projects/linux·blob master · refs

1430 lines · 47.6 KiB

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 <[email protected]>
11 * Paul E. McKenney <[email protected]>
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(&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(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(&t->rcu_node_entry, &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(&t->rcu_node_entry, &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(&t->rcu_node_entry, 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(&t->rcu_node_entry, 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(rcu_state.name,
350 t->pid,
351 (rnp->qsmask & rdp->grpmask)
352 ? rnp->gp_seq
353 : rcu_seq_snap(&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(&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_pending == DEFER_QS_PENDING)
491 rcu_defer_qs_clear(rdp);
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(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(&t->rcu_node_entry);
542 t->rcu_blocked_node = NULL;
543 trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
544 rnp->gp_seq, 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(&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 rnp->gp_seq,
566 0, rnp->qsmask,
567 rnp->level,
568 rnp->grplo,
569 rnp->grphi,
570 !!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, true);
582
583 /* Unboost if we were boosted. */
584 if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex)
585 rt_mutex_futex_unlock(&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 struct rcu_data *rdp;
618
619 if (!rcu_preempt_need_deferred_qs(t)) {
620 /*
621 * If we got here from a softirq/irq_work that fired while
622 * rcu_preempt_depth() > 0, the deferred-QS mechanism has been
623 * consumed without doing any work: rcu_preempt_need_deferred_qs()
624 * just returned false because the task is still in a reader, so
625 * the actual QS report has to wait for the next
626 * rcu_read_unlock().
627 *
628 * Clear ->defer_qs_pending here so the next outer
629 * rcu_read_unlock_special() can re-arm a fresh mechanism (in
630 * particular the irq_work path, which the local_irq_enable()
631 * recovery boundary cannot itself reschedule from).
632 *
633 * Recursion safety: rcu_preempt_depth() > 0 means we are inside
634 * an outer reader, so any inner rcu_read_unlock() reached via
635 * tracing (bpf programs attached to trace points) brings
636 * nesting to outer (> 0), never to 0, so no recursive
637 * raise_softirq_irqoff()/irq_work_queue_on() can be triggered
638 * by this clear.
639 */
640 if (rcu_preempt_depth() > 0) {
641 rdp = this_cpu_ptr(&rcu_data);
642 rcu_defer_qs_clear(rdp);
643 }
644 return;
645 }
646 local_irq_save(flags);
647 rcu_preempt_deferred_qs_irqrestore(t, flags);
648}
649
650/*
651 * Minimal handler to give the scheduler a chance to re-evaluate.
652 */
653static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp)
654{
655 struct rcu_data *rdp;
656
657 lockdep_assert_irqs_disabled();
658 rdp = container_of(iwp, struct rcu_data, defer_qs_iw);
659
660 /*
661 * If the IRQ work handler happens to run in the middle of RCU read-side
662 * critical section, it could be ineffective in getting the scheduler's
663 * attention to report a deferred quiescent state (the whole point of the
664 * IRQ work). For this reason, requeue the IRQ work.
665 *
666 * Basically, we want to avoid following situation:
667 * 1. rcu_read_unlock() queues IRQ work (state -> DEFER_QS_PENDING)
668 * 2. CPU enters new rcu_read_lock()
669 * 3. IRQ work runs but cannot report QS due to rcu_preempt_depth() > 0
670 * 4. rcu_read_unlock() does not re-queue work (state still PENDING)
671 * 5. Deferred QS reporting does not happen.
672 */
673 if (rcu_preempt_depth() > 0)
674 rcu_defer_qs_clear(rdp);
675}
676
677/*
678 * Check if expedited grace period processing during unlock is needed.
679 *
680 * This function determines whether expedited handling is required based on:
681 * 1. Task blocking an expedited grace period (based on a heuristic, could be
682 * false-positive, see below.)
683 * 2. CPU participating in an expedited grace period
684 * 3. Strict grace period mode requiring expedited handling
685 * 4. RCU priority deboosting needs when interrupts were disabled
686 *
687 * @t: The task being checked
688 * @rdp: The per-CPU RCU data
689 * @rnp: The RCU node for this CPU
690 * @irqs_were_disabled: Whether interrupts were disabled before rcu_read_unlock()
691 *
692 * Returns true if expedited processing of the rcu_read_unlock() is needed.
