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> |
62ab7072 | 30 | #include <linux/smpboot.h> |
4102adab | 31 | #include "../time/tick-internal.h" |
f41d911f | 32 | |
5b61b0ba | 33 | #ifdef CONFIG_RCU_BOOST |
61cfd097 | 34 | |
abaa93d9 | 35 | #include "../locking/rtmutex_common.h" |
21871d7e | 36 | |
61cfd097 PM |
37 | /* |
38 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
39 | * handle all flavors of RCU. | |
40 | */ | |
41 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
42 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
43 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
44 | DEFINE_PER_CPU(char, rcu_cpu_has_work); | |
45 | ||
21871d7e | 46 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
5b61b0ba | 47 | |
3fbfbf7a PM |
48 | #ifdef CONFIG_RCU_NOCB_CPU |
49 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
50 | static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */ | |
1b0048a4 | 51 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
52 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
53 | ||
26845c28 PM |
54 | /* |
55 | * Check the RCU kernel configuration parameters and print informative | |
56 | * messages about anything out of the ordinary. If you like #ifdef, you | |
57 | * will love this function. | |
58 | */ | |
59 | static void __init rcu_bootup_announce_oddness(void) | |
60 | { | |
ab6f5bd6 PM |
61 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
62 | pr_info("\tRCU debugfs-based tracing is enabled.\n"); | |
63 | if ((IS_ENABLED(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || | |
64 | (!IS_ENABLED(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32)) | |
65 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", | |
66 | CONFIG_RCU_FANOUT); | |
67 | if (IS_ENABLED(CONFIG_RCU_FANOUT_EXACT)) | |
68 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); | |
69 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
70 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
71 | if (IS_ENABLED(CONFIG_PROVE_RCU)) | |
72 | pr_info("\tRCU lockdep checking is enabled.\n"); | |
73 | if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_RUNNABLE)) | |
74 | pr_info("\tRCU torture testing starts during boot.\n"); | |
75 | if (IS_ENABLED(CONFIG_RCU_CPU_STALL_INFO)) | |
76 | pr_info("\tAdditional per-CPU info printed with stalls.\n"); | |
77 | if (NUM_RCU_LVL_4 != 0) | |
78 | pr_info("\tFour-level hierarchy is enabled.\n"); | |
f885b7f2 | 79 | if (rcu_fanout_leaf != CONFIG_RCU_FANOUT_LEAF) |
9a5739d7 | 80 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 81 | if (nr_cpu_ids != NR_CPUS) |
efc151c3 | 82 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS, nr_cpu_ids); |
ab6f5bd6 PM |
83 | if (IS_ENABLED(CONFIG_RCU_BOOST)) |
84 | pr_info("\tRCU kthread priority: %d.\n", kthread_prio); | |
26845c28 PM |
85 | } |
86 | ||
28f6569a | 87 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 88 | |
a41bfeb2 | 89 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
e534165b | 90 | static struct rcu_state *rcu_state_p = &rcu_preempt_state; |
f41d911f | 91 | |
d9a3da06 | 92 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); |
d19fb8d1 PM |
93 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
94 | bool wake); | |
d9a3da06 | 95 | |
f41d911f PM |
96 | /* |
97 | * Tell them what RCU they are running. | |
98 | */ | |
0e0fc1c2 | 99 | static void __init rcu_bootup_announce(void) |
f41d911f | 100 | { |
efc151c3 | 101 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 102 | rcu_bootup_announce_oddness(); |
f41d911f PM |
103 | } |
104 | ||
f41d911f | 105 | /* |
6cc68793 | 106 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
107 | * that this just means that the task currently running on the CPU is |
108 | * not in a quiescent state. There might be any number of tasks blocked | |
109 | * while in an RCU read-side critical section. | |
