stall sysctl. */
20static int sysctl_panic_on_rcu_stall __read_mostly;
21static int sysctl_max_rcu_stall_to_panic __read_mostly;
22
23static const struct ctl_table rcu_stall_sysctl_table[] = {
24 {
25 .procname = "panic_on_rcu_stall",
26 .data = &sysctl_panic_on_rcu_stall,
27 .maxlen = sizeof(sysctl_panic_on_rcu_stall),
28 .mode = 0644,
29 .proc_handler = proc_dointvec_minmax,
30 .extra1 = SYSCTL_ZERO,
31 .extra2 = SYSCTL_ONE,
32 },
33 {
34 .procname = "max_rcu_stall_to_panic",
35 .data = &sysctl_max_rcu_stall_to_panic,
36 .maxlen = sizeof(sysctl_max_rcu_stall_to_panic),
37 .mode = 0644,
38 .proc_handler = proc_dointvec_minmax,
39 .extra1 = SYSCTL_ONE,
40 .extra2 = SYSCTL_INT_MAX,
41 },
42};
43
44static int __init init_rcu_stall_sysctl(void)
45{
46 register_sysctl_init("kernel", rcu_stall_sysctl_table);
47 return 0;
48}
49
50subsys_initcall(init_rcu_stall_sysctl);
51
52#ifdef CONFIG_SYSFS
53
54static unsigned int rcu_stall_count;
55
56static ssize_t rcu_stall_count_show(struct kobject *kobj, struct kobj_attribute *attr,
57 char *page)
58{
59 return sysfs_emit(buf: page, fmt: "%u\n", rcu_stall_count);
60}
61
62static struct kobj_attribute rcu_stall_count_attr = __ATTR_RO(rcu_stall_count);
63
64static __init int kernel_rcu_stall_sysfs_init(void)
65{
66 sysfs_add_file_to_group(kobj: kernel_kobj, attr: &rcu_stall_count_attr.attr, NULL);
67 return 0;
68}
69
70late_initcall(kernel_rcu_stall_sysfs_init);
71
72#endif // CONFIG_SYSFS
73
74#ifdef CONFIG_PROVE_RCU
75#define RCU_STALL_DELAY_DELTA (5 * HZ)
76#else
77#define RCU_STALL_DELAY_DELTA 0
78#endif
79#define RCU_STALL_MIGHT_DIV 8
80#define RCU_STALL_MIGHT_MIN (2 * HZ)
81
82int rcu_exp_jiffies_till_stall_check(void)
83{
84 int cpu_stall_timeout = READ_ONCE(rcu_exp_cpu_stall_timeout);
85 int exp_stall_delay_delta = 0;
86 int till_stall_check;
87
88 // Zero says to use rcu_cpu_stall_timeout, but in milliseconds.
89 if (!cpu_stall_timeout)
90 cpu_stall_timeout = jiffies_to_msecs(j: rcu_jiffies_till_stall_check());
91
92 // Limit check must be consistent with the Kconfig limits for
93 // CONFIG_RCU_EXP_CPU_STALL_TIMEOUT, so check the allowed range.
94 // The minimum clamped value is "2UL", because at least one full
95 // tick has to be guaranteed.
96 till_stall_check = clamp(msecs_to_jiffies(cpu_stall_timeout), 2UL, 300UL * HZ);
97
98 if (cpu_stall_timeout && jiffies_to_msecs(j: till_stall_check) != cpu_stall_timeout)
99 WRITE_ONCE(rcu_exp_cpu_stall_timeout, jiffies_to_msecs(till_stall_check));
100
101#ifdef CONFIG_PROVE_RCU
102 /* Add extra ~25% out of till_stall_check. */
103 exp_stall_delay_delta = ((till_stall_check * 25) / 100) + 1;
104#endif
105
106 return till_stall_check + exp_stall_delay_delta;
107}
108EXPORT_SYMBOL_GPL(rcu_exp_jiffies_till_stall_check);
109
110/* Limit-check stall timeouts specified at boottime and runtime. */
111int rcu_jiffies_till_stall_check(void)
112{
113 int till_stall_check = READ_ONCE(rcu_cpu_stall_timeout);
114
115 /*
116 * Limit check must be consistent with the Kconfig limits
117 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
118 */
119 if (till_stall_check < 3) {
120 WRITE_ONCE(rcu_cpu_stall_timeout, 3);
121 till_stall_check = 3;
122 } else if (till_stall_check > 300) {
123 WRITE_ONCE(rcu_cpu_stall_timeout, 300);
124 till_stall_check = 300;
125 }
126 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
127}
128EXPORT_SYMBOL_GPL(rcu_jiffies_till_stall_check);
129
130/* Don't do RCU CPU stall warnings during long sysrq printouts. */
131void rcu_sysrq_start(void)
132{
133 if (!rcu_cpu_stall_suppress)
134 rcu_cpu_stall_suppress = 2;
135}
136
137void rcu_sysrq_end(void)
138{
139 if (rcu_cpu_stall_suppress == 2)
140 rcu_cpu_stall_suppress = 0;
141}
142
143/* Don't print RCU CPU stall warnings during a kernel panic. */
144static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
145{
146 rcu_cpu_stall_suppress = 1;
147 return NOTIFY_DONE;
148}
149
150static struct notifier_block rcu_panic_block = {
151 .notifier_call = rcu_panic,
152};
153
154static int __init check_cpu_stall_init(void)
155{
156 atomic_notifier_chain_register(nh: &panic_notifier_list, nb: &rcu_panic_block);
157 return 0;
158}
159early_initcall(check_cpu_stall_init);
160
161/* If so specified via sysctl, panic, yielding cleaner stall-warning output. */
162static void panic_on_rcu_stall(void)
163{
164 static int cpu_stall;
165
166 /*
167 * Attempt to kick out the BPF scheduler if it's installed and defer
168 * the panic to give the system a chance to recover.
