1/* SPDX-License-Identifier: GPL-2.0+ */
2/*
3 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
4 * Internal non-public definitions that provide either classic
5 * or preemptible semantics.
6 *
7 * Copyright Red Hat, 2009
8 * Copyright IBM Corporation, 2009
9 *
10 * Author: Ingo Molnar <mingo@elte.hu>
11 * Paul E. McKenney <paulmck@linux.ibm.com>
12 */
13
14#include "../locking/rtmutex_common.h"
15
16static bool rcu_rdp_is_offloaded(struct rcu_data *rdp)
17{
18 /*
19 * In order to read the offloaded state of an rdp in a safe
20 * and stable way and prevent from its value to be changed
21 * under us, we must either hold the barrier mutex, the cpu
22 * hotplug lock (read or write) or the nocb lock. Local
23 * non-preemptible reads are also safe. NOCB kthreads and
24 * timers have their own means of synchronization against the
25 * offloaded state updaters.
26 */
27 RCU_NOCB_LOCKDEP_WARN(
28 !(lockdep_is_held(&rcu_state.barrier_mutex) ||
29 (IS_ENABLED(CONFIG_HOTPLUG_CPU) && lockdep_is_cpus_held()) ||
30 lockdep_is_held(&rdp->nocb_lock) ||
31 lockdep_is_held(&rcu_state.nocb_mutex) ||
32 ((!(IS_ENABLED(CONFIG_PREEMPT_COUNT) && preemptible()) || softirq_count()) &&
33 rdp == this_cpu_ptr(&rcu_data)) ||
34 rcu_current_is_nocb_kthread(rdp)),
35 "Unsafe read of RCU_NOCB offloaded state"
36 );
37
38 return rcu_segcblist_is_offloaded(rsclp: &rdp->cblist);
39}
40
41/*
42 * Check the RCU kernel configuration parameters and print informative
43 * messages about anything out of the ordinary.
44 */
45static void __init rcu_bootup_announce_oddness(void)
46{
47 if (IS_ENABLED(CONFIG_RCU_TRACE))
48 pr_info("\tRCU event tracing is enabled.\n");
49 if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) ||
50 (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32))
51 pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n",
52 RCU_FANOUT);
53 if (rcu_fanout_exact)
54 pr_info("\tHierarchical RCU autobalancing is disabled.\n");
55 if (IS_ENABLED(CONFIG_PROVE_RCU))
56 pr_info("\tRCU lockdep checking is enabled.\n");
57 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
58 pr_info("\tRCU strict (and thus non-scalable) grace periods are enabled.\n");
59 if (RCU_NUM_LVLS >= 4)
60 pr_info("\tFour(or more)-level hierarchy is enabled.\n");
61 if (RCU_FANOUT_LEAF != 16)
62 pr_info("\tBuild-time adjustment of leaf fanout to %d.\n",
63 RCU_FANOUT_LEAF);
64 if (rcu_fanout_leaf != RCU_FANOUT_LEAF)
65 pr_info("\tBoot-time adjustment of leaf fanout to %d.\n",
66 rcu_fanout_leaf);
67 if (nr_cpu_ids != NR_CPUS)
68 pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids);
69#ifdef CONFIG_RCU_BOOST
70 pr_info("\tRCU priority boosting: priority %d delay %d ms.\n",
71 kthread_prio, CONFIG_RCU_BOOST_DELAY);
72#endif
73 if (blimit != DEFAULT_RCU_BLIMIT)
74 pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit);
75 if (qhimark != DEFAULT_RCU_QHIMARK)
76 pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark);
77 if (qlowmark != DEFAULT_RCU_QLOMARK)
78 pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark);
79 if (qovld != DEFAULT_RCU_QOVLD)
80 pr_info("\tBoot-time adjustment of callback overload level to %ld.\n", qovld);
81 if (jiffies_till_first_fqs != ULONG_MAX)
82 pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs);
83 if (jiffies_till_next_fqs != ULONG_MAX)
84 pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs);
85 if (jiffies_till_sched_qs != ULONG_MAX)
86 pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs);
87 if (rcu_kick_kthreads)
88 pr_info("\tKick kthreads if too-long grace period.\n");
89 if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD))
90 pr_info("\tRCU callback double-/use-after-free debug is enabled.\n");
91 if (gp_preinit_delay)
92 pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay);
93 if (gp_init_delay)
94 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay);
95 if (gp_cleanup_delay)
96 pr_info("\tRCU debug GP cleanup slowdown %d jiffies.\n", gp_cleanup_delay);
97 if (nohz_full_patience_delay < 0) {
98 pr_info("\tRCU NOCB CPU patience negative (%d), resetting to zero.\n", nohz_full_patience_delay);
99 nohz_full_patience_delay = 0;
100 } else if (nohz_full_patience_delay > 5 * MSEC_PER_SEC) {
101 pr_info("\tRCU NOCB CPU patience too large (%d), resetting to %ld.\n", nohz_full_patience_delay, 5 * MSEC_PER_SEC);
102 nohz_full_patience_delay = 5 * MSEC_PER_SEC;
103 } else if (nohz_full_patience_delay) {
104 pr_info("\tRCU NOCB CPU patience set to %d milliseconds.\n", nohz_full_patience_delay);
105 }
106 nohz_full_patience_delay_jiffies = msecs_to_jiffies(m: nohz_full_patience_delay);
107 if (!use_softirq)
108 pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n");
109 if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG))
110 pr_info("\tRCU debug extended QS entry/exit.\n");
111 rcupdate_announce_bootup_oddness();
112}
113
114#ifdef CONFIG_PREEMPT_RCU
115
116static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake);
117static void rcu_read_unlock_special(struct task_struct *t);
118
119/*
120 * Tell them what RCU they are running.
121 */
122static void __init rcu_bootup_announce(void)
123{
124 pr_info("Preemptible hierarchical RCU implementation.\n");
125 rcu_bootup_announce_oddness();
126}
127
128/* Flags for rcu_preempt_ctxt_queue() decision table. */
129#define RCU_GP_TASKS 0x8
130#define RCU_EXP_TASKS 0x4
131#define RCU_GP_BLKD 0x2
132#define RCU_EXP_BLKD 0x1
133
134/*
135 * Queues a task preempted within an RCU-preempt read-side critical
136 * section into the appropriate location within the ->blkd_tasks list,
137 * depending on the states of any ongoing normal and expedited grace
138 * periods. The ->gp_tasks pointer indicates which element the normal
139 * grace period is waiting on (NULL if none), and the ->exp_tasks pointer
140 * indicates which element the expedited grace period is waiting on (again,
141 * NULL if none). If a grace period is waiting on a given element in the
142 * ->blkd_tasks list, it also waits on all subsequent elements. Thus,
143 * adding a task to the tail of the list blocks any grace period that is
144 * already waiting on one of the elements. In contrast, adding a task
145 * to the head of the list won't block any grace period that is already
146 * waiting on one of the elements.
