Commit | Line | Data |
---|---|---|
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 | */ |
43 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
44 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
45 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
46 | DEFINE_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 |
62 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
1b0048a4 | 63 | static 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 | */ |
70 | static 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 | 127 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 128 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 129 | |
f41d911f PM |
130 | /* |
131 | * Tell them what RCU they are running. | |
132 | */ | |
0e0fc1c2 | 133 | static 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 |
173 | static 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 | 302 | static 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 | 328 | void 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 | 388 | EXPORT_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 | 395 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 396 | { |
12f5f524 | 397 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
398 | } |
399 | ||
5f1a6ef3 PM |
400 | /* Bias and limit values for ->rcu_read_lock_nesting. */ |
401 | #define RCU_NEST_BIAS INT_MAX | |
402 | #define RCU_NEST_NMAX (-INT_MAX / 2) | |
403 | #define RCU_NEST_PMAX (INT_MAX / 2) | |
404 | ||
0e5da22e PM |
405 | /* |
406 | * Preemptible RCU implementation for rcu_read_lock(). | |
407 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
408 | * if we block. | |
409 | */ | |
410 | void __rcu_read_lock(void) | |
411 | { | |
412 | current->rcu_read_lock_nesting++; | |
5f1a6ef3 PM |
413 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
414 | WARN_ON_ONCE(current->rcu_read_lock_nesting > RCU_NEST_PMAX); | |
0e5da22e PM |
415 | barrier(); /* critical section after entry code. */ |
416 | } | |
417 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
418 | ||
419 | /* | |
420 | * Preemptible RCU implementation for rcu_read_unlock(). | |
421 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
422 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
423 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
424 | * in an RCU read-side critical section and other special cases. | |
425 | */ | |
426 | void __rcu_read_unlock(void) | |
427 | { | |
428 | struct task_struct *t = current; | |
429 | ||
430 | if (t->rcu_read_lock_nesting != 1) { | |
431 | --t->rcu_read_lock_nesting; | |
432 | } else { | |
433 | barrier(); /* critical section before exit code. */ | |
5f1a6ef3 | 434 | t->rcu_read_lock_nesting = -RCU_NEST_BIAS; |
0e5da22e PM |
435 | barrier(); /* assign before ->rcu_read_unlock_special load */ |
436 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) | |
437 | rcu_read_unlock_special(t); | |
438 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
439 | t->rcu_read_lock_nesting = 0; | |
440 | } | |
5f1a6ef3 PM |
441 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
442 | int rrln = t->rcu_read_lock_nesting; | |
0e5da22e | 443 | |
5f1a6ef3 | 444 | WARN_ON_ONCE(rrln < 0 && rrln > RCU_NEST_NMAX); |
0e5da22e | 445 | } |
0e5da22e PM |
446 | } |
447 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
448 | ||
12f5f524 PM |
449 | /* |
450 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
451 | * returning NULL if at the end of the list. | |
452 | */ | |
453 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
454 | struct rcu_node *rnp) | |
455 | { | |
456 | struct list_head *np; | |
457 | ||
458 | np = t->rcu_node_entry.next; | |
459 | if (np == &rnp->blkd_tasks) | |
460 | np = NULL; | |
461 | return np; | |
462 | } | |
463 | ||
8af3a5e7 PM |
464 | /* |
465 | * Return true if the specified rcu_node structure has tasks that were | |
466 | * preempted within an RCU read-side critical section. | |
467 | */ | |
468 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
469 | { | |
470 | return !list_empty(&rnp->blkd_tasks); | |
471 | } | |
472 | ||
b668c9cf | 473 | /* |
3e310098 PM |
474 | * Report deferred quiescent states. The deferral time can |
475 | * be quite short, for example, in the case of the call from | |
476 | * rcu_read_unlock_special(). | |
b668c9cf | 477 | */ |
3e310098 PM |
478 | static void |
479 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 480 | { |
b6a932d1 PM |
481 | bool empty_exp; |
482 | bool empty_norm; | |
483 | bool empty_exp_now; | |
12f5f524 | 484 | struct list_head *np; |
abaa93d9 | 485 | bool drop_boost_mutex = false; |
8203d6d0 | 486 | struct rcu_data *rdp; |
f41d911f | 487 | struct rcu_node *rnp; |
1d082fd0 | 488 | union rcu_special special; |
f41d911f | 489 | |
f41d911f | 490 | /* |
8203d6d0 PM |
491 | * If RCU core is waiting for this CPU to exit its critical section, |
492 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 493 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
494 | */ |
495 | special = t->rcu_read_unlock_special; | |
da1df50d | 496 | rdp = this_cpu_ptr(&rcu_data); |
3e310098 PM |
497 | if (!special.s && !rdp->deferred_qs) { |
498 | local_irq_restore(flags); | |
499 | return; | |
500 | } | |
1d082fd0 | 501 | if (special.b.need_qs) { |
45975c7d | 502 | rcu_qs(); |
c0135d07 | 503 | t->rcu_read_unlock_special.b.need_qs = false; |
3e310098 | 504 | if (!t->rcu_read_unlock_special.s && !rdp->deferred_qs) { |
79a62f95 LJ |
505 | local_irq_restore(flags); |
506 | return; | |
507 | } | |
f41d911f PM |
508 | } |
509 | ||
8203d6d0 | 510 | /* |
3e310098 PM |
511 | * Respond to a request by an expedited grace period for a |
512 | * quiescent state from this CPU. Note that requests from | |
513 | * tasks are handled when removing the task from the | |
514 | * blocked-tasks list below. | |
8203d6d0 | 515 | */ |
fcc878e4 | 516 | if (rdp->deferred_qs) { |
63d4c8c9 | 517 | rcu_report_exp_rdp(rdp); |
8203d6d0 PM |
518 | if (!t->rcu_read_unlock_special.s) { |
519 | local_irq_restore(flags); | |
520 | return; | |
521 | } | |
522 | } | |
523 | ||
f41d911f | 524 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 PM |
525 | if (special.b.blocked) { |
526 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 527 | |
dd5d19ba | 528 | /* |
0a0ba1c9 | 529 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
530 | * now remains queued on the rcu_node corresponding to the |
531 | * CPU it first blocked on, so there is no longer any need | |
532 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 533 | */ |
8ba9153b PM |
534 | rnp = t->rcu_blocked_node; |
535 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
536 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 537 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 538 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 539 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 540 | (!empty_norm || rnp->qsmask)); |
8203d6d0 | 541 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 542 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 543 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 544 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 545 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 546 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 547 | rnp->gp_seq, t->pid); |
12f5f524 PM |
548 | if (&t->rcu_node_entry == rnp->gp_tasks) |
549 | rnp->gp_tasks = np; | |
550 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
551 | rnp->exp_tasks = np; | |
727b705b | 552 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
553 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
554 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
555 | if (&t->rcu_node_entry == rnp->boost_tasks) |
556 | rnp->boost_tasks = np; | |
727b705b | 557 | } |
f41d911f PM |
558 | |
559 | /* | |
560 | * If this was the last task on the current list, and if | |
561 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
562 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
563 | * so we must take a snapshot of the expedited state. | |
f41d911f | 564 | */ |
8203d6d0 | 565 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 566 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 567 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 568 | rnp->gp_seq, |
d4c08f2a PM |
569 | 0, rnp->qsmask, |
570 | rnp->level, | |
571 | rnp->grplo, | |
572 | rnp->grphi, | |
573 | !!rnp->gp_tasks); | |
139ad4da | 574 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 575 | } else { |
67c583a7 | 576 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 577 | } |
d9a3da06 | 578 | |
27f4d280 | 579 | /* Unboost if we were boosted. */ |
727b705b | 580 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 581 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 582 | |
d9a3da06 PM |
583 | /* |
584 | * If this was the last task on the expedited lists, | |
585 | * then we need to report up the rcu_node hierarchy. | |
586 | */ | |
389abd48 | 587 | if (!empty_exp && empty_exp_now) |
63d4c8c9 | 588 | rcu_report_exp_rnp(rnp, true); |
b668c9cf PM |
589 | } else { |
590 | local_irq_restore(flags); | |
f41d911f | 591 | } |
f41d911f PM |
592 | } |
593 | ||
3e310098 PM |
594 | /* |
595 | * Is a deferred quiescent-state pending, and are we also not in | |
596 | * an RCU read-side critical section? It is the caller's responsibility | |
597 | * to ensure it is otherwise safe to report any deferred quiescent | |
598 | * states. The reason for this is that it is safe to report a | |
599 | * quiescent state during context switch even though preemption | |
600 | * is disabled. This function cannot be expected to understand these | |
601 | * nuances, so the caller must handle them. | |
602 | */ | |
603 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
604 | { | |
117f683c | 605 | return (__this_cpu_read(rcu_data.deferred_qs) || |
3e310098 | 606 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
27c744e3 | 607 | t->rcu_read_lock_nesting <= 0; |
3e310098 PM |
608 | } |
609 | ||
610 | /* | |
611 | * Report a deferred quiescent state if needed and safe to do so. | |
612 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
613 | * not being in an RCU read-side critical section. The caller must | |
614 | * evaluate safety in terms of interrupt, softirq, and preemption | |
615 | * disabling. | |
616 | */ | |
617 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
618 | { | |
619 | unsigned long flags; | |
620 | bool couldrecurse = t->rcu_read_lock_nesting >= 0; | |
621 | ||
622 | if (!rcu_preempt_need_deferred_qs(t)) | |
623 | return; | |
624 | if (couldrecurse) | |
5f1a6ef3 | 625 | t->rcu_read_lock_nesting -= RCU_NEST_BIAS; |
3e310098 PM |
626 | local_irq_save(flags); |
627 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
628 | if (couldrecurse) | |
5f1a6ef3 | 629 | t->rcu_read_lock_nesting += RCU_NEST_BIAS; |
3e310098 PM |
630 | } |
631 | ||
632 | /* | |
633 | * Handle special cases during rcu_read_unlock(), such as needing to | |
634 | * notify RCU core processing or task having blocked during the RCU | |
635 | * read-side critical section. | |
636 | */ | |
637 | static void rcu_read_unlock_special(struct task_struct *t) | |
638 | { | |
639 | unsigned long flags; | |
640 | bool preempt_bh_were_disabled = | |
641 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
642 | bool irqs_were_disabled; | |
643 | ||
644 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
645 | if (in_nmi()) | |
646 | return; | |
647 | ||
648 | local_irq_save(flags); | |
649 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 PM |
650 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
651 | WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false); | |
3e310098 | 652 | /* Need to defer quiescent state until everything is enabled. */ |
05f41571 PM |
653 | if (irqs_were_disabled) { |
654 | /* Enabling irqs does not reschedule, so... */ | |
655 | raise_softirq_irqoff(RCU_SOFTIRQ); | |
656 | } else { | |
657 | /* Enabling BH or preempt does reschedule, so... */ | |
658 | set_tsk_need_resched(current); | |
659 | set_preempt_need_resched(); | |
660 | } | |
3e310098 PM |
661 | local_irq_restore(flags); |
662 | return; | |
663 | } | |
05f41571 | 664 | WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false); |
