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
rcu_rdp_is_offloaded(struct rcu_data * rdp)16 static 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 */
rcu_bootup_announce_oddness(void)45 static 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
116 static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake);
117 static void rcu_read_unlock_special(struct task_struct *t);
118
119 /*
120 * Tell them what RCU they are running.
121 */
rcu_bootup_announce(void)122 static 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 */
rcu_preempt_ctxt_queue(struct rcu_node * rnp,struct rcu_data * rdp)162 static 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 */
rcu_qs(void)298 static 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 */
rcu_note_context_switch(bool preempt)324 void 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 }
374 EXPORT_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 */
rcu_preempt_blocked_readers_cgp(struct rcu_node * rnp)381 static 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
rcu_preempt_read_enter(void)389 static 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
rcu_preempt_read_exit(void)394 static 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
rcu_preempt_depth_set(int val)402 static 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 */
__rcu_read_lock(void)412 void __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 }
421 EXPORT_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 */
__rcu_read_unlock(void)430 void __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 }
446 EXPORT_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 */
rcu_next_node_entry(struct task_struct * t,struct rcu_node * rnp)452 static 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 */
rcu_preempt_has_tasks(struct rcu_node * rnp)467 static 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 */
477 static notrace void
rcu_preempt_deferred_qs_irqrestore(struct task_struct * t,unsigned long flags)478 rcu_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(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 */
rcu_preempt_need_deferred_qs(struct task_struct * t)600 static 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 */
rcu_preempt_deferred_qs(struct task_struct * t)614 notrace 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 */
rcu_preempt_deferred_qs_handler(struct irq_work * iwp)627 static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp)
628 {
629 unsigned long flags;
630 struct rcu_data *rdp;
631
632 rdp = container_of(iwp, struct rcu_data, defer_qs_iw);
633 local_irq_save(flags);
634
635 /*
636 * If the IRQ work handler happens to run in the middle of RCU read-side
637 * critical section, it could be ineffective in getting the scheduler's
638 * attention to report a deferred quiescent state (the whole point of the
639 * IRQ work). For this reason, requeue the IRQ work.
640 *
641 * Basically, we want to avoid following situation:
642 * 1. rcu_read_unlock() queues IRQ work (state -> DEFER_QS_PENDING)
643 * 2. CPU enters new rcu_read_lock()
644 * 3. IRQ work runs but cannot report QS due to rcu_preempt_depth() > 0
645 * 4. rcu_read_unlock() does not re-queue work (state still PENDING)
646 * 5. Deferred QS reporting does not happen.
647 */
648 if (rcu_preempt_depth() > 0)
649 WRITE_ONCE(rdp->defer_qs_iw_pending, DEFER_QS_IDLE);
650
651 local_irq_restore(flags);
652 }
653
654 /*
655 * Check if expedited grace period processing during unlock is needed.
656 *
657 * This function determines whether expedited handling is required based on:
658 * 1. Task blocking an expedited grace period (based on a heuristic, could be
659 * false-positive, see below.)
660 * 2. CPU participating in an expedited grace period
661 * 3. Strict grace period mode requiring expedited handling
662 * 4. RCU priority deboosting needs when interrupts were disabled
663 *
664 * @t: The task being checked
665 * @rdp: The per-CPU RCU data
666 * @rnp: The RCU node for this CPU
667 * @irqs_were_disabled: Whether interrupts were disabled before rcu_read_unlock()
668 *
669 * Returns true if expedited processing of the rcu_read_unlock() is needed.
670 */
rcu_unlock_needs_exp_handling(struct task_struct * t,struct rcu_data * rdp,struct rcu_node * rnp,bool irqs_were_disabled)671 static bool rcu_unlock_needs_exp_handling(struct task_struct *t,
672 struct rcu_data *rdp,
673 struct rcu_node *rnp,
674 bool irqs_were_disabled)
675 {
676 /*
677 * Check if this task is blocking an expedited grace period. If the
678 * task was preempted within an RCU read-side critical section and is
679 * on the expedited grace period blockers list (exp_tasks), we need
680 * expedited handling to unblock the expedited GP. This is not an exact
681 * check because 't' might not be on the exp_tasks list at all - its
682 * just a fast heuristic that can be false-positive sometimes.