147 *
148 * This queuing is imprecise, and can sometimes make an ongoing grace
149 * period wait for a task that is not strictly speaking blocking it.
150 * Given the choice, we needlessly block a normal grace period rather than
151 * blocking an expedited grace period.
152 *
153 * Note that an endless sequence of expedited grace periods still cannot
154 * indefinitely postpone a normal grace period. Eventually, all of the
155 * fixed number of preempted tasks blocking the normal grace period that are
156 * not also blocking the expedited grace period will resume and complete
157 * their RCU read-side critical sections. At that point, the ->gp_tasks
158 * pointer will equal the ->exp_tasks pointer, at which point the end of
159 * the corresponding expedited grace period will also be the end of the
160 * normal grace period.
161 */
162static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp)
163 __releases(rnp->lock) /* But leaves rrupts disabled. */
164{
165 int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) +
166 (rnp->exp_tasks ? RCU_EXP_TASKS : 0) +
167 (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) +
168 (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0);
169 struct task_struct *t = current;
170
171 raw_lockdep_assert_held_rcu_node(rnp);
172 WARN_ON_ONCE(rdp->mynode != rnp);
173 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
174 /* RCU better not be waiting on newly onlined CPUs! */
175 WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask &
176 rdp->grpmask);
177
178 /*
179 * Decide where to queue the newly blocked task. In theory,
180 * this could be an if-statement. In practice, when I tried
181 * that, it was quite messy.
182 */
183 switch (blkd_state) {
184 case 0:
185 case RCU_EXP_TASKS:
186 case RCU_EXP_TASKS | RCU_GP_BLKD:
187 case RCU_GP_TASKS:
188 case RCU_GP_TASKS | RCU_EXP_TASKS:
189
190 /*
191 * Blocking neither GP, or first task blocking the normal
192 * GP but not blocking the already-waiting expedited GP.
193 * Queue at the head of the list to avoid unnecessarily
194 * blocking the already-waiting GPs.
195 */
196 list_add(new: &t->rcu_node_entry, head: &rnp->blkd_tasks);
197 break;
198
199 case RCU_EXP_BLKD:
200 case RCU_GP_BLKD:
201 case RCU_GP_BLKD | RCU_EXP_BLKD:
202 case RCU_GP_TASKS | RCU_EXP_BLKD:
203 case RCU_GP_TASKS | RCU_GP_BLKD | RCU_EXP_BLKD:
204 case RCU_GP_TASKS | RCU_EXP_TASKS | RCU_GP_BLKD | RCU_EXP_BLKD:
205
206 /*
207 * First task arriving that blocks either GP, or first task
208 * arriving that blocks the expedited GP (with the normal
209 * GP already waiting), or a task arriving that blocks
210 * both GPs with both GPs already waiting. Queue at the
211 * tail of the list to avoid any GP waiting on any of the
212 * already queued tasks that are not blocking it.
213 */
214 list_add_tail(new: &t->rcu_node_entry, head: &rnp->blkd_tasks);
215 break;
216
217 case RCU_EXP_TASKS | RCU_EXP_BLKD:
218 case RCU_EXP_TASKS | RCU_GP_BLKD | RCU_EXP_BLKD:
219 case RCU_GP_TASKS | RCU_EXP_TASKS | RCU_EXP_BLKD:
220
221 /*
222 * Second or subsequent task blocking the expedited GP.
223 * The task either does not block the normal GP, or is the
224 * first task blocking the normal GP. Queue just after
225 * the first task blocking the expedited GP.
226 */
227 list_add(new: &t->rcu_node_entry, head: rnp->exp_tasks);
228 break;
229
230 case RCU_GP_TASKS | RCU_GP_BLKD:
231 case RCU_GP_TASKS | RCU_EXP_TASKS | RCU_GP_BLKD:
232
233 /*
234 * Second or subsequent task blocking the normal GP.
235 * The task does not block the expedited GP. Queue just
236 * after the first task blocking the normal GP.
237 */
238 list_add(new: &t->rcu_node_entry, head: rnp->gp_tasks);
239 break;
240
241 default:
242
243 /* Yet another exercise in excessive paranoia. */
244 WARN_ON_ONCE(1);
245 break;
246 }
247
248 /*
249 * We have now queued the task. If it was the first one to
250 * block either grace period, update the ->gp_tasks and/or
251 * ->exp_tasks pointers, respectively, to reference the newly
252 * blocked tasks.
253 */
254 if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) {
255 WRITE_ONCE(rnp->gp_tasks, &t->rcu_node_entry);
256 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq);
257 }
258 if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD))
259 WRITE_ONCE(rnp->exp_tasks, &t->rcu_node_entry);
260 WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) !=
261 !(rnp->qsmask & rdp->grpmask));
262 WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) !=
263 !(rnp->expmask & rdp->grpmask));
264 raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */
265
266 /*
267 * Report the quiescent state for the expedited GP. This expedited
268 * GP should not be able to end until we report, so there should be
269 * no need to check for a subsequent expedited GP. (Though we are
270 * still in a quiescent state in any case.)
271 *
272 * Interrupts are disabled, so ->cpu_no_qs.b.exp cannot change.
273 */
274 if (blkd_state & RCU_EXP_BLKD && rdp->cpu_no_qs.b.exp)
275 rcu_report_exp_rdp(rdp);
276 else
277 WARN_ON_ONCE(rdp->cpu_no_qs.b.exp);
278 ASSERT_EXCLUSIVE_WRITER_SCOPED(rdp->cpu_no_qs.b.exp);
279}
280
281/*
282 * Record a preemptible-RCU quiescent state for the specified CPU.
283 * Note that this does not necessarily mean that the task currently running
284 * on the CPU is in a quiescent state: Instead, it means that the current
285 * grace period need not wait on any RCU read-side critical section that
286 * starts later on this CPU. It also means that if the current task is
287 * in an RCU read-side critical section, it has already added itself to
288 * some leaf rcu_node structure's ->blkd_tasks list. In addition to the
289 * current task, there might be any number of other tasks blocked while
290 * in an RCU read-side critical section.