25502a6c | 110 | * |
1d082fd0 PM |
111 | * As with the other rcu_*_qs() functions, callers to this function |
112 | * must disable preemption. | |
f41d911f | 113 | */ |
284a8c93 | 114 | static void rcu_preempt_qs(void) |
f41d911f | 115 | { |
284a8c93 PM |
116 | if (!__this_cpu_read(rcu_preempt_data.passed_quiesce)) { |
117 | trace_rcu_grace_period(TPS("rcu_preempt"), | |
118 | __this_cpu_read(rcu_preempt_data.gpnum), | |
119 | TPS("cpuqs")); | |
120 | __this_cpu_write(rcu_preempt_data.passed_quiesce, 1); | |
121 | barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */ | |
122 | current->rcu_read_unlock_special.b.need_qs = false; | |
123 | } | |
f41d911f PM |
124 | } |
125 | ||
126 | /* | |
c3422bea PM |
127 | * We have entered the scheduler, and the current task might soon be |
128 | * context-switched away from. If this task is in an RCU read-side | |
129 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
130 | * record that fact, so we enqueue the task on the blkd_tasks list. |
131 | * The task will dequeue itself when it exits the outermost enclosing | |
132 | * RCU read-side critical section. Therefore, the current grace period | |
133 | * cannot be permitted to complete until the blkd_tasks list entries | |
134 | * predating the current grace period drain, in other words, until | |
135 | * rnp->gp_tasks becomes NULL. | |
c3422bea PM |
136 | * |
137 | * Caller must disable preemption. | |
f41d911f | 138 | */ |
38200cf2 | 139 | static void rcu_preempt_note_context_switch(void) |
f41d911f PM |
140 | { |
141 | struct task_struct *t = current; | |
c3422bea | 142 | unsigned long flags; |
f41d911f PM |
143 | struct rcu_data *rdp; |
144 | struct rcu_node *rnp; | |
145 | ||
10f39bb1 | 146 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 147 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
148 | |
149 | /* Possibly blocking in an RCU read-side critical section. */ | |
38200cf2 | 150 | rdp = this_cpu_ptr(rcu_preempt_state.rda); |
f41d911f | 151 | rnp = rdp->mynode; |
1304afb2 | 152 | raw_spin_lock_irqsave(&rnp->lock, flags); |
6303b9c8 | 153 | smp_mb__after_unlock_lock(); |
1d082fd0 | 154 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 155 | t->rcu_blocked_node = rnp; |
f41d911f PM |
156 | |
157 | /* | |
158 | * If this CPU has already checked in, then this task | |
159 | * will hold up the next grace period rather than the | |
160 | * current grace period. Queue the task accordingly. | |
161 | * If the task is queued for the current grace period | |
162 | * (i.e., this CPU has not yet passed through a quiescent | |
163 | * state for the current grace period), then as long | |
164 | * as that task remains queued, the current grace period | |
12f5f524 PM |
165 | * cannot end. Note that there is some uncertainty as |
166 | * to exactly when the current grace period started. | |
167 | * We take a conservative approach, which can result | |
168 | * in unnecessarily waiting on tasks that started very | |
169 | * slightly after the current grace period began. C'est | |
170 | * la vie!!! | |
b0e165c0 PM |
171 | * |
172 | * But first, note that the current CPU must still be | |
173 | * on line! | |
f41d911f | 174 | */ |
b0e165c0 | 175 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); |
e7d8842e | 176 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
12f5f524 PM |
177 | if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) { |
178 | list_add(&t->rcu_node_entry, rnp->gp_tasks->prev); | |
179 | rnp->gp_tasks = &t->rcu_node_entry; | |
27f4d280 PM |
180 | #ifdef CONFIG_RCU_BOOST |
181 | if (rnp->boost_tasks != NULL) | |
182 | rnp->boost_tasks = rnp->gp_tasks; | |
183 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
12f5f524 PM |
184 | } else { |
185 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
186 | if (rnp->qsmask & rdp->grpmask) | |
