rcu: Eliminate BUG_ON() for kernel/rcu/tree.c
[linux-2.6-block.git] / kernel / rcu / tree_plugin.h
CommitLineData
f41d911f
PM
1/*
2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions that provide either classic
6cc68793 4 * or preemptible semantics.
f41d911f
PM
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
87de1cfd
PM
17 * along with this program; if not, you can access it online at
18 * http://www.gnu.org/licenses/gpl-2.0.html.
f41d911f
PM
19 *
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
22 *
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
25 */
26
d9a3da06 27#include <linux/delay.h>
3fbfbf7a 28#include <linux/gfp.h>
b626c1b6 29#include <linux/oom.h>
b17b0153 30#include <linux/sched/debug.h>
62ab7072 31#include <linux/smpboot.h>
78634061 32#include <linux/sched/isolation.h>
ae7e81c0 33#include <uapi/linux/sched/types.h>
4102adab 34#include "../time/tick-internal.h"
f41d911f 35
5b61b0ba 36#ifdef CONFIG_RCU_BOOST
61cfd097 37
abaa93d9 38#include "../locking/rtmutex_common.h"
21871d7e 39
61cfd097 40/*
0ae86a27 41 * Control variables for per-CPU and per-rcu_node kthreads.
61cfd097
PM
42 */
43static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
44DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
45DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
46DEFINE_PER_CPU(char, rcu_cpu_has_work);
47
727b705b
PM
48#else /* #ifdef CONFIG_RCU_BOOST */
49
50/*
51 * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST,
52 * all uses are in dead code. Provide a definition to keep the compiler
53 * happy, but add WARN_ON_ONCE() to complain if used in the wrong place.
54 * This probably needs to be excluded from -rt builds.
55 */
56#define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; })
b8869781 57#define rt_mutex_futex_unlock(x) WARN_ON_ONCE(1)
727b705b
PM
58
59#endif /* #else #ifdef CONFIG_RCU_BOOST */
5b61b0ba 60
3fbfbf7a
PM
61#ifdef CONFIG_RCU_NOCB_CPU
62static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */
1b0048a4 63static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */
3fbfbf7a
PM
64#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
65
26845c28
PM
66/*
67 * Check the RCU kernel configuration parameters and print informative
699d4035 68 * messages about anything out of the ordinary.
26845c28
PM
69 */
70static void __init rcu_bootup_announce_oddness(void)
71{
ab6f5bd6 72 if (IS_ENABLED(CONFIG_RCU_TRACE))
ae91aa0a 73 pr_info("\tRCU event tracing is enabled.\n");
05c5df31
PM
74 if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) ||
75 (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32))
a7538352
JP
76 pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n",
77 RCU_FANOUT);
7fa27001 78 if (rcu_fanout_exact)
ab6f5bd6
PM
79 pr_info("\tHierarchical RCU autobalancing is disabled.\n");
80 if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ))
81 pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n");
c4a09ff7 82 if (IS_ENABLED(CONFIG_PROVE_RCU))
ab6f5bd6 83 pr_info("\tRCU lockdep checking is enabled.\n");
42621697
AG
84 if (RCU_NUM_LVLS >= 4)
85 pr_info("\tFour(or more)-level hierarchy is enabled.\n");
47d631af 86 if (RCU_FANOUT_LEAF != 16)
a3bd2c09 87 pr_info("\tBuild-time adjustment of leaf fanout to %d.\n",
47d631af
PM
88 RCU_FANOUT_LEAF);
89 if (rcu_fanout_leaf != RCU_FANOUT_LEAF)
a7538352
JP
90 pr_info("\tBoot-time adjustment of leaf fanout to %d.\n",
91 rcu_fanout_leaf);
cca6f393 92 if (nr_cpu_ids != NR_CPUS)
9b130ad5 93 pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids);
17c7798b 94#ifdef CONFIG_RCU_BOOST
a7538352
JP
95 pr_info("\tRCU priority boosting: priority %d delay %d ms.\n",
96 kthread_prio, CONFIG_RCU_BOOST_DELAY);
17c7798b
PM
97#endif
98 if (blimit != DEFAULT_RCU_BLIMIT)
99 pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit);
100 if (qhimark != DEFAULT_RCU_QHIMARK)
101 pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark);
102 if (qlowmark != DEFAULT_RCU_QLOMARK)
103 pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark);
104 if (jiffies_till_first_fqs != ULONG_MAX)
105 pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs);
106 if (jiffies_till_next_fqs != ULONG_MAX)
107 pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs);
c06aed0e
PM
108 if (jiffies_till_sched_qs != ULONG_MAX)
109 pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs);
17c7798b
PM
110 if (rcu_kick_kthreads)
111 pr_info("\tKick kthreads if too-long grace period.\n");
112 if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD))
113 pr_info("\tRCU callback double-/use-after-free debug enabled.\n");
90040c9e 114 if (gp_preinit_delay)
17c7798b 115 pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay);
90040c9e 116 if (gp_init_delay)
17c7798b 117 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay);
90040c9e 118 if (gp_cleanup_delay)
17c7798b
PM
119 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay);
120 if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG))
121 pr_info("\tRCU debug extended QS entry/exit.\n");
59d80fd8 122 rcupdate_announce_bootup_oddness();
26845c28
PM
123}
124
28f6569a 125#ifdef CONFIG_PREEMPT_RCU
f41d911f 126
63d4c8c9 127static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake);
3949fa9b 128static void rcu_read_unlock_special(struct task_struct *t);
d9a3da06 129
f41d911f
PM
130/*
131 * Tell them what RCU they are running.
132 */
0e0fc1c2 133static void __init rcu_bootup_announce(void)
f41d911f 134{
efc151c3 135 pr_info("Preemptible hierarchical RCU implementation.\n");
26845c28 136 rcu_bootup_announce_oddness();
f41d911f
PM
137}
138
8203d6d0
PM
139/* Flags for rcu_preempt_ctxt_queue() decision table. */
140#define RCU_GP_TASKS 0x8
141#define RCU_EXP_TASKS 0x4
142#define RCU_GP_BLKD 0x2
143#define RCU_EXP_BLKD 0x1
144
145/*
146 * Queues a task preempted within an RCU-preempt read-side critical
147 * section into the appropriate location within the ->blkd_tasks list,
148 * depending on the states of any ongoing normal and expedited grace
149 * periods. The ->gp_tasks pointer indicates which element the normal
150 * grace period is waiting on (NULL if none), and the ->exp_tasks pointer
151 * indicates which element the expedited grace period is waiting on (again,
152 * NULL if none). If a grace period is waiting on a given element in the
153 * ->blkd_tasks list, it also waits on all subsequent elements. Thus,
154 * adding a task to the tail of the list blocks any grace period that is
155 * already waiting on one of the elements. In contrast, adding a task
156 * to the head of the list won't block any grace period that is already
157 * waiting on one of the elements.
158 *
159 * This queuing is imprecise, and can sometimes make an ongoing grace
160 * period wait for a task that is not strictly speaking blocking it.
161 * Given the choice, we needlessly block a normal grace period rather than
162 * blocking an expedited grace period.
163 *
164 * Note that an endless sequence of expedited grace periods still cannot
165 * indefinitely postpone a normal grace period. Eventually, all of the
166 * fixed number of preempted tasks blocking the normal grace period that are
167 * not also blocking the expedited grace period will resume and complete
168 * their RCU read-side critical sections. At that point, the ->gp_tasks
169 * pointer will equal the ->exp_tasks pointer, at which point the end of
170 * the corresponding expedited grace period will also be the end of the
171 * normal grace period.
172 */
46a5d164
PM
173static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp)
174 __releases(rnp->lock) /* But leaves rrupts disabled. */
8203d6d0
PM
175{
176 int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) +
177 (rnp->exp_tasks ? RCU_EXP_TASKS : 0) +
178 (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) +
179 (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0);
180 struct task_struct *t = current;
181
a32e01ee 182 raw_lockdep_assert_held_rcu_node(rnp);
2dee9404 183 WARN_ON_ONCE(rdp->mynode != rnp);
5b4c11d5 184 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
1f3e5f51
PM
185 /* RCU better not be waiting on newly onlined CPUs! */
186 WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask &
187 rdp->grpmask);
ea9b0c8a 188
8203d6d0
PM
189 /*
190 * Decide where to queue the newly blocked task. In theory,
191 * this could be an if-statement. In practice, when I tried
192 * that, it was quite messy.
193 */
194 switch (blkd_state) {
195 case 0:
196 case RCU_EXP_TASKS:
197 case RCU_EXP_TASKS + RCU_GP_BLKD:
198 case RCU_GP_TASKS:
199 case RCU_GP_TASKS + RCU_EXP_TASKS:
200
201 /*
202 * Blocking neither GP, or first task blocking the normal
203 * GP but not blocking the already-waiting expedited GP.
204 * Queue at the head of the list to avoid unnecessarily
205 * blocking the already-waiting GPs.
206 */
207 list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
208 break;
209
210 case RCU_EXP_BLKD:
211 case RCU_GP_BLKD:
212 case RCU_GP_BLKD + RCU_EXP_BLKD:
213 case RCU_GP_TASKS + RCU_EXP_BLKD:
214 case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
215 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
216
217 /*
218 * First task arriving that blocks either GP, or first task
219 * arriving that blocks the expedited GP (with the normal
220 * GP already waiting), or a task arriving that blocks
221 * both GPs with both GPs already waiting. Queue at the
222 * tail of the list to avoid any GP waiting on any of the
223 * already queued tasks that are not blocking it.
224 */
225 list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks);
226 break;
227
228 case RCU_EXP_TASKS + RCU_EXP_BLKD:
229 case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
230 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD:
231
232 /*
233 * Second or subsequent task blocking the expedited GP.
234 * The task either does not block the normal GP, or is the
235 * first task blocking the normal GP. Queue just after
236 * the first task blocking the expedited GP.
237 */
238 list_add(&t->rcu_node_entry, rnp->exp_tasks);
239 break;
240
241 case RCU_GP_TASKS + RCU_GP_BLKD:
242 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD:
243
244 /*
245 * Second or subsequent task blocking the normal GP.
246 * The task does not block the expedited GP. Queue just
247 * after the first task blocking the normal GP.
248 */
249 list_add(&t->rcu_node_entry, rnp->gp_tasks);
250 break;
251
252 default:
253
254 /* Yet another exercise in excessive paranoia. */
255 WARN_ON_ONCE(1);
256 break;
257 }
258
259 /*
260 * We have now queued the task. If it was the first one to
261 * block either grace period, update the ->gp_tasks and/or
262 * ->exp_tasks pointers, respectively, to reference the newly
263 * blocked tasks.
264 */
4bc8d555 265 if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) {
8203d6d0 266 rnp->gp_tasks = &t->rcu_node_entry;
d43a5d32 267 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq);
4bc8d555 268 }
8203d6d0
PM
269 if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD))
270 rnp->exp_tasks = &t->rcu_node_entry;
2dee9404
PM
271 WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) !=
272 !(rnp->qsmask & rdp->grpmask));
273 WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) !=
274 !(rnp->expmask & rdp->grpmask));
67c583a7 275 raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */
8203d6d0
PM
276
277 /*
278 * Report the quiescent state for the expedited GP. This expedited
279 * GP should not be able to end until we report, so there should be
280 * no need to check for a subsequent expedited GP. (Though we are
281 * still in a quiescent state in any case.)
282 */
fcc878e4 283 if (blkd_state & RCU_EXP_BLKD && rdp->deferred_qs)
63d4c8c9 284 rcu_report_exp_rdp(rdp);
fcc878e4
PM
285 else
286 WARN_ON_ONCE(rdp->deferred_qs);
8203d6d0
PM
287}
288
f41d911f 289/*
c7037ff5
PM
290 * Record a preemptible-RCU quiescent state for the specified CPU.
291 * Note that this does not necessarily mean that the task currently running
292 * on the CPU is in a quiescent state: Instead, it means that the current
293 * grace period need not wait on any RCU read-side critical section that
294 * starts later on this CPU. It also means that if the current task is
295 * in an RCU read-side critical section, it has already added itself to
296 * some leaf rcu_node structure's ->blkd_tasks list. In addition to the
297 * current task, there might be any number of other tasks blocked while
298 * in an RCU read-side critical section.
25502a6c 299 *
c7037ff5 300 * Callers to this function must disable preemption.
f41d911f 301 */
45975c7d 302static void rcu_qs(void)
f41d911f 303{
45975c7d 304 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n");
2280ee5a 305 if (__this_cpu_read(rcu_data.cpu_no_qs.s)) {
284a8c93 306 trace_rcu_grace_period(TPS("rcu_preempt"),
2280ee5a 307 __this_cpu_read(rcu_data.gp_seq),
284a8c93 308 TPS("cpuqs"));
2280ee5a 309 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
45975c7d 310 barrier(); /* Coordinate with rcu_flavor_check_callbacks(). */
284a8c93
PM
311 current->rcu_read_unlock_special.b.need_qs = false;
312 }
f41d911f
PM
313}
314
315/*
c3422bea
PM
316 * We have entered the scheduler, and the current task might soon be
317 * context-switched away from. If this task is in an RCU read-side
318 * critical section, we will no longer be able to rely on the CPU to
12f5f524
PM
319 * record that fact, so we enqueue the task on the blkd_tasks list.