291 *
292 * Unlike non-preemptible-RCU, quiescent state reports for expedited
293 * grace periods are handled separately via deferred quiescent states
294 * and context switch events.
295 *
296 * Callers to this function must disable preemption.
297 */
298static void rcu_qs(void)
299{
300 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n");
301 if (__this_cpu_read(rcu_data.cpu_no_qs.b.norm)) {
302 trace_rcu_grace_period(TPS("rcu_preempt"),
303 __this_cpu_read(rcu_data.gp_seq),
304 TPS("cpuqs"));
305 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
306 barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */
307 WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false);
308 }
309}
310
311/*
312 * We have entered the scheduler, and the current task might soon be
313 * context-switched away from. If this task is in an RCU read-side
314 * critical section, we will no longer be able to rely on the CPU to
315 * record that fact, so we enqueue the task on the blkd_tasks list.
316 * The task will dequeue itself when it exits the outermost enclosing
317 * RCU read-side critical section. Therefore, the current grace period
318 * cannot be permitted to complete until the blkd_tasks list entries
319 * predating the current grace period drain, in other words, until
320 * rnp->gp_tasks becomes NULL.
321 *
322 * Caller must disable interrupts.
323 */
324void rcu_note_context_switch(bool preempt)
325{
326 struct task_struct *t = current;
327 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
328 struct rcu_node *rnp;
329
330 trace_rcu_utilization(TPS("Start context switch"));
331 lockdep_assert_irqs_disabled();
332 WARN_ONCE(!preempt && rcu_preempt_depth() > 0, "Voluntary context switch within RCU read-side critical section!");
333 if (rcu_preempt_depth() > 0 &&
334 !t->rcu_read_unlock_special.b.blocked) {
335
336 /* Possibly blocking in an RCU read-side critical section. */
337 rnp = rdp->mynode;
338 raw_spin_lock_rcu_node(rnp);
339 t->rcu_read_unlock_special.b.blocked = true;
340 t->rcu_blocked_node = rnp;
341
342 /*
343 * Verify the CPU's sanity, trace the preemption, and
344 * then queue the task as required based on the states
345 * of any ongoing and expedited grace periods.
346 */
347 WARN_ON_ONCE(!rcu_rdp_cpu_online(rdp));
348 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
349 trace_rcu_preempt_task(rcuname: rcu_state.name,
350 pid: t->pid,
351 gp_seq: (rnp->qsmask & rdp->grpmask)
352 ? rnp->gp_seq
353 : rcu_seq_snap(sp: &rnp->gp_seq));
354 rcu_preempt_ctxt_queue(rnp, rdp);
355 } else {
356 rcu_preempt_deferred_qs(t);
357 }
358
359 /*
360 * Either we were not in an RCU read-side critical section to
361 * begin with, or we have now recorded that critical section
362 * globally. Either way, we can now note a quiescent state
363 * for this CPU. Again, if we were in an RCU read-side critical
364 * section, and if that critical section was blocking the current
365 * grace period, then the fact that the task has been enqueued
366 * means that we continue to block the current grace period.
367 */
368 rcu_qs();
369 if (rdp->cpu_no_qs.b.exp)
370 rcu_report_exp_rdp(rdp);
371 rcu_tasks_qs(current, preempt);
372 trace_rcu_utilization(TPS("End context switch"));
373}
374EXPORT_SYMBOL_GPL(rcu_note_context_switch);
375
376/*
377 * Check for preempted RCU readers blocking the current grace period
378 * for the specified rcu_node structure. If the caller needs a reliable
379 * answer, it must hold the rcu_node's ->lock.
380 */
381static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
382{
383 return READ_ONCE(rnp->gp_tasks) != NULL;
384}
385
386/* limit value for ->rcu_read_lock_nesting. */
387#define RCU_NEST_PMAX (INT_MAX / 2)
388
389static void rcu_preempt_read_enter(void)
390{
391 WRITE_ONCE(current->rcu_read_lock_nesting, READ_ONCE(current->rcu_read_lock_nesting) + 1);
392}
393
394static int rcu_preempt_read_exit(void)
395{
396 int ret = READ_ONCE(current->rcu_read_lock_nesting) - 1;
397
398 WRITE_ONCE(current->rcu_read_lock_nesting, ret);
399 return ret;
400}
401
402static void rcu_preempt_depth_set(int val)
403{
404 WRITE_ONCE(current->rcu_read_lock_nesting, val);
405}
406
407/*
408 * Preemptible RCU implementation for rcu_read_lock().
409 * Just increment ->rcu_read_lock_nesting, shared state will be updated
410 * if we block.
411 */
412void __rcu_read_lock(void)
413{
414 rcu_preempt_read_enter();
415 if (IS_ENABLED(CONFIG_PROVE_LOCKING))
416 WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX);
417 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread)
418 WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true);
419 barrier(); /* critical section after entry code. */
420}
421EXPORT_SYMBOL_GPL(__rcu_read_lock);
422
423/*
424 * Preemptible RCU implementation for rcu_read_unlock().
425 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
426 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
427 * invoke rcu_read_unlock_special() to clean up after a context switch
428 * in an RCU read-side critical section and other special cases.
429 */
430void __rcu_read_unlock(void)
431{
432 struct task_struct *t = current;
433
434 barrier(); // critical section before exit code.
435 if (rcu_preempt_read_exit() == 0) {
436 barrier(); // critical-section exit before .s check.
437 if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s)))
438 rcu_read_unlock_special(t);
439 }
440 if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
441 int rrln = rcu_preempt_depth();
442
443 WARN_ON_ONCE(rrln < 0 || rrln > RCU_NEST_PMAX);
444 }
445}
446EXPORT_SYMBOL_GPL(__rcu_read_unlock);
447
448/*
449 * Advance a ->blkd_tasks-list pointer to the next entry, instead
450 * returning NULL if at the end of the list.
451 */
452static struct list_head *rcu_next_node_entry(struct task_struct *t,
453 struct rcu_node *rnp)
454{
455 struct list_head *np;
456
457 np = t->rcu_node_entry.next;
458 if (np == &rnp->blkd_tasks)
459 np = NULL;
460 return np;
461}
462
463/*
464 * Return true if the specified rcu_node structure has tasks that were
465 * preempted within an RCU read-side critical section.
