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. */
20 int sysctl_panic_on_rcu_stall __read_mostly;
21 int sysctl_max_rcu_stall_to_panic __read_mostly;
22
23 #ifdef CONFIG_PROVE_RCU
24 #define RCU_STALL_DELAY_DELTA (5 * HZ)
25 #else
26 #define RCU_STALL_DELAY_DELTA 0
27 #endif
28 #define RCU_STALL_MIGHT_DIV 8
29 #define RCU_STALL_MIGHT_MIN (2 * HZ)
30
rcu_exp_jiffies_till_stall_check(void)31 int rcu_exp_jiffies_till_stall_check(void)
32 {
33 int cpu_stall_timeout = READ_ONCE(rcu_exp_cpu_stall_timeout);
34 int exp_stall_delay_delta = 0;
35 int till_stall_check;
36
37 // Zero says to use rcu_cpu_stall_timeout, but in milliseconds.
38 if (!cpu_stall_timeout)
39 cpu_stall_timeout = jiffies_to_msecs(rcu_jiffies_till_stall_check());
40
41 // Limit check must be consistent with the Kconfig limits for
42 // CONFIG_RCU_EXP_CPU_STALL_TIMEOUT, so check the allowed range.
43 // The minimum clamped value is "2UL", because at least one full
44 // tick has to be guaranteed.
45 till_stall_check = clamp(msecs_to_jiffies(cpu_stall_timeout), 2UL, 300UL * HZ);
46
47 if (cpu_stall_timeout && jiffies_to_msecs(till_stall_check) != cpu_stall_timeout)
48 WRITE_ONCE(rcu_exp_cpu_stall_timeout, jiffies_to_msecs(till_stall_check));
49
50 #ifdef CONFIG_PROVE_RCU
51 /* Add extra ~25% out of till_stall_check. */
52 exp_stall_delay_delta = ((till_stall_check * 25) / 100) + 1;
53 #endif
54
55 return till_stall_check + exp_stall_delay_delta;
56 }
57 EXPORT_SYMBOL_GPL(rcu_exp_jiffies_till_stall_check);
58
59 /* Limit-check stall timeouts specified at boottime and runtime. */
rcu_jiffies_till_stall_check(void)60 int rcu_jiffies_till_stall_check(void)
61 {
62 int till_stall_check = READ_ONCE(rcu_cpu_stall_timeout);
63
64 /*
65 * Limit check must be consistent with the Kconfig limits
66 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
67 */
68 if (till_stall_check < 3) {
69 WRITE_ONCE(rcu_cpu_stall_timeout, 3);
70 till_stall_check = 3;
71 } else if (till_stall_check > 300) {
72 WRITE_ONCE(rcu_cpu_stall_timeout, 300);
73 till_stall_check = 300;
74 }
75 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
76 }
77 EXPORT_SYMBOL_GPL(rcu_jiffies_till_stall_check);
78
79 /**
80 * rcu_gp_might_be_stalled - Is it likely that the grace period is stalled?
81 *
82 * Returns @true if the current grace period is sufficiently old that
83 * it is reasonable to assume that it might be stalled. This can be
84 * useful when deciding whether to allocate memory to enable RCU-mediated
85 * freeing on the one hand or just invoking synchronize_rcu() on the other.
86 * The latter is preferable when the grace period is stalled.
87 *
88 * Note that sampling of the .gp_start and .gp_seq fields must be done
89 * carefully to avoid false positives at the beginnings and ends of
90 * grace periods.
91 */
rcu_gp_might_be_stalled(void)92 bool rcu_gp_might_be_stalled(void)
93 {
94 unsigned long d = rcu_jiffies_till_stall_check() / RCU_STALL_MIGHT_DIV;
95 unsigned long j = jiffies;
96
97 if (d < RCU_STALL_MIGHT_MIN)
98 d = RCU_STALL_MIGHT_MIN;
99 smp_mb(); // jiffies before .gp_seq to avoid false positives.
100 if (!rcu_gp_in_progress())
101 return false;
102 // Long delays at this point avoids false positive, but a delay
103 // of ULONG_MAX/4 jiffies voids your no-false-positive warranty.
104 smp_mb(); // .gp_seq before second .gp_start
105 // And ditto here.
106 return !time_before(j, READ_ONCE(rcu_state.gp_start) + d);
107 }
108
109 /* Don't do RCU CPU stall warnings during long sysrq printouts. */
rcu_sysrq_start(void)110 void rcu_sysrq_start(void)
111 {
112 if (!rcu_cpu_stall_suppress)
113 rcu_cpu_stall_suppress = 2;
114 }
115
rcu_sysrq_end(void)116 void rcu_sysrq_end(void)
117 {
118 if (rcu_cpu_stall_suppress == 2)
119 rcu_cpu_stall_suppress = 0;
120 }
121
122 /* Don't print RCU CPU stall warnings during a kernel panic. */
rcu_panic(struct notifier_block * this,unsigned long ev,void * ptr)123 static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
124 {
125 rcu_cpu_stall_suppress = 1;
126 return NOTIFY_DONE;
127 }
128
129 static struct notifier_block rcu_panic_block = {
130 .notifier_call = rcu_panic,
131 };
132
check_cpu_stall_init(void)133 static int __init check_cpu_stall_init(void)
134 {
135 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
136 return 0;
137 }
138 early_initcall(check_cpu_stall_init);
139
140 /* If so specified via sysctl, panic, yielding cleaner stall-warning output. */
panic_on_rcu_stall(void)141 static void panic_on_rcu_stall(void)
142 {
143 static int cpu_stall;
144
145 if (++cpu_stall < sysctl_max_rcu_stall_to_panic)
146 return;
147
148 if (sysctl_panic_on_rcu_stall)
149 panic("RCU Stall\n");
150 }
151
152 /**
153 * rcu_cpu_stall_reset - restart stall-warning timeout for current grace period
154 *
155 * To perform the reset request from the caller, disable stall detection until
156 * 3 fqs loops have passed. This is required to ensure a fresh jiffies is
157 * loaded. It should be safe to do from the fqs loop as enough timer
158 * interrupts and context switches should have passed.