320 * The task will dequeue itself when it exits the outermost enclosing
321 * RCU read-side critical section. Therefore, the current grace period
322 * cannot be permitted to complete until the blkd_tasks list entries
323 * predating the current grace period drain, in other words, until
324 * rnp->gp_tasks becomes NULL.
c3422bea 325 *
46a5d164 326 * Caller must disable interrupts.
f41d911f 327 */
45975c7d 328void rcu_note_context_switch(bool preempt)
f41d911f
PM
329{
330 struct task_struct *t = current;
da1df50d 331 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
f41d911f
PM
332 struct rcu_node *rnp;
333
45975c7d
PM
334 barrier(); /* Avoid RCU read-side critical sections leaking down. */
335 trace_rcu_utilization(TPS("Start context switch"));
b04db8e1 336 lockdep_assert_irqs_disabled();
5b72f964 337 WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0);
10f39bb1 338 if (t->rcu_read_lock_nesting > 0 &&
1d082fd0 339 !t->rcu_read_unlock_special.b.blocked) {
f41d911f
PM
340
341 /* Possibly blocking in an RCU read-side critical section. */
f41d911f 342 rnp = rdp->mynode;
46a5d164 343 raw_spin_lock_rcu_node(rnp);
1d082fd0 344 t->rcu_read_unlock_special.b.blocked = true;
86848966 345 t->rcu_blocked_node = rnp;
f41d911f
PM
346
347 /*
8203d6d0
PM
348 * Verify the CPU's sanity, trace the preemption, and
349 * then queue the task as required based on the states
350 * of any ongoing and expedited grace periods.
f41d911f 351 */
0aa04b05 352 WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0);
e7d8842e 353 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
88d1bead 354 trace_rcu_preempt_task(rcu_state.name,
d4c08f2a
PM
355 t->pid,
356 (rnp->qsmask & rdp->grpmask)
598ce094
PM
357 ? rnp->gp_seq
358 : rcu_seq_snap(&rnp->gp_seq));
46a5d164 359 rcu_preempt_ctxt_queue(rnp, rdp);
10f39bb1 360 } else if (t->rcu_read_lock_nesting < 0 &&
1d082fd0 361 t->rcu_read_unlock_special.s) {
10f39bb1
PM
362
363 /*
364 * Complete exit from RCU read-side critical section on
365 * behalf of preempted instance of __rcu_read_unlock().
366 */
367 rcu_read_unlock_special(t);
3e310098
PM
368 rcu_preempt_deferred_qs(t);
369 } else {
370 rcu_preempt_deferred_qs(t);
f41d911f
PM
371 }
372
373 /*
374 * Either we were not in an RCU read-side critical section to
375 * begin with, or we have now recorded that critical section
376 * globally. Either way, we can now note a quiescent state
377 * for this CPU. Again, if we were in an RCU read-side critical
378 * section, and if that critical section was blocking the current
379 * grace period, then the fact that the task has been enqueued
380 * means that we continue to block the current grace period.
381 */
45975c7d 382 rcu_qs();
ba1c64c2 383 if (rdp->deferred_qs)
63d4c8c9 384 rcu_report_exp_rdp(rdp);
45975c7d
PM
385 trace_rcu_utilization(TPS("End context switch"));
386 barrier(); /* Avoid RCU read-side critical sections leaking up. */
f41d911f 387}
45975c7d 388EXPORT_SYMBOL_GPL(rcu_note_context_switch);
f41d911f 389
fc2219d4
PM
390/*
391 * Check for preempted RCU readers blocking the current grace period
392 * for the specified rcu_node structure. If the caller needs a reliable
393 * answer, it must hold the rcu_node's ->lock.
394 */
27f4d280 395static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
fc2219d4 396{
12f5f524 397 return rnp->gp_tasks != NULL;
fc2219d4
PM
398}
399
0e5da22e
PM
400/*
401 * Preemptible RCU implementation for rcu_read_lock().
402 * Just increment ->rcu_read_lock_nesting, shared state will be updated
403 * if we block.
404 */
405void __rcu_read_lock(void)
406{
407 current->rcu_read_lock_nesting++;
408 barrier(); /* critical section after entry code. */
409}
410EXPORT_SYMBOL_GPL(__rcu_read_lock);
411
412/*
413 * Preemptible RCU implementation for rcu_read_unlock().
414 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
415 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
416 * invoke rcu_read_unlock_special() to clean up after a context switch
417 * in an RCU read-side critical section and other special cases.
418 */
419void __rcu_read_unlock(void)
420{
421 struct task_struct *t = current;
422
423 if (t->rcu_read_lock_nesting != 1) {
424 --t->rcu_read_lock_nesting;
425 } else {
426 barrier(); /* critical section before exit code. */
427 t->rcu_read_lock_nesting = INT_MIN;
428 barrier(); /* assign before ->rcu_read_unlock_special load */
429 if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s)))
430 rcu_read_unlock_special(t);
431 barrier(); /* ->rcu_read_unlock_special load before assign */
432 t->rcu_read_lock_nesting = 0;
433 }
434#ifdef CONFIG_PROVE_LOCKING
435 {
436 int rrln = READ_ONCE(t->rcu_read_lock_nesting);
437
438 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
439 }
440#endif /* #ifdef CONFIG_PROVE_LOCKING */
441}
442EXPORT_SYMBOL_GPL(__rcu_read_unlock);
443
12f5f524
PM
444/*
445 * Advance a ->blkd_tasks-list pointer to the next entry, instead
446 * returning NULL if at the end of the list.
447 */
448static struct list_head *rcu_next_node_entry(struct task_struct *t,
449 struct rcu_node *rnp)
450{
451 struct list_head *np;
452
453 np = t->rcu_node_entry.next;
454 if (np == &rnp->blkd_tasks)
455 np = NULL;
456 return np;
457}
458
8af3a5e7
PM
459/*
460 * Return true if the specified rcu_node structure has tasks that were
461 * preempted within an RCU read-side critical section.
462 */
463static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
464{
465 return !list_empty(&rnp->blkd_tasks);
466}
467
b668c9cf 468/*
3e310098
PM
469 * Report deferred quiescent states. The deferral time can
470 * be quite short, for example, in the case of the call from
471 * rcu_read_unlock_special().
b668c9cf 472 */
3e310098
PM
473static void
474rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags)
f41d911f 475{
b6a932d1
PM
476 bool empty_exp;
477 bool empty_norm;
478 bool empty_exp_now;
12f5f524 479 struct list_head *np;
abaa93d9 480 bool drop_boost_mutex = false;
8203d6d0 481 struct rcu_data *rdp;
f41d911f 482 struct rcu_node *rnp;
1d082fd0 483 union rcu_special special;
f41d911f 484
f41d911f 485 /*
8203d6d0
PM
486 * If RCU core is waiting for this CPU to exit its critical section,
487 * report the fact that it has exited. Because irqs are disabled,
1d082fd0 488 * t->rcu_read_unlock_special cannot change.
f41d911f
PM
489 */
490 special = t->rcu_read_unlock_special;
da1df50d 491 rdp = this_cpu_ptr(&rcu_data);
3e310098
PM
492 if (!special.s && !rdp->deferred_qs) {
493 local_irq_restore(flags);
494 return;
495 }
1d082fd0 496 if (special.b.need_qs) {
45975c7d 497 rcu_qs();
c0135d07 498 t->rcu_read_unlock_special.b.need_qs = false;
3e310098 499 if (!t->rcu_read_unlock_special.s && !rdp->deferred_qs) {
79a62f95
LJ
500 local_irq_restore(flags);
501 return;
502 }
f41d911f
PM
503 }
504
8203d6d0 505 /*
3e310098
PM
506 * Respond to a request by an expedited grace period for a
507 * quiescent state from this CPU. Note that requests from
508 * tasks are handled when removing the task from the
509 * blocked-tasks list below.
8203d6d0 510 */
fcc878e4 511 if (rdp->deferred_qs) {
63d4c8c9 512 rcu_report_exp_rdp(rdp);
8203d6d0
PM
513 if (!t->rcu_read_unlock_special.s) {
514 local_irq_restore(flags);
515 return;
516 }
517 }
518
f41d911f 519 /* Clean up if blocked during RCU read-side critical section. */
1d082fd0
PM
520 if (special.b.blocked) {
521 t->rcu_read_unlock_special.b.blocked = false;
f41d911f 522
dd5d19ba 523 /*
0a0ba1c9 524 * Remove this task from the list it blocked on. The task
8ba9153b
PM
525 * now remains queued on the rcu_node corresponding to the
526 * CPU it first blocked on, so there is no longer any need
527 * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia.
dd5d19ba 528 */
8ba9153b
PM
529 rnp = t->rcu_blocked_node;
530 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
531 WARN_ON_ONCE(rnp != t->rcu_blocked_node);
5b4c11d5 532 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
74e871ac 533 empty_norm = !rcu_preempt_blocked_readers_cgp(rnp);
d43a5d32 534 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq &&
4bc8d555 535 (!empty_norm || rnp->qsmask));
8203d6d0 536 empty_exp = sync_rcu_preempt_exp_done(rnp);
d9a3da06 537 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
12f5f524 538 np = rcu_next_node_entry(t, rnp);
f41d911f 539 list_del_init(&t->rcu_node_entry);
82e78d80 540 t->rcu_blocked_node = NULL;
f7f7bac9 541 trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
865aa1e0 542 rnp->gp_seq, t->pid);
12f5f524
PM
543 if (&t->rcu_node_entry == rnp->gp_tasks)
544 rnp->gp_tasks = np;
545 if (&t->rcu_node_entry == rnp->exp_tasks)
546 rnp->exp_tasks = np;
727b705b 547 if (IS_ENABLED(CONFIG_RCU_BOOST)) {
727b705b
PM
548 /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
549 drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t;
2dee9404
PM
550 if (&t->rcu_node_entry == rnp->boost_tasks)
551 rnp->boost_tasks = np;
727b705b 552 }
f41d911f
PM
553
554 /*
555 * If this was the last task on the current list, and if
556 * we aren't waiting on any CPUs, report the quiescent state.
389abd48
PM
557 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
558 * so we must take a snapshot of the expedited state.
f41d911f 559 */
8203d6d0 560 empty_exp_now = sync_rcu_preempt_exp_done(rnp);
74e871ac 561 if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) {
f7f7bac9 562 trace_rcu_quiescent_state_report(TPS("preempt_rcu"),
db023296 563 rnp->gp_seq,
d4c08f2a
PM
564 0, rnp->qsmask,
565 rnp->level,
566 rnp->grplo,
567 rnp->grphi,
568 !!rnp->gp_tasks);
139ad4da 569 rcu_report_unblock_qs_rnp(rnp, flags);
c701d5d9 570 } else {
67c583a7 571 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
c701d5d9 572 }
d9a3da06 573
27f4d280 574 /* Unboost if we were boosted. */
727b705b 575 if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex)
02a7c234 576 rt_mutex_futex_unlock(&rnp->boost_mtx);
27f4d280 577
d9a3da06
PM
578 /*
579 * If this was the last task on the expedited lists,
580 * then we need to report up the rcu_node hierarchy.
581 */
389abd48 582 if (!empty_exp && empty_exp_now)
63d4c8c9 583 rcu_report_exp_rnp(rnp, true);
b668c9cf
PM
584 } else {
585 local_irq_restore(flags);
f41d911f 586 }
f41d911f
PM
587}
588
3e310098
PM
589/*
590 * Is a deferred quiescent-state pending, and are we also not in
591 * an RCU read-side critical section? It is the caller's responsibility
592 * to ensure it is otherwise safe to report any deferred quiescent
593 * states. The reason for this is that it is safe to report a
594 * quiescent state during context switch even though preemption
595 * is disabled. This function cannot be expected to understand these
596 * nuances, so the caller must handle them.
597 */
598static bool rcu_preempt_need_deferred_qs(struct task_struct *t)
599{
45975c7d 600 return (this_cpu_ptr(&rcu_data)->deferred_qs ||
3e310098 601 READ_ONCE(t->rcu_read_unlock_special.s)) &&
27c744e3 602 t->rcu_read_lock_nesting <= 0;
3e310098
PM
603}
604
605/*
606 * Report a deferred quiescent state if needed and safe to do so.
607 * As with rcu_preempt_need_deferred_qs(), "safe" involves only
608 * not being in an RCU read-side critical section. The caller must
609 * evaluate safety in terms of interrupt, softirq, and preemption
610 * disabling.
611 */
612static void rcu_preempt_deferred_qs(struct task_struct *t)
613{
614 unsigned long flags;
615 bool couldrecurse = t->rcu_read_lock_nesting >= 0;
616
617 if (!rcu_preempt_need_deferred_qs(t))
618 return;
619 if (couldrecurse)
620 t->rcu_read_lock_nesting -= INT_MIN;
621 local_irq_save(flags);
622 rcu_preempt_deferred_qs_irqrestore(t, flags);
623 if (couldrecurse)
624 t->rcu_read_lock_nesting += INT_MIN;
625}
626
627/*
628 * Handle special cases during rcu_read_unlock(), such as needing to
629 * notify RCU core processing or task having blocked during the RCU
630 * read-side critical section.
631 */
632static void rcu_read_unlock_special(struct task_struct *t)
633{
634 unsigned long flags;
635 bool preempt_bh_were_disabled =
636 !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK));
637 bool irqs_were_disabled;
638
639 /* NMI handlers cannot block and cannot safely manipulate state. */
640 if (in_nmi())
641 return;
642
643 local_irq_save(flags);
644 irqs_were_disabled = irqs_disabled_flags(flags);
645 if ((preempt_bh_were_disabled || irqs_were_disabled) &&
646 t->rcu_read_unlock_special.b.blocked) {
647 /* Need to defer quiescent state until everything is enabled. */
648 raise_softirq_irqoff(RCU_SOFTIRQ);
649 local_irq_restore(flags);
650 return;
651 }
652 rcu_preempt_deferred_qs_irqrestore(t, flags);
653}
654
1ed509a2
PM
655/*
656 * Dump detailed information for all tasks blocking the current RCU
657 * grace period on the specified rcu_node structure.
658 */
659static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
660{
661 unsigned long flags;
1ed509a2
PM
662 struct task_struct *t;
663
6cf10081 664 raw_spin_lock_irqsave_rcu_node(rnp, flags);
5fd4dc06 665 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
67c583a7 666 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
5fd4dc06
PM
667 return;
668 }
82efed06 669 t = list_entry(rnp->gp_tasks->prev,
12f5f524 670 struct task_struct, rcu_node_entry);
3caa973b
TH
671 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
672 /*
673 * We could be printing a lot while holding a spinlock.
674 * Avoid triggering hard lockup.
675 */
676 touch_nmi_watchdog();
12f5f524 677 sched_show_task(t);
3caa973b 678 }
67c583a7 679 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1ed509a2
PM
680}
681
682/*
683 * Dump detailed information for all tasks blocking the current RCU
684 * grace period.