466 */
467static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
468{
469 return !list_empty(head: &rnp->blkd_tasks);
470}
471
472/*
473 * Report deferred quiescent states. The deferral time can
474 * be quite short, for example, in the case of the call from
475 * rcu_read_unlock_special().
476 */
477static notrace void
478rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags)
479{
480 bool empty_exp;
481 bool empty_norm;
482 bool empty_exp_now;
483 struct list_head *np;
484 bool drop_boost_mutex = false;
485 struct rcu_data *rdp;
486 struct rcu_node *rnp;
487 union rcu_special special;
488
489 rdp = this_cpu_ptr(&rcu_data);
490 if (rdp->defer_qs_iw_pending == DEFER_QS_PENDING)
491 rdp->defer_qs_iw_pending = DEFER_QS_IDLE;
492
493 /*
494 * If RCU core is waiting for this CPU to exit its critical section,
495 * report the fact that it has exited. Because irqs are disabled,
496 * t->rcu_read_unlock_special cannot change.
497 */
498 special = t->rcu_read_unlock_special;
499 if (!special.s && !rdp->cpu_no_qs.b.exp) {
500 local_irq_restore(flags);
501 return;
502 }
503 t->rcu_read_unlock_special.s = 0;
504 if (special.b.need_qs) {
505 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) {
506 rdp->cpu_no_qs.b.norm = false;
507 rcu_report_qs_rdp(rdp);
508 udelay(usec: rcu_unlock_delay);
509 } else {
510 rcu_qs();
511 }
512 }
513
514 /*
515 * Respond to a request by an expedited grace period for a
516 * quiescent state from this CPU. Note that requests from
517 * tasks are handled when removing the task from the
518 * blocked-tasks list below.
519 */
520 if (rdp->cpu_no_qs.b.exp)
521 rcu_report_exp_rdp(rdp);
522
523 /* Clean up if blocked during RCU read-side critical section. */
524 if (special.b.blocked) {
525
526 /*
527 * Remove this task from the list it blocked on. The task
528 * now remains queued on the rcu_node corresponding to the
529 * CPU it first blocked on, so there is no longer any need
530 * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia.
531 */
532 rnp = t->rcu_blocked_node;
533 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
534 WARN_ON_ONCE(rnp != t->rcu_blocked_node);
535 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
536 empty_norm = !rcu_preempt_blocked_readers_cgp(rnp);
537 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq &&
538 (!empty_norm || rnp->qsmask));
539 empty_exp = sync_rcu_exp_done(rnp);
540 np = rcu_next_node_entry(t, rnp);
541 list_del_init(entry: &t->rcu_node_entry);
542 t->rcu_blocked_node = NULL;
543 trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
544 gp_seq: rnp->gp_seq, pid: t->pid);
545 if (&t->rcu_node_entry == rnp->gp_tasks)
546 WRITE_ONCE(rnp->gp_tasks, np);
547 if (&t->rcu_node_entry == rnp->exp_tasks)
548 WRITE_ONCE(rnp->exp_tasks, np);
549 if (IS_ENABLED(CONFIG_RCU_BOOST)) {
550 /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
551 drop_boost_mutex = rt_mutex_owner(lock: &rnp->boost_mtx.rtmutex) == t;
552 if (&t->rcu_node_entry == rnp->boost_tasks)
553 WRITE_ONCE(rnp->boost_tasks, np);
554 }
555
556 /*
557 * If this was the last task on the current list, and if
558 * we aren't waiting on any CPUs, report the quiescent state.
559 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
560 * so we must take a snapshot of the expedited state.
561 */
562 empty_exp_now = sync_rcu_exp_done(rnp);
563 if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) {
564 trace_rcu_quiescent_state_report(TPS("preempt_rcu"),
565 gp_seq: rnp->gp_seq,
566 mask: 0, qsmask: rnp->qsmask,
567 level: rnp->level,
568 grplo: rnp->grplo,
569 grphi: rnp->grphi,
570 gp_tasks: !!rnp->gp_tasks);
571 rcu_report_unblock_qs_rnp(rnp, flags);
572 } else {
573 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
574 }
575
576 /*
577 * If this was the last task on the expedited lists,
578 * then we need to report up the rcu_node hierarchy.
579 */
580 if (!empty_exp && empty_exp_now)
581 rcu_report_exp_rnp(rnp, wake: true);
582
583 /* Unboost if we were boosted. */
584 if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex)
585 rt_mutex_futex_unlock(lock: &rnp->boost_mtx.rtmutex);
586 } else {
587 local_irq_restore(flags);
588 }
589}
590
591/*
592 * Is a deferred quiescent-state pending, and are we also not in
593 * an RCU read-side critical section? It is the caller's responsibility
594 * to ensure it is otherwise safe to report any deferred quiescent
595 * states. The reason for this is that it is safe to report a
596 * quiescent state during context switch even though preemption
597 * is disabled. This function cannot be expected to understand these
598 * nuances, so the caller must handle them.
599 */
600static notrace bool rcu_preempt_need_deferred_qs(struct task_struct *t)
601{
602 return (__this_cpu_read(rcu_data.cpu_no_qs.b.exp) ||
603 READ_ONCE(t->rcu_read_unlock_special.s)) &&
604 rcu_preempt_depth() == 0;
605}
606
607/*
608 * Report a deferred quiescent state if needed and safe to do so.
609 * As with rcu_preempt_need_deferred_qs(), "safe" involves only
610 * not being in an RCU read-side critical section. The caller must
611 * evaluate safety in terms of interrupt, softirq, and preemption
612 * disabling.
613 */
614notrace void rcu_preempt_deferred_qs(struct task_struct *t)
615{
616 unsigned long flags;
617
618 if (!rcu_preempt_need_deferred_qs(t))
619 return;
620 local_irq_save(flags);
621 rcu_preempt_deferred_qs_irqrestore(t, flags);
622}
623
624/*
625 * Minimal handler to give the scheduler a chance to re-evaluate.
626 */
627static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp)
628{
629 struct rcu_data *rdp;
630
631 lockdep_assert_irqs_disabled();
632 rdp = container_of(iwp, struct rcu_data, defer_qs_iw);
633
634 /*
635 * If the IRQ work handler happens to run in the middle of RCU read-side
636 * critical section, it could be ineffective in getting the scheduler's
637 * attention to report a deferred quiescent state (the whole point of the
638 * IRQ work). For this reason, requeue the IRQ work.