187 | rnp->gp_tasks = &t->rcu_node_entry; | |
188 | } | |
d4c08f2a PM |
189 | trace_rcu_preempt_task(rdp->rsp->name, |
190 | t->pid, | |
191 | (rnp->qsmask & rdp->grpmask) | |
192 | ? rnp->gpnum | |
193 | : rnp->gpnum + 1); | |
1304afb2 | 194 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
10f39bb1 | 195 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 196 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
197 | |
198 | /* | |
199 | * Complete exit from RCU read-side critical section on | |
200 | * behalf of preempted instance of __rcu_read_unlock(). | |
201 | */ | |
202 | rcu_read_unlock_special(t); | |
f41d911f PM |
203 | } |
204 | ||
205 | /* | |
206 | * Either we were not in an RCU read-side critical section to | |
207 | * begin with, or we have now recorded that critical section | |
208 | * globally. Either way, we can now note a quiescent state | |
209 | * for this CPU. Again, if we were in an RCU read-side critical | |
210 | * section, and if that critical section was blocking the current | |
211 | * grace period, then the fact that the task has been enqueued | |
212 | * means that we continue to block the current grace period. | |
213 | */ | |
284a8c93 | 214 | rcu_preempt_qs(); |
f41d911f PM |
215 | } |
216 | ||
fc2219d4 PM |
217 | /* |
218 | * Check for preempted RCU readers blocking the current grace period | |
219 | * for the specified rcu_node structure. If the caller needs a reliable | |
220 | * answer, it must hold the rcu_node's ->lock. | |
221 | */ | |
27f4d280 | 222 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 223 | { |
12f5f524 | 224 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
225 | } |
226 | ||
b668c9cf PM |
227 | /* |
228 | * Record a quiescent state for all tasks that were previously queued | |
229 | * on the specified rcu_node structure and that were blocking the current | |
230 | * RCU grace period. The caller must hold the specified rnp->lock with | |
231 | * irqs disabled, and this lock is released upon return, but irqs remain | |
232 | * disabled. | |
233 | */ | |
d3f6bad3 | 234 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
235 | __releases(rnp->lock) |
236 | { | |
237 | unsigned long mask; | |
238 | struct rcu_node *rnp_p; | |
239 | ||
27f4d280 | 240 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
1304afb2 | 241 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
242 | return; /* Still need more quiescent states! */ |
243 | } | |
244 | ||
245 | rnp_p = rnp->parent; | |
246 | if (rnp_p == NULL) { | |
247 | /* | |
248 | * Either there is only one rcu_node in the tree, | |
249 | * or tasks were kicked up to root rcu_node due to | |
250 | * CPUs going offline. | |
251 | */ | |
d3f6bad3 | 252 | rcu_report_qs_rsp(&rcu_preempt_state, flags); |
b668c9cf PM |
253 | return; |
254 | } | |
255 | ||
256 | /* Report up the rest of the hierarchy. */ | |
257 | mask = rnp->grpmask; | |
1304afb2 PM |
258 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
259 | raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */ | |
6303b9c8 | 260 | smp_mb__after_unlock_lock(); |
d3f6bad3 | 261 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); |
b668c9cf PM |
262 | } |
263 | ||
12f5f524 PM |
264 | /* |
265 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
266 | * returning NULL if at the end of the list. | |
267 | */ | |
268 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
269 | struct rcu_node *rnp) | |
270 | { | |
271 | struct list_head *np; | |
272 | ||
273 | np = t->rcu_node_entry.next; | |
274 | if (np == &rnp->blkd_tasks) | |
275 | np = NULL; | |
276 | return np; | |
277 | } | |
278 | ||
8af3a5e7 PM |
279 | /* |
280 | * Return true if the specified rcu_node structure has tasks that were | |
281 | * preempted within an RCU read-side critical section. | |
282 | */ | |
283 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
284 | { | |