685 */
a2887cd8 686static void rcu_print_detail_task_stall(void)
1ed509a2 687{
336a4f6c 688 struct rcu_node *rnp = rcu_get_root();
1ed509a2
PM
689
690 rcu_print_detail_task_stall_rnp(rnp);
aedf4ba9 691 rcu_for_each_leaf_node(rnp)
1ed509a2
PM
692 rcu_print_detail_task_stall_rnp(rnp);
693}
694
a858af28
PM
695static void rcu_print_task_stall_begin(struct rcu_node *rnp)
696{
efc151c3 697 pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
a858af28
PM
698 rnp->level, rnp->grplo, rnp->grphi);
699}
700
701static void rcu_print_task_stall_end(void)
702{
efc151c3 703 pr_cont("\n");
a858af28
PM
704}
705
f41d911f
PM
706/*
707 * Scan the current list of tasks blocked within RCU read-side critical
708 * sections, printing out the tid of each.
709 */
9bc8b558 710static int rcu_print_task_stall(struct rcu_node *rnp)
f41d911f 711{
f41d911f 712 struct task_struct *t;
9bc8b558 713 int ndetected = 0;
f41d911f 714
27f4d280 715 if (!rcu_preempt_blocked_readers_cgp(rnp))
9bc8b558 716 return 0;
a858af28 717 rcu_print_task_stall_begin(rnp);
82efed06 718 t = list_entry(rnp->gp_tasks->prev,
12f5f524 719 struct task_struct, rcu_node_entry);
9bc8b558 720 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
efc151c3 721 pr_cont(" P%d", t->pid);
9bc8b558
PM
722 ndetected++;
723 }
a858af28 724 rcu_print_task_stall_end();
9bc8b558 725 return ndetected;
f41d911f
PM
726}
727
74611ecb
PM
728/*
729 * Scan the current list of tasks blocked within RCU read-side critical
730 * sections, printing out the tid of each that is blocking the current
731 * expedited grace period.
732 */
733static int rcu_print_task_exp_stall(struct rcu_node *rnp)
734{
735 struct task_struct *t;
736 int ndetected = 0;
737
738 if (!rnp->exp_tasks)
739 return 0;
740 t = list_entry(rnp->exp_tasks->prev,
741 struct task_struct, rcu_node_entry);
742 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
743 pr_cont(" P%d", t->pid);
744 ndetected++;
745 }
746 return ndetected;
747}
748
b0e165c0
PM
749/*
750 * Check that the list of blocked tasks for the newly completed grace
751 * period is in fact empty. It is a serious bug to complete a grace
752 * period that still has RCU readers blocked! This function must be
ff3bb6f4 753 * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock
b0e165c0 754 * must be held by the caller.
12f5f524
PM
755 *
756 * Also, if there are blocked tasks on the list, they automatically
757 * block the newly created grace period, so set up ->gp_tasks accordingly.
b0e165c0 758 */
81ab59a3 759static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
b0e165c0 760{
c5ebe66c
PM
761 struct task_struct *t;
762
ea9b0c8a 763 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
4bc8d555 764 if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
81ab59a3 765 dump_blkd_tasks(rnp, 10);
0b107d24
PM
766 if (rcu_preempt_has_tasks(rnp) &&
767 (rnp->qsmaskinit || rnp->wait_blkd_tasks)) {
12f5f524 768 rnp->gp_tasks = rnp->blkd_tasks.next;
c5ebe66c
PM
769 t = container_of(rnp->gp_tasks, struct task_struct,
770 rcu_node_entry);
771 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
865aa1e0 772 rnp->gp_seq, t->pid);
c5ebe66c 773 }
28ecd580 774 WARN_ON_ONCE(rnp->qsmask);
b0e165c0
PM
775}
776
f41d911f
PM
777/*
778 * Check for a quiescent state from the current CPU. When a task blocks,
779 * the task is recorded in the corresponding CPU's rcu_node structure,
780 * which is checked elsewhere.
781 *
782 * Caller must disable hard irqs.
783 */
45975c7d 784static void rcu_flavor_check_callbacks(int user)
f41d911f
PM
785{
786 struct task_struct *t = current;
787
45975c7d
PM
788 if (user || rcu_is_cpu_rrupt_from_idle()) {
789 rcu_note_voluntary_context_switch(current);
790 }
3e310098
PM
791 if (t->rcu_read_lock_nesting > 0 ||
792 (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) {
793 /* No QS, force context switch if deferred. */
fced9c8c
PM
794 if (rcu_preempt_need_deferred_qs(t)) {
795 set_tsk_need_resched(t);
796 set_preempt_need_resched();
797 }
3e310098
PM
798 } else if (rcu_preempt_need_deferred_qs(t)) {
799 rcu_preempt_deferred_qs(t); /* Report deferred QS. */
800 return;
801 } else if (!t->rcu_read_lock_nesting) {
45975c7d 802 rcu_qs(); /* Report immediate QS. */
f41d911f
PM
803 return;
804 }
3e310098
PM
805
806 /* If GP is oldish, ask for help from rcu_read_unlock_special(). */
10f39bb1 807 if (t->rcu_read_lock_nesting > 0 &&
2280ee5a
PM
808 __this_cpu_read(rcu_data.core_needs_qs) &&
809 __this_cpu_read(rcu_data.cpu_no_qs.b.norm) &&
15651201 810 !t->rcu_read_unlock_special.b.need_qs &&
564a9ae6 811 time_after(jiffies, rcu_state.gp_start + HZ))
1d082fd0 812 t->rcu_read_unlock_special.b.need_qs = true;
f41d911f
PM
813}
814
6ebb237b
PM
815/**
816 * synchronize_rcu - wait until a grace period has elapsed.
817 *
818 * Control will return to the caller some time after a full grace
819 * period has elapsed, in other words after all currently executing RCU
77d8485a
PM
820 * read-side critical sections have completed. Note, however, that
821 * upon return from synchronize_rcu(), the caller might well be executing
822 * concurrently with new RCU read-side critical sections that began while
823 * synchronize_rcu() was waiting. RCU read-side critical sections are
824 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
45975c7d
PM
825 * In addition, regions of code across which interrupts, preemption, or
826 * softirqs have been disabled also serve as RCU read-side critical
827 * sections. This includes hardware interrupt handlers, softirq handlers,
828 * and NMI handlers.
829 *
830 * Note that this guarantee implies further memory-ordering guarantees.
831 * On systems with more than one CPU, when synchronize_rcu() returns,
0ae86a27
PM
832 * each CPU is guaranteed to have executed a full memory barrier since
833 * the end of its last RCU read-side critical section whose beginning
45975c7d
PM
834 * preceded the call to synchronize_rcu(). In addition, each CPU having
835 * an RCU read-side critical section that extends beyond the return from
836 * synchronize_rcu() is guaranteed to have executed a full memory barrier
837 * after the beginning of synchronize_rcu() and before the beginning of
838 * that RCU read-side critical section. Note that these guarantees include
839 * CPUs that are offline, idle, or executing in user mode, as well as CPUs
840 * that are executing in the kernel.
f0a0e6f2 841 *
45975c7d
PM
842 * Furthermore, if CPU A invoked synchronize_rcu(), which returned
843 * to its caller on CPU B, then both CPU A and CPU B are guaranteed
844 * to have executed a full memory barrier during the execution of
845 * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but
846 * again only if the system has more than one CPU).
6ebb237b
PM
847 */
848void synchronize_rcu(void)
849{
f78f5b90
PM
850 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
851 lock_is_held(&rcu_lock_map) ||
852 lock_is_held(&rcu_sched_lock_map),
853 "Illegal synchronize_rcu() in RCU read-side critical section");
52d7e48b 854 if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
6ebb237b 855 return;
5afff48b 856 if (rcu_gp_is_expedited())
3705b88d
AM
857 synchronize_rcu_expedited();
858 else
859 wait_rcu_gp(call_rcu);
6ebb237b
PM
860}
861EXPORT_SYMBOL_GPL(synchronize_rcu);
862
2439b696
PM
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,
866 * as debug_check_no_locks_held() already does this if lockdep
867 * is enabled.
868 */
869void exit_rcu(void)
870{
871 struct task_struct *t = current;
872
873 if (likely(list_empty(&current->rcu_node_entry)))
874 return;
875 t->rcu_read_lock_nesting = 1;
876 barrier();
1d082fd0 877 t->rcu_read_unlock_special.b.blocked = true;
2439b696 878 __rcu_read_unlock();
3e310098 879 rcu_preempt_deferred_qs(current);
2439b696
PM
880}
881
4bc8d555
PM
882/*
883 * Dump the blocked-tasks state, but limit the list dump to the
884 * specified number of elements.
885 */
57738942 886static void
81ab59a3 887dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
4bc8d555 888{
57738942 889 int cpu;
4bc8d555
PM
890 int i;
891 struct list_head *lhp;
57738942
PM
892 bool onl;
893 struct rcu_data *rdp;
ff3cee39 894 struct rcu_node *rnp1;
4bc8d555 895
ce11fae8 896 raw_lockdep_assert_held_rcu_node(rnp);
ff3cee39 897 pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
77cfc7bf 898 __func__, rnp->grplo, rnp->grphi, rnp->level,
ff3cee39
PM
899 (long)rnp->gp_seq, (long)rnp->completedqs);
900 for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
901 pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n",
902 __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext);
77cfc7bf
PM
903 pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n",
904 __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks);
905 pr_info("%s: ->blkd_tasks", __func__);
4bc8d555
PM
906 i = 0;
907 list_for_each(lhp, &rnp->blkd_tasks) {
908 pr_cont(" %p", lhp);
909 if (++i >= 10)
910 break;
911 }
912 pr_cont("\n");
57738942 913 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) {
da1df50d 914 rdp = per_cpu_ptr(&rcu_data, cpu);
57738942
PM
915 onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp));
916 pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n",
917 cpu, ".o"[onl],
918 (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags,
919 (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags);
920 }
4bc8d555
PM
921}
922
28f6569a 923#else /* #ifdef CONFIG_PREEMPT_RCU */
f41d911f
PM
924
925/*
926 * Tell them what RCU they are running.
927 */
0e0fc1c2 928static void __init rcu_bootup_announce(void)
f41d911f 929{
efc151c3 930 pr_info("Hierarchical RCU implementation.\n");
26845c28 931 rcu_bootup_announce_oddness();
f41d911f
PM
932}
933
45975c7d
PM
934/*
935 * Note a quiescent state for PREEMPT=n. Because we do not need to know
936 * how many quiescent states passed, just if there was at least one since
937 * the start of the grace period, this just sets a flag. The caller must
938 * have disabled preemption.
939 */
940static void rcu_qs(void)
d28139c4 941{
45975c7d
PM
942 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!");
943 if (!__this_cpu_read(rcu_data.cpu_no_qs.s))
944 return;
945 trace_rcu_grace_period(TPS("rcu_sched"),
946 __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs"));
947 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
948 if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp))
949 return;
950 __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false);
63d4c8c9 951 rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
d28139c4
PM
952}
953
395a2f09
PM
954/*
955 * Register an urgently needed quiescent state. If there is an
956 * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight
957 * dyntick-idle quiescent state visible to other CPUs, which will in
958 * some cases serve for expedited as well as normal grace periods.
959 * Either way, register a lightweight quiescent state.
960 *
961 * The barrier() calls are redundant in the common case when this is
962 * called externally, but just in case this is called from within this
963 * file.
964 *
965 */
966void rcu_all_qs(void)
967{
968 unsigned long flags;
969
2dba13f0 970 if (!raw_cpu_read(rcu_data.rcu_urgent_qs))
395a2f09
PM
971 return;
972 preempt_disable();
973 /* Load rcu_urgent_qs before other flags. */
2dba13f0 974 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
395a2f09
PM
975 preempt_enable();
976 return;
977 }
2dba13f0 978 this_cpu_write(rcu_data.rcu_urgent_qs, false);
395a2f09 979 barrier(); /* Avoid RCU read-side critical sections leaking down. */
2dba13f0 980 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) {
395a2f09
PM
981 local_irq_save(flags);
982 rcu_momentary_dyntick_idle();
983 local_irq_restore(flags);
984 }
7e28c5af 985 rcu_qs();
395a2f09
PM
986 barrier(); /* Avoid RCU read-side critical sections leaking up. */
987 preempt_enable();
988}
989EXPORT_SYMBOL_GPL(rcu_all_qs);
990
cba6d0d6 991/*
45975c7d 992 * Note a PREEMPT=n context switch. The caller must have disabled interrupts.
cba6d0d6 993 */
45975c7d 994void rcu_note_context_switch(bool preempt)
cba6d0d6 995{
45975c7d
PM
996 barrier(); /* Avoid RCU read-side critical sections leaking down. */
997 trace_rcu_utilization(TPS("Start context switch"));
998 rcu_qs();
999 /* Load rcu_urgent_qs before other flags. */
2dba13f0 1000 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs)))
45975c7d 1001 goto out;
2dba13f0
PM
1002 this_cpu_write(rcu_data.rcu_urgent_qs, false);
1003 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs)))
45975c7d 1004 rcu_momentary_dyntick_idle();
45975c7d
PM
1005 if (!preempt)
1006 rcu_tasks_qs(current);
1007out:
1008 trace_rcu_utilization(TPS("End context switch"));
1009 barrier(); /* Avoid RCU read-side critical sections leaking up. */
cba6d0d6 1010}
45975c7d 1011EXPORT_SYMBOL_GPL(rcu_note_context_switch);
cba6d0d6 1012
fc2219d4 1013/*
6cc68793 1014 * Because preemptible RCU does not exist, there are never any preempted
fc2219d4
PM
1015 * RCU readers.
1016 */
27f4d280 1017static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
fc2219d4
PM
1018{
1019 return 0;
1020}
1021
8af3a5e7
PM
1022/*
1023 * Because there is no preemptible RCU, there can be no readers blocked.
1024 */
1025static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
b668c9cf 1026{
8af3a5e7 1027 return false;
b668c9cf
PM
1028}
1029
3e310098
PM
1030/*
1031 * Because there is no preemptible RCU, there can be no deferred quiescent
1032 * states.
1033 */
1034static bool rcu_preempt_need_deferred_qs(struct task_struct *t)
1035{
1036 return false;
1037}
1038static void rcu_preempt_deferred_qs(struct task_struct *t) { }
1039
1ed509a2 1040/*
6cc68793 1041 * Because preemptible RCU does not exist, we never have to check for
1ed509a2
PM
1042 * tasks blocked within RCU read-side critical sections.