159 *
160 * The caller must disable hard irqs.
161 */
rcu_cpu_stall_reset(void)162 void rcu_cpu_stall_reset(void)
163 {
164 WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 3);
165 WRITE_ONCE(rcu_state.jiffies_stall, ULONG_MAX);
166 }
167
168 //////////////////////////////////////////////////////////////////////////////
169 //
170 // Interaction with RCU grace periods
171
172 /* Start of new grace period, so record stall time (and forcing times). */
record_gp_stall_check_time(void)173 static void record_gp_stall_check_time(void)
174 {
175 unsigned long j = jiffies;
176 unsigned long j1;
177
178 WRITE_ONCE(rcu_state.gp_start, j);
179 j1 = rcu_jiffies_till_stall_check();
180 smp_mb(); // ->gp_start before ->jiffies_stall and caller's ->gp_seq.
181 WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 0);
182 WRITE_ONCE(rcu_state.jiffies_stall, j + j1);
183 rcu_state.jiffies_resched = j + j1 / 2;
184 rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs);
185 }
186
187 /* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */
zero_cpu_stall_ticks(struct rcu_data * rdp)188 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
189 {
190 rdp->ticks_this_gp = 0;
191 rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id());
192 WRITE_ONCE(rdp->last_fqs_resched, jiffies);
193 }
194
195 /*
196 * If too much time has passed in the current grace period, and if
197 * so configured, go kick the relevant kthreads.
198 */
rcu_stall_kick_kthreads(void)199 static void rcu_stall_kick_kthreads(void)
200 {
201 unsigned long j;
202
203 if (!READ_ONCE(rcu_kick_kthreads))
204 return;
205 j = READ_ONCE(rcu_state.jiffies_kick_kthreads);
206 if (time_after(jiffies, j) && rcu_state.gp_kthread &&
207 (rcu_gp_in_progress() || READ_ONCE(rcu_state.gp_flags))) {
208 WARN_ONCE(1, "Kicking %s grace-period kthread\n",
209 rcu_state.name);
210 rcu_ftrace_dump(DUMP_ALL);
211 wake_up_process(rcu_state.gp_kthread);
212 WRITE_ONCE(rcu_state.jiffies_kick_kthreads, j + HZ);
213 }
214 }
215
216 /*
217 * Handler for the irq_work request posted about halfway into the RCU CPU
218 * stall timeout, and used to detect excessive irq disabling. Set state
219 * appropriately, but just complain if there is unexpected state on entry.
220 */
rcu_iw_handler(struct irq_work * iwp)221 static void rcu_iw_handler(struct irq_work *iwp)
222 {
223 struct rcu_data *rdp;
224 struct rcu_node *rnp;
225
226 rdp = container_of(iwp, struct rcu_data, rcu_iw);
227 rnp = rdp->mynode;
228 raw_spin_lock_rcu_node(rnp);
229 if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) {
230 rdp->rcu_iw_gp_seq = rnp->gp_seq;
231 rdp->rcu_iw_pending = false;
232 }
233 raw_spin_unlock_rcu_node(rnp);
234 }
235
236 //////////////////////////////////////////////////////////////////////////////
237 //
238 // Printing RCU CPU stall warnings
239
240 #ifdef CONFIG_PREEMPT_RCU
241
242 /*
243 * Dump detailed information for all tasks blocking the current RCU
244 * grace period on the specified rcu_node structure.
245 */
rcu_print_detail_task_stall_rnp(struct rcu_node * rnp)246 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
247 {
248 unsigned long flags;
249 struct task_struct *t;
250
251 raw_spin_lock_irqsave_rcu_node(rnp, flags);
252 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
253 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
254 return;
255 }
256 t = list_entry(rnp->gp_tasks->prev,
257 struct task_struct, rcu_node_entry);
258 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
259 /*
260 * We could be printing a lot while holding a spinlock.
261 * Avoid triggering hard lockup.
262 */
263 touch_nmi_watchdog();
264 sched_show_task(t);
265 }
266 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
267 }
268
269 // Communicate task state back to the RCU CPU stall warning request.
270 struct rcu_stall_chk_rdr {
271 int nesting;
272 union rcu_special rs;
273 bool on_blkd_list;
274 };
275
276 /*
277 * Report out the state of a not-running task that is stalling the
278 * current RCU grace period.
279 */
check_slow_task(struct task_struct * t,void * arg)280 static int check_slow_task(struct task_struct *t, void *arg)
281 {
282 struct rcu_stall_chk_rdr *rscrp = arg;
283
284 if (task_curr(t))
285 return -EBUSY; // It is running, so decline to inspect it.
286 rscrp->nesting = t->rcu_read_lock_nesting;
287 rscrp->rs = t->rcu_read_unlock_special;
288 rscrp->on_blkd_list = !list_empty(&t->rcu_node_entry);
289 return 0;
290 }
291
292 /*
293 * Scan the current list of tasks blocked within RCU read-side critical
294 * sections, printing out the tid of each of the first few of them.