169 */
170 if (scx_rcu_cpu_stall())
171 return;
172
173 if (++cpu_stall < sysctl_max_rcu_stall_to_panic)
174 return;
175
176 if (sysctl_panic_on_rcu_stall)
177 panic(fmt: "RCU Stall\n");
178}
179
180/**
181 * rcu_cpu_stall_reset - restart stall-warning timeout for current grace period
182 *
183 * To perform the reset request from the caller, disable stall detection until
184 * 3 fqs loops have passed. This is required to ensure a fresh jiffies is
185 * loaded. It should be safe to do from the fqs loop as enough timer
186 * interrupts and context switches should have passed.
187 *
188 * The caller must disable hard irqs.
189 */
190void rcu_cpu_stall_reset(void)
191{
192 WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 3);
193 WRITE_ONCE(rcu_state.jiffies_stall, ULONG_MAX);
194}
195
196//////////////////////////////////////////////////////////////////////////////
197//
198// Interaction with RCU grace periods
199
200/* Start of new grace period, so record stall time (and forcing times). */
201static void record_gp_stall_check_time(void)
202{
203 unsigned long j = jiffies;
204 unsigned long j1;
205
206 WRITE_ONCE(rcu_state.gp_start, j);
207 j1 = rcu_jiffies_till_stall_check();
208 smp_mb(); // ->gp_start before ->jiffies_stall and caller's ->gp_seq.
209 WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 0);
210 WRITE_ONCE(rcu_state.jiffies_stall, j + j1);
211 rcu_state.jiffies_resched = j + j1 / 2;
212 rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs);
213}
214
215/* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */
216static void zero_cpu_stall_ticks(struct rcu_data *rdp)
217{
218 rdp->ticks_this_gp = 0;
219 rdp->softirq_snap = kstat_softirqs_cpu(irq: RCU_SOFTIRQ, smp_processor_id());
220 WRITE_ONCE(rdp->last_fqs_resched, jiffies);
221}
222
223/*
224 * If too much time has passed in the current grace period, and if
225 * so configured, go kick the relevant kthreads.
226 */
227static void rcu_stall_kick_kthreads(void)
228{
229 unsigned long j;
230
231 if (!READ_ONCE(rcu_kick_kthreads))
232 return;
233 j = READ_ONCE(rcu_state.jiffies_kick_kthreads);
234 if (time_after(jiffies, j) && rcu_state.gp_kthread &&
235 (rcu_gp_in_progress() || READ_ONCE(rcu_state.gp_flags))) {
236 WARN_ONCE(1, "Kicking %s grace-period kthread\n",
237 rcu_state.name);
238 rcu_ftrace_dump(DUMP_ALL);
239 wake_up_process(tsk: rcu_state.gp_kthread);
240 WRITE_ONCE(rcu_state.jiffies_kick_kthreads, j + HZ);
241 }
242}
243
244/*
245 * Handler for the irq_work request posted about halfway into the RCU CPU
246 * stall timeout, and used to detect excessive irq disabling. Set state
247 * appropriately, but just complain if there is unexpected state on entry.
248 */
249static void rcu_iw_handler(struct irq_work *iwp)
250{
251 struct rcu_data *rdp;
252 struct rcu_node *rnp;
253
254 rdp = container_of(iwp, struct rcu_data, rcu_iw);
255 rnp = rdp->mynode;
256 raw_spin_lock_rcu_node(rnp);
257 if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) {
258 rdp->rcu_iw_gp_seq = rnp->gp_seq;
259 rdp->rcu_iw_pending = false;
260 }
261 raw_spin_unlock_rcu_node(rnp);
262}
263
264//////////////////////////////////////////////////////////////////////////////
265//
266// Printing RCU CPU stall warnings
267
268#ifdef CONFIG_PREEMPT_RCU
269
270/*
271 * Dump detailed information for all tasks blocking the current RCU
272 * grace period on the specified rcu_node structure.
273 */
274static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
275{
276 unsigned long flags;
277 struct task_struct *t;
278
279 raw_spin_lock_irqsave_rcu_node(rnp, flags);
280 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
281 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
282 return;
283 }
284 t = list_entry(rnp->gp_tasks->prev,
285 struct task_struct, rcu_node_entry);
286 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
287 /*
288 * We could be printing a lot while holding a spinlock.
289 * Avoid triggering hard lockup.
290 */
291 touch_nmi_watchdog();
292 sched_show_task(p: t);
293 }
294 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
295}
296
297// Communicate task state back to the RCU CPU stall warning request.
298struct rcu_stall_chk_rdr {
299 int nesting;
300 union rcu_special rs;
301 bool on_blkd_list;
302};
303
304/*
305 * Report out the state of a not-running task that is stalling the
306 * current RCU grace period.
307 */
308static int check_slow_task(struct task_struct *t, void *arg)
309{
310 struct rcu_stall_chk_rdr *rscrp = arg;
311
312 if (task_curr(p: t))
313 return -EBUSY; // It is running, so decline to inspect it.
314 rscrp->nesting = t->rcu_read_lock_nesting;
315 rscrp->rs = t->rcu_read_unlock_special;
316 rscrp->on_blkd_list = !list_empty(head: &t->rcu_node_entry);
317 return 0;
318}
319
320/*
321 * Scan the current list of tasks blocked within RCU read-side critical
322 * sections, printing out the tid of each of the first few of them.