285 | return !list_empty(&rnp->blkd_tasks); | |
286 | } | |
287 | ||
b668c9cf PM |
288 | /* |
289 | * Handle special cases during rcu_read_unlock(), such as needing to | |
290 | * notify RCU core processing or task having blocked during the RCU | |
291 | * read-side critical section. | |
292 | */ | |
2a3fa843 | 293 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f | 294 | { |
b6a932d1 PM |
295 | bool empty; |
296 | bool empty_exp; | |
297 | bool empty_norm; | |
298 | bool empty_exp_now; | |
f41d911f | 299 | unsigned long flags; |
12f5f524 | 300 | struct list_head *np; |
82e78d80 | 301 | #ifdef CONFIG_RCU_BOOST |
abaa93d9 | 302 | bool drop_boost_mutex = false; |
82e78d80 | 303 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
f41d911f | 304 | struct rcu_node *rnp; |
1d082fd0 | 305 | union rcu_special special; |
f41d911f PM |
306 | |
307 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
308 | if (in_nmi()) | |
309 | return; | |
310 | ||
311 | local_irq_save(flags); | |
312 | ||
313 | /* | |
314 | * If RCU core is waiting for this CPU to exit critical section, | |
1d082fd0 PM |
315 | * let it know that we have done so. Because irqs are disabled, |
316 | * t->rcu_read_unlock_special cannot change. | |
f41d911f PM |
317 | */ |
318 | special = t->rcu_read_unlock_special; | |
1d082fd0 | 319 | if (special.b.need_qs) { |
284a8c93 | 320 | rcu_preempt_qs(); |
c0135d07 | 321 | t->rcu_read_unlock_special.b.need_qs = false; |
1d082fd0 | 322 | if (!t->rcu_read_unlock_special.s) { |
79a62f95 LJ |
323 | local_irq_restore(flags); |
324 | return; | |
325 | } | |
f41d911f PM |
326 | } |
327 | ||
79a62f95 | 328 | /* Hardware IRQ handlers cannot block, complain if they get here. */ |
d24209bb PM |
329 | if (in_irq() || in_serving_softirq()) { |
330 | lockdep_rcu_suspicious(__FILE__, __LINE__, | |
331 | "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n"); | |
332 | pr_alert("->rcu_read_unlock_special: %#x (b: %d, nq: %d)\n", | |
333 | t->rcu_read_unlock_special.s, | |
334 | t->rcu_read_unlock_special.b.blocked, | |
335 | t->rcu_read_unlock_special.b.need_qs); | |
f41d911f PM |
336 | local_irq_restore(flags); |
337 | return; | |
338 | } | |
339 | ||
340 | /* Clean up if blocked during RCU read-side critical section. */ | |
1d082fd0 PM |
341 | if (special.b.blocked) { |
342 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 343 | |
dd5d19ba PM |
344 | /* |
345 | * Remove this task from the list it blocked on. The | |
346 | * task can migrate while we acquire the lock, but at | |
347 | * most one time. So at most two passes through loop. | |
348 | */ | |
349 | for (;;) { | |
86848966 | 350 | rnp = t->rcu_blocked_node; |
1304afb2 | 351 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
6303b9c8 | 352 | smp_mb__after_unlock_lock(); |
86848966 | 353 | if (rnp == t->rcu_blocked_node) |
dd5d19ba | 354 | break; |
1304afb2 | 355 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
dd5d19ba | 356 | } |
b6a932d1 | 357 | empty = !rcu_preempt_has_tasks(rnp); |
74e871ac | 358 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d9a3da06 PM |
359 | empty_exp = !rcu_preempted_readers_exp(rnp); |
360 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
12f5f524 | 361 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 362 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 363 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 364 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
d4c08f2a | 365 | rnp->gpnum, t->pid); |
12f5f524 PM |
366 | if (&t->rcu_node_entry == rnp->gp_tasks) |
367 | rnp->gp_tasks = np; | |
368 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
369 | rnp->exp_tasks = np; | |
27f4d280 PM |
370 | #ifdef CONFIG_RCU_BOOST |
371 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
372 | rnp->boost_tasks = np; | |
abaa93d9 PM |
373 | /* Snapshot ->boost_mtx ownership with rcu_node lock held. */ |