295 */
rcu_print_task_stall(struct rcu_node * rnp,unsigned long flags)296 static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
297 __releases(rnp->lock)
298 {
299 int i = 0;
300 int ndetected = 0;
301 struct rcu_stall_chk_rdr rscr;
302 struct task_struct *t;
303 struct task_struct *ts[8];
304
305 lockdep_assert_irqs_disabled();
306 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
307 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
308 return 0;
309 }
310 pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
311 rnp->level, rnp->grplo, rnp->grphi);
312 t = list_entry(rnp->gp_tasks->prev,
313 struct task_struct, rcu_node_entry);
314 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
315 get_task_struct(t);
316 ts[i++] = t;
317 if (i >= ARRAY_SIZE(ts))
318 break;
319 }
320 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
321 while (i) {
322 t = ts[--i];
323 if (task_call_func(t, check_slow_task, &rscr))
324 pr_cont(" P%d", t->pid);
325 else
326 pr_cont(" P%d/%d:%c%c%c%c",
327 t->pid, rscr.nesting,
328 ".b"[rscr.rs.b.blocked],
329 ".q"[rscr.rs.b.need_qs],
330 ".e"[rscr.rs.b.exp_hint],
331 ".l"[rscr.on_blkd_list]);
332 lockdep_assert_irqs_disabled();
333 put_task_struct(t);
334 ndetected++;
335 }
336 pr_cont("\n");
337 return ndetected;
338 }
339
340 #else /* #ifdef CONFIG_PREEMPT_RCU */
341
342 /*
343 * Because preemptible RCU does not exist, we never have to check for
344 * tasks blocked within RCU read-side critical sections.
345 */
rcu_print_detail_task_stall_rnp(struct rcu_node * rnp)346 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
347 {
348 }
349
350 /*
351 * Because preemptible RCU does not exist, we never have to check for
352 * tasks blocked within RCU read-side critical sections.
353 */
rcu_print_task_stall(struct rcu_node * rnp,unsigned long flags)354 static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
355 __releases(rnp->lock)
356 {
357 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
358 return 0;
359 }
360 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
361
362 /*
363 * Dump stacks of all tasks running on stalled CPUs. First try using
364 * NMIs, but fall back to manual remote stack tracing on architectures
365 * that don't support NMI-based stack dumps. The NMI-triggered stack
366 * traces are more accurate because they are printed by the target CPU.
367 */
rcu_dump_cpu_stacks(void)368 static void rcu_dump_cpu_stacks(void)
369 {
370 int cpu;
371 unsigned long flags;
372 struct rcu_node *rnp;
373
374 rcu_for_each_leaf_node(rnp) {
375 printk_deferred_enter();
376 raw_spin_lock_irqsave_rcu_node(rnp, flags);
377 for_each_leaf_node_possible_cpu(rnp, cpu)
378 if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
379 if (cpu_is_offline(cpu))
380 pr_err("Offline CPU %d blocking current GP.\n", cpu);
381 else
382 dump_cpu_task(cpu);
383 }
384 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
385 printk_deferred_exit();
386 }
387 }
388
389 static const char * const gp_state_names[] = {
390 [RCU_GP_IDLE] = "RCU_GP_IDLE",
391 [RCU_GP_WAIT_GPS] = "RCU_GP_WAIT_GPS",
392 [RCU_GP_DONE_GPS] = "RCU_GP_DONE_GPS",
393 [RCU_GP_ONOFF] = "RCU_GP_ONOFF",
394 [RCU_GP_INIT] = "RCU_GP_INIT",
395 [RCU_GP_WAIT_FQS] = "RCU_GP_WAIT_FQS",
396 [RCU_GP_DOING_FQS] = "RCU_GP_DOING_FQS",
397 [RCU_GP_CLEANUP] = "RCU_GP_CLEANUP",
398 [RCU_GP_CLEANED] = "RCU_GP_CLEANED",
399 };
400
401 /*
402 * Convert a ->gp_state value to a character string.
403 */
gp_state_getname(short gs)404 static const char *gp_state_getname(short gs)
405 {
406 if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names))
407 return "???";
408 return gp_state_names[gs];
409 }
410
411 /* Is the RCU grace-period kthread being starved of CPU time? */
rcu_is_gp_kthread_starving(unsigned long * jp)412 static bool rcu_is_gp_kthread_starving(unsigned long *jp)
413 {
414 unsigned long j = jiffies - READ_ONCE(rcu_state.gp_activity);
415
416 if (jp)
417 *jp = j;
418 return j > 2 * HZ;
419 }
420
rcu_is_rcuc_kthread_starving(struct rcu_data * rdp,unsigned long * jp)421 static bool rcu_is_rcuc_kthread_starving(struct rcu_data *rdp, unsigned long *jp)
422 {
423 int cpu;
424 struct task_struct *rcuc;
425 unsigned long j;
426
427 rcuc = rdp->rcu_cpu_kthread_task;
428 if (!rcuc)
429 return false;
430
431 cpu = task_cpu(rcuc);
432 if (cpu_is_offline(cpu) || idle_cpu(cpu))
433 return false;
434
435 j = jiffies - READ_ONCE(rdp->rcuc_activity);
436
437 if (jp)
438 *jp = j;
439 return j > 2 * HZ;
440 }
441
print_cpu_stat_info(int cpu)442 static void print_cpu_stat_info(int cpu)
443 {
444 struct rcu_snap_record rsr, *rsrp;
445 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
446 struct kernel_cpustat *kcsp = &kcpustat_cpu(cpu);
447
448 if (!rcu_cpu_stall_cputime)
449 return;
450
451 rsrp = &rdp->snap_record;
452 if (rsrp->gp_seq != rdp->gp_seq)
453 return;
454
455 rsr.cputime_irq = kcpustat_field(kcsp, CPUTIME_IRQ, cpu);
456 rsr.cputime_softirq = kcpustat_field(kcsp, CPUTIME_SOFTIRQ, cpu);
457 rsr.cputime_system = kcpustat_field(kcsp, CPUTIME_SYSTEM, cpu);
458
459 pr_err("\t hardirqs softirqs csw/system\n");
460 pr_err("\t number: %8ld %10d %12lld\n",
461 kstat_cpu_irqs_sum(cpu) - rsrp->nr_hardirqs,
462 kstat_cpu_softirqs_sum(cpu) - rsrp->nr_softirqs,
463 nr_context_switches_cpu(cpu) - rsrp->nr_csw);
464 pr_err("\tcputime: %8lld %10lld %12lld ==> %d(ms)\n",
465 div_u64(rsr.cputime_irq - rsrp->cputime_irq, NSEC_PER_MSEC),
466 div_u64(rsr.cputime_softirq - rsrp->cputime_softirq, NSEC_PER_MSEC),
467 div_u64(rsr.cputime_system - rsrp->cputime_system, NSEC_PER_MSEC),
468 jiffies_to_msecs(jiffies - rsrp->jiffies));
469 }
470
471 /*
472 * Print out diagnostic information for the specified stalled CPU.