323 */
324static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
325 __releases(rnp->lock)
326{
327 int i = 0;
328 int ndetected = 0;
329 struct rcu_stall_chk_rdr rscr;
330 struct task_struct *t;
331 struct task_struct *ts[8];
332
333 lockdep_assert_irqs_disabled();
334 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
335 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
336 return 0;
337 }
338 pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
339 rnp->level, rnp->grplo, rnp->grphi);
340 t = list_entry(rnp->gp_tasks->prev,
341 struct task_struct, rcu_node_entry);
342 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
343 get_task_struct(t);
344 ts[i++] = t;
345 if (i >= ARRAY_SIZE(ts))
346 break;
347 }
348 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
349 while (i) {
350 t = ts[--i];
351 if (task_call_func(p: t, func: check_slow_task, arg: &rscr))
352 pr_cont(" P%d", t->pid);
353 else
354 pr_cont(" P%d/%d:%c%c%c%c",
355 t->pid, rscr.nesting,
356 ".b"[rscr.rs.b.blocked],
357 ".q"[rscr.rs.b.need_qs],
358 ".e"[rscr.rs.b.exp_hint],
359 ".l"[rscr.on_blkd_list]);
360 lockdep_assert_irqs_disabled();
361 put_task_struct(t);
362 ndetected++;
363 }
364 pr_cont("\n");
365 return ndetected;
366}
367
368#else /* #ifdef CONFIG_PREEMPT_RCU */
369
370/*
371 * Because preemptible RCU does not exist, we never have to check for
372 * tasks blocked within RCU read-side critical sections.
373 */
374static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
375{
376}
377
378/*
379 * Because preemptible RCU does not exist, we never have to check for
380 * tasks blocked within RCU read-side critical sections.
381 */
382static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
383 __releases(rnp->lock)
384{
385 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
386 return 0;
387}
388#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
389
390/*
391 * Dump stacks of all tasks running on stalled CPUs. First try using
392 * NMIs, but fall back to manual remote stack tracing on architectures
393 * that don't support NMI-based stack dumps. The NMI-triggered stack
394 * traces are more accurate because they are printed by the target CPU.
395 */
396static void rcu_dump_cpu_stacks(unsigned long gp_seq)
397{
398 int cpu;
399 unsigned long flags;
400 struct rcu_node *rnp;
401
402 rcu_for_each_leaf_node(rnp) {
403 printk_deferred_enter();
404 for_each_leaf_node_possible_cpu(rnp, cpu) {
405 if (gp_seq != data_race(rcu_state.gp_seq)) {
406 printk_deferred_exit();
407 pr_err("INFO: Stall ended during stack backtracing.\n");
408 return;
409 }
410 if (!(data_race(rnp->qsmask) & leaf_node_cpu_bit(rnp, cpu)))
411 continue;
412 raw_spin_lock_irqsave_rcu_node(rnp, flags);
413 if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
414 if (cpu_is_offline(cpu))
415 pr_err("Offline CPU %d blocking current GP.\n", cpu);
416 else
417 dump_cpu_task(cpu);
418 }
419 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
420 }
421 printk_deferred_exit();
422 }
423}
424
425static const char * const gp_state_names[] = {
426 [RCU_GP_IDLE] = "RCU_GP_IDLE",
427 [RCU_GP_WAIT_GPS] = "RCU_GP_WAIT_GPS",
428 [RCU_GP_DONE_GPS] = "RCU_GP_DONE_GPS",
429 [RCU_GP_ONOFF] = "RCU_GP_ONOFF",
430 [RCU_GP_INIT] = "RCU_GP_INIT",
431 [RCU_GP_WAIT_FQS] = "RCU_GP_WAIT_FQS",
432 [RCU_GP_DOING_FQS] = "RCU_GP_DOING_FQS",
433 [RCU_GP_CLEANUP] = "RCU_GP_CLEANUP",
434 [RCU_GP_CLEANED] = "RCU_GP_CLEANED",
435};
436
437/*
438 * Convert a ->gp_state value to a character string.
439 */
440static const char *gp_state_getname(short gs)
441{
442 if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names))
443 return "???";
444 return gp_state_names[gs];
445}
446
447/* Is the RCU grace-period kthread being starved of CPU time? */
448static bool rcu_is_gp_kthread_starving(unsigned long *jp)
449{
450 unsigned long j = jiffies - READ_ONCE(rcu_state.gp_activity);
451
452 if (jp)
453 *jp = j;
454 return j > 2 * HZ;
455}
456
457static bool rcu_is_rcuc_kthread_starving(struct rcu_data *rdp, unsigned long *jp)
458{
459 int cpu;
460 struct task_struct *rcuc;
461 unsigned long j;
462
463 rcuc = rdp->rcu_cpu_kthread_task;
464 if (!rcuc)
465 return false;
466
467 cpu = task_cpu(p: rcuc);
468 if (cpu_is_offline(cpu) || idle_cpu(cpu))
469 return false;
470
471 j = jiffies - READ_ONCE(rdp->rcuc_activity);
472
473 if (jp)
474 *jp = j;
475 return j > 2 * HZ;
476}
477
478static void print_cpu_stat_info(int cpu)
479{
480 struct rcu_snap_record rsr, *rsrp;
481 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
482 struct kernel_cpustat *kcsp = &kcpustat_cpu(cpu);
483
484 if (!rcu_cpu_stall_cputime)
485 return;
486
487 rsrp = &rdp->snap_record;
488 if (rsrp->gp_seq != rdp->gp_seq)
489 return;
490
491 rsr.cputime_irq = kcpustat_field(kcpustat: kcsp, usage: CPUTIME_IRQ, cpu);
492 rsr.cputime_softirq = kcpustat_field(kcpustat: kcsp, usage: CPUTIME_SOFTIRQ, cpu);
493 rsr.cputime_system = kcpustat_field(kcpustat: kcsp, usage: CPUTIME_SYSTEM, cpu);
494
495 pr_err("\t hardirqs softirqs csw/system\n");
496 pr_err("\t number: %8lld %10d %12lld\n",
497 kstat_cpu_irqs_sum(cpu) + arch_irq_stat_cpu(cpu) - rsrp->nr_hardirqs,
498 kstat_cpu_softirqs_sum(cpu) - rsrp->nr_softirqs,
499 nr_context_switches_cpu(cpu) - rsrp->nr_csw);
500 pr_err("\tcputime: %8lld %10lld %12lld ==> %d(ms)\n",
501 div_u64(rsr.cputime_irq - rsrp->cputime_irq, NSEC_PER_MSEC),
502 div_u64(rsr.cputime_softirq - rsrp->cputime_softirq, NSEC_PER_MSEC),
503 div_u64(rsr.cputime_system - rsrp->cputime_system, NSEC_PER_MSEC),
504 jiffies_to_msecs(jiffies - rsrp->jiffies));
505}
506
507/*
508 * Print out diagnostic information for the specified stalled CPU.