3e310098 PM |
665 | rcu_preempt_deferred_qs_irqrestore(t, flags); |
666 | } | |
667 | ||
1ed509a2 PM |
668 | /* |
669 | * Dump detailed information for all tasks blocking the current RCU | |
670 | * grace period on the specified rcu_node structure. | |
671 | */ | |
672 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
673 | { | |
674 | unsigned long flags; | |
1ed509a2 PM |
675 | struct task_struct *t; |
676 | ||
6cf10081 | 677 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 678 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 679 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
680 | return; |
681 | } | |
82efed06 | 682 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 683 | struct task_struct, rcu_node_entry); |
3caa973b TH |
684 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
685 | /* | |
686 | * We could be printing a lot while holding a spinlock. | |
687 | * Avoid triggering hard lockup. | |
688 | */ | |
689 | touch_nmi_watchdog(); | |
12f5f524 | 690 | sched_show_task(t); |
3caa973b | 691 | } |
67c583a7 | 692 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
693 | } |
694 | ||
695 | /* | |
696 | * Dump detailed information for all tasks blocking the current RCU | |
697 | * grace period. | |
698 | */ | |
a2887cd8 | 699 | static void rcu_print_detail_task_stall(void) |
1ed509a2 | 700 | { |
336a4f6c | 701 | struct rcu_node *rnp = rcu_get_root(); |
1ed509a2 PM |
702 | |
703 | rcu_print_detail_task_stall_rnp(rnp); | |
aedf4ba9 | 704 | rcu_for_each_leaf_node(rnp) |
1ed509a2 PM |
705 | rcu_print_detail_task_stall_rnp(rnp); |
706 | } | |
707 | ||
a858af28 PM |
708 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
709 | { | |
efc151c3 | 710 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
711 | rnp->level, rnp->grplo, rnp->grphi); |
712 | } | |
713 | ||
714 | static void rcu_print_task_stall_end(void) | |
715 | { | |
efc151c3 | 716 | pr_cont("\n"); |
a858af28 PM |
717 | } |
718 | ||
f41d911f PM |
719 | /* |
720 | * Scan the current list of tasks blocked within RCU read-side critical | |
721 | * sections, printing out the tid of each. | |
722 | */ | |
9bc8b558 | 723 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 724 | { |
f41d911f | 725 | struct task_struct *t; |
9bc8b558 | 726 | int ndetected = 0; |
f41d911f | 727 | |
27f4d280 | 728 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 729 | return 0; |
a858af28 | 730 | rcu_print_task_stall_begin(rnp); |
82efed06 | 731 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 732 | struct task_struct, rcu_node_entry); |
9bc8b558 | 733 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 734 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
735 | ndetected++; |
736 | } | |
a858af28 | 737 | rcu_print_task_stall_end(); |
9bc8b558 | 738 | return ndetected; |
f41d911f PM |
739 | } |
740 | ||
74611ecb PM |
741 | /* |
742 | * Scan the current list of tasks blocked within RCU read-side critical | |
743 | * sections, printing out the tid of each that is blocking the current | |
744 | * expedited grace period. | |
745 | */ | |
746 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
747 | { | |
748 | struct task_struct *t; | |
749 | int ndetected = 0; | |
750 | ||
751 | if (!rnp->exp_tasks) | |
752 | return 0; | |
753 | t = list_entry(rnp->exp_tasks->prev, | |
754 | struct task_struct, rcu_node_entry); | |
755 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
756 | pr_cont(" P%d", t->pid); | |
757 | ndetected++; | |
758 | } | |
759 | return ndetected; | |
760 | } | |
761 | ||
b0e165c0 PM |
762 | /* |
763 | * Check that the list of blocked tasks for the newly completed grace | |
764 | * period is in fact empty. It is a serious bug to complete a grace | |
765 | * period that still has RCU readers blocked! This function must be | |
ff3bb6f4 | 766 | * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock |
b0e165c0 | 767 | * must be held by the caller. |
12f5f524 PM |
768 | * |
769 | * Also, if there are blocked tasks on the list, they automatically | |
770 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 771 | */ |
81ab59a3 | 772 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 773 | { |
c5ebe66c PM |
774 | struct task_struct *t; |
775 | ||
ea9b0c8a | 776 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
4bc8d555 | 777 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 778 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
779 | if (rcu_preempt_has_tasks(rnp) && |
780 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
12f5f524 | 781 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
782 | t = container_of(rnp->gp_tasks, struct task_struct, |
783 | rcu_node_entry); | |
784 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 785 | rnp->gp_seq, t->pid); |
c5ebe66c | 786 | } |
28ecd580 | 787 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
788 | } |
789 | ||
f41d911f PM |
790 | /* |
791 | * Check for a quiescent state from the current CPU. When a task blocks, | |
792 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
793 | * which is checked elsewhere. | |
794 | * | |
795 | * Caller must disable hard irqs. | |
796 | */ | |
45975c7d | 797 | static void rcu_flavor_check_callbacks(int user) |
f41d911f PM |
798 | { |
799 | struct task_struct *t = current; | |
800 | ||
45975c7d PM |
801 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
802 | rcu_note_voluntary_context_switch(current); | |
803 | } | |
3e310098 PM |
804 | if (t->rcu_read_lock_nesting > 0 || |
805 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { | |
806 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
807 | if (rcu_preempt_need_deferred_qs(t)) { |
808 | set_tsk_need_resched(t); | |
809 | set_preempt_need_resched(); | |
810 | } | |
3e310098 PM |
811 | } else if (rcu_preempt_need_deferred_qs(t)) { |
812 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
813 | return; | |
814 | } else if (!t->rcu_read_lock_nesting) { | |
45975c7d | 815 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
816 | return; |
817 | } | |
3e310098 PM |
818 | |
819 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
10f39bb1 | 820 | if (t->rcu_read_lock_nesting > 0 && |
2280ee5a PM |
821 | __this_cpu_read(rcu_data.core_needs_qs) && |
822 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 823 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 824 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 825 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
826 | } |
827 | ||
6ebb237b PM |
828 | /** |
829 | * synchronize_rcu - wait until a grace period has elapsed. | |
830 | * | |
831 | * Control will return to the caller some time after a full grace | |
832 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
833 | * read-side critical sections have completed. Note, however, that |
834 | * upon return from synchronize_rcu(), the caller might well be executing | |
835 | * concurrently with new RCU read-side critical sections that began while | |
836 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
837 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
45975c7d PM |
838 | * In addition, regions of code across which interrupts, preemption, or |
839 | * softirqs have been disabled also serve as RCU read-side critical | |
840 | * sections. This includes hardware interrupt handlers, softirq handlers, | |
841 | * and NMI handlers. | |
842 | * | |
843 | * Note that this guarantee implies further memory-ordering guarantees. | |
844 | * On systems with more than one CPU, when synchronize_rcu() returns, | |
0ae86a27 PM |
845 | * each CPU is guaranteed to have executed a full memory barrier since |
846 | * the end of its last RCU read-side critical section whose beginning | |
45975c7d PM |
847 | * preceded the call to synchronize_rcu(). In addition, each CPU having |
848 | * an RCU read-side critical section that extends beyond the return from | |
849 | * synchronize_rcu() is guaranteed to have executed a full memory barrier | |
850 | * after the beginning of synchronize_rcu() and before the beginning of | |
851 | * that RCU read-side critical section. Note that these guarantees include | |
852 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
853 | * that are executing in the kernel. | |
f0a0e6f2 | 854 | * |
45975c7d PM |
855 | * Furthermore, if CPU A invoked synchronize_rcu(), which returned |
856 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
857 | * to have executed a full memory barrier during the execution of | |
858 | * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but | |
859 | * again only if the system has more than one CPU). | |
6ebb237b PM |
860 | */ |
861 | void synchronize_rcu(void) | |
862 | { | |
f78f5b90 PM |
863 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
864 | lock_is_held(&rcu_lock_map) || | |
865 | lock_is_held(&rcu_sched_lock_map), | |
866 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
52d7e48b | 867 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
6ebb237b | 868 | return; |
5afff48b | 869 | if (rcu_gp_is_expedited()) |
3705b88d AM |
870 | synchronize_rcu_expedited(); |
871 | else | |
872 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
873 | } |
874 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
875 | ||
2439b696 PM |
876 | /* |
877 | * Check for a task exiting while in a preemptible-RCU read-side | |
878 | * critical section, clean up if so. No need to issue warnings, | |
879 | * as debug_check_no_locks_held() already does this if lockdep | |
880 | * is enabled. | |
881 | */ | |
882 | void exit_rcu(void) | |
883 | { | |
884 | struct task_struct *t = current; | |
885 | ||
886 | if (likely(list_empty(¤t->rcu_node_entry))) | |
887 | return; | |
888 | t->rcu_read_lock_nesting = 1; | |
889 | barrier(); | |
1d082fd0 | 890 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 | 891 | __rcu_read_unlock(); |
3e310098 | 892 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
893 | } |
894 | ||
4bc8d555 PM |
895 | /* |
896 | * Dump the blocked-tasks state, but limit the list dump to the | |
897 | * specified number of elements. | |
898 | */ | |
57738942 | 899 | static void |
81ab59a3 | 900 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 901 | { |
57738942 | 902 | int cpu; |
4bc8d555 PM |
903 | int i; |
904 | struct list_head *lhp; | |
57738942 PM |
905 | bool onl; |
906 | struct rcu_data *rdp; | |
ff3cee39 | 907 | struct rcu_node *rnp1; |
4bc8d555 | 908 | |
ce11fae8 | 909 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 910 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 911 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
ff3cee39 PM |
912 | (long)rnp->gp_seq, (long)rnp->completedqs); |
913 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
914 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
915 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf PM |
916 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
917 | __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks); | |
918 | pr_info("%s: ->blkd_tasks", __func__); | |
4bc8d555 PM |
919 | i = 0; |
920 | list_for_each(lhp, &rnp->blkd_tasks) { | |
921 | pr_cont(" %p", lhp); | |
922 | if (++i >= 10) | |
923 | break; | |
924 | } | |
925 | pr_cont("\n"); | |
57738942 | 926 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 927 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 PM |
928 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); |
929 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
930 | cpu, ".o"[onl], | |
931 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
932 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
933 | } | |
4bc8d555 PM |
934 | } |
935 | ||
28f6569a | 936 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f PM |
937 | |
938 | /* | |
939 | * Tell them what RCU they are running. | |
940 | */ | |
0e0fc1c2 | 941 | static void __init rcu_bootup_announce(void) |
f41d911f | 942 | { |
efc151c3 | 943 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 944 | rcu_bootup_announce_oddness(); |
f41d911f PM |
945 | } |
946 | ||
45975c7d PM |
947 | /* |
948 | * Note a quiescent state for PREEMPT=n. Because we do not need to know | |
949 | * how many quiescent states passed, just if there was at least one since | |