374 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
27f4d280 | 375 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
f41d911f | 376 | |
b6a932d1 PM |
377 | /* |
378 | * If this was the last task on the list, go see if we | |
379 | * need to propagate ->qsmaskinit bit clearing up the | |
380 | * rcu_node tree. | |
381 | */ | |
382 | if (!empty && !rcu_preempt_has_tasks(rnp)) | |
383 | rcu_cleanup_dead_rnp(rnp); | |
384 | ||
f41d911f PM |
385 | /* |
386 | * If this was the last task on the current list, and if | |
387 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
388 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
389 | * so we must take a snapshot of the expedited state. | |
f41d911f | 390 | */ |
389abd48 | 391 | empty_exp_now = !rcu_preempted_readers_exp(rnp); |
74e871ac | 392 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 393 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
d4c08f2a PM |
394 | rnp->gpnum, |
395 | 0, rnp->qsmask, | |
396 | rnp->level, | |
397 | rnp->grplo, | |
398 | rnp->grphi, | |
399 | !!rnp->gp_tasks); | |
d3f6bad3 | 400 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 401 | } else { |
d4c08f2a | 402 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 403 | } |
d9a3da06 | 404 | |
27f4d280 PM |
405 | #ifdef CONFIG_RCU_BOOST |
406 | /* Unboost if we were boosted. */ | |
abaf3f9d | 407 | if (drop_boost_mutex) |
abaa93d9 | 408 | rt_mutex_unlock(&rnp->boost_mtx); |
27f4d280 PM |
409 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
410 | ||
d9a3da06 PM |
411 | /* |
412 | * If this was the last task on the expedited lists, | |
413 | * then we need to report up the rcu_node hierarchy. | |
414 | */ | |
389abd48 | 415 | if (!empty_exp && empty_exp_now) |
b40d293e | 416 | rcu_report_exp_rnp(&rcu_preempt_state, rnp, true); |
b668c9cf PM |
417 | } else { |
418 | local_irq_restore(flags); | |
f41d911f | 419 | } |
f41d911f PM |
420 | } |
421 | ||
1ed509a2 PM |
422 | /* |
423 | * Dump detailed information for all tasks blocking the current RCU | |
424 | * grace period on the specified rcu_node structure. | |
425 | */ | |
426 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
427 | { | |
428 | unsigned long flags; | |
1ed509a2 PM |
429 | struct task_struct *t; |
430 | ||
12f5f524 | 431 | raw_spin_lock_irqsave(&rnp->lock, flags); |
5fd4dc06 PM |
432 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
433 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
434 | return; | |
435 | } | |
12f5f524 PM |
436 | t = list_entry(rnp->gp_tasks, |
437 | struct task_struct, rcu_node_entry); | |
438 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) | |
439 | sched_show_task(t); | |
440 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1ed509a2 PM |
441 | } |
442 | ||
443 | /* | |
444 | * Dump detailed information for all tasks blocking the current RCU | |
445 | * grace period. | |
446 | */ | |
447 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
448 | { | |
449 | struct rcu_node *rnp = rcu_get_root(rsp); | |
450 | ||
451 | rcu_print_detail_task_stall_rnp(rnp); | |
452 | rcu_for_each_leaf_node(rsp, rnp) | |
453 | rcu_print_detail_task_stall_rnp(rnp); | |
454 | } | |
455 | ||
a858af28 PM |
456 | #ifdef CONFIG_RCU_CPU_STALL_INFO |
457 | ||
458 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
459 | { | |
efc151c3 | 460 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
461 | rnp->level, rnp->grplo, rnp->grphi); |
462 | } | |
463 | ||
464 | static void rcu_print_task_stall_end(void) | |
465 | { | |
efc151c3 | 466 | pr_cont("\n"); |
a858af28 PM |
467 | } |
468 | ||
469 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
470 | ||
471 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
472 | { | |
473 | } | |
474 | ||
475 | static void rcu_print_task_stall_end(void) | |
476 | { | |
477 | } | |
478 | ||