683 */
684 if (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks))
685 return true;
686
687 /*
688 * Check if this CPU is participating in an expedited grace period.
689 * The expmask bitmap tracks which CPUs need to check in for the
690 * current expedited GP. If our CPU's bit is set, we need expedited
691 * handling to help complete the expedited GP.
692 */
693 if (rdp->grpmask & READ_ONCE(rnp->expmask))
694 return true;
695
696 /*
697 * In CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, all grace periods
698 * are treated as short for testing purposes even if that means
699 * disturbing the system more. Check if either:
700 * - This CPU has not yet reported a quiescent state, or
701 * - This task was preempted within an RCU critical section
702 * In either case, require expedited handling for strict GP mode.
703 */
704 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) &&
705 ((rdp->grpmask & READ_ONCE(rnp->qsmask)) || t->rcu_blocked_node))
706 return true;
707
708 /*
709 * RCU priority boosting case: If a task is subject to RCU priority
710 * boosting and exits an RCU read-side critical section with interrupts
711 * disabled, we need expedited handling to ensure timely deboosting.
712 * Without this, a low-priority task could incorrectly run at high
713 * real-time priority for an extended period degrading real-time
714 * responsiveness. This applies to all CONFIG_RCU_BOOST=y kernels,
715 * not just to PREEMPT_RT.
716 */
717 if (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && t->rcu_blocked_node)
718 return true;
719
720 return false;
721 }
722
723 /*
724 * Handle special cases during rcu_read_unlock(), such as needing to
725 * notify RCU core processing or task having blocked during the RCU
726 * read-side critical section.
727 */
rcu_read_unlock_special(struct task_struct * t)728 static void rcu_read_unlock_special(struct task_struct *t)
729 {
730 unsigned long flags;
731 bool irqs_were_disabled;
732 bool preempt_bh_were_disabled =
733 !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK));
734
735 /* NMI handlers cannot block and cannot safely manipulate state. */
736 if (in_nmi())
737 return;
738
739 local_irq_save(flags);
740 irqs_were_disabled = irqs_disabled_flags(flags);
741 if (preempt_bh_were_disabled || irqs_were_disabled) {
742 bool needs_exp; // Expedited handling needed.
743 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
744 struct rcu_node *rnp = rdp->mynode;
745
746 needs_exp = rcu_unlock_needs_exp_handling(t, rdp, rnp, irqs_were_disabled);
747
748 // Need to defer quiescent state until everything is enabled.
749 if (use_softirq && (in_hardirq() || (needs_exp && !irqs_were_disabled))) {
750 // Using softirq, safe to awaken, and either the
751 // wakeup is free or there is either an expedited
752 // GP in flight or a potential need to deboost.
753 raise_softirq_irqoff(RCU_SOFTIRQ);
754 } else {
755 // Enabling BH or preempt does reschedule, so...
756 // Also if no expediting and no possible deboosting,
757 // slow is OK. Plus nohz_full CPUs eventually get
758 // tick enabled.
759 set_tsk_need_resched(current);
760 set_preempt_need_resched();
761 if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled &&
762 needs_exp && rdp->defer_qs_iw_pending != DEFER_QS_PENDING &&
763 cpu_online(rdp->cpu)) {
764 // Get scheduler to re-evaluate and call hooks.
765 // If !IRQ_WORK, FQS scan will eventually IPI.
766 rdp->defer_qs_iw =
767 IRQ_WORK_INIT_HARD(rcu_preempt_deferred_qs_handler);
768 rdp->defer_qs_iw_pending = DEFER_QS_PENDING;
769 irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu);
770 }
771 }
772 local_irq_restore(flags);
773 return;
774 }
775 rcu_preempt_deferred_qs_irqrestore(t, flags);
776 }
777
778 /*
779 * Check that the list of blocked tasks for the newly completed grace
780 * period is in fact empty. It is a serious bug to complete a grace
781 * period that still has RCU readers blocked! This function must be
782 * invoked -before- updating this rnp's ->gp_seq.
783 *
784 * Also, if there are blocked tasks on the list, they automatically
785 * block the newly created grace period, so set up ->gp_tasks accordingly.