950 | * the start of the grace period, this just sets a flag. The caller must | |
951 | * have disabled preemption. | |
952 | */ | |
953 | static void rcu_qs(void) | |
d28139c4 | 954 | { |
45975c7d PM |
955 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
956 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
957 | return; | |
958 | trace_rcu_grace_period(TPS("rcu_sched"), | |
959 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
960 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
961 | if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) | |
962 | return; | |
963 | __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false); | |
63d4c8c9 | 964 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
d28139c4 PM |
965 | } |
966 | ||
395a2f09 PM |
967 | /* |
968 | * Register an urgently needed quiescent state. If there is an | |
969 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
970 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
971 | * some cases serve for expedited as well as normal grace periods. | |
972 | * Either way, register a lightweight quiescent state. | |
973 | * | |
974 | * The barrier() calls are redundant in the common case when this is | |
975 | * called externally, but just in case this is called from within this | |
976 | * file. | |
977 | * | |
978 | */ | |
979 | void rcu_all_qs(void) | |
980 | { | |
981 | unsigned long flags; | |
982 | ||
2dba13f0 | 983 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
984 | return; |
985 | preempt_disable(); | |
986 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 987 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
988 | preempt_enable(); |
989 | return; | |
990 | } | |
2dba13f0 | 991 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
395a2f09 | 992 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
2dba13f0 | 993 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
994 | local_irq_save(flags); |
995 | rcu_momentary_dyntick_idle(); | |
996 | local_irq_restore(flags); | |
997 | } | |
7e28c5af | 998 | rcu_qs(); |
395a2f09 PM |
999 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
1000 | preempt_enable(); | |
1001 | } | |
1002 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
1003 | ||
cba6d0d6 | 1004 | /* |
45975c7d | 1005 | * Note a PREEMPT=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 1006 | */ |
45975c7d | 1007 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 1008 | { |
45975c7d PM |
1009 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
1010 | trace_rcu_utilization(TPS("Start context switch")); | |
1011 | rcu_qs(); | |
1012 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 1013 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 1014 | goto out; |
2dba13f0 PM |
1015 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
1016 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 1017 | rcu_momentary_dyntick_idle(); |
45975c7d PM |
1018 | if (!preempt) |
1019 | rcu_tasks_qs(current); | |
1020 | out: | |
1021 | trace_rcu_utilization(TPS("End context switch")); | |
1022 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ | |
cba6d0d6 | 1023 | } |
45975c7d | 1024 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 1025 | |
fc2219d4 | 1026 | /* |
6cc68793 | 1027 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
1028 | * RCU readers. |
1029 | */ | |
27f4d280 | 1030 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
1031 | { |
1032 | return 0; | |
1033 | } | |
1034 | ||
8af3a5e7 PM |
1035 | /* |
1036 | * Because there is no preemptible RCU, there can be no readers blocked. | |
1037 | */ | |
1038 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 1039 | { |
8af3a5e7 | 1040 | return false; |
b668c9cf PM |
1041 | } |
1042 | ||
3e310098 PM |
1043 | /* |
1044 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
1045 | * states. | |
1046 | */ | |
1047 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
1048 | { | |
1049 | return false; | |
1050 | } | |
1051 | static void rcu_preempt_deferred_qs(struct task_struct *t) { } | |
1052 | ||
1ed509a2 | 1053 | /* |
6cc68793 | 1054 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
1055 | * tasks blocked within RCU read-side critical sections. |
1056 | */ | |
a2887cd8 | 1057 | static void rcu_print_detail_task_stall(void) |
1ed509a2 PM |
1058 | { |
1059 | } | |
1060 | ||
f41d911f | 1061 | /* |
6cc68793 | 1062 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
1063 | * tasks blocked within RCU read-side critical sections. |
1064 | */ | |
9bc8b558 | 1065 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 1066 | { |
9bc8b558 | 1067 | return 0; |
f41d911f PM |
1068 | } |
1069 | ||
74611ecb PM |
1070 | /* |
1071 | * Because preemptible RCU does not exist, we never have to check for | |
1072 | * tasks blocked within RCU read-side critical sections that are | |
1073 | * blocking the current expedited grace period. | |
1074 | */ | |
1075 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
1076 | { | |
1077 | return 0; | |
1078 | } | |
1079 | ||
b0e165c0 | 1080 | /* |
6cc68793 | 1081 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
1082 | * so there is no need to check for blocked tasks. So check only for |
1083 | * bogus qsmask values. | |
b0e165c0 | 1084 | */ |
81ab59a3 | 1085 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 1086 | { |
49e29126 | 1087 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
1088 | } |
1089 | ||
f41d911f | 1090 | /* |
45975c7d PM |
1091 | * Check to see if this CPU is in a non-context-switch quiescent state |
1092 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1093 | * Also schedule RCU core processing. | |
1094 | * | |
1095 | * This function must be called from hardirq context. It is normally | |
1096 | * invoked from the scheduling-clock interrupt. | |
f41d911f | 1097 | */ |
45975c7d | 1098 | static void rcu_flavor_check_callbacks(int user) |
f41d911f | 1099 | { |
45975c7d | 1100 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 1101 | |
45975c7d PM |
1102 | /* |
1103 | * Get here if this CPU took its interrupt from user | |
1104 | * mode or from the idle loop, and if this is not a | |
1105 | * nested interrupt. In this case, the CPU is in | |
1106 | * a quiescent state, so note it. | |
1107 | * | |
1108 | * No memory barrier is required here because rcu_qs() | |
1109 | * references only CPU-local variables that other CPUs | |
1110 | * neither access nor modify, at least not while the | |
1111 | * corresponding CPU is online. | |
1112 | */ | |
1113 | ||
1114 | rcu_qs(); | |
1115 | } | |
e74f4c45 | 1116 | } |
e74f4c45 | 1117 | |
45975c7d PM |
1118 | /* PREEMPT=n implementation of synchronize_rcu(). */ |
1119 | void synchronize_rcu(void) | |
1eba8f84 | 1120 | { |
45975c7d PM |
1121 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
1122 | lock_is_held(&rcu_lock_map) || | |
1123 | lock_is_held(&rcu_sched_lock_map), | |
0ae86a27 | 1124 | "Illegal synchronize_rcu() in RCU read-side critical section"); |
45975c7d PM |
1125 | if (rcu_blocking_is_gp()) |
1126 | return; | |
1127 | if (rcu_gp_is_expedited()) | |
1128 | synchronize_rcu_expedited(); | |
1129 | else | |
1130 | wait_rcu_gp(call_rcu); | |
1eba8f84 | 1131 | } |
45975c7d | 1132 | EXPORT_SYMBOL_GPL(synchronize_rcu); |
1eba8f84 | 1133 | |
2439b696 PM |
1134 | /* |
1135 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
1136 | * while in preemptible RCU read-side critical sections. | |
1137 | */ | |
1138 | void exit_rcu(void) | |
1139 | { | |
1140 | } | |
1141 | ||
4bc8d555 PM |
1142 | /* |
1143 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
1144 | */ | |
57738942 | 1145 | static void |
81ab59a3 | 1146 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
1147 | { |
1148 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
1149 | } | |
1150 | ||
28f6569a | 1151 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 1152 | |
27f4d280 PM |
1153 | #ifdef CONFIG_RCU_BOOST |
1154 | ||
5d01bbd1 TG |
1155 | static void rcu_wake_cond(struct task_struct *t, int status) |
1156 | { | |
1157 | /* | |
1158 | * If the thread is yielding, only wake it when this | |
1159 | * is invoked from idle | |
1160 | */ | |
1161 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1162 | wake_up_process(t); | |
1163 | } | |
1164 | ||
27f4d280 PM |
1165 | /* |
1166 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1167 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1168 | * ->blkd_tasks list. | |
1169 | * | |
1170 | * Note that irqs must be enabled: boosting the task can block. | |
1171 | * Returns 1 if there are more tasks needing to be boosted. | |
1172 | */ | |
1173 | static int rcu_boost(struct rcu_node *rnp) | |
1174 | { | |
1175 | unsigned long flags; | |
27f4d280 PM |
1176 | struct task_struct *t; |
1177 | struct list_head *tb; | |
1178 | ||
7d0ae808 PM |
1179 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1180 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1181 | return 0; /* Nothing left to boost. */ |
1182 | ||
2a67e741 | 1183 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1184 | |
1185 | /* | |
1186 | * Recheck under the lock: all tasks in need of boosting | |
1187 | * might exit their RCU read-side critical sections on their own. | |
1188 | */ | |
1189 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1190 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1191 | return 0; |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Preferentially boost tasks blocking expedited grace periods. | |
1196 | * This cannot starve the normal grace periods because a second | |
1197 | * expedited grace period must boost all blocked tasks, including | |
1198 | * those blocking the pre-existing normal grace period. | |
1199 | */ | |
bec06785 | 1200 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1201 | tb = rnp->exp_tasks; |
bec06785 | 1202 | else |
27f4d280 PM |
1203 | tb = rnp->boost_tasks; |
1204 | ||
1205 | /* | |
1206 | * We boost task t by manufacturing an rt_mutex that appears to | |
1207 | * be held by task t. We leave a pointer to that rt_mutex where | |
1208 | * task t can find it, and task t will release the mutex when it | |
1209 | * exits its outermost RCU read-side critical section. Then | |
1210 | * simply acquiring this artificial rt_mutex will boost task | |
1211 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1212 | * | |
1213 | * Note that task t must acquire rnp->lock to remove itself from | |
1214 | * the ->blkd_tasks list, which it will do from exit() if from | |
1215 | * nowhere else. We therefore are guaranteed that task t will | |
1216 | * stay around at least until we drop rnp->lock. Note that | |
1217 | * rnp->lock also resolves races between our priority boosting | |
1218 | * and task t's exiting its outermost RCU read-side critical | |
1219 | * section. | |
1220 | */ | |
1221 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1222 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1223 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1224 | /* Lock only for side effect: boosts task t's priority. */ |
1225 | rt_mutex_lock(&rnp->boost_mtx); | |
1226 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1227 | |
7d0ae808 PM |
1228 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1229 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1230 | } |
1231 | ||
27f4d280 | 1232 | /* |
bc17ea10 | 1233 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1234 | */ |
1235 | static int rcu_boost_kthread(void *arg) | |
1236 | { | |
1237 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1238 | int spincnt = 0; | |
1239 | int more2boost; | |
1240 | ||
f7f7bac9 | 1241 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1242 | for (;;) { |
d71df90e | 1243 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1244 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1245 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1246 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1247 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1248 | more2boost = rcu_boost(rnp); |
1249 | if (more2boost) | |
1250 | spincnt++; | |
1251 | else | |
1252 | spincnt = 0; | |
1253 | if (spincnt > 10) { | |