509 *
510 * If the specified CPU is aware of the current RCU grace period, then
511 * print the number of scheduling clock interrupts the CPU has taken
512 * during the time that it has been aware. Otherwise, print the number
513 * of RCU grace periods that this CPU is ignorant of, for example, "1"
514 * if the CPU was aware of the previous grace period.
515 *
516 * Also print out idle info.
517 */
518static void print_cpu_stall_info(int cpu)
519{
520 unsigned long delta;
521 bool falsepositive;
522 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
523 char *ticks_title;
524 unsigned long ticks_value;
525 bool rcuc_starved;
526 unsigned long j;
527 char buf[32];
528
529 /*
530 * We could be printing a lot while holding a spinlock. Avoid
531 * triggering hard lockup.
532 */
533 touch_nmi_watchdog();
534
535 ticks_value = rcu_seq_ctr(s: rcu_state.gp_seq - rdp->gp_seq);
536 if (ticks_value) {
537 ticks_title = "GPs behind";
538 } else {
539 ticks_title = "ticks this GP";
540 ticks_value = rdp->ticks_this_gp;
541 }
542 delta = rcu_seq_ctr(s: rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq);
543 falsepositive = rcu_is_gp_kthread_starving(NULL) &&
544 rcu_watching_snap_in_eqs(snap: ct_rcu_watching_cpu(cpu));
545 rcuc_starved = rcu_is_rcuc_kthread_starving(rdp, jp: &j);
546 if (rcuc_starved)
547 // Print signed value, as negative values indicate a probable bug.
548 snprintf(buf, size: sizeof(buf), fmt: " rcuc=%ld jiffies(starved)", j);
549 pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%04x/%ld/%#lx softirq=%u/%u fqs=%ld%s%s\n",
550 cpu,
551 "O."[!!cpu_online(cpu)],
552 "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
553 "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)],
554 !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' :
555 rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' :
556 "!."[!delta],
557 ticks_value, ticks_title,
558 ct_rcu_watching_cpu(cpu) & 0xffff,
559 ct_nesting_cpu(cpu), ct_nmi_nesting_cpu(cpu),
560 rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
561 data_race(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
562 rcuc_starved ? buf : "",
563 falsepositive ? " (false positive?)" : "");
564
565 print_cpu_stat_info(cpu);
566}
567
568/* Complain about starvation of grace-period kthread. */
569static void rcu_check_gp_kthread_starvation(void)
570{
571 int cpu;
572 struct task_struct *gpk = rcu_state.gp_kthread;
573 unsigned long j;
574
575 if (rcu_is_gp_kthread_starving(jp: &j)) {
576 cpu = gpk ? task_cpu(p: gpk) : -1;
577 pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x ->cpu=%d\n",
578 rcu_state.name, j,
579 (long)rcu_seq_current(&rcu_state.gp_seq),
580 data_race(READ_ONCE(rcu_state.gp_flags)),
581 gp_state_getname(rcu_state.gp_state),
582 data_race(READ_ONCE(rcu_state.gp_state)),
583 gpk ? data_race(READ_ONCE(gpk->__state)) : ~0, cpu);
584 if (gpk) {
585 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
586
587 pr_err("\tUnless %s kthread gets sufficient CPU time, OOM is now expected behavior.\n", rcu_state.name);
588 pr_err("RCU grace-period kthread stack dump:\n");
589 sched_show_task(p: gpk);
590 if (cpu_is_offline(cpu)) {
591 pr_err("RCU GP kthread last ran on offline CPU %d.\n", cpu);
592 } else if (!(data_race(READ_ONCE(rdp->mynode->qsmask)) & rdp->grpmask)) {
593 pr_err("Stack dump where RCU GP kthread last ran:\n");
594 dump_cpu_task(cpu);
595 }
596 wake_up_process(tsk: gpk);
597 }
598 }
599}
600
601/* Complain about missing wakeups from expired fqs wait timer */
602static void rcu_check_gp_kthread_expired_fqs_timer(void)
603{
604 struct task_struct *gpk = rcu_state.gp_kthread;
605 short gp_state;
606 unsigned long jiffies_fqs;
607 int cpu;
608
609 /*
610 * Order reads of .gp_state and .jiffies_force_qs.
611 * Matching smp_wmb() is present in rcu_gp_fqs_loop().