473 *
474 * If the specified CPU is aware of the current RCU grace period, then
475 * print the number of scheduling clock interrupts the CPU has taken
476 * during the time that it has been aware. Otherwise, print the number
477 * of RCU grace periods that this CPU is ignorant of, for example, "1"
478 * if the CPU was aware of the previous grace period.
479 *
480 * Also print out idle info.
481 */
print_cpu_stall_info(int cpu)482 static void print_cpu_stall_info(int cpu)
483 {
484 unsigned long delta;
485 bool falsepositive;
486 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
487 char *ticks_title;
488 unsigned long ticks_value;
489 bool rcuc_starved;
490 unsigned long j;
491 char buf[32];
492
493 /*
494 * We could be printing a lot while holding a spinlock. Avoid
495 * triggering hard lockup.
496 */
497 touch_nmi_watchdog();
498
499 ticks_value = rcu_seq_ctr(rcu_state.gp_seq - rdp->gp_seq);
500 if (ticks_value) {
501 ticks_title = "GPs behind";
502 } else {
503 ticks_title = "ticks this GP";
504 ticks_value = rdp->ticks_this_gp;
505 }
506 delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq);
507 falsepositive = rcu_is_gp_kthread_starving(NULL) &&
508 rcu_watching_snap_in_eqs(ct_rcu_watching_cpu(cpu));
509 rcuc_starved = rcu_is_rcuc_kthread_starving(rdp, &j);
510 if (rcuc_starved)
511 // Print signed value, as negative values indicate a probable bug.
512 snprintf(buf, sizeof(buf), " rcuc=%ld jiffies(starved)", j);
513 pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%04x/%ld/%#lx softirq=%u/%u fqs=%ld%s%s\n",
514 cpu,
515 "O."[!!cpu_online(cpu)],
516 "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
517 "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)],
518 !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' :
519 rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' :
520 "!."[!delta],
521 ticks_value, ticks_title,
522 ct_rcu_watching_cpu(cpu) & 0xffff,
523 ct_nesting_cpu(cpu), ct_nmi_nesting_cpu(cpu),
524 rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
525 data_race(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
526 rcuc_starved ? buf : "",
527 falsepositive ? " (false positive?)" : "");
528
529 print_cpu_stat_info(cpu);
530 }
531
532 /* Complain about starvation of grace-period kthread. */
rcu_check_gp_kthread_starvation(void)533 static void rcu_check_gp_kthread_starvation(void)
534 {
535 int cpu;
536 struct task_struct *gpk = rcu_state.gp_kthread;
537 unsigned long j;
538
539 if (rcu_is_gp_kthread_starving(&j)) {
540 cpu = gpk ? task_cpu(gpk) : -1;
541 pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x ->cpu=%d\n",
542 rcu_state.name, j,
543 (long)rcu_seq_current(&rcu_state.gp_seq),
544 data_race(READ_ONCE(rcu_state.gp_flags)),
545 gp_state_getname(rcu_state.gp_state),
546 data_race(READ_ONCE(rcu_state.gp_state)),
547 gpk ? data_race(READ_ONCE(gpk->__state)) : ~0, cpu);
548 if (gpk) {
549 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
550
551 pr_err("\tUnless %s kthread gets sufficient CPU time, OOM is now expected behavior.\n", rcu_state.name);
552 pr_err("RCU grace-period kthread stack dump:\n");
553 sched_show_task(gpk);
554 if (cpu_is_offline(cpu)) {
555 pr_err("RCU GP kthread last ran on offline CPU %d.\n", cpu);
556 } else if (!(data_race(READ_ONCE(rdp->mynode->qsmask)) & rdp->grpmask)) {
557 pr_err("Stack dump where RCU GP kthread last ran:\n");
558 dump_cpu_task(cpu);
559 }
560 wake_up_process(gpk);
561 }
562 }
563 }
564
565 /* Complain about missing wakeups from expired fqs wait timer */
rcu_check_gp_kthread_expired_fqs_timer(void)566 static void rcu_check_gp_kthread_expired_fqs_timer(void)
567 {
568 struct task_struct *gpk = rcu_state.gp_kthread;
569 short gp_state;
570 unsigned long jiffies_fqs;
571 int cpu;
572
573 /*
574 * Order reads of .gp_state and .jiffies_force_qs.
575 * Matching smp_wmb() is present in rcu_gp_fqs_loop().
576 */
577 gp_state = smp_load_acquire(&rcu_state.gp_state);
578 jiffies_fqs = READ_ONCE(rcu_state.jiffies_force_qs);
579
580 if (gp_state == RCU_GP_WAIT_FQS &&
581 time_after(jiffies, jiffies_fqs + RCU_STALL_MIGHT_MIN) &&
582 gpk && !READ_ONCE(gpk->on_rq)) {
583 cpu = task_cpu(gpk);
584 pr_err("%s kthread timer wakeup didn't happen for %ld jiffies! g%ld f%#x %s(%d) ->state=%#x\n",
585 rcu_state.name, (jiffies - jiffies_fqs),
586 (long)rcu_seq_current(&rcu_state.gp_seq),
587 data_race(READ_ONCE(rcu_state.gp_flags)), // Diagnostic read
588 gp_state_getname(RCU_GP_WAIT_FQS), RCU_GP_WAIT_FQS,
589 data_race(READ_ONCE(gpk->__state)));
590 pr_err("\tPossible timer handling issue on cpu=%d timer-softirq=%u\n",
591 cpu, kstat_softirqs_cpu(TIMER_SOFTIRQ, cpu));
592 }
593 }
594
print_other_cpu_stall(unsigned long gp_seq,unsigned long gps)595 static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
596 {
597 int cpu;
598 unsigned long flags;
599 unsigned long gpa;
600 unsigned long j;
601 int ndetected = 0;
602 struct rcu_node *rnp;
603 long totqlen = 0;
604
605 lockdep_assert_irqs_disabled();
606
607 /* Kick and suppress, if so configured. */
608 rcu_stall_kick_kthreads();
609 if (rcu_stall_is_suppressed())
610 return;
611
612 nbcon_cpu_emergency_enter();
613
614 /*
615 * OK, time to rat on our buddy...