639 *
640 * Basically, we want to avoid following situation:
641 * 1. rcu_read_unlock() queues IRQ work (state -> DEFER_QS_PENDING)
642 * 2. CPU enters new rcu_read_lock()
643 * 3. IRQ work runs but cannot report QS due to rcu_preempt_depth() > 0
644 * 4. rcu_read_unlock() does not re-queue work (state still PENDING)
645 * 5. Deferred QS reporting does not happen.
646 */
647 if (rcu_preempt_depth() > 0)
648 WRITE_ONCE(rdp->defer_qs_iw_pending, DEFER_QS_IDLE);
649}
650
651/*
652 * Check if expedited grace period processing during unlock is needed.
653 *
654 * This function determines whether expedited handling is required based on:
655 * 1. Task blocking an expedited grace period (based on a heuristic, could be
656 * false-positive, see below.)
657 * 2. CPU participating in an expedited grace period
658 * 3. Strict grace period mode requiring expedited handling
659 * 4. RCU priority deboosting needs when interrupts were disabled
660 *
661 * @t: The task being checked
662 * @rdp: The per-CPU RCU data
663 * @rnp: The RCU node for this CPU
664 * @irqs_were_disabled: Whether interrupts were disabled before rcu_read_unlock()
665 *
666 * Returns true if expedited processing of the rcu_read_unlock() is needed.
667 */
668static bool rcu_unlock_needs_exp_handling(struct task_struct *t,
669 struct rcu_data *rdp,
670 struct rcu_node *rnp,
671 bool irqs_were_disabled)
672{
673 /*
674 * Check if this task is blocking an expedited grace period. If the
675 * task was preempted within an RCU read-side critical section and is
676 * on the expedited grace period blockers list (exp_tasks), we need
677 * expedited handling to unblock the expedited GP. This is not an exact
678 * check because 't' might not be on the exp_tasks list at all - its
679 * just a fast heuristic that can be false-positive sometimes.
680 */
681 if (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks))
682 return true;
683
684 /*
685 * Check if this CPU is participating in an expedited grace period.
686 * The expmask bitmap tracks which CPUs need to check in for the
687 * current expedited GP. If our CPU's bit is set, we need expedited
688 * handling to help complete the expedited GP.
689 */
690 if (rdp->grpmask & READ_ONCE(rnp->expmask))
691 return true;
692
693 /*
694 * In CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, all grace periods
695 * are treated as short for testing purposes even if that means
696 * disturbing the system more. Check if either:
697 * - This CPU has not yet reported a quiescent state, or
698 * - This task was preempted within an RCU critical section
699 * In either case, require expedited handling for strict GP mode.
700 */
701 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) &&
702 ((rdp->grpmask & READ_ONCE(rnp->qsmask)) || t->rcu_blocked_node))
703 return true;
704
705 /*
706 * RCU priority boosting case: If a task is subject to RCU priority
707 * boosting and exits an RCU read-side critical section with interrupts
708 * disabled, we need expedited handling to ensure timely deboosting.
709 * Without this, a low-priority task could incorrectly run at high
710 * real-time priority for an extended period degrading real-time
711 * responsiveness. This applies to all CONFIG_RCU_BOOST=y kernels,
712 * not just to PREEMPT_RT.
713 */
714 if (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && t->rcu_blocked_node)
715 return true;
716
717 return false;
718}
719
720/*
721 * Handle special cases during rcu_read_unlock(), such as needing to
722 * notify RCU core processing or task having blocked during the RCU
723 * read-side critical section.
724 */
725static void rcu_read_unlock_special(struct task_struct *t)
726{
727 unsigned long flags;
728 bool irqs_were_disabled;
729 bool preempt_bh_were_disabled =
730 !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK));
731
732 /* NMI handlers cannot block and cannot safely manipulate state. */
733 if (in_nmi())
734 return;
735
736 local_irq_save(flags);
737 irqs_were_disabled = irqs_disabled_flags(flags);
738 if (preempt_bh_were_disabled || irqs_were_disabled) {
739 bool needs_exp; // Expedited handling needed.
740 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
741 struct rcu_node *rnp = rdp->mynode;
742
743 needs_exp = rcu_unlock_needs_exp_handling(t, rdp, rnp, irqs_were_disabled);
744
745 // Need to defer quiescent state until everything is enabled.
746 if (use_softirq && (in_hardirq() || (needs_exp && !irqs_were_disabled))) {
747 // Using softirq, safe to awaken, and either the
748 // wakeup is free or there is either an expedited
749 // GP in flight or a potential need to deboost.
750 raise_softirq_irqoff(nr: RCU_SOFTIRQ);
751 } else {
752 // Enabling BH or preempt does reschedule, so...
753 // Also if no expediting and no possible deboosting,
754 // slow is OK. Plus nohz_full CPUs eventually get
755 // tick enabled.
756 set_tsk_need_resched(current);
757 set_preempt_need_resched();
758 if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled &&
759 needs_exp && rdp->defer_qs_iw_pending != DEFER_QS_PENDING &&
760 cpu_online(cpu: rdp->cpu)) {
761 // Get scheduler to re-evaluate and call hooks.
762 // If !IRQ_WORK, FQS scan will eventually IPI.
763 rdp->defer_qs_iw_pending = DEFER_QS_PENDING;
764 irq_work_queue_on(work: &rdp->defer_qs_iw, cpu: rdp->cpu);
765 }
766 }
767 local_irq_restore(flags);
768 return;
769 }
770 rcu_preempt_deferred_qs_irqrestore(t, flags);
771}
772
773/*
774 * Check that the list of blocked tasks for the newly completed grace
775 * period is in fact empty. It is a serious bug to complete a grace
776 * period that still has RCU readers blocked! This function must be
777 * invoked -before- updating this rnp's ->gp_seq.
778 *
779 * Also, if there are blocked tasks on the list, they automatically
780 * block the newly created grace period, so set up ->gp_tasks accordingly.