612 */
613 gp_state = smp_load_acquire(&rcu_state.gp_state);
614 jiffies_fqs = READ_ONCE(rcu_state.jiffies_force_qs);
615
616 if (gp_state == RCU_GP_WAIT_FQS &&
617 time_after(jiffies, jiffies_fqs + RCU_STALL_MIGHT_MIN) &&
618 gpk && !READ_ONCE(gpk->on_rq)) {
619 cpu = task_cpu(p: gpk);
620 pr_err("%s kthread timer wakeup didn't happen for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x\n",
621 rcu_state.name, (jiffies - jiffies_fqs),
622 (long)rcu_seq_current(&rcu_state.gp_seq),
623 data_race(READ_ONCE(rcu_state.gp_flags)), // Diagnostic read
624 gp_state_getname(RCU_GP_WAIT_FQS), RCU_GP_WAIT_FQS,
625 data_race(READ_ONCE(gpk->__state)));
626 pr_err("\tPossible timer handling issue on cpu=%d timer-softirq=%u\n",
627 cpu, kstat_softirqs_cpu(TIMER_SOFTIRQ, cpu));
628 }
629}
630
631static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
632{
633 int cpu;
634 unsigned long flags;
635 unsigned long gpa;
636 unsigned long j;
637 int ndetected = 0;
638 struct rcu_node *rnp;
639 long totqlen = 0;
640
641 lockdep_assert_irqs_disabled();
642
643 /* Kick and suppress, if so configured. */
644 rcu_stall_kick_kthreads();
645 if (rcu_stall_is_suppressed())
646 return;
647
648 nbcon_cpu_emergency_enter();
649
650 /*
651 * OK, time to rat on our buddy...
652 * See Documentation/RCU/stallwarn.rst for info on how to debug
653 * RCU CPU stall warnings.
654 */
655 trace_rcu_stall_warning(rcuname: rcu_state.name, TPS("StallDetected"));
656 pr_err("INFO: %s detected stalls on CPUs/tasks:\n", rcu_state.name);
657 rcu_for_each_leaf_node(rnp) {
658 raw_spin_lock_irqsave_rcu_node(rnp, flags);
659 if (rnp->qsmask != 0) {
660 for_each_leaf_node_possible_cpu(rnp, cpu)
661 if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
662 print_cpu_stall_info(cpu);
663 ndetected++;
664 }
665 }
666 ndetected += rcu_print_task_stall(rnp, flags); // Releases rnp->lock.
667 lockdep_assert_irqs_disabled();
668 }
669
670 for_each_possible_cpu(cpu)
671 totqlen += rcu_get_n_cbs_cpu(cpu);
672 pr_err("\t(detected by %d, t=%ld jiffies, g=%ld, q=%lu ncpus=%d)\n",
673 smp_processor_id(), (long)(jiffies - gps),
674 (long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
675 data_race(rcu_state.n_online_cpus)); // Diagnostic read
676 if (ndetected) {
677 rcu_dump_cpu_stacks(gp_seq);
678
679 /* Complain about tasks blocking the grace period. */
680 rcu_for_each_leaf_node(rnp)
681 rcu_print_detail_task_stall_rnp(rnp);
682 } else {
683 if (rcu_seq_current(sp: &rcu_state.gp_seq) != gp_seq) {
684 pr_err("INFO: Stall ended before state dump start\n");
685 } else {
686 j = jiffies;
687 gpa = data_race(READ_ONCE(rcu_state.gp_activity));
688 pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
689 rcu_state.name, j - gpa, j, gpa,
690 data_race(READ_ONCE(jiffies_till_next_fqs)),
691 data_race(READ_ONCE(rcu_get_root()->qsmask)));
692 }
693 }
694 /* Rewrite if needed in case of slow consoles. */
695 if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
696 WRITE_ONCE(rcu_state.jiffies_stall,
697 jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
698
699 rcu_check_gp_kthread_expired_fqs_timer();
700 rcu_check_gp_kthread_starvation();
701
702 nbcon_cpu_emergency_exit();
703
704 panic_on_rcu_stall();
705
706 rcu_force_quiescent_state(); /* Kick them all. */
707}
708
709static void print_cpu_stall(unsigned long gp_seq, unsigned long gps)
710{
711 int cpu;
712 unsigned long flags;
713 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
714 struct rcu_node *rnp = rcu_get_root();
715 long totqlen = 0;
716
717 lockdep_assert_irqs_disabled();
718
719 /* Kick and suppress, if so configured. */
720 rcu_stall_kick_kthreads();
721 if (rcu_stall_is_suppressed())
722 return;
723
724 nbcon_cpu_emergency_enter();
725
726 /*
727 * OK, time to rat on ourselves...
728 * See Documentation/RCU/stallwarn.rst for info on how to debug
729 * RCU CPU stall warnings.
730 */
731 trace_rcu_stall_warning(rcuname: rcu_state.name, TPS("SelfDetected"));
732 pr_err("INFO: %s self-detected stall on CPU\n", rcu_state.name);
733 raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags);
734 print_cpu_stall_info(smp_processor_id());
735 raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags);
736 for_each_possible_cpu(cpu)
737 totqlen += rcu_get_n_cbs_cpu(cpu);
738 pr_err("\t(t=%lu jiffies g=%ld q=%lu ncpus=%d)\n",
739 jiffies - gps,
740 (long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
741 data_race(rcu_state.n_online_cpus)); // Diagnostic read
742
743 rcu_check_gp_kthread_expired_fqs_timer();
744 rcu_check_gp_kthread_starvation();
745
746 rcu_dump_cpu_stacks(gp_seq);
747
748 raw_spin_lock_irqsave_rcu_node(rnp, flags);
749 /* Rewrite if needed in case of slow consoles. */
750 if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
751 WRITE_ONCE(rcu_state.jiffies_stall,
752 jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
753 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
754
755 nbcon_cpu_emergency_exit();
756
757 panic_on_rcu_stall();
758
759 /*
760 * Attempt to revive the RCU machinery by forcing a context switch.
761 *
762 * A context switch would normally allow the RCU state machine to make
763 * progress and it could be we're stuck in kernel space without context
764 * switches for an entirely unreasonable amount of time.