616 * See Documentation/RCU/stallwarn.rst for info on how to debug
617 * RCU CPU stall warnings.
618 */
619 trace_rcu_stall_warning(rcu_state.name, TPS("StallDetected"));
620 pr_err("INFO: %s detected stalls on CPUs/tasks:\n", rcu_state.name);
621 rcu_for_each_leaf_node(rnp) {
622 raw_spin_lock_irqsave_rcu_node(rnp, flags);
623 if (rnp->qsmask != 0) {
624 for_each_leaf_node_possible_cpu(rnp, cpu)
625 if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
626 print_cpu_stall_info(cpu);
627 ndetected++;
628 }
629 }
630 ndetected += rcu_print_task_stall(rnp, flags); // Releases rnp->lock.
631 lockdep_assert_irqs_disabled();
632 }
633
634 for_each_possible_cpu(cpu)
635 totqlen += rcu_get_n_cbs_cpu(cpu);
636 pr_err("\t(detected by %d, t=%ld jiffies, g=%ld, q=%lu ncpus=%d)\n",
637 smp_processor_id(), (long)(jiffies - gps),
638 (long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
639 data_race(rcu_state.n_online_cpus)); // Diagnostic read
640 if (ndetected) {
641 rcu_dump_cpu_stacks();
642
643 /* Complain about tasks blocking the grace period. */
644 rcu_for_each_leaf_node(rnp)
645 rcu_print_detail_task_stall_rnp(rnp);
646 } else {
647 if (rcu_seq_current(&rcu_state.gp_seq) != gp_seq) {
648 pr_err("INFO: Stall ended before state dump start\n");
649 } else {
650 j = jiffies;
651 gpa = data_race(READ_ONCE(rcu_state.gp_activity));
652 pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
653 rcu_state.name, j - gpa, j, gpa,
654 data_race(READ_ONCE(jiffies_till_next_fqs)),
655 data_race(READ_ONCE(rcu_get_root()->qsmask)));
656 }
657 }
658 /* Rewrite if needed in case of slow consoles. */
659 if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
660 WRITE_ONCE(rcu_state.jiffies_stall,
661 jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
662
663 rcu_check_gp_kthread_expired_fqs_timer();
664 rcu_check_gp_kthread_starvation();
665
666 nbcon_cpu_emergency_exit();
667
668 panic_on_rcu_stall();
669
670 rcu_force_quiescent_state(); /* Kick them all. */
671 }
672
print_cpu_stall(unsigned long gps)673 static void print_cpu_stall(unsigned long gps)
674 {
675 int cpu;
676 unsigned long flags;
677 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
678 struct rcu_node *rnp = rcu_get_root();
679 long totqlen = 0;
680
681 lockdep_assert_irqs_disabled();
682
683 /* Kick and suppress, if so configured. */
684 rcu_stall_kick_kthreads();
685 if (rcu_stall_is_suppressed())
686 return;
687
688 nbcon_cpu_emergency_enter();
689
690 /*
691 * OK, time to rat on ourselves...
692 * See Documentation/RCU/stallwarn.rst for info on how to debug
693 * RCU CPU stall warnings.
694 */
695 trace_rcu_stall_warning(rcu_state.name, TPS("SelfDetected"));
696 pr_err("INFO: %s self-detected stall on CPU\n", rcu_state.name);
697 raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags);
698 print_cpu_stall_info(smp_processor_id());
699 raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags);
700 for_each_possible_cpu(cpu)
701 totqlen += rcu_get_n_cbs_cpu(cpu);
702 pr_err("\t(t=%lu jiffies g=%ld q=%lu ncpus=%d)\n",
703 jiffies - gps,
704 (long)rcu_seq_current(&rcu_state.gp_seq), totqlen,
705 data_race(rcu_state.n_online_cpus)); // Diagnostic read
706
707 rcu_check_gp_kthread_expired_fqs_timer();
708 rcu_check_gp_kthread_starvation();
709
710 rcu_dump_cpu_stacks();
711
712 raw_spin_lock_irqsave_rcu_node(rnp, flags);
713 /* Rewrite if needed in case of slow consoles. */
714 if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall)))
715 WRITE_ONCE(rcu_state.jiffies_stall,
716 jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
717 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
718
719 nbcon_cpu_emergency_exit();
720
721 panic_on_rcu_stall();
722
723 /*
724 * Attempt to revive the RCU machinery by forcing a context switch.
725 *
726 * A context switch would normally allow the RCU state machine to make
727 * progress and it could be we're stuck in kernel space without context
728 * switches for an entirely unreasonable amount of time.
729 */
730 set_tsk_need_resched(current);
731 set_preempt_need_resched();
732 }
733
734 static bool csd_lock_suppress_rcu_stall;
735 module_param(csd_lock_suppress_rcu_stall, bool, 0644);
736
check_cpu_stall(struct rcu_data * rdp)737 static void check_cpu_stall(struct rcu_data *rdp)
738 {
739 bool self_detected;
740 unsigned long gs1;
741 unsigned long gs2;
742 unsigned long gps;
743 unsigned long j;
744 unsigned long jn;
745 unsigned long js;
746 struct rcu_node *rnp;
747
748 lockdep_assert_irqs_disabled();
749 if ((rcu_stall_is_suppressed() && !READ_ONCE(rcu_kick_kthreads)) ||
750 !rcu_gp_in_progress())
751 return;
752 rcu_stall_kick_kthreads();
753
754 /*
755 * Check if it was requested (via rcu_cpu_stall_reset()) that the FQS
756 * loop has to set jiffies to ensure a non-stale jiffies value. This
757 * is required to have good jiffies value after coming out of long
758 * breaks of jiffies updates. Not doing so can cause false positives.