786 */
rcu_preempt_check_blocked_tasks(struct rcu_node * rnp)787 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
788 {
789 struct task_struct *t;
790
791 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
792 raw_lockdep_assert_held_rcu_node(rnp);
793 if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
794 dump_blkd_tasks(rnp, 10);
795 if (rcu_preempt_has_tasks(rnp) &&
796 (rnp->qsmaskinit || rnp->wait_blkd_tasks)) {
797 WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next);
798 t = container_of(rnp->gp_tasks, struct task_struct,
799 rcu_node_entry);
800 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
801 rnp->gp_seq, t->pid);
802 }
803 WARN_ON_ONCE(rnp->qsmask);
804 }
805
806 /*
807 * Check for a quiescent state from the current CPU, including voluntary
808 * context switches for Tasks RCU. When a task blocks, the task is
809 * recorded in the corresponding CPU's rcu_node structure, which is checked
810 * elsewhere, hence this function need only check for quiescent states
811 * related to the current CPU, not to those related to tasks.
812 */
rcu_flavor_sched_clock_irq(int user)813 static void rcu_flavor_sched_clock_irq(int user)
814 {
815 struct task_struct *t = current;
816
817 lockdep_assert_irqs_disabled();
818 if (rcu_preempt_depth() > 0 ||
819 (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) {
820 /* No QS, force context switch if deferred. */
821 if (rcu_preempt_need_deferred_qs(t)) {
822 set_tsk_need_resched(t);
823 set_preempt_need_resched();
824 }
825 } else if (rcu_preempt_need_deferred_qs(t)) {
826 rcu_preempt_deferred_qs(t); /* Report deferred QS. */
827 return;
828 } else if (!WARN_ON_ONCE(rcu_preempt_depth())) {
829 rcu_qs(); /* Report immediate QS. */
830 return;
831 }
832
833 /* If GP is oldish, ask for help from rcu_read_unlock_special(). */
834 if (rcu_preempt_depth() > 0 &&
835 __this_cpu_read(rcu_data.core_needs_qs) &&
836 __this_cpu_read(rcu_data.cpu_no_qs.b.norm) &&
837 !t->rcu_read_unlock_special.b.need_qs &&
838 time_after(jiffies, rcu_state.gp_start + HZ))
839 t->rcu_read_unlock_special.b.need_qs = true;
840 }
841
842 /*
843 * Check for a task exiting while in a preemptible-RCU read-side
844 * critical section, clean up if so. No need to issue warnings, as
845 * debug_check_no_locks_held() already does this if lockdep is enabled.
846 * Besides, if this function does anything other than just immediately
847 * return, there was a bug of some sort. Spewing warnings from this
848 * function is like as not to simply obscure important prior warnings.
849 */
exit_rcu(void)850 void exit_rcu(void)
851 {
852 struct task_struct *t = current;
853
854 if (unlikely(!list_empty(¤t->rcu_node_entry))) {
855 rcu_preempt_depth_set(1);
856 barrier();
857 WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true);
858 } else if (unlikely(rcu_preempt_depth())) {
859 rcu_preempt_depth_set(1);
860 } else {
861 return;
862 }
863 __rcu_read_unlock();
864 rcu_preempt_deferred_qs(current);
865 }
866
867 /*
868 * Dump the blocked-tasks state, but limit the list dump to the
869 * specified number of elements.
870 */
871 static void
dump_blkd_tasks(struct rcu_node * rnp,int ncheck)872 dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
873 {
874 int cpu;
875 int i;
876 struct list_head *lhp;
877 struct rcu_data *rdp;
878 struct rcu_node *rnp1;
879
880 raw_lockdep_assert_held_rcu_node(rnp);
881 pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
882 __func__, rnp->grplo, rnp->grphi, rnp->level,
883 (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs);
884 for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
885 pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n",
886 __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext);
887 pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n",
888 __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks),
889 READ_ONCE(rnp->exp_tasks));
890 pr_info("%s: ->blkd_tasks", __func__);
891 i = 0;
892 list_for_each(lhp, &rnp->blkd_tasks) {
893 pr_cont(" %p", lhp);
894 if (++i >= ncheck)
895 break;
896 }
897 pr_cont("\n");
898 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) {
899 rdp = per_cpu_ptr(&rcu_data, cpu);
900 pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n",
901 cpu, ".o"[rcu_rdp_cpu_online(rdp)],
902 (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_state,
903 (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_state);
904 }
905 }
906
907 #else /* #ifdef CONFIG_PREEMPT_RCU */
908
909 /*
910 * If strict grace periods are enabled, and if the calling
911 * __rcu_read_unlock() marks the beginning of a quiescent state, immediately
912 * report that quiescent state and, if requested, spin for a bit.