693 */
694static bool rcu_unlock_needs_exp_handling(struct task_struct *t,
695 struct rcu_data *rdp,
696 struct rcu_node *rnp,
697 bool irqs_were_disabled)
698{
699 /*
700 * Check if this task is blocking an expedited grace period. If the
701 * task was preempted within an RCU read-side critical section and is
702 * on the expedited grace period blockers list (exp_tasks), we need
703 * expedited handling to unblock the expedited GP. This is not an exact
704 * check because 't' might not be on the exp_tasks list at all - its
705 * just a fast heuristic that can be false-positive sometimes.
706 */
707 if (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks))
708 return true;
709
710 /*
711 * Check if this CPU is participating in an expedited grace period.
712 * The expmask bitmap tracks which CPUs need to check in for the
713 * current expedited GP. If our CPU's bit is set, we need expedited
714 * handling to help complete the expedited GP.
715 */
716 if (rdp->grpmask & READ_ONCE(rnp->expmask))
717 return true;
718
719 /*
720 * In CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, all grace periods
721 * are treated as short for testing purposes even if that means
722 * disturbing the system more. Check if either:
723 * - This CPU has not yet reported a quiescent state, or
724 * - This task was preempted within an RCU critical section
725 * In either case, require expedited handling for strict GP mode.
726 */
727 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) &&
728 ((rdp->grpmask & READ_ONCE(rnp->qsmask)) || t->rcu_blocked_node))
729 return true;
730
731 /*
732 * RCU priority boosting case: If a task is subject to RCU priority
733 * boosting and exits an RCU read-side critical section with interrupts
734 * disabled, we need expedited handling to ensure timely deboosting.
735 * Without this, a low-priority task could incorrectly run at high
736 * real-time priority for an extended period degrading real-time
737 * responsiveness. This applies to all CONFIG_RCU_BOOST=y kernels,
738 * not just to PREEMPT_RT.
739 */
740 if (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && t->rcu_blocked_node)
741 return true;
742
743 return false;
744}
745
746/*
747 * Handle special cases during rcu_read_unlock(), such as needing to
748 * notify RCU core processing or task having blocked during the RCU
749 * read-side critical section.
750 */
751static void rcu_read_unlock_special(struct task_struct *t)
752{
753 unsigned long flags;
754 bool irqs_were_disabled;
755 bool preempt_bh_were_disabled =
756 !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK));
757
758 /* NMI handlers cannot block and cannot safely manipulate state. */
759 if (in_nmi())
760 return;
761
762 local_irq_save(flags);
763 irqs_were_disabled = irqs_disabled_flags(flags);
764 if (preempt_bh_were_disabled || irqs_were_disabled) {
765 bool needs_exp; // Expedited handling needed.
766 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
767 struct rcu_node *rnp = rdp->mynode;
768
769 needs_exp = rcu_unlock_needs_exp_handling(t, rdp, rnp, irqs_were_disabled);
770
771 // Need to defer quiescent state until everything is enabled.
772 if (use_softirq && (in_hardirq() || (needs_exp && !irqs_were_disabled))) {
773 // Using softirq, safe to awaken, and either the
774 // wakeup is free or there is either an expedited
775 // GP in flight or a potential need to deboost.
776 if (rdp->defer_qs_pending != DEFER_QS_PENDING) {
777 rdp->defer_qs_pending = DEFER_QS_PENDING;
778 raise_softirq_irqoff(RCU_SOFTIRQ);
779 }
780 } else {
781 // Enabling BH or preempt does reschedule, so...
782 // Also if no expediting and no possible deboosting,
783 // slow is OK. Plus nohz_full CPUs eventually get
784 // tick enabled.
785 set_need_resched_current();
786 if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled &&
787 needs_exp && rdp->defer_qs_pending != DEFER_QS_PENDING &&
788 cpu_online(rdp->cpu)) {
789 // Get scheduler to re-evaluate and call hooks.
790 // If !IRQ_WORK, FQS scan will eventually IPI.
791 rdp->defer_qs_pending = DEFER_QS_PENDING;
792 irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu);
793 }
794 }
795 local_irq_restore(flags);
796 return;
797 }
798 rcu_preempt_deferred_qs_irqrestore(t, flags);
799}
800
801/*
802 * Check that the list of blocked tasks for the newly completed grace
803 * period is in fact empty. It is a serious bug to complete a grace
804 * period that still has RCU readers blocked! This function must be
805 * invoked -before- updating this rnp's ->gp_seq.