1043 */
a2887cd8 1044static void rcu_print_detail_task_stall(void)
1ed509a2
PM
1045{
1046}
1047
f41d911f 1048/*
6cc68793 1049 * Because preemptible RCU does not exist, we never have to check for
f41d911f
PM
1050 * tasks blocked within RCU read-side critical sections.
1051 */
9bc8b558 1052static int rcu_print_task_stall(struct rcu_node *rnp)
f41d911f 1053{
9bc8b558 1054 return 0;
f41d911f
PM
1055}
1056
74611ecb
PM
1057/*
1058 * Because preemptible RCU does not exist, we never have to check for
1059 * tasks blocked within RCU read-side critical sections that are
1060 * blocking the current expedited grace period.
1061 */
1062static int rcu_print_task_exp_stall(struct rcu_node *rnp)
1063{
1064 return 0;
1065}
1066
b0e165c0 1067/*
6cc68793 1068 * Because there is no preemptible RCU, there can be no readers blocked,
49e29126
PM
1069 * so there is no need to check for blocked tasks. So check only for
1070 * bogus qsmask values.
b0e165c0 1071 */
81ab59a3 1072static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
b0e165c0 1073{
49e29126 1074 WARN_ON_ONCE(rnp->qsmask);
b0e165c0
PM
1075}
1076
f41d911f 1077/*
45975c7d
PM
1078 * Check to see if this CPU is in a non-context-switch quiescent state
1079 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1080 * Also schedule RCU core processing.
1081 *
1082 * This function must be called from hardirq context. It is normally
1083 * invoked from the scheduling-clock interrupt.
f41d911f 1084 */
45975c7d 1085static void rcu_flavor_check_callbacks(int user)
f41d911f 1086{
45975c7d 1087 if (user || rcu_is_cpu_rrupt_from_idle()) {
f41d911f 1088
45975c7d
PM
1089 /*
1090 * Get here if this CPU took its interrupt from user
1091 * mode or from the idle loop, and if this is not a
1092 * nested interrupt. In this case, the CPU is in
1093 * a quiescent state, so note it.
1094 *
1095 * No memory barrier is required here because rcu_qs()
1096 * references only CPU-local variables that other CPUs
1097 * neither access nor modify, at least not while the
1098 * corresponding CPU is online.
1099 */
1100
1101 rcu_qs();
1102 }
e74f4c45 1103}
e74f4c45 1104
45975c7d
PM
1105/* PREEMPT=n implementation of synchronize_rcu(). */
1106void synchronize_rcu(void)
1eba8f84 1107{
45975c7d
PM
1108 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
1109 lock_is_held(&rcu_lock_map) ||
1110 lock_is_held(&rcu_sched_lock_map),
0ae86a27 1111 "Illegal synchronize_rcu() in RCU read-side critical section");
45975c7d
PM
1112 if (rcu_blocking_is_gp())
1113 return;
1114 if (rcu_gp_is_expedited())
1115 synchronize_rcu_expedited();
1116 else
1117 wait_rcu_gp(call_rcu);
1eba8f84 1118}
45975c7d 1119EXPORT_SYMBOL_GPL(synchronize_rcu);
1eba8f84 1120
2439b696
PM
1121/*
1122 * Because preemptible RCU does not exist, tasks cannot possibly exit
1123 * while in preemptible RCU read-side critical sections.
1124 */
1125void exit_rcu(void)
1126{
1127}
1128
4bc8d555
PM
1129/*
1130 * Dump the guaranteed-empty blocked-tasks state. Trust but verify.
1131 */
57738942 1132static void
81ab59a3 1133dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
4bc8d555
PM
1134{
1135 WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks));
1136}
1137
28f6569a 1138#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
8bd93a2c 1139
27f4d280
PM
1140#ifdef CONFIG_RCU_BOOST
1141
5d01bbd1
TG
1142static void rcu_wake_cond(struct task_struct *t, int status)
1143{
1144 /*
1145 * If the thread is yielding, only wake it when this
1146 * is invoked from idle
1147 */
1148 if (status != RCU_KTHREAD_YIELDING || is_idle_task(current))
1149 wake_up_process(t);
1150}
1151
27f4d280
PM
1152/*
1153 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1154 * or ->boost_tasks, advancing the pointer to the next task in the
1155 * ->blkd_tasks list.
1156 *
1157 * Note that irqs must be enabled: boosting the task can block.
1158 * Returns 1 if there are more tasks needing to be boosted.
1159 */
1160static int rcu_boost(struct rcu_node *rnp)
1161{
1162 unsigned long flags;
27f4d280
PM
1163 struct task_struct *t;
1164 struct list_head *tb;
1165
7d0ae808
PM
1166 if (READ_ONCE(rnp->exp_tasks) == NULL &&
1167 READ_ONCE(rnp->boost_tasks) == NULL)
27f4d280
PM
1168 return 0; /* Nothing left to boost. */
1169
2a67e741 1170 raw_spin_lock_irqsave_rcu_node(rnp, flags);
27f4d280
PM
1171
1172 /*
1173 * Recheck under the lock: all tasks in need of boosting
1174 * might exit their RCU read-side critical sections on their own.
1175 */
1176 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
67c583a7 1177 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
27f4d280
PM
1178 return 0;
1179 }
1180
1181 /*
1182 * Preferentially boost tasks blocking expedited grace periods.
1183 * This cannot starve the normal grace periods because a second
1184 * expedited grace period must boost all blocked tasks, including
1185 * those blocking the pre-existing normal grace period.
1186 */
bec06785 1187 if (rnp->exp_tasks != NULL)
27f4d280 1188 tb = rnp->exp_tasks;
bec06785 1189 else
27f4d280
PM
1190 tb = rnp->boost_tasks;
1191
1192 /*
1193 * We boost task t by manufacturing an rt_mutex that appears to
1194 * be held by task t. We leave a pointer to that rt_mutex where
1195 * task t can find it, and task t will release the mutex when it
1196 * exits its outermost RCU read-side critical section. Then
1197 * simply acquiring this artificial rt_mutex will boost task
1198 * t's priority. (Thanks to tglx for suggesting this approach!)
1199 *
1200 * Note that task t must acquire rnp->lock to remove itself from
1201 * the ->blkd_tasks list, which it will do from exit() if from
1202 * nowhere else. We therefore are guaranteed that task t will
1203 * stay around at least until we drop rnp->lock. Note that
1204 * rnp->lock also resolves races between our priority boosting
1205 * and task t's exiting its outermost RCU read-side critical
1206 * section.
1207 */
1208 t = container_of(tb, struct task_struct, rcu_node_entry);
abaa93d9 1209 rt_mutex_init_proxy_locked(&rnp->boost_mtx, t);
67c583a7 1210 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
abaa93d9
PM
1211 /* Lock only for side effect: boosts task t's priority. */
1212 rt_mutex_lock(&rnp->boost_mtx);
1213 rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
27f4d280 1214
7d0ae808
PM
1215 return READ_ONCE(rnp->exp_tasks) != NULL ||
1216 READ_ONCE(rnp->boost_tasks) != NULL;
27f4d280
PM
1217}
1218
27f4d280 1219/*
bc17ea10 1220 * Priority-boosting kthread, one per leaf rcu_node.
27f4d280
PM
1221 */
1222static int rcu_boost_kthread(void *arg)
1223{
1224 struct rcu_node *rnp = (struct rcu_node *)arg;
1225 int spincnt = 0;
1226 int more2boost;
1227
f7f7bac9 1228 trace_rcu_utilization(TPS("Start boost kthread@init"));
27f4d280 1229 for (;;) {
d71df90e 1230 rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
f7f7bac9 1231 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
08bca60a 1232 rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
f7f7bac9 1233 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
d71df90e 1234 rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
27f4d280
PM
1235 more2boost = rcu_boost(rnp);
1236 if (more2boost)
1237 spincnt++;
1238 else
1239 spincnt = 0;
1240 if (spincnt > 10) {
5d01bbd1 1241 rnp->boost_kthread_status = RCU_KTHREAD_YIELDING;
f7f7bac9 1242 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
5d01bbd1 1243 schedule_timeout_interruptible(2);
f7f7bac9 1244 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
27f4d280
PM
1245 spincnt = 0;
1246 }
1247 }
1217ed1b 1248 /* NOTREACHED */
f7f7bac9 1249 trace_rcu_utilization(TPS("End boost kthread@notreached"));
27f4d280
PM
1250 return 0;
1251}
1252
1253/*
1254 * Check to see if it is time to start boosting RCU readers that are
1255 * blocking the current grace period, and, if so, tell the per-rcu_node
1256 * kthread to start boosting them. If there is an expedited grace
1257 * period in progress, it is always time to boost.
1258 *
b065a853
PM
1259 * The caller must hold rnp->lock, which this function releases.
1260 * The ->boost_kthread_task is immortal, so we don't need to worry
1261 * about it going away.
27f4d280 1262 */
1217ed1b 1263static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
615e41c6 1264 __releases(rnp->lock)
27f4d280
PM
1265{
1266 struct task_struct *t;
1267
a32e01ee 1268 raw_lockdep_assert_held_rcu_node(rnp);
0ea1f2eb 1269 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
67c583a7 1270 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
27f4d280 1271 return;
0ea1f2eb 1272 }
27f4d280
PM
1273 if (rnp->exp_tasks != NULL ||
1274 (rnp->gp_tasks != NULL &&
1275 rnp->boost_tasks == NULL &&
1276 rnp->qsmask == 0 &&
1277 ULONG_CMP_GE(jiffies, rnp->boost_time))) {
1278 if (rnp->exp_tasks == NULL)
1279 rnp->boost_tasks = rnp->gp_tasks;
67c583a7 1280 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
27f4d280 1281 t = rnp->boost_kthread_task;
5d01bbd1
TG
1282 if (t)
1283 rcu_wake_cond(t, rnp->boost_kthread_status);
1217ed1b 1284 } else {
67c583a7 1285 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1217ed1b 1286 }
27f4d280
PM
1287}
1288
a46e0899
PM
1289/*
1290 * Wake up the per-CPU kthread to invoke RCU callbacks.
1291 */
1292static void invoke_rcu_callbacks_kthread(void)
1293{
1294 unsigned long flags;
1295
1296 local_irq_save(flags);
1297 __this_cpu_write(rcu_cpu_has_work, 1);
1eb52121 1298 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
5d01bbd1
TG
1299 current != __this_cpu_read(rcu_cpu_kthread_task)) {
1300 rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task),
1301 __this_cpu_read(rcu_cpu_kthread_status));
1302 }
a46e0899
PM
1303 local_irq_restore(flags);
1304}
1305
dff1672d
PM
1306/*
1307 * Is the current CPU running the RCU-callbacks kthread?
1308 * Caller must have preemption disabled.
1309 */
1310static bool rcu_is_callbacks_kthread(void)
1311{
c9d4b0af 1312 return __this_cpu_read(rcu_cpu_kthread_task) == current;
dff1672d
PM
1313}
1314
27f4d280
PM
1315#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1316
1317/*
1318 * Do priority-boost accounting for the start of a new grace period.
1319 */
1320static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1321{
1322 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1323}
1324
27f4d280
PM
1325/*
1326 * Create an RCU-boost kthread for the specified node if one does not
1327 * already exist. We only create this kthread for preemptible RCU.
1328 * Returns zero if all is well, a negated errno otherwise.
1329 */
6dbfdc14 1330static int rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
27f4d280 1331{
6dbfdc14 1332 int rnp_index = rnp - rcu_get_root();
27f4d280
PM
1333 unsigned long flags;
1334 struct sched_param sp;
1335 struct task_struct *t;
1336
6dbfdc14 1337 if (!IS_ENABLED(CONFIG_PREEMPT_RCU))
27f4d280 1338 return 0;
5d01bbd1 1339
0aa04b05 1340 if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0)
5d01bbd1
TG
1341 return 0;
1342
6dbfdc14 1343 rcu_state.boost = 1;
27f4d280
PM
1344 if (rnp->boost_kthread_task != NULL)
1345 return 0;
1346 t = kthread_create(rcu_boost_kthread, (void *)rnp,
5b61b0ba 1347 "rcub/%d", rnp_index);
27f4d280
PM
1348 if (IS_ERR(t))
1349 return PTR_ERR(t);
2a67e741 1350 raw_spin_lock_irqsave_rcu_node(rnp, flags);
27f4d280 1351 rnp->boost_kthread_task = t;
67c583a7 1352 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
21871d7e 1353 sp.sched_priority = kthread_prio;
27f4d280 1354 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
9a432736 1355 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
27f4d280
PM
1356 return 0;
1357}
1358
f8b7fc6b
PM
1359static void rcu_kthread_do_work(void)
1360{
5bb5d09c 1361 rcu_do_batch(this_cpu_ptr(&rcu_data));
f8b7fc6b
PM
1362}
1363
62ab7072 1364static void rcu_cpu_kthread_setup(unsigned int cpu)
f8b7fc6b 1365{
f8b7fc6b 1366 struct sched_param sp;
f8b7fc6b 1367
21871d7e 1368 sp.sched_priority = kthread_prio;
62ab7072 1369 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
f8b7fc6b
PM
1370}
1371
62ab7072 1372static void rcu_cpu_kthread_park(unsigned int cpu)
f8b7fc6b 1373{
62ab7072 1374 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
f8b7fc6b
PM
1375}
1376
62ab7072 1377static int rcu_cpu_kthread_should_run(unsigned int cpu)
f8b7fc6b 1378{
c9d4b0af 1379 return __this_cpu_read(rcu_cpu_has_work);
f8b7fc6b
PM
1380}
1381
1382/*
0ae86a27
PM
1383 * Per-CPU kernel thread that invokes RCU callbacks. This replaces
1384 * the RCU softirq used in configurations of RCU that do not support RCU
1385 * priority boosting.
f8b7fc6b 1386 */
62ab7072 1387static void rcu_cpu_kthread(unsigned int cpu)
f8b7fc6b 1388{
c9d4b0af
CL
1389 unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status);
1390 char work, *workp = this_cpu_ptr(&rcu_cpu_has_work);
62ab7072 1391 int spincnt;
f8b7fc6b 1392
62ab7072 1393 for (spincnt = 0; spincnt < 10; spincnt++) {
f7f7bac9 1394 trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
f8b7fc6b 1395 local_bh_disable();
f8b7fc6b 1396 *statusp = RCU_KTHREAD_RUNNING;
62ab7072
PM
1397 this_cpu_inc(rcu_cpu_kthread_loops);
1398 local_irq_disable();
f8b7fc6b
PM
1399 work = *workp;
1400 *workp = 0;
62ab7072 1401 local_irq_enable();
f8b7fc6b
PM
1402 if (work)
1403 rcu_kthread_do_work();
1404 local_bh_enable();
62ab7072 1405 if (*workp == 0) {
f7f7bac9 1406 trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
62ab7072
PM
1407 *statusp = RCU_KTHREAD_WAITING;
1408 return;
f8b7fc6b
PM
1409 }
1410 }
62ab7072 1411 *statusp = RCU_KTHREAD_YIELDING;
f7f7bac9 1412 trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
62ab7072 1413 schedule_timeout_interruptible(2);
f7f7bac9 1414 trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
62ab7072 1415 *statusp = RCU_KTHREAD_WAITING;
f8b7fc6b
PM
1416}
1417
1418/*
1419 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1420 * served by the rcu_node in question. The CPU hotplug lock is still
1421 * held, so the value of rnp->qsmaskinit will be stable.