913 */
rcu_read_unlock_strict(void)914 void rcu_read_unlock_strict(void)
915 {
916 struct rcu_data *rdp;
917
918 if (irqs_disabled() || in_atomic_preempt_off() || !rcu_state.gp_kthread)
919 return;
920
921 /*
922 * rcu_report_qs_rdp() can only be invoked with a stable rdp and
923 * from the local CPU.
924 *
925 * The in_atomic_preempt_off() check ensures that we come here holding
926 * the last preempt_count (which will get dropped once we return to
927 * __rcu_read_unlock().
928 */
929 rdp = this_cpu_ptr(&rcu_data);
930 rdp->cpu_no_qs.b.norm = false;
931 rcu_report_qs_rdp(rdp);
932 udelay(rcu_unlock_delay);
933 }
934 EXPORT_SYMBOL_GPL(rcu_read_unlock_strict);
935
936 /*
937 * Tell them what RCU they are running.
938 */
rcu_bootup_announce(void)939 static void __init rcu_bootup_announce(void)
940 {
941 pr_info("Hierarchical RCU implementation.\n");
942 rcu_bootup_announce_oddness();
943 }
944
945 /*
946 * Note a quiescent state for PREEMPTION=n. Because we do not need to know
947 * how many quiescent states passed, just if there was at least one since
948 * the start of the grace period, this just sets a flag. The caller must
949 * have disabled preemption.
950 */
rcu_qs(void)951 static void rcu_qs(void)
952 {
953 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!");
954 if (!__this_cpu_read(rcu_data.cpu_no_qs.s))
955 return;
956 trace_rcu_grace_period(TPS("rcu_sched"),
957 __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs"));
958 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
959 if (__this_cpu_read(rcu_data.cpu_no_qs.b.exp))
960 rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
961 }
962
963 /*
964 * Register an urgently needed quiescent state. If there is an
965 * emergency, invoke rcu_momentary_eqs() to do a heavy-weight
966 * dyntick-idle quiescent state visible to other CPUs, which will in
967 * some cases serve for expedited as well as normal grace periods.
968 * Either way, register a lightweight quiescent state.
969 */
rcu_all_qs(void)970 void rcu_all_qs(void)
971 {
972 unsigned long flags;
973
974 if (!raw_cpu_read(rcu_data.rcu_urgent_qs))
975 return;
976 preempt_disable(); // For CONFIG_PREEMPT_COUNT=y kernels
977 /* Load rcu_urgent_qs before other flags. */
978 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
979 preempt_enable();
980 return;
981 }
982 this_cpu_write(rcu_data.rcu_urgent_qs, false);
983 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) {
984 local_irq_save(flags);
985 rcu_momentary_eqs();
986 local_irq_restore(flags);
987 }
988 rcu_qs();
989 preempt_enable();
990 }
991 EXPORT_SYMBOL_GPL(rcu_all_qs);
992
993 /*
994 * Note a PREEMPTION=n context switch. The caller must have disabled interrupts.
995 */
rcu_note_context_switch(bool preempt)996 void rcu_note_context_switch(bool preempt)
997 {
998 trace_rcu_utilization(TPS("Start context switch"));
999 rcu_qs();
1000 /* Load rcu_urgent_qs before other flags. */
1001 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs)))
1002 goto out;
1003 this_cpu_write(rcu_data.rcu_urgent_qs, false);
1004 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs)))
1005 rcu_momentary_eqs();
1006 out:
1007 rcu_tasks_qs(current, preempt);
1008 trace_rcu_utilization(TPS("End context switch"));
1009 }
1010 EXPORT_SYMBOL_GPL(rcu_note_context_switch);
1011
1012 /*
1013 * Because preemptible RCU does not exist, there are never any preempted
1014 * RCU readers.
1015 */
rcu_preempt_blocked_readers_cgp(struct rcu_node * rnp)1016 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
1017 {
1018 return 0;
1019 }
1020
1021 /*
1022 * Because there is no preemptible RCU, there can be no readers blocked.