781 */
782static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
783{
784 struct task_struct *t;
785
786 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
787 raw_lockdep_assert_held_rcu_node(rnp);
788 if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
789 dump_blkd_tasks(rnp, ncheck: 10);
790 if (rcu_preempt_has_tasks(rnp) &&
791 (rnp->qsmaskinit || rnp->wait_blkd_tasks)) {
792 WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next);
793 t = container_of(rnp->gp_tasks, struct task_struct,
794 rcu_node_entry);
795 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
796 gp_seq: rnp->gp_seq, pid: t->pid);
797 }
798 WARN_ON_ONCE(rnp->qsmask);
799}
800
801/*
802 * Check for a quiescent state from the current CPU, including voluntary
803 * context switches for Tasks RCU. When a task blocks, the task is
804 * recorded in the corresponding CPU's rcu_node structure, which is checked
805 * elsewhere, hence this function need only check for quiescent states
806 * related to the current CPU, not to those related to tasks.
807 */
808static void rcu_flavor_sched_clock_irq(int user)
809{
810 struct task_struct *t = current;
811
812 lockdep_assert_irqs_disabled();
813 if (rcu_preempt_depth() > 0 ||
814 (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) {
815 /* No QS, force context switch if deferred. */
816 if (rcu_preempt_need_deferred_qs(t)) {
817 set_tsk_need_resched(t);
818 set_preempt_need_resched();
819 }
820 } else if (rcu_preempt_need_deferred_qs(t)) {
821 rcu_preempt_deferred_qs(t); /* Report deferred QS. */
822 return;
823 } else if (!WARN_ON_ONCE(rcu_preempt_depth())) {
824 rcu_qs(); /* Report immediate QS. */
825 return;
826 }
827
828 /* If GP is oldish, ask for help from rcu_read_unlock_special(). */
829 if (rcu_preempt_depth() > 0 &&
830 __this_cpu_read(rcu_data.core_needs_qs) &&
831 __this_cpu_read(rcu_data.cpu_no_qs.b.norm) &&
832 !t->rcu_read_unlock_special.b.need_qs &&
833 time_after(jiffies, rcu_state.gp_start + HZ))
834 t->rcu_read_unlock_special.b.need_qs = true;
835}
836
837/*
838 * Check for a task exiting while in a preemptible-RCU read-side
839 * critical section, clean up if so. No need to issue warnings, as
840 * debug_check_no_locks_held() already does this if lockdep is enabled.
841 * Besides, if this function does anything other than just immediately
842 * return, there was a bug of some sort. Spewing warnings from this
843 * function is like as not to simply obscure important prior warnings.
844 */
845void exit_rcu(void)
846{
847 struct task_struct *t = current;
848
849 if (unlikely(!list_empty(&current->rcu_node_entry))) {
850 rcu_preempt_depth_set(val: 1);
851 barrier();
852 WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true);
853 } else if (unlikely(rcu_preempt_depth())) {
854 rcu_preempt_depth_set(val: 1);
855 } else {
856 return;
857 }
858 __rcu_read_unlock();
859 rcu_preempt_deferred_qs(current);
860}
861
862/*
863 * Dump the blocked-tasks state, but limit the list dump to the
864 * specified number of elements.
865 */
866static void
867dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
868{
869 int cpu;
870 int i;
871 struct list_head *lhp;
872 struct rcu_data *rdp;
873 struct rcu_node *rnp1;
874
875 raw_lockdep_assert_held_rcu_node(rnp);
876 pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
877 __func__, rnp->grplo, rnp->grphi, rnp->level,
878 (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs);
879 for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
880 pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n",
881 __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext);
882 pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n",
883 __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks),
884 READ_ONCE(rnp->exp_tasks));
885 pr_info("%s: ->blkd_tasks", __func__);
886 i = 0;
887 list_for_each(lhp, &rnp->blkd_tasks) {
888 pr_cont(" %p", lhp);
889 if (++i >= ncheck)
890 break;
891 }
892 pr_cont("\n");
893 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) {
894 rdp = per_cpu_ptr(&rcu_data, cpu);
895 pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n",
896 cpu, ".o"[rcu_rdp_cpu_online(rdp)],
897 (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_state,
898 (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_state);
899 }
900}
901
902static void rcu_preempt_deferred_qs_init(struct rcu_data *rdp)
903{
904 rdp->defer_qs_iw = IRQ_WORK_INIT_HARD(rcu_preempt_deferred_qs_handler);
905}
906#else /* #ifdef CONFIG_PREEMPT_RCU */
907
908/*
909 * If strict grace periods are enabled, and if the calling
910 * __rcu_read_unlock() marks the beginning of a quiescent state, immediately
911 * report that quiescent state and, if requested, spin for a bit.
912 */
913void rcu_read_unlock_strict(void)
914{
915 struct rcu_data *rdp;
916
917 if (irqs_disabled() || in_atomic_preempt_off() || !rcu_state.gp_kthread)
918 return;
919
920 /*
921 * rcu_report_qs_rdp() can only be invoked with a stable rdp and
922 * from the local CPU.
923 *
924 * The in_atomic_preempt_off() check ensures that we come here holding
925 * the last preempt_count (which will get dropped once we return to
926 * __rcu_read_unlock().
927 */
928 rdp = this_cpu_ptr(&rcu_data);
929 rdp->cpu_no_qs.b.norm = false;
930 rcu_report_qs_rdp(rdp);
931 udelay(rcu_unlock_delay);
932}
933EXPORT_SYMBOL_GPL(rcu_read_unlock_strict);
934
935/*
936 * Tell them what RCU they are running.
937 */
938static void __init rcu_bootup_announce(void)
939{
940 pr_info("Hierarchical RCU implementation.\n");
941 rcu_bootup_announce_oddness();
942}
943
944/*
945 * Note a quiescent state for PREEMPTION=n. Because we do not need to know
946 * how many quiescent states passed, just if there was at least one since
947 * the start of the grace period, this just sets a flag. The caller must
948 * have disabled preemption.
949 */
950static void rcu_qs(void)
951{
952 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!");
953 if (!__this_cpu_read(rcu_data.cpu_no_qs.s))
954 return;
955 trace_rcu_grace_period(TPS("rcu_sched"),
956 __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs"));
957 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
958 if (__this_cpu_read(rcu_data.cpu_no_qs.b.exp))
959 rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
960}
961
962/*
963 * Register an urgently needed quiescent state. If there is an
964 * emergency, invoke rcu_momentary_eqs() to do a heavy-weight
965 * dyntick-idle quiescent state visible to other CPUs, which will in
966 * some cases serve for expedited as well as normal grace periods.
967 * Either way, register a lightweight quiescent state.