765 */
766 set_tsk_need_resched(current);
767 set_preempt_need_resched();
768}
769
770static bool csd_lock_suppress_rcu_stall;
771module_param(csd_lock_suppress_rcu_stall, bool, 0644);
772
773static void check_cpu_stall(struct rcu_data *rdp)
774{
775 bool self_detected;
776 unsigned long gs1;
777 unsigned long gs2;
778 unsigned long gps;
779 unsigned long j;
780 unsigned long jn;
781 unsigned long js;
782 struct rcu_node *rnp;
783
784 lockdep_assert_irqs_disabled();
785 if ((rcu_stall_is_suppressed() && !READ_ONCE(rcu_kick_kthreads)) ||
786 !rcu_gp_in_progress())
787 return;
788 rcu_stall_kick_kthreads();
789
790 /*
791 * Check if it was requested (via rcu_cpu_stall_reset()) that the FQS
792 * loop has to set jiffies to ensure a non-stale jiffies value. This
793 * is required to have good jiffies value after coming out of long
794 * breaks of jiffies updates. Not doing so can cause false positives.
795 */
796 if (READ_ONCE(rcu_state.nr_fqs_jiffies_stall) > 0)
797 return;
798
799 j = jiffies;
800
801 /*
802 * Lots of memory barriers to reject false positives.
803 *
804 * The idea is to pick up rcu_state.gp_seq, then
805 * rcu_state.jiffies_stall, then rcu_state.gp_start, and finally
806 * another copy of rcu_state.gp_seq. These values are updated in
807 * the opposite order with memory barriers (or equivalent) during
808 * grace-period initialization and cleanup. Now, a false positive
809 * can occur if we get an new value of rcu_state.gp_start and a old
810 * value of rcu_state.jiffies_stall. But given the memory barriers,
811 * the only way that this can happen is if one grace period ends
812 * and another starts between these two fetches. This is detected
813 * by comparing the second fetch of rcu_state.gp_seq with the
814 * previous fetch from rcu_state.gp_seq.
815 *
816 * Given this check, comparisons of jiffies, rcu_state.jiffies_stall,
817 * and rcu_state.gp_start suffice to forestall false positives.
818 */
819 gs1 = READ_ONCE(rcu_state.gp_seq);
820 smp_rmb(); /* Pick up ->gp_seq first... */
821 js = READ_ONCE(rcu_state.jiffies_stall);
822 smp_rmb(); /* ...then ->jiffies_stall before the rest... */
823 gps = READ_ONCE(rcu_state.gp_start);
824 smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */
825 gs2 = READ_ONCE(rcu_state.gp_seq);
826 if (gs1 != gs2 ||
827 ULONG_CMP_LT(j, js) ||
828 ULONG_CMP_GE(gps, js) ||
829 !rcu_seq_state(s: gs2))
830 return; /* No stall or GP completed since entering function. */
831 rnp = rdp->mynode;
832 jn = jiffies + ULONG_MAX / 2;
833 self_detected = READ_ONCE(rnp->qsmask) & rdp->grpmask;
834 if (rcu_gp_in_progress() &&
835 (self_detected || ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) &&
836 cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) {
837 /*
838 * If a virtual machine is stopped by the host it can look to
839 * the watchdog like an RCU stall. Check to see if the host
840 * stopped the vm.
841 */
842 if (kvm_check_and_clear_guest_paused())
843 return;
844
845#ifdef CONFIG_SYSFS
846 ++rcu_stall_count;
847#endif
848
849 rcu_stall_notifier_call_chain(RCU_STALL_NOTIFY_NORM, v: (void *)j - gps);
850 if (READ_ONCE(csd_lock_suppress_rcu_stall) && csd_lock_is_stuck()) {
851 pr_err("INFO: %s detected stall, but suppressed full report due to a stuck CSD-lock.\n", rcu_state.name);
852 } else if (self_detected) {
853 /* We haven't checked in, so go dump stack. */
854 print_cpu_stall(gp_seq: gs2, gps);
855 } else {
856 /* They had a few time units to dump stack, so complain. */
857 print_other_cpu_stall(gp_seq: gs2, gps);
858 }
859
860 if (READ_ONCE(rcu_cpu_stall_ftrace_dump))
861 rcu_ftrace_dump(DUMP_ALL);
862
863 if (READ_ONCE(rcu_state.jiffies_stall) == jn) {
864 jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
865 WRITE_ONCE(rcu_state.jiffies_stall, jn);
866 }
867 }
868}
869
870//////////////////////////////////////////////////////////////////////////////
871//
872// RCU forward-progress mechanisms, including for callback invocation.
873
874
875/*
876 * Check to see if a failure to end RCU priority inversion was due to
877 * a CPU not passing through a quiescent state. When this happens, there
878 * is nothing that RCU priority boosting can do to help, so we shouldn't
879 * count this as an RCU priority boosting failure. A return of true says
880 * RCU priority boosting is to blame, and false says otherwise. If false
881 * is returned, the first of the CPUs to blame is stored through cpup.
882 * If there was no CPU blocking the current grace period, but also nothing
883 * in need of being boosted, *cpup is set to -1. This can happen in case
884 * of vCPU preemption while the last CPU is reporting its quiscent state,
885 * for example.
886 *
887 * If cpup is NULL, then a lockless quick check is carried out, suitable
888 * for high-rate usage. On the other hand, if cpup is non-NULL, each
889 * rcu_node structure's ->lock is acquired, ruling out high-rate usage.
890 */
891bool rcu_check_boost_fail(unsigned long gp_state, int *cpup)
892{
893 bool atb = false;
894 int cpu;
895 unsigned long flags;
896 struct rcu_node *rnp;
897
898 rcu_for_each_leaf_node(rnp) {
899 if (!cpup) {
900 if (data_race(READ_ONCE(rnp->qsmask))) {
901 return false;
902 } else {
903 if (READ_ONCE(rnp->gp_tasks))
904 atb = true;
905 continue;
906 }
907 }
908 *cpup = -1;
909 raw_spin_lock_irqsave_rcu_node(rnp, flags);
910 if (rnp->gp_tasks)
911 atb = true;
912 if (!rnp->qsmask) {
913 // No CPUs without quiescent states for this rnp.