479 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
480 | ||
f41d911f PM |
481 | /* |
482 | * Scan the current list of tasks blocked within RCU read-side critical | |
483 | * sections, printing out the tid of each. | |
484 | */ | |
9bc8b558 | 485 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 486 | { |
f41d911f | 487 | struct task_struct *t; |
9bc8b558 | 488 | int ndetected = 0; |
f41d911f | 489 | |
27f4d280 | 490 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 491 | return 0; |
a858af28 | 492 | rcu_print_task_stall_begin(rnp); |
12f5f524 PM |
493 | t = list_entry(rnp->gp_tasks, |
494 | struct task_struct, rcu_node_entry); | |
9bc8b558 | 495 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 496 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
497 | ndetected++; |
498 | } | |
a858af28 | 499 | rcu_print_task_stall_end(); |
9bc8b558 | 500 | return ndetected; |
f41d911f PM |
501 | } |
502 | ||
b0e165c0 PM |
503 | /* |
504 | * Check that the list of blocked tasks for the newly completed grace | |
505 | * period is in fact empty. It is a serious bug to complete a grace | |
506 | * period that still has RCU readers blocked! This function must be | |
507 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
508 | * must be held by the caller. | |
12f5f524 PM |
509 | * |
510 | * Also, if there are blocked tasks on the list, they automatically | |
511 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
512 | */ |
513 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
514 | { | |
27f4d280 | 515 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
96e92021 | 516 | if (rcu_preempt_has_tasks(rnp)) |
12f5f524 | 517 | rnp->gp_tasks = rnp->blkd_tasks.next; |
28ecd580 | 518 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
519 | } |
520 | ||
33f76148 PM |
521 | #ifdef CONFIG_HOTPLUG_CPU |
522 | ||
e5601400 PM |
523 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
524 | ||
f41d911f PM |
525 | /* |
526 | * Check for a quiescent state from the current CPU. When a task blocks, | |
527 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
528 | * which is checked elsewhere. | |
529 | * | |
530 | * Caller must disable hard irqs. | |
531 | */ | |
86aea0e6 | 532 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
533 | { |
534 | struct task_struct *t = current; | |
535 | ||
536 | if (t->rcu_read_lock_nesting == 0) { | |
284a8c93 | 537 | rcu_preempt_qs(); |
f41d911f PM |
538 | return; |
539 | } | |
10f39bb1 | 540 | if (t->rcu_read_lock_nesting > 0 && |
86aea0e6 PM |
541 | __this_cpu_read(rcu_preempt_data.qs_pending) && |
542 | !__this_cpu_read(rcu_preempt_data.passed_quiesce)) | |
1d082fd0 | 543 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
544 | } |
545 | ||
a46e0899 PM |
546 | #ifdef CONFIG_RCU_BOOST |
547 | ||
09223371 SL |
548 | static void rcu_preempt_do_callbacks(void) |
549 | { | |
c9d4b0af | 550 | rcu_do_batch(&rcu_preempt_state, this_cpu_ptr(&rcu_preempt_data)); |
09223371 SL |
551 | } |
552 | ||
a46e0899 PM |
553 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
554 | ||
f41d911f | 555 | /* |
6cc68793 | 556 | * Queue a preemptible-RCU callback for invocation after a grace period. |
f41d911f PM |
557 | */ |
558 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
559 | { | |
3fbfbf7a | 560 | __call_rcu(head, func, &rcu_preempt_state, -1, 0); |
f41d911f PM |
561 | } |
562 | EXPORT_SYMBOL_GPL(call_rcu); | |
563 | ||
6ebb237b PM |
564 | /** |
565 | * synchronize_rcu - wait until a grace period has elapsed. | |
566 | * | |
567 | * Control will return to the caller some time after a full grace | |
568 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
569 | * read-side critical sections have completed. Note, however, that |
570 | * upon return from synchronize_rcu(), the caller might well be executing | |
571 | * concurrently with new RCU read-side critical sections that began while | |