5d01bbd1 | 1254 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1255 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1256 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1257 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1258 | spincnt = 0; |
1259 | } | |
1260 | } | |
1217ed1b | 1261 | /* NOTREACHED */ |
f7f7bac9 | 1262 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1263 | return 0; |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * Check to see if it is time to start boosting RCU readers that are | |
1268 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1269 | * kthread to start boosting them. If there is an expedited grace | |
1270 | * period in progress, it is always time to boost. | |
1271 | * | |
b065a853 PM |
1272 | * The caller must hold rnp->lock, which this function releases. |
1273 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1274 | * about it going away. | |
27f4d280 | 1275 | */ |
1217ed1b | 1276 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1277 | __releases(rnp->lock) |
27f4d280 PM |
1278 | { |
1279 | struct task_struct *t; | |
1280 | ||
a32e01ee | 1281 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1282 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1283 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1284 | return; |
0ea1f2eb | 1285 | } |
27f4d280 PM |
1286 | if (rnp->exp_tasks != NULL || |
1287 | (rnp->gp_tasks != NULL && | |
1288 | rnp->boost_tasks == NULL && | |
1289 | rnp->qsmask == 0 && | |
1290 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1291 | if (rnp->exp_tasks == NULL) | |
1292 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1293 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1294 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1295 | if (t) |
1296 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1297 | } else { |
67c583a7 | 1298 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1299 | } |
27f4d280 PM |
1300 | } |
1301 | ||
a46e0899 PM |
1302 | /* |
1303 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1304 | */ | |
1305 | static void invoke_rcu_callbacks_kthread(void) | |
1306 | { | |
1307 | unsigned long flags; | |
1308 | ||
1309 | local_irq_save(flags); | |
1310 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1311 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1312 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1313 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1314 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1315 | } | |
a46e0899 PM |
1316 | local_irq_restore(flags); |
1317 | } | |
1318 | ||
dff1672d PM |
1319 | /* |
1320 | * Is the current CPU running the RCU-callbacks kthread? | |
1321 | * Caller must have preemption disabled. | |
1322 | */ | |
1323 | static bool rcu_is_callbacks_kthread(void) | |
1324 | { | |
c9d4b0af | 1325 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1326 | } |
1327 | ||
27f4d280 PM |
1328 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1329 | ||
1330 | /* | |
1331 | * Do priority-boost accounting for the start of a new grace period. | |
1332 | */ | |
1333 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1334 | { | |
1335 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1336 | } | |
1337 | ||
27f4d280 PM |
1338 | /* |
1339 | * Create an RCU-boost kthread for the specified node if one does not | |
1340 | * already exist. We only create this kthread for preemptible RCU. | |
1341 | * Returns zero if all is well, a negated errno otherwise. | |
1342 | */ | |
6dbfdc14 | 1343 | static int rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 | 1344 | { |
6dbfdc14 | 1345 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1346 | unsigned long flags; |
1347 | struct sched_param sp; | |
1348 | struct task_struct *t; | |
1349 | ||
6dbfdc14 | 1350 | if (!IS_ENABLED(CONFIG_PREEMPT_RCU)) |
27f4d280 | 1351 | return 0; |
5d01bbd1 | 1352 | |
0aa04b05 | 1353 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1354 | return 0; |
1355 | ||
6dbfdc14 | 1356 | rcu_state.boost = 1; |
27f4d280 PM |
1357 | if (rnp->boost_kthread_task != NULL) |
1358 | return 0; | |
1359 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1360 | "rcub/%d", rnp_index); |
27f4d280 PM |
1361 | if (IS_ERR(t)) |
1362 | return PTR_ERR(t); | |
2a67e741 | 1363 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1364 | rnp->boost_kthread_task = t; |
67c583a7 | 1365 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1366 | sp.sched_priority = kthread_prio; |
27f4d280 | 1367 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1368 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1369 | return 0; |
1370 | } | |
1371 | ||
f8b7fc6b PM |
1372 | static void rcu_kthread_do_work(void) |
1373 | { | |
5bb5d09c | 1374 | rcu_do_batch(this_cpu_ptr(&rcu_data)); |
f8b7fc6b PM |
1375 | } |
1376 | ||
62ab7072 | 1377 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1378 | { |
f8b7fc6b | 1379 | struct sched_param sp; |
f8b7fc6b | 1380 | |
21871d7e | 1381 | sp.sched_priority = kthread_prio; |
62ab7072 | 1382 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1383 | } |
1384 | ||
62ab7072 | 1385 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1386 | { |
62ab7072 | 1387 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1388 | } |
1389 | ||
62ab7072 | 1390 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1391 | { |
c9d4b0af | 1392 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1393 | } |
1394 | ||
1395 | /* | |
0ae86a27 PM |
1396 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces |
1397 | * the RCU softirq used in configurations of RCU that do not support RCU | |
1398 | * priority boosting. | |
f8b7fc6b | 1399 | */ |
62ab7072 | 1400 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1401 | { |
c9d4b0af CL |
1402 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1403 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1404 | int spincnt; |
f8b7fc6b | 1405 | |
62ab7072 | 1406 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1407 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1408 | local_bh_disable(); |
f8b7fc6b | 1409 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1410 | this_cpu_inc(rcu_cpu_kthread_loops); |
1411 | local_irq_disable(); | |
f8b7fc6b PM |
1412 | work = *workp; |
1413 | *workp = 0; | |
62ab7072 | 1414 | local_irq_enable(); |
f8b7fc6b PM |
1415 | if (work) |
1416 | rcu_kthread_do_work(); | |
1417 | local_bh_enable(); | |
62ab7072 | 1418 | if (*workp == 0) { |
f7f7bac9 | 1419 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1420 | *statusp = RCU_KTHREAD_WAITING; |
1421 | return; | |
f8b7fc6b PM |
1422 | } |
1423 | } | |
62ab7072 | 1424 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1425 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1426 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1427 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1428 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1429 | } |
1430 | ||
1431 | /* | |
1432 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1433 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1434 | * held, so the value of rnp->qsmaskinit will be stable. | |
1435 | * | |
1436 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1437 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1438 | * this function allows the kthread to execute on any CPU. | |
1439 | */ | |
5d01bbd1 | 1440 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1441 | { |
5d01bbd1 | 1442 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1443 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1444 | cpumask_var_t cm; |
1445 | int cpu; | |
f8b7fc6b | 1446 | |
5d01bbd1 | 1447 | if (!t) |
f8b7fc6b | 1448 | return; |
5d01bbd1 | 1449 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1450 | return; |
bc75e999 MR |
1451 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1452 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1453 | cpu != outgoingcpu) | |
f8b7fc6b | 1454 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1455 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1456 | cpumask_setall(cm); |
5d01bbd1 | 1457 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1458 | free_cpumask_var(cm); |
1459 | } | |
1460 | ||
62ab7072 PM |
1461 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1462 | .store = &rcu_cpu_kthread_task, | |
1463 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1464 | .thread_fn = rcu_cpu_kthread, | |
1465 | .thread_comm = "rcuc/%u", | |
1466 | .setup = rcu_cpu_kthread_setup, | |
1467 | .park = rcu_cpu_kthread_park, | |
1468 | }; | |
f8b7fc6b PM |
1469 | |
1470 | /* | |
9386c0b7 | 1471 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1472 | */ |
9386c0b7 | 1473 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1474 | { |
f8b7fc6b | 1475 | struct rcu_node *rnp; |
5d01bbd1 | 1476 | int cpu; |
f8b7fc6b | 1477 | |
62ab7072 | 1478 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1479 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
9213784b PM |
1480 | if (WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), "%s: Could not start rcub kthread, OOM is now expected behavior\n", __func__)) |
1481 | return; | |
aedf4ba9 | 1482 | rcu_for_each_leaf_node(rnp) |
6dbfdc14 | 1483 | (void)rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b | 1484 | } |
f8b7fc6b | 1485 | |
49fb4c62 | 1486 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1487 | { |
da1df50d | 1488 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
f8b7fc6b PM |
1489 | struct rcu_node *rnp = rdp->mynode; |
1490 | ||
1491 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1492 | if (rcu_scheduler_fully_active) |
6dbfdc14 | 1493 | (void)rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b PM |
1494 | } |
1495 | ||
27f4d280 PM |
1496 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1497 | ||
1217ed1b | 1498 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1499 | __releases(rnp->lock) |
27f4d280 | 1500 | { |
67c583a7 | 1501 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1502 | } |
1503 | ||
a46e0899 | 1504 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1505 | { |
a46e0899 | 1506 | WARN_ON_ONCE(1); |
27f4d280 PM |
1507 | } |
1508 | ||
dff1672d PM |
1509 | static bool rcu_is_callbacks_kthread(void) |
1510 | { | |
1511 | return false; | |
1512 | } | |
1513 | ||
27f4d280 PM |
1514 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1515 | { | |
1516 | } | |
1517 | ||
5d01bbd1 | 1518 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1519 | { |
1520 | } | |
1521 | ||
9386c0b7 | 1522 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1523 | { |
b0d30417 | 1524 | } |
b0d30417 | 1525 | |
49fb4c62 | 1526 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1527 | { |
1528 | } | |
1529 | ||
27f4d280 PM |
1530 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1531 | ||
8bd93a2c PM |
1532 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1533 | ||
1534 | /* | |
1535 | * Check to see if any future RCU-related work will need to be done | |
1536 | * by the current CPU, even if none need be done immediately, returning | |
1537 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1538 | * an exported member of the RCU API. | |
1539 | * | |
0ae86a27 PM |
1540 | * Because we not have RCU_FAST_NO_HZ, just check whether or not this |
1541 | * CPU has RCU callbacks queued. | |
8bd93a2c | 1542 | */ |
c1ad348b | 1543 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1544 | { |
c1ad348b | 1545 | *nextevt = KTIME_MAX; |
44c65ff2 | 1546 | return rcu_cpu_has_callbacks(NULL); |
7cb92499 PM |
1547 | } |
1548 | ||
1549 | /* | |
1550 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1551 | * after it. | |
1552 | */ | |
8fa7845d | 1553 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1554 | { |
1555 | } | |
1556 | ||
aea1b35e | 1557 | /* |
a858af28 | 1558 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1559 | * is nothing. |
1560 | */ | |
198bbf81 | 1561 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1562 | { |
1563 | } | |
1564 | ||
c57afe80 PM |
1565 | /* |
1566 | * Don't bother keeping a running count of the number of RCU callbacks | |
1567 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1568 | */ | |
1569 | static void rcu_idle_count_callbacks_posted(void) | |
1570 | { | |
1571 | } | |
1572 | ||
8bd93a2c PM |