759 */
760 if (READ_ONCE(rcu_state.nr_fqs_jiffies_stall) > 0)
761 return;
762
763 j = jiffies;
764
765 /*
766 * Lots of memory barriers to reject false positives.
767 *
768 * The idea is to pick up rcu_state.gp_seq, then
769 * rcu_state.jiffies_stall, then rcu_state.gp_start, and finally
770 * another copy of rcu_state.gp_seq. These values are updated in
771 * the opposite order with memory barriers (or equivalent) during
772 * grace-period initialization and cleanup. Now, a false positive
773 * can occur if we get an new value of rcu_state.gp_start and a old
774 * value of rcu_state.jiffies_stall. But given the memory barriers,
775 * the only way that this can happen is if one grace period ends
776 * and another starts between these two fetches. This is detected
777 * by comparing the second fetch of rcu_state.gp_seq with the
778 * previous fetch from rcu_state.gp_seq.
779 *
780 * Given this check, comparisons of jiffies, rcu_state.jiffies_stall,
781 * and rcu_state.gp_start suffice to forestall false positives.
782 */
783 gs1 = READ_ONCE(rcu_state.gp_seq);
784 smp_rmb(); /* Pick up ->gp_seq first... */
785 js = READ_ONCE(rcu_state.jiffies_stall);
786 smp_rmb(); /* ...then ->jiffies_stall before the rest... */
787 gps = READ_ONCE(rcu_state.gp_start);
788 smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */
789 gs2 = READ_ONCE(rcu_state.gp_seq);
790 if (gs1 != gs2 ||
791 ULONG_CMP_LT(j, js) ||
792 ULONG_CMP_GE(gps, js))
793 return; /* No stall or GP completed since entering function. */
794 rnp = rdp->mynode;
795 jn = jiffies + ULONG_MAX / 2;
796 self_detected = READ_ONCE(rnp->qsmask) & rdp->grpmask;
797 if (rcu_gp_in_progress() &&
798 (self_detected || ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) &&
799 cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) {
800 /*
801 * If a virtual machine is stopped by the host it can look to
802 * the watchdog like an RCU stall. Check to see if the host
803 * stopped the vm.
804 */
805 if (kvm_check_and_clear_guest_paused())
806 return;
807
808 rcu_stall_notifier_call_chain(RCU_STALL_NOTIFY_NORM, (void *)j - gps);
809 if (READ_ONCE(csd_lock_suppress_rcu_stall) && csd_lock_is_stuck()) {
810 pr_err("INFO: %s detected stall, but suppressed full report due to a stuck CSD-lock.\n", rcu_state.name);
811 } else if (self_detected) {
812 /* We haven't checked in, so go dump stack. */
813 print_cpu_stall(gps);
814 } else {
815 /* They had a few time units to dump stack, so complain. */
816 print_other_cpu_stall(gs2, gps);
817 }
818
819 if (READ_ONCE(rcu_cpu_stall_ftrace_dump))
820 rcu_ftrace_dump(DUMP_ALL);
821
822 if (READ_ONCE(rcu_state.jiffies_stall) == jn) {
823 jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
824 WRITE_ONCE(rcu_state.jiffies_stall, jn);
825 }
826 }
827 }
828
829 //////////////////////////////////////////////////////////////////////////////
830 //
831 // RCU forward-progress mechanisms, including for callback invocation.
832
833
834 /*
835 * Check to see if a failure to end RCU priority inversion was due to
836 * a CPU not passing through a quiescent state. When this happens, there
837 * is nothing that RCU priority boosting can do to help, so we shouldn't
838 * count this as an RCU priority boosting failure. A return of true says
839 * RCU priority boosting is to blame, and false says otherwise. If false
840 * is returned, the first of the CPUs to blame is stored through cpup.
841 * If there was no CPU blocking the current grace period, but also nothing
842 * in need of being boosted, *cpup is set to -1. This can happen in case
843 * of vCPU preemption while the last CPU is reporting its quiscent state,
844 * for example.
845 *
846 * If cpup is NULL, then a lockless quick check is carried out, suitable
847 * for high-rate usage. On the other hand, if cpup is non-NULL, each
848 * rcu_node structure's ->lock is acquired, ruling out high-rate usage.
849 */
rcu_check_boost_fail(unsigned long gp_state,int * cpup)850 bool rcu_check_boost_fail(unsigned long gp_state, int *cpup)
851 {
852 bool atb = false;
853 int cpu;
854 unsigned long flags;
855 struct rcu_node *rnp;
856
857 rcu_for_each_leaf_node(rnp) {
858 if (!cpup) {
859 if (data_race(READ_ONCE(rnp->qsmask))) {
860 return false;
861 } else {
862 if (READ_ONCE(rnp->gp_tasks))
863 atb = true;
864 continue;
865 }
866 }
867 *cpup = -1;
868 raw_spin_lock_irqsave_rcu_node(rnp, flags);
869 if (rnp->gp_tasks)
870 atb = true;
871 if (!rnp->qsmask) {
872 // No CPUs without quiescent states for this rnp.
873 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
874 continue;
875 }
876 // Find the first holdout CPU.
877 for_each_leaf_node_possible_cpu(rnp, cpu) {
878 if (rnp->qsmask & (1UL << (cpu - rnp->grplo))) {
879 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
880 *cpup = cpu;
881 return false;
882 }
883 }
884 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
885 }
886 // Can't blame CPUs, so must blame RCU priority boosting.