572 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
573 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 PM |
574 | * |
575 | * See the description of synchronize_sched() for more detailed information | |
576 | * on memory ordering guarantees. | |
6ebb237b PM |
577 | */ |
578 | void synchronize_rcu(void) | |
579 | { | |
fe15d706 PM |
580 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
581 | !lock_is_held(&rcu_lock_map) && | |
582 | !lock_is_held(&rcu_sched_lock_map), | |
583 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
6ebb237b PM |
584 | if (!rcu_scheduler_active) |
585 | return; | |
3705b88d AM |
586 | if (rcu_expedited) |
587 | synchronize_rcu_expedited(); | |
588 | else | |
589 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
590 | } |
591 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
592 | ||
d9a3da06 | 593 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); |
bcfa57ce | 594 | static unsigned long sync_rcu_preempt_exp_count; |
d9a3da06 PM |
595 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); |
596 | ||
597 | /* | |
598 | * Return non-zero if there are any tasks in RCU read-side critical | |
599 | * sections blocking the current preemptible-RCU expedited grace period. | |
600 | * If there is no preemptible-RCU expedited grace period currently in | |
601 | * progress, returns zero unconditionally. | |
602 | */ | |
603 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
604 | { | |
12f5f524 | 605 | return rnp->exp_tasks != NULL; |
d9a3da06 PM |
606 | } |
607 | ||
608 | /* | |
609 | * return non-zero if there is no RCU expedited grace period in progress | |
610 | * for the specified rcu_node structure, in other words, if all CPUs and | |
611 | * tasks covered by the specified rcu_node structure have done their bit | |
612 | * for the current expedited grace period. Works only for preemptible | |
613 | * RCU -- other RCU implementation use other means. | |
614 | * | |
615 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
616 | */ | |
617 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
618 | { | |
619 | return !rcu_preempted_readers_exp(rnp) && | |
620 | ACCESS_ONCE(rnp->expmask) == 0; | |
621 | } | |
622 | ||
623 | /* | |
624 | * Report the exit from RCU read-side critical section for the last task | |
625 | * that queued itself during or before the current expedited preemptible-RCU | |
626 | * grace period. This event is reported either to the rcu_node structure on | |
627 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
628 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
629 | * iteratively!) | |
630 | * | |
b40d293e TG |
631 | * Most callers will set the "wake" flag, but the task initiating the |
632 | * expedited grace period need not wake itself. | |
633 | * | |
d9a3da06 PM |
634 | * Caller must hold sync_rcu_preempt_exp_mutex. |
635 | */ | |
b40d293e TG |
636 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
637 | bool wake) | |
d9a3da06 PM |
638 | { |
639 | unsigned long flags; | |
640 | unsigned long mask; | |
641 | ||
1304afb2 | 642 | raw_spin_lock_irqsave(&rnp->lock, flags); |
6303b9c8 | 643 | smp_mb__after_unlock_lock(); |
d9a3da06 | 644 | for (;;) { |
131906b0 PM |
645 | if (!sync_rcu_preempt_exp_done(rnp)) { |
646 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
d9a3da06 | 647 | break; |
131906b0 | 648 | } |
d9a3da06 | 649 | if (rnp->parent == NULL) { |
131906b0 | 650 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
78e4bc34 PM |
651 | if (wake) { |
652 | smp_mb(); /* EGP done before wake_up(). */ | |
b40d293e | 653 | wake_up(&sync_rcu_preempt_exp_wq); |
78e4bc34 | 654 | } |
d9a3da06 PM |
655 | break; |
656 | } | |
657 | mask = rnp->grpmask; | |
1304afb2 | 658 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
d9a3da06 | 659 | rnp = rnp->parent; |