968 */
969void rcu_all_qs(void)
970{
971 unsigned long flags;
972
973 if (!raw_cpu_read(rcu_data.rcu_urgent_qs))
974 return;
975 preempt_disable(); // For CONFIG_PREEMPT_COUNT=y kernels
976 /* Load rcu_urgent_qs before other flags. */
977 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
978 preempt_enable();
979 return;
980 }
981 this_cpu_write(rcu_data.rcu_urgent_qs, false);
982 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) {
983 local_irq_save(flags);
984 rcu_momentary_eqs();
985 local_irq_restore(flags);
986 }
987 rcu_qs();
988 preempt_enable();
989}
990EXPORT_SYMBOL_GPL(rcu_all_qs);
991
992/*
993 * Note a PREEMPTION=n context switch. The caller must have disabled interrupts.
994 */
995void rcu_note_context_switch(bool preempt)
996{
997 trace_rcu_utilization(TPS("Start context switch"));
998 rcu_qs();
999 /* Load rcu_urgent_qs before other flags. */
1000 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs)))
1001 goto out;
1002 this_cpu_write(rcu_data.rcu_urgent_qs, false);
1003 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs)))
1004 rcu_momentary_eqs();
1005out:
1006 rcu_tasks_qs(current, preempt);
1007 trace_rcu_utilization(TPS("End context switch"));
1008}
1009EXPORT_SYMBOL_GPL(rcu_note_context_switch);
1010
1011/*
1012 * Because preemptible RCU does not exist, there are never any preempted
1013 * RCU readers.
1014 */
1015static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
1016{
1017 return 0;
1018}
1019
1020/*
1021 * Because there is no preemptible RCU, there can be no readers blocked.
1022 */
1023static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
1024{
1025 return false;
1026}
1027
1028/*
1029 * Because there is no preemptible RCU, there can be no deferred quiescent
1030 * states.
1031 */
1032static notrace bool rcu_preempt_need_deferred_qs(struct task_struct *t)
1033{
1034 return false;
1035}
1036
1037// Except that we do need to respond to a request by an expedited
1038// grace period for a quiescent state from this CPU. Note that in
1039// non-preemptible kernels, there can be no context switches within RCU
1040// read-side critical sections, which in turn means that the leaf rcu_node
1041// structure's blocked-tasks list is always empty. is therefore no need to
1042// actually check it. Instead, a quiescent state from this CPU suffices,
1043// and this function is only called from such a quiescent state.
1044notrace void rcu_preempt_deferred_qs(struct task_struct *t)
1045{
1046 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
1047
1048 if (READ_ONCE(rdp->cpu_no_qs.b.exp))
1049 rcu_report_exp_rdp(rdp);
1050}
1051
1052/*
1053 * Because there is no preemptible RCU, there can be no readers blocked,
1054 * so there is no need to check for blocked tasks. So check only for
1055 * bogus qsmask values.
1056 */
1057static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
1058{
1059 WARN_ON_ONCE(rnp->qsmask);
1060}
1061
1062/*
1063 * Check to see if this CPU is in a non-context-switch quiescent state,
1064 * namely user mode and idle loop.
1065 */
1066static void rcu_flavor_sched_clock_irq(int user)
1067{
1068 if (user || rcu_is_cpu_rrupt_from_idle() ||
1069 (IS_ENABLED(CONFIG_PREEMPT_COUNT) &&
1070 (preempt_count() == HARDIRQ_OFFSET))) {
1071
1072 /*
1073 * Get here if this CPU took its interrupt from user
1074 * mode, from the idle loop without this being a nested
1075 * interrupt, or while not holding the task preempt count
1076 * (with PREEMPT_COUNT=y). In this case, the CPU is in a
1077 * quiescent state, so note it.
1078 *
1079 * No memory barrier is required here because rcu_qs()
1080 * references only CPU-local variables that other CPUs
1081 * neither access nor modify, at least not while the
1082 * corresponding CPU is online.
1083 */
1084 rcu_qs();
1085 }
1086}
1087
1088/*
1089 * Because preemptible RCU does not exist, tasks cannot possibly exit
1090 * while in preemptible RCU read-side critical sections.
1091 */
1092void exit_rcu(void)
1093{
1094}
1095
1096/*
1097 * Dump the guaranteed-empty blocked-tasks state. Trust but verify.
1098 */
1099static void
1100dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
1101{
1102 WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks));
1103}
1104
1105static void rcu_preempt_deferred_qs_init(struct rcu_data *rdp) { }
1106
1107#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
1108
1109/*
1110 * If boosting, set rcuc kthreads to realtime priority.
1111 */
1112static void rcu_cpu_kthread_setup(unsigned int cpu)
1113{
1114 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
1115#ifdef CONFIG_RCU_BOOST
1116 struct sched_param sp;
1117
1118 sp.sched_priority = kthread_prio;
1119 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1120#endif /* #ifdef CONFIG_RCU_BOOST */
1121
1122 WRITE_ONCE(rdp->rcuc_activity, jiffies);
1123}
1124
1125static bool rcu_is_callbacks_nocb_kthread(struct rcu_data *rdp)
1126{
1127#ifdef CONFIG_RCU_NOCB_CPU
1128 return rdp->nocb_cb_kthread == current;
1129#else
1130 return false;
1131#endif
1132}
1133
1134/*
1135 * Is the current CPU running the RCU-callbacks kthread?
1136 * Caller must have preemption disabled.
1137 */
1138static bool rcu_is_callbacks_kthread(struct rcu_data *rdp)
1139{
1140 return rdp->rcu_cpu_kthread_task == current ||
1141 rcu_is_callbacks_nocb_kthread(rdp);
1142}
1143
1144#ifdef CONFIG_RCU_BOOST
1145
1146/*
1147 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1148 * or ->boost_tasks, advancing the pointer to the next task in the
1149 * ->blkd_tasks list.
1150 *
1151 * Note that irqs must be enabled: boosting the task can block.
1152 * Returns 1 if there are more tasks needing to be boosted.
1153 */
1154static int rcu_boost(struct rcu_node *rnp)
1155{
1156 unsigned long flags;
1157 struct task_struct *t;
1158 struct list_head *tb;
1159
1160 if (READ_ONCE(rnp->exp_tasks) == NULL &&
1161 READ_ONCE(rnp->boost_tasks) == NULL)
1162 return 0; /* Nothing left to boost. */
1163
1164 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1165
1166 /*
1167 * Recheck under the lock: all tasks in need of boosting
1168 * might exit their RCU read-side critical sections on their own.