1023 */
rcu_preempt_has_tasks(struct rcu_node * rnp)1024 static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
1025 {
1026 return false;
1027 }
1028
1029 /*
1030 * Because there is no preemptible RCU, there can be no deferred quiescent
1031 * states.
1032 */
rcu_preempt_need_deferred_qs(struct task_struct * t)1033 static notrace bool rcu_preempt_need_deferred_qs(struct task_struct *t)
1034 {
1035 return false;
1036 }
1037
1038 // Except that we do need to respond to a request by an expedited
1039 // grace period for a quiescent state from this CPU. Note that in
1040 // non-preemptible kernels, there can be no context switches within RCU
1041 // read-side critical sections, which in turn means that the leaf rcu_node
1042 // structure's blocked-tasks list is always empty. is therefore no need to
1043 // actually check it. Instead, a quiescent state from this CPU suffices,
1044 // and this function is only called from such a quiescent state.
rcu_preempt_deferred_qs(struct task_struct * t)1045 notrace void rcu_preempt_deferred_qs(struct task_struct *t)
1046 {
1047 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
1048
1049 if (READ_ONCE(rdp->cpu_no_qs.b.exp))
1050 rcu_report_exp_rdp(rdp);
1051 }
1052
1053 /*
1054 * Because there is no preemptible RCU, there can be no readers blocked,
1055 * so there is no need to check for blocked tasks. So check only for
1056 * bogus qsmask values.
1057 */
rcu_preempt_check_blocked_tasks(struct rcu_node * rnp)1058 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
1059 {
1060 WARN_ON_ONCE(rnp->qsmask);
1061 }
1062
1063 /*
1064 * Check to see if this CPU is in a non-context-switch quiescent state,
1065 * namely user mode and idle loop.
1066 */
rcu_flavor_sched_clock_irq(int user)1067 static void rcu_flavor_sched_clock_irq(int user)
1068 {
1069 if (user || rcu_is_cpu_rrupt_from_idle() ||
1070 (IS_ENABLED(CONFIG_PREEMPT_COUNT) &&
1071 (preempt_count() == HARDIRQ_OFFSET))) {
1072
1073 /*
1074 * Get here if this CPU took its interrupt from user
1075 * mode, from the idle loop without this being a nested
1076 * interrupt, or while not holding the task preempt count
1077 * (with PREEMPT_COUNT=y). In this case, the CPU is in a
1078 * quiescent state, so note it.
1079 *
1080 * No memory barrier is required here because rcu_qs()
1081 * references only CPU-local variables that other CPUs
1082 * neither access nor modify, at least not while the
1083 * corresponding CPU is online.
1084 */
1085 rcu_qs();
1086 }
1087 }
1088
1089 /*
1090 * Because preemptible RCU does not exist, tasks cannot possibly exit
1091 * while in preemptible RCU read-side critical sections.
1092 */
exit_rcu(void)1093 void exit_rcu(void)
1094 {
1095 }
1096
1097 /*
1098 * Dump the guaranteed-empty blocked-tasks state. Trust but verify.
1099 */
1100 static void
dump_blkd_tasks(struct rcu_node * rnp,int ncheck)1101 dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
1102 {
1103 WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks));
1104 }
1105
1106 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
1107
1108 /*
1109 * If boosting, set rcuc kthreads to realtime priority.
1110 */
rcu_cpu_kthread_setup(unsigned int cpu)1111 static void rcu_cpu_kthread_setup(unsigned int cpu)
1112 {
1113 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
1114 #ifdef CONFIG_RCU_BOOST
1115 struct sched_param sp;
1116
1117 sp.sched_priority = kthread_prio;
1118 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1119 #endif /* #ifdef CONFIG_RCU_BOOST */
1120
1121 WRITE_ONCE(rdp->rcuc_activity, jiffies);
1122 }
1123
rcu_is_callbacks_nocb_kthread(struct rcu_data * rdp)1124 static bool rcu_is_callbacks_nocb_kthread(struct rcu_data *rdp)
1125 {
1126 #ifdef CONFIG_RCU_NOCB_CPU
1127 return rdp->nocb_cb_kthread == current;
1128 #else
1129 return false;
1130 #endif
1131 }
1132
1133 /*
1134 * Is the current CPU running the RCU-callbacks kthread?