1573 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1574 | ||
f23f7fa1 PM |
1575 | /* |
1576 | * This code is invoked when a CPU goes idle, at which point we want | |
1577 | * to have the CPU do everything required for RCU so that it can enter | |
1578 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1579 | * state machine implemented by rcu_prepare_for_idle() below. | |
1580 | * | |
1581 | * The following three proprocessor symbols control this state machine: | |
1582 | * | |
f23f7fa1 PM |
1583 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1584 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1585 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1586 | * benchmarkers who might otherwise be tempted to set this to a large | |
1587 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1588 | * system. And if you are -that- concerned about energy efficiency, | |
1589 | * just power the system down and be done with it! | |
778d250a PM |
1590 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1591 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1592 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1593 | * |
1594 | * The values below work well in practice. If future workloads require | |
1595 | * adjustment, they can be converted into kernel config parameters, though | |
1596 | * making the state machine smarter might be a better option. | |
1597 | */ | |
e84c48ae | 1598 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1599 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1600 | |
5e44ce35 PM |
1601 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1602 | module_param(rcu_idle_gp_delay, int, 0644); | |
1603 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1604 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1605 | |
486e2593 | 1606 | /* |
0ae86a27 PM |
1607 | * Try to advance callbacks on the current CPU, but only if it has been |
1608 | * awhile since the last time we did so. Afterwards, if there are any | |
1609 | * callbacks ready for immediate invocation, return true. | |
486e2593 | 1610 | */ |
f1f399d1 | 1611 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1612 | { |
c0f4dfd4 | 1613 | bool cbs_ready = false; |
5998a75a | 1614 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1615 | struct rcu_node *rnp; |
486e2593 | 1616 | |
c229828c | 1617 | /* Exit early if we advanced recently. */ |
5998a75a | 1618 | if (jiffies == rdp->last_advance_all) |
d0bc90fd | 1619 | return false; |
5998a75a | 1620 | rdp->last_advance_all = jiffies; |
c229828c | 1621 | |
b97d23c5 | 1622 | rnp = rdp->mynode; |
486e2593 | 1623 | |
b97d23c5 PM |
1624 | /* |
1625 | * Don't bother checking unless a grace period has | |
1626 | * completed since we last checked and there are | |
1627 | * callbacks not yet ready to invoke. | |
1628 | */ | |
1629 | if ((rcu_seq_completed_gp(rdp->gp_seq, | |
1630 | rcu_seq_current(&rnp->gp_seq)) || | |
1631 | unlikely(READ_ONCE(rdp->gpwrap))) && | |
1632 | rcu_segcblist_pend_cbs(&rdp->cblist)) | |
1633 | note_gp_changes(rdp); | |
1634 | ||
1635 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
1636 | cbs_ready = true; | |
c0f4dfd4 | 1637 | return cbs_ready; |
486e2593 PM |
1638 | } |
1639 | ||
aa9b1630 | 1640 | /* |
c0f4dfd4 PM |
1641 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1642 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1643 | * caller to set the timeout based on whether or not there are non-lazy | |
1644 | * callbacks. | |
aa9b1630 | 1645 | * |
c0f4dfd4 | 1646 | * The caller must have disabled interrupts. |
aa9b1630 | 1647 | */ |
c1ad348b | 1648 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1649 | { |
5998a75a | 1650 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c1ad348b | 1651 | unsigned long dj; |
aa9b1630 | 1652 | |
b04db8e1 | 1653 | lockdep_assert_irqs_disabled(); |
3382adbc | 1654 | |
c0f4dfd4 | 1655 | /* Snapshot to detect later posting of non-lazy callback. */ |
c458a89e | 1656 | rdp->nonlazy_posted_snap = rdp->nonlazy_posted; |
c0f4dfd4 | 1657 | |
aa9b1630 | 1658 | /* If no callbacks, RCU doesn't need the CPU. */ |
c458a89e | 1659 | if (!rcu_cpu_has_callbacks(&rdp->all_lazy)) { |
c1ad348b | 1660 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1661 | return 0; |
1662 | } | |
c0f4dfd4 PM |
1663 | |
1664 | /* Attempt to advance callbacks. */ | |
1665 | if (rcu_try_advance_all_cbs()) { | |
1666 | /* Some ready to invoke, so initiate later invocation. */ | |
1667 | invoke_rcu_core(); | |
aa9b1630 PM |
1668 | return 1; |
1669 | } | |
5998a75a | 1670 | rdp->last_accelerate = jiffies; |
c0f4dfd4 PM |
1671 | |
1672 | /* Request timer delay depending on laziness, and round. */ | |
c458a89e | 1673 | if (!rdp->all_lazy) { |
c1ad348b | 1674 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1675 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1676 | } else { |
c1ad348b | 1677 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1678 | } |
c1ad348b | 1679 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1680 | return 0; |
1681 | } | |
1682 | ||
21e52e15 | 1683 | /* |
c0f4dfd4 PM |
1684 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1685 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1686 | * The second major task is to check to see if a non-lazy callback has | |
1687 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1688 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1689 | * any recently arrived callbacks. | |
aea1b35e PM |
1690 | * |
1691 | * The caller must have disabled interrupts. | |
8bd93a2c | 1692 | */ |
198bbf81 | 1693 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1694 | { |
48a7639c | 1695 | bool needwake; |
0fd79e75 | 1696 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1697 | struct rcu_node *rnp; |
9d2ad243 PM |
1698 | int tne; |
1699 | ||
b04db8e1 | 1700 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1701 | if (rcu_is_nocb_cpu(smp_processor_id())) |
3382adbc PM |
1702 | return; |
1703 | ||
9d2ad243 | 1704 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1705 | tne = READ_ONCE(tick_nohz_active); |
0fd79e75 | 1706 | if (tne != rdp->tick_nohz_enabled_snap) { |
aa6da514 | 1707 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 | 1708 | invoke_rcu_core(); /* force nohz to see update. */ |
0fd79e75 | 1709 | rdp->tick_nohz_enabled_snap = tne; |
9d2ad243 PM |
1710 | return; |
1711 | } | |
1712 | if (!tne) | |
1713 | return; | |
f511fc62 | 1714 | |
c57afe80 | 1715 | /* |
c0f4dfd4 PM |
1716 | * If a non-lazy callback arrived at a CPU having only lazy |
1717 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1718 | * idle duration on re-entry to idle. | |
c57afe80 | 1719 | */ |
c458a89e PM |
1720 | if (rdp->all_lazy && |
1721 | rdp->nonlazy_posted != rdp->nonlazy_posted_snap) { | |
1722 | rdp->all_lazy = false; | |
1723 | rdp->nonlazy_posted_snap = rdp->nonlazy_posted; | |
c0f4dfd4 | 1724 | invoke_rcu_core(); |
c57afe80 PM |
1725 | return; |
1726 | } | |
c57afe80 | 1727 | |
3084f2f8 | 1728 | /* |
c0f4dfd4 PM |
1729 | * If we have not yet accelerated this jiffy, accelerate all |
1730 | * callbacks on this CPU. | |
3084f2f8 | 1731 | */ |
5998a75a | 1732 | if (rdp->last_accelerate == jiffies) |
aea1b35e | 1733 | return; |
5998a75a | 1734 | rdp->last_accelerate = jiffies; |
b97d23c5 | 1735 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
c0f4dfd4 | 1736 | rnp = rdp->mynode; |
2a67e741 | 1737 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
02f50142 | 1738 | needwake = rcu_accelerate_cbs(rnp, rdp); |
67c583a7 | 1739 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c | 1740 | if (needwake) |
532c00c9 | 1741 | rcu_gp_kthread_wake(); |
77e38ed3 | 1742 | } |
c0f4dfd4 | 1743 | } |
3084f2f8 | 1744 | |
c0f4dfd4 PM |
1745 | /* |
1746 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1747 | * any grace periods that elapsed while the CPU was idle, and if any | |
1748 | * callbacks are now ready to invoke, initiate invocation. | |
1749 | */ | |
8fa7845d | 1750 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1751 | { |
b04db8e1 | 1752 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1753 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1754 | return; |
7a497c96 PM |
1755 | if (rcu_try_advance_all_cbs()) |
1756 | invoke_rcu_core(); | |
8bd93a2c PM |
1757 | } |
1758 | ||
c57afe80 | 1759 | /* |
98248a0e PM |
1760 | * Keep a running count of the number of non-lazy callbacks posted |
1761 | * on this CPU. This running counter (which is never decremented) allows | |
1762 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1763 | * posts a callback, even if an equal number of callbacks are invoked. | |
1764 | * Of course, callbacks should only be posted from within a trace event | |
1765 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1766 | */ |
1767 | static void rcu_idle_count_callbacks_posted(void) | |
1768 | { | |
c458a89e | 1769 | __this_cpu_add(rcu_data.nonlazy_posted, 1); |
c57afe80 PM |
1770 | } |
1771 | ||
8bd93a2c | 1772 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 | 1773 | |
a858af28 PM |
1774 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1775 | ||
1776 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1777 | { | |
0fd79e75 | 1778 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
c458a89e | 1779 | unsigned long nlpd = rdp->nonlazy_posted - rdp->nonlazy_posted_snap; |
a858af28 | 1780 | |
c0f4dfd4 | 1781 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
5998a75a | 1782 | rdp->last_accelerate & 0xffff, jiffies & 0xffff, |
c0f4dfd4 | 1783 | ulong2long(nlpd), |
c458a89e | 1784 | rdp->all_lazy ? 'L' : '.', |
0fd79e75 | 1785 | rdp->tick_nohz_enabled_snap ? '.' : 'D'); |
a858af28 PM |
1786 | } |
1787 | ||
1788 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1789 | ||
1790 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1791 | { | |
1c17e4d4 | 1792 | *cp = '\0'; |
a858af28 PM |
1793 | } |
1794 | ||
1795 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1796 | ||
1797 | /* Initiate the stall-info list. */ | |
1798 | static void print_cpu_stall_info_begin(void) | |
1799 | { | |
efc151c3 | 1800 | pr_cont("\n"); |
a858af28 PM |
1801 | } |
1802 | ||
1803 | /* | |
1804 | * Print out diagnostic information for the specified stalled CPU. | |
1805 | * | |
564a9ae6 PM |
1806 | * If the specified CPU is aware of the current RCU grace period, then |
1807 | * print the number of scheduling clock interrupts the CPU has taken | |
1808 | * during the time that it has been aware. Otherwise, print the number | |
1809 | * of RCU grace periods that this CPU is ignorant of, for example, "1" | |
1810 | * if the CPU was aware of the previous grace period. | |
a858af28 PM |
1811 | * |
1812 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1813 | */ | |
b21ebed9 | 1814 | static void print_cpu_stall_info(int cpu) |
a858af28 | 1815 | { |
9b9500da | 1816 | unsigned long delta; |
a858af28 | 1817 | char fast_no_hz[72]; |
da1df50d | 1818 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
a858af28 PM |
1819 | char *ticks_title; |
1820 | unsigned long ticks_value; | |
1821 | ||
3caa973b TH |
1822 | /* |
1823 | * We could be printing a lot while holding a spinlock. Avoid | |
1824 | * triggering hard lockup. | |
1825 | */ | |
1826 | touch_nmi_watchdog(); | |
1827 | ||
b21ebed9 | 1828 | ticks_value = rcu_seq_ctr(rcu_state.gp_seq - rdp->gp_seq); |
471f87c3 PM |
1829 | if (ticks_value) { |
1830 | ticks_title = "GPs behind"; | |
1831 | } else { | |
a858af28 PM |
1832 | ticks_title = "ticks this GP"; |
1833 | ticks_value = rdp->ticks_this_gp; | |
a858af28 PM |
1834 | } |
1835 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
8aa670cd | 1836 | delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq); |
89b4cd4b | 1837 | pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%#lx softirq=%u/%u fqs=%ld %s\n", |
7f21aeef PM |
1838 | cpu, |
1839 | "O."[!!cpu_online(cpu)], | |
1840 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1841 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
9b9500da PM |
1842 | !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' : |
1843 | rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : | |
1844 | "!."[!delta], | |
7f21aeef | 1845 | ticks_value, ticks_title, |
dc5a4f29 | 1846 | rcu_dynticks_snap(rdp) & 0xfff, |