1169 */
1170 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1171 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1172 return 0;
1173 }
1174
1175 /*
1176 * Preferentially boost tasks blocking expedited grace periods.
1177 * This cannot starve the normal grace periods because a second
1178 * expedited grace period must boost all blocked tasks, including
1179 * those blocking the pre-existing normal grace period.
1180 */
1181 if (rnp->exp_tasks != NULL)
1182 tb = rnp->exp_tasks;
1183 else
1184 tb = rnp->boost_tasks;
1185
1186 /*
1187 * We boost task t by manufacturing an rt_mutex that appears to
1188 * be held by task t. We leave a pointer to that rt_mutex where
1189 * task t can find it, and task t will release the mutex when it
1190 * exits its outermost RCU read-side critical section. Then
1191 * simply acquiring this artificial rt_mutex will boost task
1192 * t's priority. (Thanks to tglx for suggesting this approach!)
1193 *
1194 * Note that task t must acquire rnp->lock to remove itself from
1195 * the ->blkd_tasks list, which it will do from exit() if from
1196 * nowhere else. We therefore are guaranteed that task t will
1197 * stay around at least until we drop rnp->lock. Note that
1198 * rnp->lock also resolves races between our priority boosting
1199 * and task t's exiting its outermost RCU read-side critical
1200 * section.
1201 */
1202 t = container_of(tb, struct task_struct, rcu_node_entry);
1203 rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t);
1204 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1205 /* Lock only for side effect: boosts task t's priority. */
1206 rt_mutex_lock(&rnp->boost_mtx);
1207 rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
1208 rnp->n_boosts++;
1209
1210 return READ_ONCE(rnp->exp_tasks) != NULL ||
1211 READ_ONCE(rnp->boost_tasks) != NULL;
1212}
1213
1214/*
1215 * Priority-boosting kthread, one per leaf rcu_node.
1216 */
1217static int rcu_boost_kthread(void *arg)
1218{
1219 struct rcu_node *rnp = (struct rcu_node *)arg;
1220 int spincnt = 0;
1221 int more2boost;
1222
1223 trace_rcu_utilization(TPS("Start boost kthread@init"));
1224 for (;;) {
1225 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING);
1226 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
1227 rcu_wait(READ_ONCE(rnp->boost_tasks) ||
1228 READ_ONCE(rnp->exp_tasks));
1229 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
1230 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING);
1231 more2boost = rcu_boost(rnp);
1232 if (more2boost)
1233 spincnt++;
1234 else
1235 spincnt = 0;
1236 if (spincnt > 10) {
1237 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING);
1238 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
1239 schedule_timeout_idle(2);
1240 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
1241 spincnt = 0;
1242 }
1243 }
1244 /* NOTREACHED */
1245 trace_rcu_utilization(TPS("End boost kthread@notreached"));
1246 return 0;
1247}
1248
1249/*
1250 * Check to see if it is time to start boosting RCU readers that are
1251 * blocking the current grace period, and, if so, tell the per-rcu_node
1252 * kthread to start boosting them. If there is an expedited grace
1253 * period in progress, it is always time to boost.
1254 *
1255 * The caller must hold rnp->lock, which this function releases.
1256 * The ->boost_kthread_task is immortal, so we don't need to worry
1257 * about it going away.
1258 */
1259static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1260 __releases(rnp->lock)
1261{
1262 raw_lockdep_assert_held_rcu_node(rnp);
1263 if (!rnp->boost_kthread_task ||
1264 (!rcu_preempt_blocked_readers_cgp(rnp) && !rnp->exp_tasks)) {
1265 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1266 return;
1267 }
1268 if (rnp->exp_tasks != NULL ||
1269 (rnp->gp_tasks != NULL &&
1270 rnp->boost_tasks == NULL &&
1271 rnp->qsmask == 0 &&
1272 (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld ||
1273 IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)))) {
1274 if (rnp->exp_tasks == NULL)
1275 WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks);
1276 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1277 rcu_wake_cond(rnp->boost_kthread_task,
1278 READ_ONCE(rnp->boost_kthread_status));
1279 } else {
1280 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1281 }
1282}
1283
1284#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1285
1286/*
1287 * Do priority-boost accounting for the start of a new grace period.
1288 */
1289static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1290{
1291 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1292}
1293
1294/*
1295 * Create an RCU-boost kthread for the specified node if one does not
1296 * already exist. We only create this kthread for preemptible RCU.
1297 */
1298static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1299{
1300 unsigned long flags;
1301 int rnp_index = rnp - rcu_get_root();
1302 struct sched_param sp;
1303 struct task_struct *t;
1304
1305 if (rnp->boost_kthread_task)
1306 return;
1307
1308 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1309 "rcub/%d", rnp_index);
1310 if (WARN_ON_ONCE(IS_ERR(t)))
1311 return;
1312
1313 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1314 rnp->boost_kthread_task = t;
1315 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1316
1317 sp.sched_priority = kthread_prio;
1318 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1319 rcu_thread_affine_rnp(t, rnp);
1320 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1321}
1322
1323#else /* #ifdef CONFIG_RCU_BOOST */
1324
1325static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1326 __releases(rnp->lock)
1327{
1328 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1329}
1330
1331static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1332{
1333}
1334
1335static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1336{
1337}
1338
1339#endif /* #else #ifdef CONFIG_RCU_BOOST */
1340
1341/*
1342 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
1343 * grace-period kthread will do force_quiescent_state() processing?
1344 * The idea is to avoid waking up RCU core processing on such a
1345 * CPU unless the grace period has extended for too long.
1346 *
1347 * This code relies on the fact that all NO_HZ_FULL CPUs are also
1348 * RCU_NOCB_CPU CPUs.
1349 */
1350static bool rcu_nohz_full_cpu(void)
1351{
1352#ifdef CONFIG_NO_HZ_FULL
1353 if (tick_nohz_full_cpu(smp_processor_id()) &&
1354 (!rcu_gp_in_progress() ||
1355 time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ)))
1356 return true;
1357#endif /* #ifdef CONFIG_NO_HZ_FULL */
1358 return false;
1359}
1360
1361/*
1362 * Bind the RCU grace-period kthreads to the housekeeping CPU.
1363 */
1364static void rcu_bind_gp_kthread(void)
1365{
1366 if (!tick_nohz_full_enabled())
1367 return;
1368 housekeeping_affine(current, type: HK_TYPE_RCU);
1369}
1370