1135 * Caller must have preemption disabled.
1136 */
rcu_is_callbacks_kthread(struct rcu_data * rdp)1137 static bool rcu_is_callbacks_kthread(struct rcu_data *rdp)
1138 {
1139 return rdp->rcu_cpu_kthread_task == current ||
1140 rcu_is_callbacks_nocb_kthread(rdp);
1141 }
1142
1143 #ifdef CONFIG_RCU_BOOST
1144
1145 /*
1146 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1147 * or ->boost_tasks, advancing the pointer to the next task in the
1148 * ->blkd_tasks list.
1149 *
1150 * Note that irqs must be enabled: boosting the task can block.
1151 * Returns 1 if there are more tasks needing to be boosted.
1152 */
rcu_boost(struct rcu_node * rnp)1153 static int rcu_boost(struct rcu_node *rnp)
1154 {
1155 unsigned long flags;
1156 struct task_struct *t;
1157 struct list_head *tb;
1158
1159 if (READ_ONCE(rnp->exp_tasks) == NULL &&
1160 READ_ONCE(rnp->boost_tasks) == NULL)
1161 return 0; /* Nothing left to boost. */
1162
1163 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1164
1165 /*
1166 * Recheck under the lock: all tasks in need of boosting
1167 * might exit their RCU read-side critical sections on their own.
1168 */
1169 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1170 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1171 return 0;
1172 }
1173
1174 /*
1175 * Preferentially boost tasks blocking expedited grace periods.
1176 * This cannot starve the normal grace periods because a second
1177 * expedited grace period must boost all blocked tasks, including
1178 * those blocking the pre-existing normal grace period.
1179 */
1180 if (rnp->exp_tasks != NULL)
1181 tb = rnp->exp_tasks;
1182 else
1183 tb = rnp->boost_tasks;
1184
1185 /*
1186 * We boost task t by manufacturing an rt_mutex that appears to
1187 * be held by task t. We leave a pointer to that rt_mutex where
1188 * task t can find it, and task t will release the mutex when it
1189 * exits its outermost RCU read-side critical section. Then
1190 * simply acquiring this artificial rt_mutex will boost task
1191 * t's priority. (Thanks to tglx for suggesting this approach!)
1192 *
1193 * Note that task t must acquire rnp->lock to remove itself from
1194 * the ->blkd_tasks list, which it will do from exit() if from
1195 * nowhere else. We therefore are guaranteed that task t will
1196 * stay around at least until we drop rnp->lock. Note that
1197 * rnp->lock also resolves races between our priority boosting
1198 * and task t's exiting its outermost RCU read-side critical
1199 * section.
1200 */
1201 t = container_of(tb, struct task_struct, rcu_node_entry);
1202 rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t);
1203 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1204 /* Lock only for side effect: boosts task t's priority. */
1205 rt_mutex_lock(&rnp->boost_mtx);
1206 rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
1207 rnp->n_boosts++;
1208
1209 return READ_ONCE(rnp->exp_tasks) != NULL ||
1210 READ_ONCE(rnp->boost_tasks) != NULL;
1211 }
1212
1213 /*
1214 * Priority-boosting kthread, one per leaf rcu_node.
1215 */
rcu_boost_kthread(void * arg)1216 static int rcu_boost_kthread(void *arg)
1217 {
1218 struct rcu_node *rnp = (struct rcu_node *)arg;
1219 int spincnt = 0;
1220 int more2boost;
1221
1222 trace_rcu_utilization(TPS("Start boost kthread@init"));
1223 for (;;) {
1224 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING);
1225 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
1226 rcu_wait(READ_ONCE(rnp->boost_tasks) ||
1227 READ_ONCE(rnp->exp_tasks));
1228 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
1229 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING);
1230 more2boost = rcu_boost(rnp);
1231 if (more2boost)
1232 spincnt++;
1233 else
1234 spincnt = 0;
1235 if (spincnt > 10) {
1236 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING);
1237 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
1238 schedule_timeout_idle(2);
1239 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
1240 spincnt = 0;
1241 }
1242 }
1243 /* NOTREACHED */
1244 trace_rcu_utilization(TPS("End boost kthread@notreached"));
1245 return 0;
1246 }
1247
1248 /*
1249 * Check to see if it is time to start boosting RCU readers that are
1250 * blocking the current grace period, and, if so, tell the per-rcu_node
1251 * kthread to start boosting them. If there is an expedited grace
1252 * period in progress, it is always time to boost.