4c5273bf | 1847 | rdp->dynticks_nesting, rdp->dynticks_nmi_nesting, |
6231069b | 1848 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
b21ebed9 | 1849 | READ_ONCE(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart, |
a858af28 PM |
1850 | fast_no_hz); |
1851 | } | |
1852 | ||
1853 | /* Terminate the stall-info list. */ | |
1854 | static void print_cpu_stall_info_end(void) | |
1855 | { | |
efc151c3 | 1856 | pr_err("\t"); |
a858af28 PM |
1857 | } |
1858 | ||
0ae86a27 | 1859 | /* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */ |
a858af28 PM |
1860 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) |
1861 | { | |
1862 | rdp->ticks_this_gp = 0; | |
6231069b | 1863 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
d3052109 | 1864 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
a858af28 PM |
1865 | } |
1866 | ||
3fbfbf7a PM |
1867 | #ifdef CONFIG_RCU_NOCB_CPU |
1868 | ||
1869 | /* | |
1870 | * Offload callback processing from the boot-time-specified set of CPUs | |
1871 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1872 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1873 | * waits for a grace period to elapse, and invokes the callbacks. | |
1874 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1875 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1876 | * has been specified, in which case each kthread actively polls its | |
1877 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1878 | * reduce RCU's overhead on that CPU.) | |
1879 | * | |
1880 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1881 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1882 | * running CPU-bound user-mode computations. | |
1883 | * | |
1884 | * Offloading of callback processing could also in theory be used as | |
1885 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1886 | * queued are more aggressive about entering dyntick-idle mode. | |
1887 | */ | |
1888 | ||
1889 | ||
1890 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1891 | static int __init rcu_nocb_setup(char *str) | |
1892 | { | |
1893 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
3fbfbf7a PM |
1894 | cpulist_parse(str, rcu_nocb_mask); |
1895 | return 1; | |
1896 | } | |
1897 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1898 | ||
1b0048a4 PG |
1899 | static int __init parse_rcu_nocb_poll(char *arg) |
1900 | { | |
5455a7f6 | 1901 | rcu_nocb_poll = true; |
1b0048a4 PG |
1902 | return 0; |
1903 | } | |
1904 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1905 | ||
dae6e64d | 1906 | /* |
0446be48 PM |
1907 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1908 | * grace period. | |
dae6e64d | 1909 | */ |
abedf8e2 | 1910 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1911 | { |
abedf8e2 | 1912 | swake_up_all(sq); |
dae6e64d PM |
1913 | } |
1914 | ||
abedf8e2 | 1915 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1916 | { |
e0da2374 | 1917 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1918 | } |
1919 | ||
dae6e64d | 1920 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1921 | { |
abedf8e2 PG |
1922 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1923 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1924 | } |
1925 | ||
24342c96 | 1926 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1927 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1928 | { |
84b12b75 | 1929 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1930 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1931 | return false; | |
1932 | } | |
1933 | ||
fbce7497 | 1934 | /* |
8be6e1b1 PM |
1935 | * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock |
1936 | * and this function releases it. | |
fbce7497 | 1937 | */ |
8be6e1b1 PM |
1938 | static void __wake_nocb_leader(struct rcu_data *rdp, bool force, |
1939 | unsigned long flags) | |
1940 | __releases(rdp->nocb_lock) | |
fbce7497 PM |
1941 | { |
1942 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1943 | ||
8be6e1b1 PM |
1944 | lockdep_assert_held(&rdp->nocb_lock); |
1945 | if (!READ_ONCE(rdp_leader->nocb_kthread)) { | |
1946 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 1947 | return; |
8be6e1b1 PM |
1948 | } |
1949 | if (rdp_leader->nocb_leader_sleep || force) { | |
39953dfd | 1950 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1951 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
8be6e1b1 PM |
1952 | del_timer(&rdp->nocb_timer); |
1953 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
b3dae109 PZ |
1954 | smp_mb(); /* ->nocb_leader_sleep before swake_up_one(). */ |
1955 | swake_up_one(&rdp_leader->nocb_wq); | |
8be6e1b1 PM |
1956 | } else { |
1957 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 PM |
1958 | } |
1959 | } | |
1960 | ||
8be6e1b1 PM |
1961 | /* |
1962 | * Kick the leader kthread for this NOCB group, but caller has not | |
1963 | * acquired locks. | |
1964 | */ | |
1965 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1966 | { | |
1967 | unsigned long flags; | |
1968 | ||
1969 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1970 | __wake_nocb_leader(rdp, force, flags); | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * Arrange to wake the leader kthread for this NOCB group at some | |
1975 | * future time when it is safe to do so. | |
1976 | */ | |
1977 | static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype, | |
1978 | const char *reason) | |
1979 | { | |
1980 | unsigned long flags; | |
1981 | ||
1982 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1983 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) | |
1984 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
1985 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
88d1bead | 1986 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, reason); |
8be6e1b1 PM |
1987 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
1988 | } | |
1989 | ||
d7e29933 | 1990 | /* |
0ae86a27 | 1991 | * Does the specified CPU need an RCU callback for this invocation |
d7e29933 PM |
1992 | * of rcu_barrier()? |
1993 | */ | |
4580b054 | 1994 | static bool rcu_nocb_cpu_needs_barrier(int cpu) |
d7e29933 | 1995 | { |
da1df50d | 1996 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
41050a00 PM |
1997 | unsigned long ret; |
1998 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1999 | struct rcu_head *rhp; |
41050a00 | 2000 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 2001 | |
41050a00 PM |
2002 | /* |
2003 | * Check count of all no-CBs callbacks awaiting invocation. | |
2004 | * There needs to be a barrier before this function is called, | |
2005 | * but associated with a prior determination that no more | |
2006 | * callbacks would be posted. In the worst case, the first | |
dd46a788 | 2007 | * barrier in rcu_barrier() suffices (but the caller cannot |
41050a00 PM |
2008 | * necessarily rely on this, not a substitute for the caller |
2009 | * getting the concurrency design right!). There must also be | |
2010 | * a barrier between the following load an posting of a callback | |
2011 | * (if a callback is in fact needed). This is associated with an | |
2012 | * atomic_inc() in the caller. | |
2013 | */ | |
903ee83d | 2014 | ret = rcu_get_n_cbs_nocb_cpu(rdp); |
d7e29933 | 2015 | |
41050a00 | 2016 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 2017 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 2018 | if (!rhp) |
7d0ae808 | 2019 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 2020 | if (!rhp) |
7d0ae808 | 2021 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
2022 | |
2023 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 2024 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 2025 | rcu_scheduler_fully_active) { |
d7e29933 PM |
2026 | /* RCU callback enqueued before CPU first came online??? */ |
2027 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
2028 | cpu, rhp->func); | |
2029 | WARN_ON_ONCE(1); | |
2030 | } | |
41050a00 | 2031 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 2032 | |
41050a00 | 2033 | return !!ret; |
d7e29933 PM |
2034 | } |
2035 | ||
3fbfbf7a PM |
2036 | /* |
2037 | * Enqueue the specified string of rcu_head structures onto the specified | |
2038 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
2039 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
2040 | * counts are supplied by rhcount and rhcount_lazy. | |
2041 | * | |
2042 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
2043 | */ | |
2044 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
2045 | struct rcu_head *rhp, | |
2046 | struct rcu_head **rhtp, | |
96d3fd0d PM |
2047 | int rhcount, int rhcount_lazy, |
2048 | unsigned long flags) | |
3fbfbf7a PM |
2049 | { |
2050 | int len; | |
2051 | struct rcu_head **old_rhpp; | |
2052 | struct task_struct *t; | |
2053 | ||
2054 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
2055 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
2056 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 2057 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 2058 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 2059 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 2060 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
2061 | |
2062 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 2063 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 2064 | if (rcu_nocb_poll || !t) { |
88d1bead | 2065 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9261dd0d | 2066 | TPS("WakeNotPoll")); |
3fbfbf7a | 2067 | return; |
9261dd0d | 2068 | } |
903ee83d | 2069 | len = rcu_get_n_cbs_nocb_cpu(rdp); |
3fbfbf7a | 2070 | if (old_rhpp == &rdp->nocb_head) { |
96d3fd0d | 2071 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
2072 | /* ... if queue was empty ... */ |
2073 | wake_nocb_leader(rdp, false); | |
88d1bead | 2074 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
96d3fd0d PM |
2075 | TPS("WakeEmpty")); |
2076 | } else { | |
8be6e1b1 PM |
2077 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
2078 | TPS("WakeEmptyIsDeferred")); | |
96d3fd0d | 2079 | } |
3fbfbf7a PM |
2080 | rdp->qlen_last_fqs_check = 0; |
2081 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 2082 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
2083 | if (!irqs_disabled_flags(flags)) { |
2084 | wake_nocb_leader(rdp, true); | |
88d1bead | 2085 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9fdd3bc9 PM |
2086 | TPS("WakeOvf")); |
2087 | } else { | |
efcd2d54 | 2088 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE_FORCE, |
8be6e1b1 | 2089 | TPS("WakeOvfIsDeferred")); |
9fdd3bc9 | 2090 | } |
3fbfbf7a | 2091 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d | 2092 | } else { |
88d1bead | 2093 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WakeNot")); |
3fbfbf7a PM |
2094 | } |
2095 | return; | |
2096 | } | |
2097 | ||
2098 | /* | |
2099 | * This is a helper for __call_rcu(), which invokes this when the normal | |
2100 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
2101 | * function returns failure back to __call_rcu(), which can complain | |
2102 | * appropriately. | |
2103 | * | |
2104 | * Otherwise, this function queues the callback where the corresponding | |
2105 | * "rcuo" kthread can find it. | |
2106 | */ | |
2107 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 2108 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
2109 | { |
2110 | ||
d1e43fa5 | 2111 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 2112 | return false; |
96d3fd0d | 2113 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 | 2114 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
88d1bead | 2115 | trace_rcu_kfree_callback(rcu_state.name, rhp, |
21e7a608 | 2116 | (unsigned long)rhp->func, |
756cbf6b | 2117 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2118 | -rcu_get_n_cbs_nocb_cpu(rdp)); |
21e7a608 | 2119 | else |
88d1bead | 2120 | trace_rcu_callback(rcu_state.name, rhp, |
756cbf6b | 2121 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2122 | -rcu_get_n_cbs_nocb_cpu(rdp)); |
1772947b PM |
2123 | |
2124 | /* | |
2125 | * If called from an extended quiescent state with interrupts | |
2126 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2127 | * deferred-wakeup check to function. | |
2128 | */ | |
2129 | if (irqs_disabled_flags(flags) && | |