806 *
807 * Also, if there are blocked tasks on the list, they automatically
808 * block the newly created grace period, so set up ->gp_tasks accordingly.
809 */
810static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
811{
812 struct task_struct *t;
813
814 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
815 raw_lockdep_assert_held_rcu_node(rnp);
816 if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
817 dump_blkd_tasks(rnp, 10);
818 if (rcu_preempt_has_tasks(rnp) &&
819 (rnp->qsmaskinit || rnp->wait_blkd_tasks)) {
820 WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next);
821 t = container_of(rnp->gp_tasks, struct task_struct,
822 rcu_node_entry);
823 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
824 rnp->gp_seq, t->pid);
825 }
826 WARN_ON_ONCE(rnp->qsmask);
827}
828
829/*
830 * Check for a quiescent state from the current CPU, including voluntary
831 * context switches for Tasks RCU. When a task blocks, the task is
832 * recorded in the corresponding CPU's rcu_node structure, which is checked
833 * elsewhere, hence this function need only check for quiescent states
834 * related to the current CPU, not to those related to tasks.
835 */
836static void rcu_flavor_sched_clock_irq(int user)
837{
838 struct task_struct *t = current;
839
840 lockdep_assert_irqs_disabled();
841 if (rcu_preempt_depth() > 0 ||
842 (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) {
843 /* No QS, force context switch if deferred. */
844 if (rcu_preempt_need_deferred_qs(t))
845 set_need_resched_current();
846 } else if (rcu_preempt_need_deferred_qs(t)) {
847 rcu_preempt_deferred_qs(t); /* Report deferred QS. */
848 return;
849 } else if (!WARN_ON_ONCE(rcu_preempt_depth())) {
850 rcu_qs(); /* Report immediate QS. */
851 return;
852 }
853
854 /* If GP is oldish, ask for help from rcu_read_unlock_special(). */
855 if (rcu_preempt_depth() > 0 &&
856 __this_cpu_read(rcu_data.core_needs_qs) &&
857 __this_cpu_read(rcu_data.cpu_no_qs.b.norm) &&
858 !t->rcu_read_unlock_special.b.need_qs &&
859 time_after(jiffies, rcu_state.gp_start + HZ))
860 t->rcu_read_unlock_special.b.need_qs = true;
861}
862
863/*
864 * Check for a task exiting while in a preemptible-RCU read-side
865 * critical section, clean up if so. No need to issue warnings, as
866 * debug_check_no_locks_held() already does this if lockdep is enabled.
867 * Besides, if this function does anything other than just immediately
868 * return, there was a bug of some sort. Spewing warnings from this
869 * function is like as not to simply obscure important prior warnings.
870 */
871void exit_rcu(void)
872{
873 struct task_struct *t = current;
874
875 if (unlikely(!list_empty(&current->rcu_node_entry))) {
876 rcu_preempt_depth_set(1);
877 barrier();
878 WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true);
879 } else if (unlikely(rcu_preempt_depth())) {
880 rcu_preempt_depth_set(1);
881 } else {
882 return;
883 }
884 __rcu_read_unlock();
885 rcu_preempt_deferred_qs(current);
886}
887
888/*
889 * Dump the blocked-tasks state, but limit the list dump to the
890 * specified number of elements.
891 */
892static void
893dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
894{
895 int cpu;
896 int i;
897 struct list_head *lhp;
898 struct rcu_data *rdp;
899 struct rcu_node *rnp1;
900
901 raw_lockdep_assert_held_rcu_node(rnp);
902 pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
903 __func__, rnp->grplo, rnp->grphi, rnp->level,
904 (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs);
905 for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
906 pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n",
907 __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext);
908 pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n",
909 __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks),
910 READ_ONCE(rnp->exp_tasks));
911 pr_info("%s: ->blkd_tasks", __func__);
912 i = 0;
913 list_for_each(lhp, &rnp->blkd_tasks) {
914 pr_cont(" %p", lhp);
915 if (++i >= ncheck)
916 break;
917 }
918 pr_cont("\n");
919 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) {
920 rdp = per_cpu_ptr(&rcu_data, cpu);
921 pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n",
922 cpu, ".o"[rcu_rdp_cpu_online(rdp)],
923 (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_state,
924 (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_state);
925 }
926}
927
928static void rcu_preempt_deferred_qs_init(struct rcu_data *rdp)
929{
930 rdp->defer_qs_iw = IRQ_WORK_INIT_HARD(rcu_preempt_deferred_qs_handler);
931}
932#else /* #ifdef CONFIG_PREEMPT_RCU */
933
934/*
935 * If strict grace periods are enabled, and if the calling
936 * __rcu_read_unlock() marks the beginning of a quiescent state, immediately
937 * report that quiescent state and, if requested, spin for a bit.