887 return atb;
888 }
889 EXPORT_SYMBOL_GPL(rcu_check_boost_fail);
890
891 /*
892 * Show the state of the grace-period kthreads.
893 */
show_rcu_gp_kthreads(void)894 void show_rcu_gp_kthreads(void)
895 {
896 unsigned long cbs = 0;
897 int cpu;
898 unsigned long j;
899 unsigned long ja;
900 unsigned long jr;
901 unsigned long js;
902 unsigned long jw;
903 struct rcu_data *rdp;
904 struct rcu_node *rnp;
905 struct task_struct *t = READ_ONCE(rcu_state.gp_kthread);
906
907 j = jiffies;
908 ja = j - data_race(READ_ONCE(rcu_state.gp_activity));
909 jr = j - data_race(READ_ONCE(rcu_state.gp_req_activity));
910 js = j - data_race(READ_ONCE(rcu_state.gp_start));
911 jw = j - data_race(READ_ONCE(rcu_state.gp_wake_time));
912 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",
913 rcu_state.name, gp_state_getname(rcu_state.gp_state),
914 data_race(READ_ONCE(rcu_state.gp_state)),
915 t ? data_race(READ_ONCE(t->__state)) : 0x1ffff, t ? t->rt_priority : 0xffU,
916 js, ja, jr, jw, (long)data_race(READ_ONCE(rcu_state.gp_wake_seq)),
917 (long)data_race(READ_ONCE(rcu_state.gp_seq)),
918 (long)data_race(READ_ONCE(rcu_get_root()->gp_seq_needed)),
919 data_race(READ_ONCE(rcu_state.gp_max)),
920 data_race(READ_ONCE(rcu_state.gp_flags)));
921 rcu_for_each_node_breadth_first(rnp) {
922 if (ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq), READ_ONCE(rnp->gp_seq_needed)) &&
923 !data_race(READ_ONCE(rnp->qsmask)) && !data_race(READ_ONCE(rnp->boost_tasks)) &&
924 !data_race(READ_ONCE(rnp->exp_tasks)) && !data_race(READ_ONCE(rnp->gp_tasks)))
925 continue;
926 pr_info("\trcu_node %d:%d ->gp_seq %ld ->gp_seq_needed %ld ->qsmask %#lx %c%c%c%c ->n_boosts %ld\n",
927 rnp->grplo, rnp->grphi,
928 (long)data_race(READ_ONCE(rnp->gp_seq)),
929 (long)data_race(READ_ONCE(rnp->gp_seq_needed)),
930 data_race(READ_ONCE(rnp->qsmask)),
931 ".b"[!!data_race(READ_ONCE(rnp->boost_kthread_task))],
932 ".B"[!!data_race(READ_ONCE(rnp->boost_tasks))],
933 ".E"[!!data_race(READ_ONCE(rnp->exp_tasks))],
934 ".G"[!!data_race(READ_ONCE(rnp->gp_tasks))],
935 data_race(READ_ONCE(rnp->n_boosts)));
936 if (!rcu_is_leaf_node(rnp))
937 continue;
938 for_each_leaf_node_possible_cpu(rnp, cpu) {
939 rdp = per_cpu_ptr(&rcu_data, cpu);
940 if (READ_ONCE(rdp->gpwrap) ||
941 ULONG_CMP_GE(READ_ONCE(rcu_state.gp_seq),
942 READ_ONCE(rdp->gp_seq_needed)))
943 continue;
944 pr_info("\tcpu %d ->gp_seq_needed %ld\n",
945 cpu, (long)data_race(READ_ONCE(rdp->gp_seq_needed)));
946 }
947 }
948 for_each_possible_cpu(cpu) {
949 rdp = per_cpu_ptr(&rcu_data, cpu);
950 cbs += data_race(READ_ONCE(rdp->n_cbs_invoked));
951 if (rcu_segcblist_is_offloaded(&rdp->cblist))
952 show_rcu_nocb_state(rdp);
953 }
954 pr_info("RCU callbacks invoked since boot: %lu\n", cbs);
955 show_rcu_tasks_gp_kthreads();
956 }
957 EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
958
959 /*
960 * This function checks for grace-period requests that fail to motivate
961 * RCU to come out of its idle mode.
962 */
rcu_check_gp_start_stall(struct rcu_node * rnp,struct rcu_data * rdp,const unsigned long gpssdelay)963 static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
964 const unsigned long gpssdelay)
965 {
966 unsigned long flags;
967 unsigned long j;
968 struct rcu_node *rnp_root = rcu_get_root();
969 static atomic_t warned = ATOMIC_INIT(0);
970
971 if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() ||
972 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
973 READ_ONCE(rnp_root->gp_seq_needed)) ||
974 !smp_load_acquire(&rcu_state.gp_kthread)) // Get stable kthread.
975 return;
976 j = jiffies; /* Expensive access, and in common case don't get here. */
977 if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
978 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
979 atomic_read(&warned))
980 return;
981
982 raw_spin_lock_irqsave_rcu_node(rnp, flags);
983 j = jiffies;
984 if (rcu_gp_in_progress() ||
985 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
986 READ_ONCE(rnp_root->gp_seq_needed)) ||
987 time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
988 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
989 atomic_read(&warned)) {
990 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
991 return;
992 }
993 /* Hold onto the leaf lock to make others see warned==1. */
994
995 if (rnp_root != rnp)
996 raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
997 j = jiffies;
998 if (rcu_gp_in_progress() ||
999 ULONG_CMP_GE(READ_ONCE(rnp_root->gp_seq),
1000 READ_ONCE(rnp_root->gp_seq_needed)) ||
1001 time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) ||
1002 time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) ||
1003 atomic_xchg(&warned, 1)) {
1004 if (rnp_root != rnp)
1005 /* irqs remain disabled. */
1006 raw_spin_unlock_rcu_node(rnp_root);
1007 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1008 return;
1009 }
1010 WARN_ON(1);
1011 if (rnp_root != rnp)
1012 raw_spin_unlock_rcu_node(rnp_root);
1013 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1014 show_rcu_gp_kthreads();
1015 }
1016
1017 /*
1018 * Do a forward-progress check for rcutorture. This is normally invoked
1019 * due to an OOM event. The argument "j" gives the time period during
1020 * which rcutorture would like progress to have been made.