2130 | !rcu_is_watching() && | |
2131 | cpu_online(smp_processor_id())) | |
2132 | invoke_rcu_core(); | |
2133 | ||
c271d3a9 | 2134 | return true; |
3fbfbf7a PM |
2135 | } |
2136 | ||
2137 | /* | |
2138 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2139 | * not a no-CBs CPU. | |
2140 | */ | |
b1a2d79f | 2141 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2142 | struct rcu_data *rdp, |
2143 | unsigned long flags) | |
3fbfbf7a | 2144 | { |
b04db8e1 | 2145 | lockdep_assert_irqs_disabled(); |
d1e43fa5 | 2146 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
b1a2d79f PM |
2147 | return false; /* Not NOCBs CPU, caller must migrate CBs. */ |
2148 | __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist), | |
2149 | rcu_segcblist_tail(&rdp->cblist), | |
2150 | rcu_segcblist_n_cbs(&rdp->cblist), | |
2151 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags); | |
2152 | rcu_segcblist_init(&rdp->cblist); | |
2153 | rcu_segcblist_disable(&rdp->cblist); | |
0a9e1e11 | 2154 | return true; |
3fbfbf7a PM |
2155 | } |
2156 | ||
2157 | /* | |
34ed6246 PM |
2158 | * If necessary, kick off a new grace period, and either way wait |
2159 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2160 | */ |
34ed6246 | 2161 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2162 | { |
34ed6246 | 2163 | unsigned long c; |
dae6e64d | 2164 | bool d; |
34ed6246 | 2165 | unsigned long flags; |
48a7639c | 2166 | bool needwake; |
34ed6246 PM |
2167 | struct rcu_node *rnp = rdp->mynode; |
2168 | ||
ab5e869c | 2169 | local_irq_save(flags); |
88d1bead | 2170 | c = rcu_seq_snap(&rcu_state.gp_seq); |
ab5e869c PM |
2171 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
2172 | local_irq_restore(flags); | |
2173 | } else { | |
2174 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
2175 | needwake = rcu_start_this_gp(rnp, rdp, c); | |
2176 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2177 | if (needwake) | |
532c00c9 | 2178 | rcu_gp_kthread_wake(); |
ab5e869c | 2179 | } |
3fbfbf7a PM |
2180 | |
2181 | /* | |
34ed6246 PM |
2182 | * Wait for the grace period. Do so interruptibly to avoid messing |
2183 | * up the load average. | |
3fbfbf7a | 2184 | */ |
41e80595 | 2185 | trace_rcu_this_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2186 | for (;;) { |
b3dae109 | 2187 | swait_event_interruptible_exclusive( |
29365e56 PM |
2188 | rnp->nocb_gp_wq[rcu_seq_ctr(c) & 0x1], |
2189 | (d = rcu_seq_done(&rnp->gp_seq, c))); | |
dae6e64d | 2190 | if (likely(d)) |
34ed6246 | 2191 | break; |
73a860cd | 2192 | WARN_ON(signal_pending(current)); |
41e80595 | 2193 | trace_rcu_this_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2194 | } |
41e80595 | 2195 | trace_rcu_this_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2196 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2197 | } |
2198 | ||
fbce7497 PM |
2199 | /* |
2200 | * Leaders come here to wait for additional callbacks to show up. | |
2201 | * This function does not return until callbacks appear. | |
2202 | */ | |
2203 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2204 | { | |
2205 | bool firsttime = true; | |
8be6e1b1 | 2206 | unsigned long flags; |
fbce7497 PM |
2207 | bool gotcbs; |
2208 | struct rcu_data *rdp; | |
2209 | struct rcu_head **tail; | |
2210 | ||
2211 | wait_again: | |
2212 | ||
2213 | /* Wait for callbacks to appear. */ | |
2214 | if (!rcu_nocb_poll) { | |
88d1bead | 2215 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Sleep")); |
b3dae109 | 2216 | swait_event_interruptible_exclusive(my_rdp->nocb_wq, |
7d0ae808 | 2217 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
8be6e1b1 PM |
2218 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); |
2219 | my_rdp->nocb_leader_sleep = true; | |
2220 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); | |
2221 | del_timer(&my_rdp->nocb_timer); | |
2222 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); | |
fbce7497 PM |
2223 | } else if (firsttime) { |
2224 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
88d1bead | 2225 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Poll")); |
fbce7497 PM |
2226 | } |
2227 | ||
2228 | /* | |
2229 | * Each pass through the following loop checks a follower for CBs. | |
2230 | * We are our own first follower. Any CBs found are moved to | |
2231 | * nocb_gp_head, where they await a grace period. | |
2232 | */ | |
2233 | gotcbs = false; | |
8be6e1b1 | 2234 | smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */ |
fbce7497 | 2235 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2236 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2237 | if (!rdp->nocb_gp_head) |
2238 | continue; /* No CBs here, try next follower. */ | |
2239 | ||
2240 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2241 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2242 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2243 | gotcbs = true; |
2244 | } | |
2245 | ||
8be6e1b1 | 2246 | /* No callbacks? Sleep a bit if polling, and go retry. */ |
fbce7497 | 2247 | if (unlikely(!gotcbs)) { |
73a860cd | 2248 | WARN_ON(signal_pending(current)); |
8be6e1b1 PM |
2249 | if (rcu_nocb_poll) { |
2250 | schedule_timeout_interruptible(1); | |
2251 | } else { | |
88d1bead | 2252 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, |
bedbb648 | 2253 | TPS("WokeEmpty")); |
8be6e1b1 | 2254 | } |
fbce7497 PM |
2255 | goto wait_again; |
2256 | } | |
2257 | ||
2258 | /* Wait for one grace period. */ | |
2259 | rcu_nocb_wait_gp(my_rdp); | |
2260 | ||
fbce7497 PM |
2261 | /* Each pass through the following loop wakes a follower, if needed. */ |
2262 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
8be6e1b1 PM |
2263 | if (!rcu_nocb_poll && |
2264 | READ_ONCE(rdp->nocb_head) && | |
2265 | READ_ONCE(my_rdp->nocb_leader_sleep)) { | |
2266 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); | |
11ed7f93 | 2267 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
8be6e1b1 PM |
2268 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); |
2269 | } | |
fbce7497 PM |
2270 | if (!rdp->nocb_gp_head) |
2271 | continue; /* No CBs, so no need to wake follower. */ | |
2272 | ||
2273 | /* Append callbacks to follower's "done" list. */ | |
8be6e1b1 PM |
2274 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2275 | tail = rdp->nocb_follower_tail; | |
2276 | rdp->nocb_follower_tail = rdp->nocb_gp_tail; | |
fbce7497 | 2277 | *tail = rdp->nocb_gp_head; |
8be6e1b1 | 2278 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
fbce7497 | 2279 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
8be6e1b1 | 2280 | /* List was empty, so wake up the follower. */ |
b3dae109 | 2281 | swake_up_one(&rdp->nocb_wq); |
fbce7497 PM |
2282 | } |
2283 | } | |
2284 | ||
2285 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2286 | if (!my_rdp->nocb_follower_head) | |
2287 | goto wait_again; | |
2288 | } | |
2289 | ||
2290 | /* | |
2291 | * Followers come here to wait for additional callbacks to show up. | |
2292 | * This function does not return until callbacks appear. | |
2293 | */ | |
2294 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2295 | { | |
fbce7497 | 2296 | for (;;) { |
88d1bead | 2297 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("FollowerSleep")); |
b3dae109 | 2298 | swait_event_interruptible_exclusive(rdp->nocb_wq, |
8be6e1b1 | 2299 | READ_ONCE(rdp->nocb_follower_head)); |
fbce7497 PM |
2300 | if (smp_load_acquire(&rdp->nocb_follower_head)) { |
2301 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2302 | return; | |
2303 | } | |
73a860cd | 2304 | WARN_ON(signal_pending(current)); |
88d1bead | 2305 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty")); |
fbce7497 PM |
2306 | } |
2307 | } | |
2308 | ||
3fbfbf7a PM |
2309 | /* |
2310 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2311 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2312 | * an optional leader-follower relationship so that the grace-period | |
2313 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2314 | */ |
2315 | static int rcu_nocb_kthread(void *arg) | |
2316 | { | |
2317 | int c, cl; | |
8be6e1b1 | 2318 | unsigned long flags; |
3fbfbf7a PM |
2319 | struct rcu_head *list; |
2320 | struct rcu_head *next; | |
2321 | struct rcu_head **tail; | |
2322 | struct rcu_data *rdp = arg; | |
2323 | ||
2324 | /* Each pass through this loop invokes one batch of callbacks */ | |
2325 | for (;;) { | |
fbce7497 PM |
2326 | /* Wait for callbacks. */ |
2327 | if (rdp->nocb_leader == rdp) | |
2328 | nocb_leader_wait(rdp); | |
2329 | else | |
2330 | nocb_follower_wait(rdp); | |
2331 | ||
2332 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
8be6e1b1 PM |
2333 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2334 | list = rdp->nocb_follower_head; | |
2335 | rdp->nocb_follower_head = NULL; | |
2336 | tail = rdp->nocb_follower_tail; | |
2337 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; | |
2338 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
9213784b PM |
2339 | if (WARN_ON_ONCE(!list)) |
2340 | continue; | |
88d1bead | 2341 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeNonEmpty")); |
3fbfbf7a PM |
2342 | |
2343 | /* Each pass through the following loop invokes a callback. */ | |
88d1bead | 2344 | trace_rcu_batch_start(rcu_state.name, |
41050a00 | 2345 | atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2346 | rcu_get_n_cbs_nocb_cpu(rdp), -1); |
3fbfbf7a PM |
2347 | c = cl = 0; |
2348 | while (list) { | |
2349 | next = list->next; | |
2350 | /* Wait for enqueuing to complete, if needed. */ | |
2351 | while (next == NULL && &list->next != tail) { | |
88d1bead | 2352 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
69a79bb1 | 2353 | TPS("WaitQueue")); |
3fbfbf7a | 2354 | schedule_timeout_interruptible(1); |
88d1bead | 2355 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
69a79bb1 | 2356 | TPS("WokeQueue")); |
3fbfbf7a PM |
2357 | next = list->next; |
2358 | } | |
2359 | debug_rcu_head_unqueue(list); | |
2360 | local_bh_disable(); | |
88d1bead | 2361 | if (__rcu_reclaim(rcu_state.name, list)) |
3fbfbf7a PM |
2362 | cl++; |
2363 | c++; | |
2364 | local_bh_enable(); | |
cee43939 | 2365 | cond_resched_tasks_rcu_qs(); |
3fbfbf7a PM |
2366 | list = next; |
2367 | } | |
88d1bead | 2368 | trace_rcu_batch_end(rcu_state.name, c, !!list, 0, 0, 1); |
41050a00 PM |
2369 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2370 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2371 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
3fbfbf7a PM |
2372 | } |
2373 | return 0; | |
2374 | } | |
2375 | ||
96d3fd0d | 2376 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2377 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2378 | { |
7d0ae808 | 2379 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2380 | } |
2381 | ||
2382 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2383 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2384 | { |
8be6e1b1 | 2385 | unsigned long flags; |
9fdd3bc9 PM |
2386 | int ndw; |
2387 | ||
8be6e1b1 PM |
2388 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2389 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { | |
2390 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
96d3fd0d | 2391 | return; |
8be6e1b1 | 2392 | } |
7d0ae808 | 2393 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2394 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
8be6e1b1 | 2395 | __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
88d1bead | 2396 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2397 | } |
2398 | ||
8be6e1b1 | 2399 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2400 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2401 | { |
fd30b717 KC |
2402 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2403 | ||
2404 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2405 | } |
2406 | ||
2407 | /* | |
2408 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2409 | * This means we do an inexact common-case check. Note that if | |
2410 | * we miss, ->nocb_timer will eventually clean things up. | |
2411 | */ | |
2412 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2413 | { | |
2414 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2415 | do_nocb_deferred_wakeup_common(rdp); | |