938 */
939void rcu_read_unlock_strict(void)
940{
941 struct rcu_data *rdp;
942
943 if (irqs_disabled() || in_atomic_preempt_off() || !rcu_state.gp_kthread)
944 return;
945
946 /*
947 * rcu_report_qs_rdp() can only be invoked with a stable rdp and
948 * from the local CPU.
949 *
950 * The in_atomic_preempt_off() check ensures that we come here holding
951 * the last preempt_count (which will get dropped once we return to
952 * __rcu_read_unlock()).
953 */
954 rdp = this_cpu_ptr(&rcu_data);
955 WRITE_ONCE(rdp->cpu_no_qs.b.norm, false);
956 rcu_report_qs_rdp(rdp);
957 udelay(rcu_unlock_delay);
958}
959EXPORT_SYMBOL_GPL(rcu_read_unlock_strict);
960
961/*
962 * Tell them what RCU they are running.
963 */
964static void __init rcu_bootup_announce(void)
965{
966 pr_info("Hierarchical RCU implementation.\n");
967 rcu_bootup_announce_oddness();
968}
969
970/*
971 * Note a quiescent state for PREEMPTION=n. Because we do not need to know
972 * how many quiescent states passed, just if there was at least one since
973 * the start of the grace period, this just sets a flag. The caller must
974 * have disabled preemption.
975 */
976static void rcu_qs(void)
977{
978 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!");
979 if (!this_cpu_read(rcu_data.cpu_no_qs.s))
980 return;
981 trace_rcu_grace_period(TPS("rcu_sched"),
982 __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs"));
983 this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
984 if (this_cpu_read(rcu_data.cpu_no_qs.b.exp))
985 rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
986}
987
988/*
989 * Register an urgently needed quiescent state. If there is an
990 * emergency, invoke rcu_momentary_eqs() to do a heavy-weight
991 * dyntick-idle quiescent state visible to other CPUs, which will in
992 * some cases serve for expedited as well as normal grace periods.
993 * Either way, register a lightweight quiescent state.
994 */
995void rcu_all_qs(void)
996{
997 unsigned long flags;
998
999 if (!READ_ONCE(*raw_cpu_ptr(&rcu_data.rcu_urgent_qs)))
1000 return;
1001 preempt_disable(); // For CONFIG_PREEMPT_COUNT=y kernels
1002 /* Load rcu_urgent_qs before other flags. */
1003 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
1004 preempt_enable();
1005 return;
1006 }
1007 WRITE_ONCE(*this_cpu_ptr(&rcu_data.rcu_urgent_qs), false);
1008 if (unlikely(READ_ONCE(*this_cpu_ptr(&rcu_data.rcu_need_heavy_qs)))) {
1009 local_irq_save(flags);
1010 rcu_momentary_eqs();
1011 local_irq_restore(flags);
1012 }
1013 rcu_qs();
1014 preempt_enable();
1015}
1016EXPORT_SYMBOL_GPL(rcu_all_qs);
1017
1018/*
1019 * Note a PREEMPTION=n context switch. The caller must have disabled interrupts.
1020 */
1021void rcu_note_context_switch(bool preempt)
1022{
1023 trace_rcu_utilization(TPS("Start context switch"));
1024 rcu_qs();
1025 /* Load rcu_urgent_qs before other flags. */
1026 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs)))
1027 goto out;
1028 WRITE_ONCE(*this_cpu_ptr(&rcu_data.rcu_urgent_qs), false);
1029 if (unlikely(READ_ONCE(*this_cpu_ptr(&rcu_data.rcu_need_heavy_qs))))
1030 rcu_momentary_eqs();
1031out:
1032 rcu_tasks_qs(current, preempt);
1033 trace_rcu_utilization(TPS("End context switch"));
1034}
1035EXPORT_SYMBOL_GPL(rcu_note_context_switch);
1036
1037/*
1038 * Because preemptible RCU does not exist, there are never any preempted
1039 * RCU readers.