914 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
915 continue;
916 }
917 // Find the first holdout CPU.
918 for_each_leaf_node_possible_cpu(rnp, cpu) {
919 if (rnp->qsmask & (1UL << (cpu - rnp->grplo))) {
920 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
921 *cpup = cpu;
922 return false;
923 }
924 }
925 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
926 }
927 // Can't blame CPUs, so must blame RCU priority boosting.
928 return atb;
929}
930EXPORT_SYMBOL_GPL(rcu_check_boost_fail);
931
932/*
933 * Show the state of the grace-period kthreads.
934 */
935void show_rcu_gp_kthreads(void)
936{
937 unsigned long cbs = 0;
938 int cpu;
939 unsigned long j;
940 unsigned long ja;
941 unsigned long jr;
942 unsigned long js;
943 unsigned long jw;
944 struct rcu_data *rdp;
945 struct rcu_node *rnp;
946 struct task_struct *t = READ_ONCE(rcu_state.gp_kthread);
947
948 j = jiffies;
949 ja = j - data_race(READ_ONCE(rcu_state.gp_activity));
950 jr = j - data_race(READ_ONCE(rcu_state.gp_req_activity));
951 js = j - data_race(READ_ONCE(rcu_state.gp_start));
952 jw = j - data_race(READ_ONCE(rcu_state.gp_wake_time));
953 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",
954 rcu_state.name, gp_state_getname(rcu_state.gp_state),
955 data_race(READ_ONCE(rcu_state.gp_state)),
956 t ? data_race(READ_ONCE(t->__state)) : 0x1ffff, t ? t->rt_priority : 0xffU,
957 js, ja, jr, jw, (long)data_race(READ_ONCE(rcu_state.gp_wake_seq)),
958 (long)data_race(READ_ONCE(rcu_state.gp_seq)),
959 (long)data_race(READ_ONCE(rcu_get_root()->gp_seq_needed)),
960 data_race(READ_ONCE(rcu_state.gp_max)),
961 data_race(READ_ONCE(rcu_state.gp_flags)));
962 rcu_for_each_node_breadth_first(rnp) {
963 if (ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq), READ_ONCE(rnp->gp_seq_needed)) &&
964 !data_race(READ_ONCE(rnp->qsmask)) && !data_race(READ_ONCE(rnp->boost_tasks)) &&
965 !data_race(READ_ONCE(rnp->exp_tasks)) && !data_race(READ_ONCE(rnp->gp_tasks)))
966 continue;
967 pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld ->qsmask %#lx %c%c%c%c ->n_boosts %ld\n",
968 rnp->grplo, rnp->grphi,
969 (long)data_race(READ_ONCE(rnp->gp_seq)),
970 (long)data_race(READ_ONCE(rnp->gp_seq_needed)),
971 data_race(READ_ONCE(rnp->qsmask)),
972 ".b"[!!data_race(READ_ONCE(rnp->boost_kthread_task))],
973 ".B"[!!data_race(READ_ONCE(rnp->boost_tasks))],
974 ".E"[!!data_race(READ_ONCE(rnp->exp_tasks))],
975 ".G"[!!data_race(READ_ONCE(rnp->gp_tasks))],
976 data_race(READ_ONCE(rnp->n_boosts)));
977 if (!rcu_is_leaf_node(rnp))
978 continue;
979 for_each_leaf_node_possible_cpu(rnp, cpu) {
980 rdp = per_cpu_ptr(&rcu_data, cpu);
981 if (READ_ONCE(rdp->gpwrap) ||
982 ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq),
983 READ_ONCE(rdp->gp_seq_needed)))
984 continue;
985 pr_info("\tcpu %d ->gp_seq_needed %ld\n",
986 cpu, (long)data_race(READ_ONCE(rdp->gp_seq_needed)));
987 }
988 }
989 for_each_possible_cpu(cpu) {
990 rdp = per_cpu_ptr(&rcu_data, cpu);
991 cbs += data_race(READ_ONCE(rdp->n_cbs_invoked));
992 show_rcu_nocb_state(rdp);
993 }
994 pr_info("RCU callbacks invoked since boot: %lu\n", cbs);
995 show_rcu_tasks_gp_kthreads();
996}
997EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
998
999/*
1000 * This function checks for grace-period requests that fail to motivate
1001 * RCU to come out of its idle mode.
1002 */
1003static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
1004 const unsigned long gpssdelay)
1005{
1006 unsigned long flags;
1007 unsigned long j;
1008 struct rcu_node *rnp_root = rcu_get_root();
1009 static atomic_t warned = ATOMIC_INIT(0);
1010
1011 if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() ||
1012 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
1013 READ_ONCE(rnp_root->gp_seq_needed)) ||
1014 !smp_load_acquire(&rcu_state.gp_kthread)) // Get stable kthread.
1015 return;
1016 j = jiffies; /* Expensive access, and in common case don't get here. */
1017 if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
1018 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
1019 atomic_read(v: &warned))
1020 return;
1021
1022 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1023 j = jiffies;
1024 if (rcu_gp_in_progress() ||
1025 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
1026 READ_ONCE(rnp_root->gp_seq_needed)) ||
1027 time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
1028 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
1029 atomic_read(v: &warned)) {
1030 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1031 return;
1032 }
1033 /* Hold onto the leaf lock to make others see warned==1. */
1034
1035 if (rnp_root != rnp)
1036 raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
1037 j = jiffies;
1038 if (rcu_gp_in_progress() ||
1039 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
1040 READ_ONCE(rnp_root->gp_seq_needed)) ||
1041 time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
1042 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
1043 atomic_xchg(v: &warned, new: 1)) {
1044 if (rnp_root != rnp)
1045 /* irqs remain disabled. */
1046 raw_spin_unlock_rcu_node(rnp_root);
1047 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1048 return;
1049 }
1050 WARN_ON(1);
1051 if (rnp_root != rnp)
1052 raw_spin_unlock_rcu_node(rnp_root);
1053 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1054 show_rcu_gp_kthreads();
1055}
1056
1057/*
1058 * Do a forward-progress check for rcutorture. This is normally invoked
1059 * due to an OOM event. The argument "j" gives the time period during
1060 * which rcutorture would like progress to have been made.