2416 | } | |
2417 | ||
f4579fc5 PM |
2418 | void __init rcu_init_nohz(void) |
2419 | { | |
2420 | int cpu; | |
ef126206 | 2421 | bool need_rcu_nocb_mask = false; |
f4579fc5 | 2422 | |
f4579fc5 PM |
2423 | #if defined(CONFIG_NO_HZ_FULL) |
2424 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2425 | need_rcu_nocb_mask = true; | |
2426 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2427 | ||
84b12b75 | 2428 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
2429 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2430 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2431 | return; | |
2432 | } | |
f4579fc5 | 2433 | } |
84b12b75 | 2434 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
2435 | return; |
2436 | ||
f4579fc5 PM |
2437 | #if defined(CONFIG_NO_HZ_FULL) |
2438 | if (tick_nohz_full_running) | |
2439 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2440 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2441 | ||
2442 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 2443 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
2444 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
2445 | rcu_nocb_mask); | |
2446 | } | |
3016611e PM |
2447 | if (cpumask_empty(rcu_nocb_mask)) |
2448 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
2449 | else | |
2450 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
2451 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2452 | if (rcu_nocb_poll) |
2453 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2454 | ||
b97d23c5 PM |
2455 | for_each_cpu(cpu, rcu_nocb_mask) |
2456 | init_nocb_callback_list(per_cpu_ptr(&rcu_data, cpu)); | |
2457 | rcu_organize_nocb_kthreads(); | |
96d3fd0d PM |
2458 | } |
2459 | ||
3fbfbf7a PM |
2460 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2461 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2462 | { | |
2463 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2464 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2465 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
8be6e1b1 | 2466 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 2467 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
2468 | } |
2469 | ||
35ce7f29 PM |
2470 | /* |
2471 | * If the specified CPU is a no-CBs CPU that does not already have its | |
0ae86a27 PM |
2472 | * rcuo kthread, spawn it. If the CPUs are brought online out of order, |
2473 | * this can require re-organizing the leader-follower relationships. | |
35ce7f29 | 2474 | */ |
4580b054 | 2475 | static void rcu_spawn_one_nocb_kthread(int cpu) |
35ce7f29 PM |
2476 | { |
2477 | struct rcu_data *rdp; | |
2478 | struct rcu_data *rdp_last; | |
2479 | struct rcu_data *rdp_old_leader; | |
da1df50d | 2480 | struct rcu_data *rdp_spawn = per_cpu_ptr(&rcu_data, cpu); |
35ce7f29 PM |
2481 | struct task_struct *t; |
2482 | ||
2483 | /* | |
2484 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2485 | * then nothing to do. | |
2486 | */ | |
2487 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2488 | return; | |
2489 | ||
2490 | /* If we didn't spawn the leader first, reorganize! */ | |
2491 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2492 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2493 | rdp_last = NULL; | |
2494 | rdp = rdp_old_leader; | |
2495 | do { | |
2496 | rdp->nocb_leader = rdp_spawn; | |
2497 | if (rdp_last && rdp != rdp_spawn) | |
2498 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2499 | if (rdp == rdp_spawn) { |
2500 | rdp = rdp->nocb_next_follower; | |
2501 | } else { | |
2502 | rdp_last = rdp; | |
2503 | rdp = rdp->nocb_next_follower; | |
2504 | rdp_last->nocb_next_follower = NULL; | |
2505 | } | |
35ce7f29 PM |
2506 | } while (rdp); |
2507 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2508 | } | |
2509 | ||
0ae86a27 | 2510 | /* Spawn the kthread for this CPU. */ |
35ce7f29 | 2511 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, |
4580b054 | 2512 | "rcuo%c/%d", rcu_state.abbr, cpu); |
9213784b PM |
2513 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo kthread, OOM is now expected behavior\n", __func__)) |
2514 | return; | |
7d0ae808 | 2515 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2520 | * rcuo kthreads, spawn them. | |
2521 | */ | |
2522 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2523 | { | |
35ce7f29 | 2524 | if (rcu_scheduler_fully_active) |
b97d23c5 | 2525 | rcu_spawn_one_nocb_kthread(cpu); |
35ce7f29 PM |
2526 | } |
2527 | ||
2528 | /* | |
2529 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2530 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2531 | * non-boot CPUs come online -- if this changes, we will need to add | |
2532 | * some mutual exclusion. | |
2533 | */ | |
2534 | static void __init rcu_spawn_nocb_kthreads(void) | |
2535 | { | |
2536 | int cpu; | |
2537 | ||
2538 | for_each_online_cpu(cpu) | |
2539 | rcu_spawn_all_nocb_kthreads(cpu); | |
2540 | } | |
2541 | ||
fbce7497 PM |
2542 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2543 | static int rcu_nocb_leader_stride = -1; | |
2544 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2545 | ||
2546 | /* | |
35ce7f29 | 2547 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2548 | */ |
4580b054 | 2549 | static void __init rcu_organize_nocb_kthreads(void) |
3fbfbf7a PM |
2550 | { |
2551 | int cpu; | |
fbce7497 PM |
2552 | int ls = rcu_nocb_leader_stride; |
2553 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2554 | struct rcu_data *rdp; |
fbce7497 PM |
2555 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2556 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2557 | |
84b12b75 | 2558 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2559 | return; |
fbce7497 PM |
2560 | if (ls == -1) { |
2561 | ls = int_sqrt(nr_cpu_ids); | |
2562 | rcu_nocb_leader_stride = ls; | |
2563 | } | |
2564 | ||
2565 | /* | |
9831ce3b PM |
2566 | * Each pass through this loop sets up one rcu_data structure. |
2567 | * Should the corresponding CPU come online in the future, then | |
2568 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2569 | */ |
3fbfbf7a | 2570 | for_each_cpu(cpu, rcu_nocb_mask) { |
da1df50d | 2571 | rdp = per_cpu_ptr(&rcu_data, cpu); |
fbce7497 PM |
2572 | if (rdp->cpu >= nl) { |
2573 | /* New leader, set up for followers & next leader. */ | |
2574 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2575 | rdp->nocb_leader = rdp; | |
2576 | rdp_leader = rdp; | |
2577 | } else { | |
2578 | /* Another follower, link to previous leader. */ | |
2579 | rdp->nocb_leader = rdp_leader; | |
2580 | rdp_prev->nocb_next_follower = rdp; | |
2581 | } | |
2582 | rdp_prev = rdp; | |
3fbfbf7a PM |
2583 | } |
2584 | } | |
2585 | ||
2586 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2587 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2588 | { |
22c2f669 | 2589 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2590 | return false; |
22c2f669 | 2591 | |
34404ca8 | 2592 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2593 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2594 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2595 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2596 | atomic_long_set(&rdp->nocb_q_count, | |
2597 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2598 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2599 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2600 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2601 | } |
15fecf89 | 2602 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2603 | return true; |
3fbfbf7a PM |
2604 | } |
2605 | ||
5ab7ab83 PM |
2606 | /* |
2607 | * Bind the current task to the offloaded CPUs. If there are no offloaded | |
2608 | * CPUs, leave the task unbound. Splat if the bind attempt fails. | |
2609 | */ | |
2610 | void rcu_bind_current_to_nocb(void) | |
2611 | { | |
2612 | if (cpumask_available(rcu_nocb_mask) && cpumask_weight(rcu_nocb_mask)) | |
2613 | WARN_ON(sched_setaffinity(current->pid, rcu_nocb_mask)); | |
2614 | } | |
2615 | EXPORT_SYMBOL_GPL(rcu_bind_current_to_nocb); | |
2616 | ||
903ee83d PM |
2617 | /* |
2618 | * Return the number of RCU callbacks still queued from the specified | |
2619 | * CPU, which must be a nocbs CPU. | |
2620 | */ | |
2621 | static unsigned long rcu_get_n_cbs_nocb_cpu(struct rcu_data *rdp) | |
2622 | { | |
2623 | return atomic_long_read(&rdp->nocb_q_count); | |
2624 | } | |
2625 | ||
34ed6246 PM |
2626 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2627 | ||
4580b054 | 2628 | static bool rcu_nocb_cpu_needs_barrier(int cpu) |
d7e29933 PM |
2629 | { |
2630 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2631 | return false; | |
2632 | } | |
2633 | ||
abedf8e2 | 2634 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2635 | { |
3fbfbf7a PM |
2636 | } |
2637 | ||
abedf8e2 | 2638 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2639 | { |
2640 | return NULL; | |
2641 | } | |
2642 | ||
dae6e64d PM |
2643 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2644 | { | |
2645 | } | |
3fbfbf7a | 2646 | |
3fbfbf7a | 2647 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2648 | bool lazy, unsigned long flags) |
3fbfbf7a | 2649 | { |
4afc7e26 | 2650 | return false; |
3fbfbf7a PM |
2651 | } |
2652 | ||
b1a2d79f | 2653 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2654 | struct rcu_data *rdp, |
2655 | unsigned long flags) | |
3fbfbf7a | 2656 | { |
f4aa84ba | 2657 | return false; |
3fbfbf7a PM |
2658 | } |
2659 | ||
3fbfbf7a PM |
2660 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2661 | { | |
2662 | } | |
2663 | ||
9fdd3bc9 | 2664 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2665 | { |
2666 | return false; | |
2667 | } | |
2668 | ||
2669 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2670 | { | |
2671 | } | |
2672 | ||
35ce7f29 PM |
2673 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2674 | { | |
2675 | } | |
2676 | ||
2677 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2678 | { |
2679 | } | |
2680 | ||
34ed6246 | 2681 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2682 | { |
34ed6246 | 2683 | return false; |
3fbfbf7a PM |
2684 | } |
2685 | ||
903ee83d PM |
2686 | static unsigned long rcu_get_n_cbs_nocb_cpu(struct rcu_data *rdp) |
2687 | { | |
2688 | return 0; | |
2689 | } | |
2690 | ||
3fbfbf7a | 2691 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ |
65d798f0 | 2692 | |
a096932f PM |
2693 | /* |
2694 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2695 | * grace-period kthread will do force_quiescent_state() processing? | |
2696 | * The idea is to avoid waking up RCU core processing on such a | |
2697 | * CPU unless the grace period has extended for too long. | |
2698 | * | |
2699 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2700 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f | 2701 | */ |
4580b054 | 2702 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
2703 | { |
2704 | #ifdef CONFIG_NO_HZ_FULL | |
2705 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 2706 | (!rcu_gp_in_progress() || |
4580b054 | 2707 | ULONG_CMP_LT(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 2708 | return true; |
a096932f | 2709 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2710 | return false; |
a096932f | 2711 | } |
5057f55e PM |
2712 | |
2713 | /* | |
265f5f28 | 2714 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2715 | */ |
2716 | static void rcu_bind_gp_kthread(void) | |
2717 | { | |
c0f489d2 | 2718 | if (!tick_nohz_full_enabled()) |
5057f55e | 2719 | return; |
de201559 | 2720 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2721 | } |
176f8f7a PM |
2722 | |
2723 | /* Record the current task on dyntick-idle entry. */ | |
2724 | static void rcu_dynticks_task_enter(void) | |
2725 | { | |
2726 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2727 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2728 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2729 | } | |
2730 | ||
2731 | /* Record no current task on dyntick-idle exit. */ | |
2732 | static void rcu_dynticks_task_exit(void) | |
2733 | { | |
2734 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2735 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2736 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2737 | } |