1040 */
1041static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
1042{
1043 return 0;
1044}
1045
1046/*
1047 * Because there is no preemptible RCU, there can be no readers blocked.
1048 */
1049static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
1050{
1051 return false;
1052}
1053
1054/*
1055 * Because there is no preemptible RCU, there can be no deferred quiescent
1056 * states.
1057 */
1058static notrace bool rcu_preempt_need_deferred_qs(struct task_struct *t)
1059{
1060 return false;
1061}
1062
1063// Except that we do need to respond to a request by an expedited
1064// grace period for a quiescent state from this CPU. Note that in
1065// non-preemptible kernels, there can be no context switches within RCU
1066// read-side critical sections, which in turn means that the leaf rcu_node
1067// structure's blocked-tasks list is always empty. is therefore no need to
1068// actually check it. Instead, a quiescent state from this CPU suffices,
1069// and this function is only called from such a quiescent state.
1070notrace void rcu_preempt_deferred_qs(struct task_struct *t)
1071{
1072 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
1073
1074 if (READ_ONCE(rdp->cpu_no_qs.b.exp))
1075 rcu_report_exp_rdp(rdp);
1076}
1077
1078/*
1079 * Because there is no preemptible RCU, there can be no readers blocked,
1080 * so there is no need to check for blocked tasks. So check only for
1081 * bogus qsmask values.
1082 */
1083static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
1084{
1085 WARN_ON_ONCE(rnp->qsmask);
1086}
1087
1088/*
1089 * Check to see if this CPU is in a non-context-switch quiescent state,
1090 * namely user mode and idle loop.
1091 */
1092static void rcu_flavor_sched_clock_irq(int user)
1093{
1094 if (user || rcu_is_cpu_rrupt_from_idle() ||
1095 (IS_ENABLED(CONFIG_PREEMPT_COUNT) &&
1096 (preempt_count() == HARDIRQ_OFFSET))) {
1097
1098 /*
1099 * Get here if this CPU took its interrupt from user
1100 * mode, from the idle loop without this being a nested
1101 * interrupt, or while not holding the task preempt count
1102 * (with PREEMPT_COUNT=y). In this case, the CPU is in a
1103 * quiescent state, so note it.
1104 *
1105 * No memory barrier is required here because rcu_qs()
1106 * references only CPU-local variables that other CPUs
1107 * neither access nor modify, at least not while the
1108 * corresponding CPU is online.
1109 */
1110 rcu_qs();
1111 }
1112}
1113
1114/*
1115 * Because preemptible RCU does not exist, tasks cannot possibly exit
1116 * while in preemptible RCU read-side critical sections.
1117 */
1118void exit_rcu(void)
1119{
1120}
1121
1122/*
1123 * Dump the guaranteed-empty blocked-tasks state. Trust but verify.
1124 */
1125static void
1126dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
1127{
1128 WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks));
1129}
1130
1131static void rcu_preempt_deferred_qs_init(struct rcu_data *rdp) { }
1132
1133#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
1134
1135/*
1136 * If boosting, set rcuc kthreads to realtime priority.
1137 */
1138static void rcu_cpu_kthread_setup(unsigned int cpu)
1139{
1140 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
1141#ifdef CONFIG_RCU_BOOST
1142 struct sched_param sp;
1143
1144 sp.sched_priority = kthread_prio;
1145 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1146#endif /* #ifdef CONFIG_RCU_BOOST */
1147
1148 WRITE_ONCE(rdp->rcuc_activity, jiffies);
1149}
1150
1151static bool rcu_is_callbacks_nocb_kthread(struct rcu_data *rdp)
1152{
1153#ifdef CONFIG_RCU_NOCB_CPU
1154 return rdp->nocb_cb_kthread == current;
1155#else
1156 return false;
1157#endif
1158}
1159
1160/*
1161 * Is the current CPU running the RCU-callbacks kthread?
1162 * Caller must have preemption disabled.
1163 */
1164static bool rcu_is_callbacks_kthread(struct rcu_data *rdp)
1165{
1166 return rdp->rcu_cpu_kthread_task == current ||
1167 rcu_is_callbacks_nocb_kthread(rdp);
1168}
1169
1170#ifdef CONFIG_RCU_BOOST
1171
1172/*
1173 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1174 * or ->boost_tasks, advancing the pointer to the next task in the
1175 * ->blkd_tasks list.