1253 *
1254 * The caller must hold rnp->lock, which this function releases.
1255 * The ->boost_kthread_task is immortal, so we don't need to worry
1256 * about it going away.
1257 */
rcu_initiate_boost(struct rcu_node * rnp,unsigned long flags)1258 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1259 __releases(rnp->lock)
1260 {
1261 raw_lockdep_assert_held_rcu_node(rnp);
1262 if (!rnp->boost_kthread_task ||
1263 (!rcu_preempt_blocked_readers_cgp(rnp) && !rnp->exp_tasks)) {
1264 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1265 return;
1266 }
1267 if (rnp->exp_tasks != NULL ||
1268 (rnp->gp_tasks != NULL &&
1269 rnp->boost_tasks == NULL &&
1270 rnp->qsmask == 0 &&
1271 (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld ||
1272 IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)))) {
1273 if (rnp->exp_tasks == NULL)
1274 WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks);
1275 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1276 rcu_wake_cond(rnp->boost_kthread_task,
1277 READ_ONCE(rnp->boost_kthread_status));
1278 } else {
1279 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1280 }
1281 }
1282
1283 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1284
1285 /*
1286 * Do priority-boost accounting for the start of a new grace period.
1287 */
rcu_preempt_boost_start_gp(struct rcu_node * rnp)1288 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1289 {
1290 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1291 }
1292
1293 /*
1294 * Create an RCU-boost kthread for the specified node if one does not
1295 * already exist. We only create this kthread for preemptible RCU.
1296 */
rcu_spawn_one_boost_kthread(struct rcu_node * rnp)1297 static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1298 {
1299 unsigned long flags;
1300 int rnp_index = rnp - rcu_get_root();
1301 struct sched_param sp;
1302 struct task_struct *t;
1303
1304 if (rnp->boost_kthread_task)
1305 return;
1306
1307 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1308 "rcub/%d", rnp_index);
1309 if (WARN_ON_ONCE(IS_ERR(t)))
1310 return;
1311
1312 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1313 rnp->boost_kthread_task = t;
1314 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1315
1316 sp.sched_priority = kthread_prio;
1317 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1318 rcu_thread_affine_rnp(t, rnp);
1319 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1320 }
1321
1322 #else /* #ifdef CONFIG_RCU_BOOST */
1323
rcu_initiate_boost(struct rcu_node * rnp,unsigned long flags)1324 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1325 __releases(rnp->lock)
1326 {
1327 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1328 }
1329
rcu_preempt_boost_start_gp(struct rcu_node * rnp)1330 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1331 {
1332 }
1333
rcu_spawn_one_boost_kthread(struct rcu_node * rnp)1334 static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1335 {
1336 }
1337
1338 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1339
1340 /*
1341 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
1342 * grace-period kthread will do force_quiescent_state() processing?
1343 * The idea is to avoid waking up RCU core processing on such a
1344 * CPU unless the grace period has extended for too long.
1345 *
1346 * This code relies on the fact that all NO_HZ_FULL CPUs are also
1347 * RCU_NOCB_CPU CPUs.
1348 */
rcu_nohz_full_cpu(void)1349 static bool rcu_nohz_full_cpu(void)
1350 {
1351 #ifdef CONFIG_NO_HZ_FULL
1352 if (tick_nohz_full_cpu(smp_processor_id()) &&
1353 (!rcu_gp_in_progress() ||
1354 time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ)))
1355 return true;
1356 #endif /* #ifdef CONFIG_NO_HZ_FULL */
1357 return false;
1358 }
1359
1360 /*
1361 * Bind the RCU grace-period kthreads to the housekeeping CPU.
1362 */
rcu_bind_gp_kthread(void)1363 static void rcu_bind_gp_kthread(void)
1364 {
1365 if (!tick_nohz_full_enabled())
1366 return;
1367 housekeeping_affine(current, HK_TYPE_RCU);
1368 }
1369