1422 *
1423 * We don't include outgoingcpu in the affinity set, use -1 if there is
1424 * no outgoing CPU. If there are no CPUs left in the affinity set,
1425 * this function allows the kthread to execute on any CPU.
1426 */
5d01bbd1 1427static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
f8b7fc6b 1428{
5d01bbd1 1429 struct task_struct *t = rnp->boost_kthread_task;
0aa04b05 1430 unsigned long mask = rcu_rnp_online_cpus(rnp);
f8b7fc6b
PM
1431 cpumask_var_t cm;
1432 int cpu;
f8b7fc6b 1433
5d01bbd1 1434 if (!t)
f8b7fc6b 1435 return;
5d01bbd1 1436 if (!zalloc_cpumask_var(&cm, GFP_KERNEL))
f8b7fc6b 1437 return;
bc75e999
MR
1438 for_each_leaf_node_possible_cpu(rnp, cpu)
1439 if ((mask & leaf_node_cpu_bit(rnp, cpu)) &&
1440 cpu != outgoingcpu)
f8b7fc6b 1441 cpumask_set_cpu(cpu, cm);
5d0b0249 1442 if (cpumask_weight(cm) == 0)
f8b7fc6b 1443 cpumask_setall(cm);
5d01bbd1 1444 set_cpus_allowed_ptr(t, cm);
f8b7fc6b
PM
1445 free_cpumask_var(cm);
1446}
1447
62ab7072
PM
1448static struct smp_hotplug_thread rcu_cpu_thread_spec = {
1449 .store = &rcu_cpu_kthread_task,
1450 .thread_should_run = rcu_cpu_kthread_should_run,
1451 .thread_fn = rcu_cpu_kthread,
1452 .thread_comm = "rcuc/%u",
1453 .setup = rcu_cpu_kthread_setup,
1454 .park = rcu_cpu_kthread_park,
1455};
f8b7fc6b
PM
1456
1457/*
9386c0b7 1458 * Spawn boost kthreads -- called as soon as the scheduler is running.
f8b7fc6b 1459 */
9386c0b7 1460static void __init rcu_spawn_boost_kthreads(void)
f8b7fc6b 1461{
f8b7fc6b 1462 struct rcu_node *rnp;
5d01bbd1 1463 int cpu;
f8b7fc6b 1464
62ab7072 1465 for_each_possible_cpu(cpu)
f8b7fc6b 1466 per_cpu(rcu_cpu_has_work, cpu) = 0;
62ab7072 1467 BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec));
aedf4ba9 1468 rcu_for_each_leaf_node(rnp)
6dbfdc14 1469 (void)rcu_spawn_one_boost_kthread(rnp);
f8b7fc6b 1470}
f8b7fc6b 1471
49fb4c62 1472static void rcu_prepare_kthreads(int cpu)
f8b7fc6b 1473{
da1df50d 1474 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
f8b7fc6b
PM
1475 struct rcu_node *rnp = rdp->mynode;
1476
1477 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
62ab7072 1478 if (rcu_scheduler_fully_active)
6dbfdc14 1479 (void)rcu_spawn_one_boost_kthread(rnp);
f8b7fc6b
PM
1480}
1481
27f4d280
PM
1482#else /* #ifdef CONFIG_RCU_BOOST */
1483
1217ed1b 1484static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
615e41c6 1485 __releases(rnp->lock)
27f4d280 1486{
67c583a7 1487 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
27f4d280
PM
1488}
1489
a46e0899 1490static void invoke_rcu_callbacks_kthread(void)
27f4d280 1491{
a46e0899 1492 WARN_ON_ONCE(1);
27f4d280
PM
1493}
1494
dff1672d
PM
1495static bool rcu_is_callbacks_kthread(void)
1496{
1497 return false;
1498}
1499
27f4d280
PM
1500static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1501{
1502}
1503
5d01bbd1 1504static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
f8b7fc6b
PM
1505{
1506}
1507
9386c0b7 1508static void __init rcu_spawn_boost_kthreads(void)
b0d30417 1509{
b0d30417 1510}
b0d30417 1511
49fb4c62 1512static void rcu_prepare_kthreads(int cpu)
f8b7fc6b
PM
1513{
1514}
1515
27f4d280
PM
1516#endif /* #else #ifdef CONFIG_RCU_BOOST */
1517
8bd93a2c
PM
1518#if !defined(CONFIG_RCU_FAST_NO_HZ)
1519
1520/*
1521 * Check to see if any future RCU-related work will need to be done
1522 * by the current CPU, even if none need be done immediately, returning
1523 * 1 if so. This function is part of the RCU implementation; it is -not-
1524 * an exported member of the RCU API.
1525 *
0ae86a27
PM
1526 * Because we not have RCU_FAST_NO_HZ, just check whether or not this
1527 * CPU has RCU callbacks queued.
8bd93a2c 1528 */
c1ad348b 1529int rcu_needs_cpu(u64 basemono, u64 *nextevt)
8bd93a2c 1530{
c1ad348b 1531 *nextevt = KTIME_MAX;
44c65ff2 1532 return rcu_cpu_has_callbacks(NULL);
7cb92499
PM
1533}
1534
1535/*
1536 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1537 * after it.
1538 */
8fa7845d 1539static void rcu_cleanup_after_idle(void)
7cb92499
PM
1540{
1541}
1542
aea1b35e 1543/*
a858af28 1544 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
aea1b35e
PM
1545 * is nothing.
1546 */
198bbf81 1547static void rcu_prepare_for_idle(void)
aea1b35e
PM
1548{
1549}
1550
c57afe80
PM
1551/*
1552 * Don't bother keeping a running count of the number of RCU callbacks
1553 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1554 */
1555static void rcu_idle_count_callbacks_posted(void)
1556{
1557}
1558
8bd93a2c
PM
1559#else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1560
f23f7fa1
PM
1561/*
1562 * This code is invoked when a CPU goes idle, at which point we want
1563 * to have the CPU do everything required for RCU so that it can enter
1564 * the energy-efficient dyntick-idle mode. This is handled by a
1565 * state machine implemented by rcu_prepare_for_idle() below.
1566 *
1567 * The following three proprocessor symbols control this state machine:
1568 *
f23f7fa1
PM
1569 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1570 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1571 * is sized to be roughly one RCU grace period. Those energy-efficiency
1572 * benchmarkers who might otherwise be tempted to set this to a large
1573 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1574 * system. And if you are -that- concerned about energy efficiency,
1575 * just power the system down and be done with it!
778d250a
PM
1576 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1577 * permitted to sleep in dyntick-idle mode with only lazy RCU
1578 * callbacks pending. Setting this too high can OOM your system.
f23f7fa1
PM
1579 *
1580 * The values below work well in practice. If future workloads require
1581 * adjustment, they can be converted into kernel config parameters, though
1582 * making the state machine smarter might be a better option.
1583 */
e84c48ae 1584#define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
778d250a 1585#define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
f23f7fa1 1586
5e44ce35
PM
1587static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY;
1588module_param(rcu_idle_gp_delay, int, 0644);
1589static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY;
1590module_param(rcu_idle_lazy_gp_delay, int, 0644);
486e2593 1591
486e2593 1592/*
0ae86a27
PM
1593 * Try to advance callbacks on the current CPU, but only if it has been
1594 * awhile since the last time we did so. Afterwards, if there are any
1595 * callbacks ready for immediate invocation, return true.
486e2593 1596 */
f1f399d1 1597static bool __maybe_unused rcu_try_advance_all_cbs(void)
486e2593 1598{
c0f4dfd4 1599 bool cbs_ready = false;
5998a75a 1600 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
c0f4dfd4 1601 struct rcu_node *rnp;
486e2593 1602
c229828c 1603 /* Exit early if we advanced recently. */
5998a75a 1604 if (jiffies == rdp->last_advance_all)
d0bc90fd 1605 return false;
5998a75a 1606 rdp->last_advance_all = jiffies;
c229828c 1607
b97d23c5 1608 rnp = rdp->mynode;
486e2593 1609
b97d23c5
PM
1610 /*
1611 * Don't bother checking unless a grace period has
1612 * completed since we last checked and there are
1613 * callbacks not yet ready to invoke.
1614 */
1615 if ((rcu_seq_completed_gp(rdp->gp_seq,
1616 rcu_seq_current(&rnp->gp_seq)) ||
1617 unlikely(READ_ONCE(rdp->gpwrap))) &&
1618 rcu_segcblist_pend_cbs(&rdp->cblist))
1619 note_gp_changes(rdp);
1620
1621 if (rcu_segcblist_ready_cbs(&rdp->cblist))
1622 cbs_ready = true;
c0f4dfd4 1623 return cbs_ready;
486e2593
PM
1624}
1625
aa9b1630 1626/*
c0f4dfd4
PM
1627 * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready
1628 * to invoke. If the CPU has callbacks, try to advance them. Tell the
1629 * caller to set the timeout based on whether or not there are non-lazy
1630 * callbacks.
aa9b1630 1631 *
c0f4dfd4 1632 * The caller must have disabled interrupts.
aa9b1630 1633 */
c1ad348b 1634int rcu_needs_cpu(u64 basemono, u64 *nextevt)
aa9b1630 1635{
5998a75a 1636 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
c1ad348b 1637 unsigned long dj;
aa9b1630 1638
b04db8e1 1639 lockdep_assert_irqs_disabled();
3382adbc 1640
c0f4dfd4 1641 /* Snapshot to detect later posting of non-lazy callback. */
c458a89e 1642 rdp->nonlazy_posted_snap = rdp->nonlazy_posted;
c0f4dfd4 1643
aa9b1630 1644 /* If no callbacks, RCU doesn't need the CPU. */
c458a89e 1645 if (!rcu_cpu_has_callbacks(&rdp->all_lazy)) {
c1ad348b 1646 *nextevt = KTIME_MAX;
aa9b1630
PM
1647 return 0;
1648 }
c0f4dfd4
PM
1649
1650 /* Attempt to advance callbacks. */
1651 if (rcu_try_advance_all_cbs()) {
1652 /* Some ready to invoke, so initiate later invocation. */
1653 invoke_rcu_core();
aa9b1630
PM
1654 return 1;
1655 }
5998a75a 1656 rdp->last_accelerate = jiffies;
c0f4dfd4
PM
1657
1658 /* Request timer delay depending on laziness, and round. */
c458a89e 1659 if (!rdp->all_lazy) {
c1ad348b 1660 dj = round_up(rcu_idle_gp_delay + jiffies,
c0f4dfd4 1661 rcu_idle_gp_delay) - jiffies;
e84c48ae 1662 } else {
c1ad348b 1663 dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies;
e84c48ae 1664 }
c1ad348b 1665 *nextevt = basemono + dj * TICK_NSEC;
aa9b1630
PM
1666 return 0;
1667}
1668
21e52e15 1669/*
c0f4dfd4
PM
1670 * Prepare a CPU for idle from an RCU perspective. The first major task
1671 * is to sense whether nohz mode has been enabled or disabled via sysfs.
1672 * The second major task is to check to see if a non-lazy callback has
1673 * arrived at a CPU that previously had only lazy callbacks. The third
1674 * major task is to accelerate (that is, assign grace-period numbers to)
1675 * any recently arrived callbacks.
aea1b35e
PM
1676 *
1677 * The caller must have disabled interrupts.
8bd93a2c 1678 */
198bbf81 1679static void rcu_prepare_for_idle(void)
8bd93a2c 1680{
48a7639c 1681 bool needwake;
0fd79e75 1682 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
c0f4dfd4 1683 struct rcu_node *rnp;
9d2ad243
PM
1684 int tne;
1685
b04db8e1 1686 lockdep_assert_irqs_disabled();
44c65ff2 1687 if (rcu_is_nocb_cpu(smp_processor_id()))
3382adbc
PM
1688 return;
1689
9d2ad243 1690 /* Handle nohz enablement switches conservatively. */
7d0ae808 1691 tne = READ_ONCE(tick_nohz_active);
0fd79e75 1692 if (tne != rdp->tick_nohz_enabled_snap) {
aa6da514 1693 if (rcu_cpu_has_callbacks(NULL))
9d2ad243 1694 invoke_rcu_core(); /* force nohz to see update. */
0fd79e75 1695 rdp->tick_nohz_enabled_snap = tne;
9d2ad243
PM
1696 return;
1697 }
1698 if (!tne)
1699 return;
f511fc62 1700
c57afe80 1701 /*
c0f4dfd4
PM
1702 * If a non-lazy callback arrived at a CPU having only lazy
1703 * callbacks, invoke RCU core for the side-effect of recalculating
1704 * idle duration on re-entry to idle.
c57afe80 1705 */
c458a89e
PM
1706 if (rdp->all_lazy &&
1707 rdp->nonlazy_posted != rdp->nonlazy_posted_snap) {
1708 rdp->all_lazy = false;
1709 rdp->nonlazy_posted_snap = rdp->nonlazy_posted;
c0f4dfd4 1710 invoke_rcu_core();
c57afe80
PM
1711 return;
1712 }
c57afe80 1713
3084f2f8 1714 /*
c0f4dfd4
PM
1715 * If we have not yet accelerated this jiffy, accelerate all
1716 * callbacks on this CPU.