1176 *
1177 * Note that irqs must be enabled: boosting the task can block.
1178 * Returns 1 if there are more tasks needing to be boosted.
1179 */
1180static int rcu_boost(struct rcu_node *rnp)
1181{
1182 unsigned long flags;
1183 struct task_struct *t;
1184 struct list_head *tb;
1185
1186 if (READ_ONCE(rnp->exp_tasks) == NULL &&
1187 READ_ONCE(rnp->boost_tasks) == NULL)
1188 return 0; /* Nothing left to boost. */
1189
1190 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1191
1192 /*
1193 * Recheck under the lock: all tasks in need of boosting
1194 * might exit their RCU read-side critical sections on their own.
1195 */
1196 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1197 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1198 return 0;
1199 }
1200
1201 /*
1202 * Preferentially boost tasks blocking expedited grace periods.
1203 * This cannot starve the normal grace periods because a second
1204 * expedited grace period must boost all blocked tasks, including
1205 * those blocking the pre-existing normal grace period.
1206 */
1207 if (rnp->exp_tasks != NULL)
1208 tb = rnp->exp_tasks;
1209 else
1210 tb = rnp->boost_tasks;
1211
1212 /*
1213 * We boost task t by manufacturing an rt_mutex that appears to
1214 * be held by task t. We leave a pointer to that rt_mutex where
1215 * task t can find it, and task t will release the mutex when it
1216 * exits its outermost RCU read-side critical section. Then
1217 * simply acquiring this artificial rt_mutex will boost task
1218 * t's priority. (Thanks to tglx for suggesting this approach!)
1219 *
1220 * Note that task t must acquire rnp->lock to remove itself from
1221 * the ->blkd_tasks list, which it will do from exit() if from
1222 * nowhere else. We therefore are guaranteed that task t will
1223 * stay around at least until we drop rnp->lock. Note that
1224 * rnp->lock also resolves races between our priority boosting
1225 * and task t's exiting its outermost RCU read-side critical
1226 * section.
1227 */
1228 t = container_of(tb, struct task_struct, rcu_node_entry);
1229 rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t);
1230 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1231 /* Lock only for side effect: boosts task t's priority. */
1232 rt_mutex_lock(&rnp->boost_mtx);
1233 rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
1234 rnp->n_boosts++;
1235
1236 return READ_ONCE(rnp->exp_tasks) != NULL ||
1237 READ_ONCE(rnp->boost_tasks) != NULL;
1238}
1239
1240/*
1241 * Priority-boosting kthread, one per leaf rcu_node.
1242 */
1243static int rcu_boost_kthread(void *arg)
1244{
1245 struct rcu_node *rnp = (struct rcu_node *)arg;
1246 int spincnt = 0;
1247 int more2boost;
1248
1249 trace_rcu_utilization(TPS("Start boost kthread@init"));
1250 for (;;) {
1251 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING);
1252 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
1253 rcu_wait(READ_ONCE(rnp->boost_tasks) ||
1254 READ_ONCE(rnp->exp_tasks));
1255 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
1256 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING);
1257 more2boost = rcu_boost(rnp);
1258 if (more2boost)
1259 spincnt++;
1260 else
1261 spincnt = 0;
1262 if (spincnt > 10) {
1263 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING);
1264 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
1265 schedule_timeout_idle(2);
1266 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
1267 spincnt = 0;
1268 }
1269 }
1270 /* NOTREACHED */
1271 trace_rcu_utilization(TPS("End boost kthread@notreached"));
1272 return 0;
1273}
1274
1275/*
1276 * Check to see if it is time to start boosting RCU readers that are
1277 * blocking the current grace period, and, if so, tell the per-rcu_node
1278 * kthread to start boosting them. If there is an expedited grace
1279 * period in progress, it is always time to boost.
1280 *
1281 * The caller must hold rnp->lock, which this function releases.
1282 * The ->boost_kthread_task is immortal, so we don't need to worry
1283 * about it going away.