1061 */
1062void rcu_fwd_progress_check(unsigned long j)
1063{
1064 unsigned long cbs;
1065 int cpu;
1066 unsigned long max_cbs = 0;
1067 int max_cpu = -1;
1068 struct rcu_data *rdp;
1069
1070 if (rcu_gp_in_progress()) {
1071 pr_info("%s: GP age %lu jiffies\n",
1072 __func__, jiffies - data_race(READ_ONCE(rcu_state.gp_start)));
1073 show_rcu_gp_kthreads();
1074 } else {
1075 pr_info("%s: Last GP end %lu jiffies ago\n",
1076 __func__, jiffies - data_race(READ_ONCE(rcu_state.gp_end)));
1077 preempt_disable();
1078 rdp = this_cpu_ptr(&rcu_data);
1079 rcu_check_gp_start_stall(rnp: rdp->mynode, rdp, gpssdelay: j);
1080 preempt_enable();
1081 }
1082 for_each_possible_cpu(cpu) {
1083 cbs = rcu_get_n_cbs_cpu(cpu);
1084 if (!cbs)
1085 continue;
1086 if (max_cpu < 0)
1087 pr_info("%s: callbacks", __func__);
1088 pr_cont(" %d: %lu", cpu, cbs);
1089 if (cbs <= max_cbs)
1090 continue;
1091 max_cbs = cbs;
1092 max_cpu = cpu;
1093 }
1094 if (max_cpu >= 0)
1095 pr_cont("\n");
1096}
1097EXPORT_SYMBOL_GPL(rcu_fwd_progress_check);
1098
1099/* Commandeer a sysrq key to dump RCU's tree. */
1100static bool sysrq_rcu;
1101module_param(sysrq_rcu, bool, 0444);
1102
1103/* Dump grace-period-request information due to commandeered sysrq. */
1104static void sysrq_show_rcu(u8 key)
1105{
1106 show_rcu_gp_kthreads();
1107}
1108
1109static const struct sysrq_key_op sysrq_rcudump_op = {
1110 .handler = sysrq_show_rcu,
1111 .help_msg = "show-rcu(y)",
1112 .action_msg = "Show RCU tree",
1113 .enable_mask = SYSRQ_ENABLE_DUMP,
1114};
1115
1116static int __init rcu_sysrq_init(void)
1117{
1118 if (sysrq_rcu)
1119 return register_sysrq_key(key: 'y', op: &sysrq_rcudump_op);
1120 return 0;
1121}
1122early_initcall(rcu_sysrq_init);
1123
1124#ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1125
1126//////////////////////////////////////////////////////////////////////////////
1127//
1128// RCU CPU stall-warning notifiers
1129
1130static ATOMIC_NOTIFIER_HEAD(rcu_cpu_stall_notifier_list);
1131
1132/**
1133 * rcu_stall_chain_notifier_register - Add an RCU CPU stall notifier
1134 * @n: Entry to add.
1135 *
1136 * Adds an RCU CPU stall notifier to an atomic notifier chain.
1137 * The @action passed to a notifier will be @RCU_STALL_NOTIFY_NORM or
1138 * friends. The @data will be the duration of the stalled grace period,
1139 * in jiffies, coerced to a void* pointer.
1140 *
1141 * Returns 0 on success, %-EEXIST on error.
1142 */
1143int rcu_stall_chain_notifier_register(struct notifier_block *n)
1144{
1145 int rcsn = rcu_cpu_stall_notifiers;
1146
1147 WARN(1, "Adding %pS() to RCU stall notifier list (%s).\n", n->notifier_call,
1148 rcsn ? "possibly suppressing RCU CPU stall warnings" : "failed, so all is well");
1149 if (rcsn)
1150 return atomic_notifier_chain_register(&rcu_cpu_stall_notifier_list, n);
1151 return -EEXIST;
1152}
1153EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_register);
1154
1155/**
1156 * rcu_stall_chain_notifier_unregister - Remove an RCU CPU stall notifier
1157 * @n: Entry to add.
1158 *
1159 * Removes an RCU CPU stall notifier from an atomic notifier chain.
1160 *
1161 * Returns zero on success, %-ENOENT on failure.
1162 */
1163int rcu_stall_chain_notifier_unregister(struct notifier_block *n)
1164{
1165 return atomic_notifier_chain_unregister(&rcu_cpu_stall_notifier_list, n);
1166}
1167EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_unregister);
1168
1169/*
1170 * rcu_stall_notifier_call_chain - Call functions in an RCU CPU stall notifier chain
1171 * @val: Value passed unmodified to notifier function
1172 * @v: Pointer passed unmodified to notifier function
1173 *
1174 * Calls each function in the RCU CPU stall notifier chain in turn, which
1175 * is an atomic call chain. See atomic_notifier_call_chain() for more
1176 * information.
1177 *
1178 * This is for use within RCU, hence the omission of the extra asterisk
1179 * to indicate a non-kerneldoc format header comment.
1180 */
1181int rcu_stall_notifier_call_chain(unsigned long val, void *v)
1182{
1183 return atomic_notifier_call_chain(&rcu_cpu_stall_notifier_list, val, v);
1184}
1185
1186#endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1187