3084f2f8 1717 */
5998a75a 1718 if (rdp->last_accelerate == jiffies)
aea1b35e 1719 return;
5998a75a 1720 rdp->last_accelerate = jiffies;
b97d23c5 1721 if (rcu_segcblist_pend_cbs(&rdp->cblist)) {
c0f4dfd4 1722 rnp = rdp->mynode;
2a67e741 1723 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
02f50142 1724 needwake = rcu_accelerate_cbs(rnp, rdp);
67c583a7 1725 raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
48a7639c 1726 if (needwake)
532c00c9 1727 rcu_gp_kthread_wake();
77e38ed3 1728 }
c0f4dfd4 1729}
3084f2f8 1730
c0f4dfd4
PM
1731/*
1732 * Clean up for exit from idle. Attempt to advance callbacks based on
1733 * any grace periods that elapsed while the CPU was idle, and if any
1734 * callbacks are now ready to invoke, initiate invocation.
1735 */
8fa7845d 1736static void rcu_cleanup_after_idle(void)
c0f4dfd4 1737{
b04db8e1 1738 lockdep_assert_irqs_disabled();
44c65ff2 1739 if (rcu_is_nocb_cpu(smp_processor_id()))
aea1b35e 1740 return;
7a497c96
PM
1741 if (rcu_try_advance_all_cbs())
1742 invoke_rcu_core();
8bd93a2c
PM
1743}
1744
c57afe80 1745/*
98248a0e
PM
1746 * Keep a running count of the number of non-lazy callbacks posted
1747 * on this CPU. This running counter (which is never decremented) allows
1748 * rcu_prepare_for_idle() to detect when something out of the idle loop
1749 * posts a callback, even if an equal number of callbacks are invoked.
1750 * Of course, callbacks should only be posted from within a trace event
1751 * designed to be called from idle or from within RCU_NONIDLE().
c57afe80
PM
1752 */
1753static void rcu_idle_count_callbacks_posted(void)
1754{
c458a89e 1755 __this_cpu_add(rcu_data.nonlazy_posted, 1);
c57afe80
PM
1756}
1757
8bd93a2c 1758#endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
a858af28 1759
a858af28
PM
1760#ifdef CONFIG_RCU_FAST_NO_HZ
1761
1762static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
1763{
0fd79e75 1764 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
c458a89e 1765 unsigned long nlpd = rdp->nonlazy_posted - rdp->nonlazy_posted_snap;
a858af28 1766
c0f4dfd4 1767 sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c",
5998a75a 1768 rdp->last_accelerate & 0xffff, jiffies & 0xffff,
c0f4dfd4 1769 ulong2long(nlpd),
c458a89e 1770 rdp->all_lazy ? 'L' : '.',
0fd79e75 1771 rdp->tick_nohz_enabled_snap ? '.' : 'D');
a858af28
PM
1772}
1773
1774#else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
1775
1776static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
1777{
1c17e4d4 1778 *cp = '\0';
a858af28
PM
1779}
1780
1781#endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
1782
1783/* Initiate the stall-info list. */
1784static void print_cpu_stall_info_begin(void)
1785{
efc151c3 1786 pr_cont("\n");
a858af28
PM
1787}
1788
1789/*
1790 * Print out diagnostic information for the specified stalled CPU.
1791 *
564a9ae6
PM
1792 * If the specified CPU is aware of the current RCU grace period, then
1793 * print the number of scheduling clock interrupts the CPU has taken
1794 * during the time that it has been aware. Otherwise, print the number
1795 * of RCU grace periods that this CPU is ignorant of, for example, "1"
1796 * if the CPU was aware of the previous grace period.
a858af28
PM
1797 *
1798 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
1799 */
b21ebed9 1800static void print_cpu_stall_info(int cpu)
a858af28 1801{
9b9500da 1802 unsigned long delta;
a858af28 1803 char fast_no_hz[72];
da1df50d 1804 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
a858af28
PM
1805 char *ticks_title;
1806 unsigned long ticks_value;
1807
3caa973b
TH
1808 /*
1809 * We could be printing a lot while holding a spinlock. Avoid
1810 * triggering hard lockup.
1811 */
1812 touch_nmi_watchdog();
1813
b21ebed9 1814 ticks_value = rcu_seq_ctr(rcu_state.gp_seq - rdp->gp_seq);
471f87c3
PM
1815 if (ticks_value) {
1816 ticks_title = "GPs behind";
1817 } else {
a858af28
PM
1818 ticks_title = "ticks this GP";
1819 ticks_value = rdp->ticks_this_gp;
a858af28
PM
1820 }
1821 print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
8aa670cd 1822 delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq);
89b4cd4b 1823 pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%#lx softirq=%u/%u fqs=%ld %s\n",
7f21aeef
PM
1824 cpu,
1825 "O."[!!cpu_online(cpu)],
1826 "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
1827 "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)],
9b9500da
PM
1828 !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' :
1829 rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' :
1830 "!."[!delta],
7f21aeef 1831 ticks_value, ticks_title,
dc5a4f29 1832 rcu_dynticks_snap(rdp) & 0xfff,
4c5273bf 1833 rdp->dynticks_nesting, rdp->dynticks_nmi_nesting,
6231069b 1834 rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
b21ebed9 1835 READ_ONCE(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
a858af28
PM
1836 fast_no_hz);
1837}
1838
1839/* Terminate the stall-info list. */
1840static void print_cpu_stall_info_end(void)
1841{
efc151c3 1842 pr_err("\t");
a858af28
PM
1843}
1844
0ae86a27 1845/* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */
a858af28
PM
1846static void zero_cpu_stall_ticks(struct rcu_data *rdp)
1847{
1848 rdp->ticks_this_gp = 0;
6231069b 1849 rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id());
d3052109 1850 WRITE_ONCE(rdp->last_fqs_resched, jiffies);
a858af28
PM
1851}
1852
3fbfbf7a
PM
1853#ifdef CONFIG_RCU_NOCB_CPU
1854
1855/*
1856 * Offload callback processing from the boot-time-specified set of CPUs
1857 * specified by rcu_nocb_mask. For each CPU in the set, there is a
1858 * kthread created that pulls the callbacks from the corresponding CPU,
1859 * waits for a grace period to elapse, and invokes the callbacks.
1860 * The no-CBs CPUs do a wake_up() on their kthread when they insert
1861 * a callback into any empty list, unless the rcu_nocb_poll boot parameter
1862 * has been specified, in which case each kthread actively polls its
1863 * CPU. (Which isn't so great for energy efficiency, but which does
1864 * reduce RCU's overhead on that CPU.)
1865 *
1866 * This is intended to be used in conjunction with Frederic Weisbecker's
1867 * adaptive-idle work, which would seriously reduce OS jitter on CPUs
1868 * running CPU-bound user-mode computations.
1869 *
1870 * Offloading of callback processing could also in theory be used as
1871 * an energy-efficiency measure because CPUs with no RCU callbacks
1872 * queued are more aggressive about entering dyntick-idle mode.
1873 */
1874
1875
1876/* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */
1877static int __init rcu_nocb_setup(char *str)
1878{
1879 alloc_bootmem_cpumask_var(&rcu_nocb_mask);
3fbfbf7a
PM
1880 cpulist_parse(str, rcu_nocb_mask);
1881 return 1;
1882}
1883__setup("rcu_nocbs=", rcu_nocb_setup);
1884
1b0048a4
PG
1885static int __init parse_rcu_nocb_poll(char *arg)
1886{
5455a7f6 1887 rcu_nocb_poll = true;
1b0048a4
PG
1888 return 0;
1889}
1890early_param("rcu_nocb_poll", parse_rcu_nocb_poll);
1891
dae6e64d 1892/*
0446be48
PM
1893 * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended
1894 * grace period.
dae6e64d 1895 */
abedf8e2 1896static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
dae6e64d 1897{
abedf8e2 1898 swake_up_all(sq);
dae6e64d
PM
1899}
1900
abedf8e2 1901static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
065bb78c 1902{
e0da2374 1903 return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1];
065bb78c
DW
1904}
1905
dae6e64d 1906static void rcu_init_one_nocb(struct rcu_node *rnp)
34ed6246 1907{
abedf8e2
PG
1908 init_swait_queue_head(&rnp->nocb_gp_wq[0]);
1909 init_swait_queue_head(&rnp->nocb_gp_wq[1]);
34ed6246
PM
1910}
1911
24342c96 1912/* Is the specified CPU a no-CBs CPU? */
d1e43fa5 1913bool rcu_is_nocb_cpu(int cpu)
3fbfbf7a 1914{
84b12b75 1915 if (cpumask_available(rcu_nocb_mask))
3fbfbf7a
PM
1916 return cpumask_test_cpu(cpu, rcu_nocb_mask);
1917 return false;
1918}
1919
fbce7497 1920/*
8be6e1b1
PM
1921 * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock
1922 * and this function releases it.
fbce7497 1923 */
8be6e1b1
PM
1924static void __wake_nocb_leader(struct rcu_data *rdp, bool force,
1925 unsigned long flags)
1926 __releases(rdp->nocb_lock)
fbce7497
PM
1927{
1928 struct rcu_data *rdp_leader = rdp->nocb_leader;
1929
8be6e1b1
PM
1930 lockdep_assert_held(&rdp->nocb_lock);
1931 if (!READ_ONCE(rdp_leader->nocb_kthread)) {
1932 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
fbce7497 1933 return;
8be6e1b1
PM
1934 }
1935 if (rdp_leader->nocb_leader_sleep || force) {
39953dfd 1936 /* Prior smp_mb__after_atomic() orders against prior enqueue. */
7d0ae808 1937 WRITE_ONCE(rdp_leader->nocb_leader_sleep, false);
8be6e1b1
PM
1938 del_timer(&rdp->nocb_timer);
1939 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
b3dae109
PZ
1940 smp_mb(); /* ->nocb_leader_sleep before swake_up_one(). */
1941 swake_up_one(&rdp_leader->nocb_wq);
8be6e1b1
PM
1942 } else {
1943 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
fbce7497
PM
1944 }
1945}
1946
8be6e1b1
PM
1947/*
1948 * Kick the leader kthread for this NOCB group, but caller has not
1949 * acquired locks.
1950 */
1951static void wake_nocb_leader(struct rcu_data *rdp, bool force)
1952{
1953 unsigned long flags;
1954
1955 raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
1956 __wake_nocb_leader(rdp, force, flags);
1957}
1958
1959/*
1960 * Arrange to wake the leader kthread for this NOCB group at some
1961 * future time when it is safe to do so.
1962 */
1963static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype,
1964 const char *reason)
1965{
1966 unsigned long flags;
1967
1968 raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
1969 if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT)
1970 mod_timer(&rdp->nocb_timer, jiffies + 1);
1971 WRITE_ONCE(rdp->nocb_defer_wakeup, waketype);
88d1bead 1972 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, reason);
8be6e1b1
PM
1973 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
1974}
1975
d7e29933 1976/*
0ae86a27 1977 * Does the specified CPU need an RCU callback for this invocation
d7e29933
PM
1978 * of rcu_barrier()?
1979 */
4580b054 1980static bool rcu_nocb_cpu_needs_barrier(int cpu)
d7e29933 1981{
da1df50d 1982 struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
41050a00
PM
1983 unsigned long ret;
1984#ifdef CONFIG_PROVE_RCU
d7e29933 1985 struct rcu_head *rhp;
41050a00 1986#endif /* #ifdef CONFIG_PROVE_RCU */
d7e29933 1987
41050a00
PM
1988 /*
1989 * Check count of all no-CBs callbacks awaiting invocation.
1990 * There needs to be a barrier before this function is called,
1991 * but associated with a prior determination that no more
1992 * callbacks would be posted. In the worst case, the first
dd46a788 1993 * barrier in rcu_barrier() suffices (but the caller cannot
41050a00
PM
1994 * necessarily rely on this, not a substitute for the caller
1995 * getting the concurrency design right!). There must also be
1996 * a barrier between the following load an posting of a callback
1997 * (if a callback is in fact needed). This is associated with an
1998 * atomic_inc() in the caller.
1999 */
2000 ret = atomic_long_read(&rdp->nocb_q_count);
d7e29933 2001
41050a00 2002#ifdef CONFIG_PROVE_RCU
7d0ae808 2003 rhp = READ_ONCE(rdp->nocb_head);
d7e29933 2004 if (!rhp)
7d0ae808 2005 rhp = READ_ONCE(rdp->nocb_gp_head);
d7e29933 2006 if (!rhp)
7d0ae808 2007 rhp = READ_ONCE(rdp->nocb_follower_head);
d7e29933
PM
2008
2009 /* Having no rcuo kthread but CBs after scheduler starts is bad! */
7d0ae808 2010 if (!READ_ONCE(rdp->nocb_kthread) && rhp &&
59f792d1 2011 rcu_scheduler_fully_active) {
d7e29933
PM
2012 /* RCU callback enqueued before CPU first came online??? */
2013 pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n",
2014 cpu, rhp->func);
2015 WARN_ON_ONCE(1);
2016 }
41050a00 2017#endif /* #ifdef CONFIG_PROVE_RCU */
d7e29933 2018
41050a00 2019 return !!ret;
d7e29933
PM
2020}
2021
3fbfbf7a
PM
2022/*
2023 * Enqueue the specified string of rcu_head structures onto the specified
2024 * CPU's no-CBs lists. The CPU is specified by rdp, the head of the
2025 * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy
2026 * counts are supplied by rhcount and rhcount_lazy.
2027 *
2028 * If warranted, also wake up the kthread servicing this CPUs queues.