1284 */
1285static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1286 __releases(rnp->lock)
1287{
1288 raw_lockdep_assert_held_rcu_node(rnp);
1289 if (!rnp->boost_kthread_task ||
1290 (!rcu_preempt_blocked_readers_cgp(rnp) && !rnp->exp_tasks)) {
1291 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1292 return;
1293 }
1294 if (rnp->exp_tasks != NULL ||
1295 (rnp->gp_tasks != NULL &&
1296 rnp->boost_tasks == NULL &&
1297 rnp->qsmask == 0 &&
1298 (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld ||
1299 IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)))) {
1300 if (rnp->exp_tasks == NULL)
1301 WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks);
1302 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1303 rcu_wake_cond(rnp->boost_kthread_task,
1304 READ_ONCE(rnp->boost_kthread_status));
1305 } else {
1306 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1307 }
1308}
1309
1310#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1311
1312/*
1313 * Do priority-boost accounting for the start of a new grace period.
1314 */
1315static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1316{
1317 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1318}
1319
1320/*
1321 * Create an RCU-boost kthread for the specified node if one does not
1322 * already exist. We only create this kthread for preemptible RCU.
1323 */
1324static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1325{
1326 unsigned long flags;
1327 int rnp_index = rnp - rcu_get_root();
1328 struct sched_param sp;
1329 struct task_struct *t;
1330
1331 if (rnp->boost_kthread_task)
1332 return;
1333
1334 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1335 "rcub/%d", rnp_index);
1336 if (WARN_ON_ONCE(IS_ERR(t)))
1337 return;
1338
1339 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1340 rnp->boost_kthread_task = t;
1341 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1342
1343 sp.sched_priority = kthread_prio;
1344 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1345 rcu_thread_affine_rnp(t, rnp);
1346 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1347}
1348
1349#ifdef CONFIG_RCU_TORTURE_TEST
1350
1351/*
1352 * Is the current task RCU priority boosted? This is used by
1353 * rcutorture to check that tasks are always deboosted once then exit
1354 * an RCU read-side critical section, no matter how many overlapping
1355 * segments of rcu_read_lock(), preempt_disable(), local_bh_disable(),
1356 * or local_irq_disable() made up that reader.
1357 *
1358 * The lockless accesses in rt_mutex_owner(&rnp->boost_mtx.rtmutex)
1359 * are safe because tasks release ->boost_mtx when they own it, they
1360 * cannot be boosted unless current->rcu_blocked_node is non-NULL,
1361 * current->rcu_blocked_node is modified only by the current task,
1362 * rt_mutex_owner() uses READ_ONCE() on the ->owner field, and the owner
1363 * switching among other tasks cannot force an equality comparison.
1364 */
1365bool rcu_is_task_rcu_boosted(void)
1366{
1367 bool ret;
1368 struct rcu_node *rnp;
1369 struct task_struct *t = current;
1370
1371 preempt_disable(); // Stabilize ->rcu_blocked_node
1372 rnp = t->rcu_blocked_node;
1373 if (!rnp)
1374 ret = false;
1375 else
1376 ret = (rt_mutex_owner(&rnp->boost_mtx.rtmutex) == t);
1377 preempt_enable();
1378 return ret;
1379}
1380EXPORT_SYMBOL_GPL(rcu_is_task_rcu_boosted);
1381
1382#endif // #ifdef CONFIG_RCU_TORTURE_TEST
1383
1384#else /* #ifdef CONFIG_RCU_BOOST */
1385
1386static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1387 __releases(rnp->lock)
1388{
1389 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1390}
1391
1392static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1393{
1394}
1395
1396static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1397{
1398}
1399
1400#endif /* #else #ifdef CONFIG_RCU_BOOST */
1401
1402/*
1403 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
1404 * grace-period kthread will do force_quiescent_state() processing?
1405 * The idea is to avoid waking up RCU core processing on such a
1406 * CPU unless the grace period has extended for too long.
1407 *
1408 * This code relies on the fact that all NO_HZ_FULL CPUs are also
1409 * RCU_NOCB_CPU CPUs.
1410 */
1411static bool rcu_nohz_full_cpu(void)
1412{
1413#ifdef CONFIG_NO_HZ_FULL
1414 if (tick_nohz_full_cpu(smp_processor_id()) &&
1415 (!rcu_gp_in_progress() ||
1416 time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ)))
1417 return true;
1418#endif /* #ifdef CONFIG_NO_HZ_FULL */
1419 return false;
1420}
1421
1422/*
1423 * Bind the RCU grace-period kthreads to the housekeeping CPU.
1424 */
1425static void rcu_bind_gp_kthread(void)
1426{
1427 if (!tick_nohz_full_enabled())
1428 return;
1429 housekeeping_affine(current, HK_TYPE_RCU);
1430}