1021 */
rcu_fwd_progress_check(unsigned long j)1022 void rcu_fwd_progress_check(unsigned long j)
1023 {
1024 unsigned long cbs;
1025 int cpu;
1026 unsigned long max_cbs = 0;
1027 int max_cpu = -1;
1028 struct rcu_data *rdp;
1029
1030 if (rcu_gp_in_progress()) {
1031 pr_info("%s: GP age %lu jiffies\n",
1032 __func__, jiffies - data_race(READ_ONCE(rcu_state.gp_start)));
1033 show_rcu_gp_kthreads();
1034 } else {
1035 pr_info("%s: Last GP end %lu jiffies ago\n",
1036 __func__, jiffies - data_race(READ_ONCE(rcu_state.gp_end)));
1037 preempt_disable();
1038 rdp = this_cpu_ptr(&rcu_data);
1039 rcu_check_gp_start_stall(rdp->mynode, rdp, j);
1040 preempt_enable();
1041 }
1042 for_each_possible_cpu(cpu) {
1043 cbs = rcu_get_n_cbs_cpu(cpu);
1044 if (!cbs)
1045 continue;
1046 if (max_cpu < 0)
1047 pr_info("%s: callbacks", __func__);
1048 pr_cont(" %d: %lu", cpu, cbs);
1049 if (cbs <= max_cbs)
1050 continue;
1051 max_cbs = cbs;
1052 max_cpu = cpu;
1053 }
1054 if (max_cpu >= 0)
1055 pr_cont("\n");
1056 }
1057 EXPORT_SYMBOL_GPL(rcu_fwd_progress_check);
1058
1059 /* Commandeer a sysrq key to dump RCU's tree. */
1060 static bool sysrq_rcu;
1061 module_param(sysrq_rcu, bool, 0444);
1062
1063 /* Dump grace-period-request information due to commandeered sysrq. */
sysrq_show_rcu(u8 key)1064 static void sysrq_show_rcu(u8 key)
1065 {
1066 show_rcu_gp_kthreads();
1067 }
1068
1069 static const struct sysrq_key_op sysrq_rcudump_op = {
1070 .handler = sysrq_show_rcu,
1071 .help_msg = "show-rcu(y)",
1072 .action_msg = "Show RCU tree",
1073 .enable_mask = SYSRQ_ENABLE_DUMP,
1074 };
1075
rcu_sysrq_init(void)1076 static int __init rcu_sysrq_init(void)
1077 {
1078 if (sysrq_rcu)
1079 return register_sysrq_key('y', &sysrq_rcudump_op);
1080 return 0;
1081 }
1082 early_initcall(rcu_sysrq_init);
1083
1084 #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1085
1086 //////////////////////////////////////////////////////////////////////////////
1087 //
1088 // RCU CPU stall-warning notifiers
1089
1090 static ATOMIC_NOTIFIER_HEAD(rcu_cpu_stall_notifier_list);
1091
1092 /**
1093 * rcu_stall_chain_notifier_register - Add an RCU CPU stall notifier
1094 * @n: Entry to add.
1095 *
1096 * Adds an RCU CPU stall notifier to an atomic notifier chain.
1097 * The @action passed to a notifier will be @RCU_STALL_NOTIFY_NORM or
1098 * friends. The @data will be the duration of the stalled grace period,
1099 * in jiffies, coerced to a void* pointer.
1100 *
1101 * Returns 0 on success, %-EEXIST on error.
1102 */
rcu_stall_chain_notifier_register(struct notifier_block * n)1103 int rcu_stall_chain_notifier_register(struct notifier_block *n)
1104 {
1105 int rcsn = rcu_cpu_stall_notifiers;
1106
1107 WARN(1, "Adding %pS() to RCU stall notifier list (%s).\n", n->notifier_call,
1108 rcsn ? "possibly suppressing RCU CPU stall warnings" : "failed, so all is well");
1109 if (rcsn)
1110 return atomic_notifier_chain_register(&rcu_cpu_stall_notifier_list, n);
1111 return -EEXIST;
1112 }
1113 EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_register);
1114
1115 /**
1116 * rcu_stall_chain_notifier_unregister - Remove an RCU CPU stall notifier
1117 * @n: Entry to add.
1118 *
1119 * Removes an RCU CPU stall notifier from an atomic notifier chain.
1120 *
1121 * Returns zero on success, %-ENOENT on failure.
1122 */
rcu_stall_chain_notifier_unregister(struct notifier_block * n)1123 int rcu_stall_chain_notifier_unregister(struct notifier_block *n)
1124 {
1125 return atomic_notifier_chain_unregister(&rcu_cpu_stall_notifier_list, n);
1126 }
1127 EXPORT_SYMBOL_GPL(rcu_stall_chain_notifier_unregister);
1128
1129 /*
1130 * rcu_stall_notifier_call_chain - Call functions in an RCU CPU stall notifier chain
1131 * @val: Value passed unmodified to notifier function
1132 * @v: Pointer passed unmodified to notifier function
1133 *
1134 * Calls each function in the RCU CPU stall notifier chain in turn, which
1135 * is an atomic call chain. See atomic_notifier_call_chain() for more
1136 * information.
1137 *
1138 * This is for use within RCU, hence the omission of the extra asterisk
1139 * to indicate a non-kerneldoc format header comment.
1140 */
rcu_stall_notifier_call_chain(unsigned long val,void * v)1141 int rcu_stall_notifier_call_chain(unsigned long val, void *v)
1142 {
1143 return atomic_notifier_call_chain(&rcu_cpu_stall_notifier_list, val, v);
1144 }
1145
1146 #endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
1147