2029 */
2030static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
2031 struct rcu_head *rhp,
2032 struct rcu_head **rhtp,
96d3fd0d
PM
2033 int rhcount, int rhcount_lazy,
2034 unsigned long flags)
3fbfbf7a
PM
2035{
2036 int len;
2037 struct rcu_head **old_rhpp;
2038 struct task_struct *t;
2039
2040 /* Enqueue the callback on the nocb list and update counts. */
41050a00
PM
2041 atomic_long_add(rhcount, &rdp->nocb_q_count);
2042 /* rcu_barrier() relies on ->nocb_q_count add before xchg. */
3fbfbf7a 2043 old_rhpp = xchg(&rdp->nocb_tail, rhtp);
7d0ae808 2044 WRITE_ONCE(*old_rhpp, rhp);
3fbfbf7a 2045 atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy);
39953dfd 2046 smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */
3fbfbf7a
PM
2047
2048 /* If we are not being polled and there is a kthread, awaken it ... */
7d0ae808 2049 t = READ_ONCE(rdp->nocb_kthread);
25e03a74 2050 if (rcu_nocb_poll || !t) {
88d1bead 2051 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
9261dd0d 2052 TPS("WakeNotPoll"));
3fbfbf7a 2053 return;
9261dd0d 2054 }
3fbfbf7a
PM
2055 len = atomic_long_read(&rdp->nocb_q_count);
2056 if (old_rhpp == &rdp->nocb_head) {
96d3fd0d 2057 if (!irqs_disabled_flags(flags)) {
fbce7497
PM
2058 /* ... if queue was empty ... */
2059 wake_nocb_leader(rdp, false);
88d1bead 2060 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
96d3fd0d
PM
2061 TPS("WakeEmpty"));
2062 } else {
8be6e1b1
PM
2063 wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE,
2064 TPS("WakeEmptyIsDeferred"));
96d3fd0d 2065 }
3fbfbf7a
PM
2066 rdp->qlen_last_fqs_check = 0;
2067 } else if (len > rdp->qlen_last_fqs_check + qhimark) {
fbce7497 2068 /* ... or if many callbacks queued. */
9fdd3bc9
PM
2069 if (!irqs_disabled_flags(flags)) {
2070 wake_nocb_leader(rdp, true);
88d1bead 2071 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
9fdd3bc9
PM
2072 TPS("WakeOvf"));
2073 } else {
efcd2d54 2074 wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE_FORCE,
8be6e1b1 2075 TPS("WakeOvfIsDeferred"));
9fdd3bc9 2076 }
3fbfbf7a 2077 rdp->qlen_last_fqs_check = LONG_MAX / 2;
9261dd0d 2078 } else {
88d1bead 2079 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WakeNot"));
3fbfbf7a
PM
2080 }
2081 return;
2082}
2083
2084/*
2085 * This is a helper for __call_rcu(), which invokes this when the normal
2086 * callback queue is inoperable. If this is not a no-CBs CPU, this
2087 * function returns failure back to __call_rcu(), which can complain
2088 * appropriately.
2089 *
2090 * Otherwise, this function queues the callback where the corresponding
2091 * "rcuo" kthread can find it.
2092 */
2093static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
96d3fd0d 2094 bool lazy, unsigned long flags)
3fbfbf7a
PM
2095{
2096
d1e43fa5 2097 if (!rcu_is_nocb_cpu(rdp->cpu))
c271d3a9 2098 return false;
96d3fd0d 2099 __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags);
21e7a608 2100 if (__is_kfree_rcu_offset((unsigned long)rhp->func))
88d1bead 2101 trace_rcu_kfree_callback(rcu_state.name, rhp,
21e7a608 2102 (unsigned long)rhp->func,
756cbf6b
PM
2103 -atomic_long_read(&rdp->nocb_q_count_lazy),
2104 -atomic_long_read(&rdp->nocb_q_count));
21e7a608 2105 else
88d1bead 2106 trace_rcu_callback(rcu_state.name, rhp,
756cbf6b
PM
2107 -atomic_long_read(&rdp->nocb_q_count_lazy),
2108 -atomic_long_read(&rdp->nocb_q_count));
1772947b
PM
2109
2110 /*
2111 * If called from an extended quiescent state with interrupts
2112 * disabled, invoke the RCU core in order to allow the idle-entry
2113 * deferred-wakeup check to function.
2114 */
2115 if (irqs_disabled_flags(flags) &&
2116 !rcu_is_watching() &&
2117 cpu_online(smp_processor_id()))
2118 invoke_rcu_core();
2119
c271d3a9 2120 return true;
3fbfbf7a
PM
2121}
2122
2123/*
2124 * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is
2125 * not a no-CBs CPU.
2126 */
b1a2d79f 2127static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
96d3fd0d
PM
2128 struct rcu_data *rdp,
2129 unsigned long flags)
3fbfbf7a 2130{
b04db8e1 2131 lockdep_assert_irqs_disabled();
d1e43fa5 2132 if (!rcu_is_nocb_cpu(smp_processor_id()))
b1a2d79f
PM
2133 return false; /* Not NOCBs CPU, caller must migrate CBs. */
2134 __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist),
2135 rcu_segcblist_tail(&rdp->cblist),
2136 rcu_segcblist_n_cbs(&rdp->cblist),
2137 rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags);
2138 rcu_segcblist_init(&rdp->cblist);
2139 rcu_segcblist_disable(&rdp->cblist);
0a9e1e11 2140 return true;
3fbfbf7a
PM
2141}
2142
2143/*
34ed6246
PM
2144 * If necessary, kick off a new grace period, and either way wait
2145 * for a subsequent grace period to complete.
3fbfbf7a 2146 */
34ed6246 2147static void rcu_nocb_wait_gp(struct rcu_data *rdp)
3fbfbf7a 2148{
34ed6246 2149 unsigned long c;
dae6e64d 2150 bool d;
34ed6246 2151 unsigned long flags;
48a7639c 2152 bool needwake;
34ed6246
PM
2153 struct rcu_node *rnp = rdp->mynode;
2154
ab5e869c 2155 local_irq_save(flags);
88d1bead 2156 c = rcu_seq_snap(&rcu_state.gp_seq);
ab5e869c
PM
2157 if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) {
2158 local_irq_restore(flags);
2159 } else {
2160 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
2161 needwake = rcu_start_this_gp(rnp, rdp, c);
2162 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2163 if (needwake)
532c00c9 2164 rcu_gp_kthread_wake();
ab5e869c 2165 }
3fbfbf7a
PM
2166
2167 /*
34ed6246
PM
2168 * Wait for the grace period. Do so interruptibly to avoid messing
2169 * up the load average.
3fbfbf7a 2170 */
41e80595 2171 trace_rcu_this_gp(rnp, rdp, c, TPS("StartWait"));
34ed6246 2172 for (;;) {
b3dae109 2173 swait_event_interruptible_exclusive(
29365e56
PM
2174 rnp->nocb_gp_wq[rcu_seq_ctr(c) & 0x1],
2175 (d = rcu_seq_done(&rnp->gp_seq, c)));
dae6e64d 2176 if (likely(d))
34ed6246 2177 break;
73a860cd 2178 WARN_ON(signal_pending(current));
41e80595 2179 trace_rcu_this_gp(rnp, rdp, c, TPS("ResumeWait"));
34ed6246 2180 }
41e80595 2181 trace_rcu_this_gp(rnp, rdp, c, TPS("EndWait"));
34ed6246 2182 smp_mb(); /* Ensure that CB invocation happens after GP end. */
3fbfbf7a
PM
2183}
2184
fbce7497
PM
2185/*
2186 * Leaders come here to wait for additional callbacks to show up.
2187 * This function does not return until callbacks appear.
2188 */
2189static void nocb_leader_wait(struct rcu_data *my_rdp)
2190{
2191 bool firsttime = true;
8be6e1b1 2192 unsigned long flags;
fbce7497
PM
2193 bool gotcbs;
2194 struct rcu_data *rdp;
2195 struct rcu_head **tail;
2196
2197wait_again:
2198
2199 /* Wait for callbacks to appear. */
2200 if (!rcu_nocb_poll) {
88d1bead 2201 trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Sleep"));
b3dae109 2202 swait_event_interruptible_exclusive(my_rdp->nocb_wq,
7d0ae808 2203 !READ_ONCE(my_rdp->nocb_leader_sleep));
8be6e1b1
PM
2204 raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags);
2205 my_rdp->nocb_leader_sleep = true;
2206 WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT);
2207 del_timer(&my_rdp->nocb_timer);
2208 raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags);
fbce7497
PM
2209 } else if (firsttime) {
2210 firsttime = false; /* Don't drown trace log with "Poll"! */
88d1bead 2211 trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Poll"));
fbce7497
PM
2212 }
2213
2214 /*
2215 * Each pass through the following loop checks a follower for CBs.
2216 * We are our own first follower. Any CBs found are moved to
2217 * nocb_gp_head, where they await a grace period.
2218 */
2219 gotcbs = false;
8be6e1b1 2220 smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */
fbce7497 2221 for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) {
7d0ae808 2222 rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head);
fbce7497
PM
2223 if (!rdp->nocb_gp_head)
2224 continue; /* No CBs here, try next follower. */
2225
2226 /* Move callbacks to wait-for-GP list, which is empty. */
7d0ae808 2227 WRITE_ONCE(rdp->nocb_head, NULL);
fbce7497 2228 rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head);
fbce7497
PM
2229 gotcbs = true;
2230 }
2231
8be6e1b1 2232 /* No callbacks? Sleep a bit if polling, and go retry. */
fbce7497 2233 if (unlikely(!gotcbs)) {
73a860cd 2234 WARN_ON(signal_pending(current));
8be6e1b1
PM
2235 if (rcu_nocb_poll) {
2236 schedule_timeout_interruptible(1);
2237 } else {
88d1bead 2238 trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu,
bedbb648 2239 TPS("WokeEmpty"));
8be6e1b1 2240 }
fbce7497
PM
2241 goto wait_again;
2242 }
2243
2244 /* Wait for one grace period. */
2245 rcu_nocb_wait_gp(my_rdp);
2246
fbce7497
PM
2247 /* Each pass through the following loop wakes a follower, if needed. */
2248 for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) {
8be6e1b1
PM
2249 if (!rcu_nocb_poll &&
2250 READ_ONCE(rdp->nocb_head) &&
2251 READ_ONCE(my_rdp->nocb_leader_sleep)) {
2252 raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags);
11ed7f93 2253 my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/
8be6e1b1
PM
2254 raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags);
2255 }
fbce7497
PM
2256 if (!rdp->nocb_gp_head)
2257 continue; /* No CBs, so no need to wake follower. */
2258
2259 /* Append callbacks to follower's "done" list. */
8be6e1b1
PM
2260 raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
2261 tail = rdp->nocb_follower_tail;
2262 rdp->nocb_follower_tail = rdp->nocb_gp_tail;
fbce7497 2263 *tail = rdp->nocb_gp_head;
8be6e1b1 2264 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
fbce7497 2265 if (rdp != my_rdp && tail == &rdp->nocb_follower_head) {
8be6e1b1 2266 /* List was empty, so wake up the follower. */
b3dae109 2267 swake_up_one(&rdp->nocb_wq);
fbce7497
PM
2268 }
2269 }
2270
2271 /* If we (the leader) don't have CBs, go wait some more. */
2272 if (!my_rdp->nocb_follower_head)
2273 goto wait_again;
2274}
2275
2276/*
2277 * Followers come here to wait for additional callbacks to show up.
2278 * This function does not return until callbacks appear.
2279 */
2280static void nocb_follower_wait(struct rcu_data *rdp)
2281{
fbce7497 2282 for (;;) {
88d1bead 2283 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("FollowerSleep"));
b3dae109 2284 swait_event_interruptible_exclusive(rdp->nocb_wq,
8be6e1b1 2285 READ_ONCE(rdp->nocb_follower_head));
fbce7497
PM
2286 if (smp_load_acquire(&rdp->nocb_follower_head)) {
2287 /* ^^^ Ensure CB invocation follows _head test. */
2288 return;
2289 }
73a860cd 2290 WARN_ON(signal_pending(current));
88d1bead 2291 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty"));
fbce7497
PM
2292 }
2293}
2294
3fbfbf7a
PM
2295/*
2296 * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes
fbce7497
PM
2297 * callbacks queued by the corresponding no-CBs CPU, however, there is
2298 * an optional leader-follower relationship so that the grace-period
2299 * kthreads don't have to do quite so many wakeups.
3fbfbf7a
PM
2300 */
2301static int rcu_nocb_kthread(void *arg)
2302{
2303 int c, cl;
8be6e1b1 2304 unsigned long flags;
3fbfbf7a
PM
2305 struct rcu_head *list;
2306 struct rcu_head *next;
2307 struct rcu_head **tail;
2308 struct rcu_data *rdp = arg;
2309
2310 /* Each pass through this loop invokes one batch of callbacks */
2311 for (;;) {
fbce7497
PM
2312 /* Wait for callbacks. */
2313 if (rdp->nocb_leader == rdp)
2314 nocb_leader_wait(rdp);
2315 else
2316 nocb_follower_wait(rdp);
2317
2318 /* Pull the ready-to-invoke callbacks onto local list. */
8be6e1b1
PM
2319 raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
2320 list = rdp->nocb_follower_head;
2321 rdp->nocb_follower_head = NULL;
2322 tail = rdp->nocb_follower_tail;
2323 rdp->nocb_follower_tail = &rdp->nocb_follower_head;
2324 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
fbce7497 2325 BUG_ON(!list);
88d1bead 2326 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeNonEmpty"));
3fbfbf7a
PM
2327
2328 /* Each pass through the following loop invokes a callback. */
88d1bead 2329 trace_rcu_batch_start(rcu_state.name,
41050a00
PM
2330 atomic_long_read(&rdp->nocb_q_count_lazy),
2331 atomic_long_read(&rdp->nocb_q_count), -1);
3fbfbf7a
PM
2332 c = cl = 0;
2333 while (list) {
2334 next = list->next;
2335 /* Wait for enqueuing to complete, if needed. */
2336 while (next == NULL && &list->next != tail) {
88d1bead 2337 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
69a79bb1 2338 TPS("WaitQueue"));
3fbfbf7a 2339 schedule_timeout_interruptible(1);
88d1bead 2340 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
69a79bb1 2341 TPS("WokeQueue"));
3fbfbf7a
PM
2342 next = list->next;
2343 }
2344 debug_rcu_head_unqueue(list);
2345 local_bh_disable();
88d1bead 2346 if (__rcu_reclaim(rcu_state.name, list))
3fbfbf7a
PM
2347 cl++;
2348 c++;
2349 local_bh_enable();
cee43939 2350 cond_resched_tasks_rcu_qs();
3fbfbf7a
PM
2351 list = next;
2352 }
88d1bead 2353 trace_rcu_batch_end(rcu_state.name, c, !!list, 0, 0, 1);
41050a00
PM
2354 smp_mb__before_atomic(); /* _add after CB invocation. */
2355 atomic_long_add(-c, &rdp->nocb_q_count);
2356 atomic_long_add(-cl, &rdp->nocb_q_count_lazy);
3fbfbf7a
PM
2357 }
2358 return 0;
2359}
2360
96d3fd0d 2361/* Is a deferred wakeup of rcu_nocb_kthread() required? */
9fdd3bc9 2362static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
96d3fd0d 2363{
7d0ae808 2364 return READ_ONCE(rdp->nocb_defer_wakeup);
96d3fd0d
PM
2365}
2366
2367/* Do a deferred wakeup of rcu_nocb_kthread(). */
8be6e1b1 2368static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp)
96d3fd0d 2369{
8be6e1b1 2370 unsigned long flags;
9fdd3bc9
PM
2371 int ndw;
2372
8be6e1b1
PM
2373 raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
2374 if (!rcu_nocb_need_deferred_wakeup(rdp)) {
2375 raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
96d3fd0d 2376 return;
8be6e1b1 2377 }
7d0ae808 2378 ndw = READ_ONCE(rdp->nocb_defer_wakeup);
511324e4 2379 WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT);
8be6e1b1 2380 __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags);
88d1bead 2381 trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DeferredWake"));
96d3fd0d
PM
2382}
2383
8be6e1b1 2384/* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */
fd30b717 2385static void do_nocb_deferred_wakeup_timer(struct timer_list *t)
8be6e1b1 2386{
fd30b717
KC
2387 struct rcu_data *rdp = from_timer(rdp, t, nocb_timer);
2388
2389 do_nocb_deferred_wakeup_common(rdp);
8be6e1b1
PM
2390}
2391
2392/*
2393 * Do a deferred wakeup of rcu_nocb_kthread() from fastpath.
2394 * This means we do an inexact common-case check. Note that if
2395 * we miss, ->nocb_timer will eventually clean things up.
2396 */
2397static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
2398{
2399 if (rcu_nocb_need_deferred_wakeup(rdp))
2400 do_nocb_deferred_wakeup_common(rdp);
2401}
2402
f4579fc5
PM
2403void __init rcu_init_nohz(void)
2404{
2405 int cpu;
ef126206 2406 bool need_rcu_nocb_mask = false;
f4579fc5 2407
f4579fc5
PM
2408#if defined(CONFIG_NO_HZ_FULL)
2409 if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask))
2410 need_rcu_nocb_mask = true;
2411#endif /* #if defined(CONFIG_NO_HZ_FULL) */
2412
84b12b75 2413 if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) {
949cccdb
PK
2414 if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) {
2415 pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n");
2416 return;
2417 }
f4579fc5 2418 }
84b12b75 2419 if (!cpumask_available(rcu_nocb_mask))
f4579fc5
PM
2420 return;
2421
f4579fc5
PM
2422#if defined(CONFIG_NO_HZ_FULL)
2423 if (tick_nohz_full_running)
2424 cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask);
2425#endif /* #if defined(CONFIG_NO_HZ_FULL) */
2426
2427 if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) {
ef126206 2428 pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n");
f4579fc5
PM
2429 cpumask_and(rcu_nocb_mask, cpu_possible_mask,
2430 rcu_nocb_mask);
2431 }
3016611e
PM
2432 if (cpumask_empty(rcu_nocb_mask))
2433 pr_info("\tOffload RCU callbacks from CPUs: (none).\n");
2434 else
2435 pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n",
2436 cpumask_pr_args(rcu_nocb_mask));
f4579fc5
PM
2437 if (rcu_nocb_poll)
2438 pr_info("\tPoll for callbacks from no-CBs CPUs.\n");
2439
b97d23c5
PM
2440 for_each_cpu(cpu, rcu_nocb_mask)
2441 init_nocb_callback_list(per_cpu_ptr(&rcu_data, cpu));
2442 rcu_organize_nocb_kthreads();
96d3fd0d
PM
2443}
2444
3fbfbf7a
PM
2445/* Initialize per-rcu_data variables for no-CBs CPUs. */
2446static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
2447{
2448 rdp->nocb_tail = &rdp->nocb_head;
abedf8e2 2449 init_swait_queue_head(&rdp->nocb_wq);
fbce7497 2450 rdp->nocb_follower_tail = &rdp->nocb_follower_head;
8be6e1b1 2451 raw_spin_lock_init(&rdp->nocb_lock);
fd30b717 2452 timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0);
3fbfbf7a
PM
2453}
2454
35ce7f29
PM
2455/*
2456 * If the specified CPU is a no-CBs CPU that does not already have its
0ae86a27
PM
2457 * rcuo kthread, spawn it. If the CPUs are brought online out of order,
2458 * this can require re-organizing the leader-follower relationships.
35ce7f29 2459 */
4580b054 2460static void rcu_spawn_one_nocb_kthread(int cpu)
35ce7f29
PM
2461{
2462 struct rcu_data *rdp;
2463 struct rcu_data *rdp_last;
2464 struct rcu_data *rdp_old_leader;
da1df50d 2465 struct rcu_data *rdp_spawn = per_cpu_ptr(&rcu_data, cpu);
35ce7f29
PM
2466 struct task_struct *t;
2467
2468 /*
2469 * If this isn't a no-CBs CPU or if it already has an rcuo kthread,
2470 * then nothing to do.
2471 */
2472 if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread)
2473 return;
2474
2475 /* If we didn't spawn the leader first, reorganize! */
2476 rdp_old_leader = rdp_spawn->nocb_leader;
2477 if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) {
2478 rdp_last = NULL;
2479 rdp = rdp_old_leader;
2480 do {
2481 rdp->nocb_leader = rdp_spawn;
2482 if (rdp_last && rdp != rdp_spawn)
2483 rdp_last->nocb_next_follower = rdp;
bbe5d7a9
PM
2484 if (rdp == rdp_spawn) {
2485 rdp = rdp->nocb_next_follower;
2486 } else {
2487 rdp_last = rdp;
2488 rdp = rdp->nocb_next_follower;
2489 rdp_last->nocb_next_follower = NULL;
2490 }
35ce7f29
PM
2491 } while (rdp);
2492 rdp_spawn->nocb_next_follower = rdp_old_leader;
2493 }
2494
0ae86a27 2495 /* Spawn the kthread for this CPU. */
35ce7f29 2496 t = kthread_run(rcu_nocb_kthread, rdp_spawn,
4580b054 2497 "rcuo%c/%d", rcu_state.abbr, cpu);
35ce7f29 2498 BUG_ON(IS_ERR(t));
7d0ae808 2499 WRITE_ONCE(rdp_spawn->nocb_kthread, t);
35ce7f29
PM
2500}
2501
2502/*
2503 * If the specified CPU is a no-CBs CPU that does not already have its
2504 * rcuo kthreads, spawn them.
2505 */
2506static void rcu_spawn_all_nocb_kthreads(int cpu)
2507{
35ce7f29 2508 if (rcu_scheduler_fully_active)
b97d23c5 2509 rcu_spawn_one_nocb_kthread(cpu);
35ce7f29
PM
2510}
2511
2512/*
2513 * Once the scheduler is running, spawn rcuo kthreads for all online
2514 * no-CBs CPUs. This assumes that the early_initcall()s happen before
2515 * non-boot CPUs come online -- if this changes, we will need to add
2516 * some mutual exclusion.
2517 */
2518static void __init rcu_spawn_nocb_kthreads(void)
2519{
2520 int cpu;
2521
2522 for_each_online_cpu(cpu)
2523 rcu_spawn_all_nocb_kthreads(cpu);
2524}
2525
fbce7497
PM
2526/* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */
2527static int rcu_nocb_leader_stride = -1;
2528module_param(rcu_nocb_leader_stride, int, 0444);
2529
2530/*
35ce7f29 2531 * Initialize leader-follower relationships for all no-CBs CPU.
fbce7497 2532 */
4580b054 2533static void __init rcu_organize_nocb_kthreads(void)
3fbfbf7a
PM
2534{
2535 int cpu;
fbce7497
PM
2536 int ls = rcu_nocb_leader_stride;
2537 int nl = 0; /* Next leader. */
3fbfbf7a 2538 struct rcu_data *rdp;
fbce7497
PM
2539 struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */
2540 struct rcu_data *rdp_prev = NULL;
3fbfbf7a 2541
84b12b75 2542 if (!cpumask_available(rcu_nocb_mask))
3fbfbf7a 2543 return;
fbce7497
PM
2544 if (ls == -1) {
2545 ls = int_sqrt(nr_cpu_ids);
2546 rcu_nocb_leader_stride = ls;
2547 }
2548
2549 /*
9831ce3b
PM
2550 * Each pass through this loop sets up one rcu_data structure.
2551 * Should the corresponding CPU come online in the future, then
2552 * we will spawn the needed set of rcu_nocb_kthread() kthreads.
fbce7497 2553 */
3fbfbf7a 2554 for_each_cpu(cpu, rcu_nocb_mask) {
da1df50d 2555 rdp = per_cpu_ptr(&rcu_data, cpu);
fbce7497
PM
2556 if (rdp->cpu >= nl) {
2557 /* New leader, set up for followers & next leader. */
2558 nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls;
2559 rdp->nocb_leader = rdp;
2560 rdp_leader = rdp;
2561 } else {
2562 /* Another follower, link to previous leader. */
2563 rdp->nocb_leader = rdp_leader;
2564 rdp_prev->nocb_next_follower = rdp;
2565 }
2566 rdp_prev = rdp;
3fbfbf7a
PM
2567 }
2568}
2569
2570/* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */
34ed6246 2571static bool init_nocb_callback_list(struct rcu_data *rdp)
3fbfbf7a 2572{
22c2f669 2573 if (!rcu_is_nocb_cpu(rdp->cpu))
34ed6246 2574 return false;
22c2f669 2575
34404ca8 2576 /* If there are early-boot callbacks, move them to nocb lists. */
15fecf89
PM
2577 if (!rcu_segcblist_empty(&rdp->cblist)) {
2578 rdp->nocb_head = rcu_segcblist_head(&rdp->cblist);
2579 rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist);
2580 atomic_long_set(&rdp->nocb_q_count,
2581 rcu_segcblist_n_cbs(&rdp->cblist));
2582 atomic_long_set(&rdp->nocb_q_count_lazy,
2583 rcu_segcblist_n_lazy_cbs(&rdp->cblist));
2584 rcu_segcblist_init(&rdp->cblist);
34404ca8 2585 }
15fecf89 2586 rcu_segcblist_disable(&rdp->cblist);
34ed6246 2587 return true;
3fbfbf7a
PM
2588}
2589
34ed6246
PM
2590#else /* #ifdef CONFIG_RCU_NOCB_CPU */
2591
4580b054 2592static bool rcu_nocb_cpu_needs_barrier(int cpu)
d7e29933
PM
2593{
2594 WARN_ON_ONCE(1); /* Should be dead code. */
2595 return false;
2596}
2597
abedf8e2 2598static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
3fbfbf7a 2599{
3fbfbf7a
PM
2600}
2601
abedf8e2 2602static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
065bb78c
DW
2603{
2604 return NULL;
2605}
2606
dae6e64d
PM
2607static void rcu_init_one_nocb(struct rcu_node *rnp)
2608{
2609}
3fbfbf7a 2610
3fbfbf7a 2611static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
96d3fd0d 2612 bool lazy, unsigned long flags)
3fbfbf7a 2613{
4afc7e26 2614 return false;
3fbfbf7a
PM
2615}
2616
b1a2d79f 2617static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
96d3fd0d
PM
2618 struct rcu_data *rdp,
2619 unsigned long flags)
3fbfbf7a 2620{
f4aa84ba 2621 return false;
3fbfbf7a
PM
2622}
2623
3fbfbf7a
PM
2624static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
2625{
2626}
2627
9fdd3bc9 2628static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
96d3fd0d
PM
2629{
2630 return false;
2631}
2632
2633static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
2634{
2635}
2636
35ce7f29
PM
2637static void rcu_spawn_all_nocb_kthreads(int cpu)
2638{
2639}
2640
2641static void __init rcu_spawn_nocb_kthreads(void)
3fbfbf7a
PM
2642{
2643}
2644
34ed6246 2645static bool init_nocb_callback_list(struct rcu_data *rdp)
3fbfbf7a 2646{
34ed6246 2647 return false;
3fbfbf7a
PM
2648}
2649
2650#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
65d798f0 2651
a096932f
PM
2652/*
2653 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
2654 * grace-period kthread will do force_quiescent_state() processing?
2655 * The idea is to avoid waking up RCU core processing on such a
2656 * CPU unless the grace period has extended for too long.
2657 *
2658 * This code relies on the fact that all NO_HZ_FULL CPUs are also
52e2bb95 2659 * CONFIG_RCU_NOCB_CPU CPUs.
a096932f 2660 */
4580b054 2661static bool rcu_nohz_full_cpu(void)
a096932f
PM
2662{
2663#ifdef CONFIG_NO_HZ_FULL
2664 if (tick_nohz_full_cpu(smp_processor_id()) &&
de8e8730 2665 (!rcu_gp_in_progress() ||
4580b054 2666 ULONG_CMP_LT(jiffies, READ_ONCE(rcu_state.gp_start) + HZ)))
5ce035fb 2667 return true;
a096932f 2668#endif /* #ifdef CONFIG_NO_HZ_FULL */
5ce035fb 2669 return false;
a096932f 2670}
5057f55e
PM
2671
2672/*
265f5f28 2673 * Bind the RCU grace-period kthreads to the housekeeping CPU.
5057f55e
PM
2674 */
2675static void rcu_bind_gp_kthread(void)
2676{
c0f489d2 2677 if (!tick_nohz_full_enabled())
5057f55e 2678 return;
de201559 2679 housekeeping_affine(current, HK_FLAG_RCU);
5057f55e 2680}
176f8f7a
PM
2681
2682/* Record the current task on dyntick-idle entry. */
2683static void rcu_dynticks_task_enter(void)
2684{
2685#if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
7d0ae808 2686 WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id());
176f8f7a
PM
2687#endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
2688}
2689
2690/* Record no current task on dyntick-idle exit. */
2691static void rcu_dynticks_task_exit(void)
2692{
2693#if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
7d0ae808 2694 WRITE_ONCE(current->rcu_tasks_idle_cpu, -1);
176f8f7a
PM
2695#endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
2696}