1304afb2 | 660 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
6303b9c8 | 661 | smp_mb__after_unlock_lock(); |
d9a3da06 PM |
662 | rnp->expmask &= ~mask; |
663 | } | |
d9a3da06 PM |
664 | } |
665 | ||
666 | /* | |
667 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
668 | * grace period for the specified rcu_node structure. If there are no such | |
669 | * tasks, report it up the rcu_node hierarchy. | |
670 | * | |
7b2e6011 PM |
671 | * Caller must hold sync_rcu_preempt_exp_mutex and must exclude |
672 | * CPU hotplug operations. | |
d9a3da06 PM |
673 | */ |
674 | static void | |
675 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
676 | { | |
1217ed1b | 677 | unsigned long flags; |
12f5f524 | 678 | int must_wait = 0; |
d9a3da06 | 679 | |
1217ed1b | 680 | raw_spin_lock_irqsave(&rnp->lock, flags); |
6303b9c8 | 681 | smp_mb__after_unlock_lock(); |
96e92021 | 682 | if (!rcu_preempt_has_tasks(rnp)) { |
1217ed1b | 683 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 684 | } else { |
12f5f524 | 685 | rnp->exp_tasks = rnp->blkd_tasks.next; |
1217ed1b | 686 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
12f5f524 PM |
687 | must_wait = 1; |
688 | } | |
d9a3da06 | 689 | if (!must_wait) |
b40d293e | 690 | rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */ |
d9a3da06 PM |
691 | } |
692 | ||
236fefaf PM |
693 | /** |
694 | * synchronize_rcu_expedited - Brute-force RCU grace period | |
695 | * | |
696 | * Wait for an RCU-preempt grace period, but expedite it. The basic | |
697 | * idea is to invoke synchronize_sched_expedited() to push all the tasks to | |
698 | * the ->blkd_tasks lists and wait for this list to drain. This consumes | |
699 | * significant time on all CPUs and is unfriendly to real-time workloads, | |
700 | * so is thus not recommended for any sort of common-case code. | |
701 | * In fact, if you are using synchronize_rcu_expedited() in a loop, | |
702 | * please restructure your code to batch your updates, and then Use a | |
703 | * single synchronize_rcu() instead. | |
019129d5 PM |
704 | */ |
705 | void synchronize_rcu_expedited(void) | |
706 | { | |
d9a3da06 PM |
707 | unsigned long flags; |
708 | struct rcu_node *rnp; | |
709 | struct rcu_state *rsp = &rcu_preempt_state; | |
bcfa57ce | 710 | unsigned long snap; |
d9a3da06 PM |
711 | int trycount = 0; |
712 | ||
713 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
714 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
715 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
716 | ||
1943c89d PM |
717 | /* |
718 | * Block CPU-hotplug operations. This means that any CPU-hotplug | |
719 | * operation that finds an rcu_node structure with tasks in the | |
720 | * process of being boosted will know that all tasks blocking | |
721 | * this expedited grace period will already be in the process of | |
722 | * being boosted. This simplifies the process of moving tasks | |
723 | * from leaf to root rcu_node structures. | |
724 | */ | |
dd56af42 PM |
725 | if (!try_get_online_cpus()) { |
726 | /* CPU-hotplug operation in flight, fall back to normal GP. */ | |
727 | wait_rcu_gp(call_rcu); | |
728 | return; | |
729 | } | |
1943c89d | 730 | |
d9a3da06 PM |
731 | /* |
732 | * Acquire lock, falling back to synchronize_rcu() if too many | |
733 | * lock-acquisition failures. Of course, if someone does the | |
734 | * expedited grace period for us, just leave. | |
735 | */ | |
736 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
1943c89d PM |
737 | if (ULONG_CMP_LT(snap, |
738 | ACCESS_ONCE(sync_rcu_preempt_exp_count))) { | |
739 | put_online_cpus(); | |
740 | goto mb_ret; /* Others did our work for us. */ | |
741 | } | |
c701d5d9 | 742 | if (trycount++ < 10) { |
d9a3da06 | 743 | udelay(trycount * num_online_cpus()); |
c701d5d9 | 744 | } else { |
1943c89d | 745 | put_online_cpus(); |
3705b88d | 746 | wait_rcu_gp(call_rcu); |
d9a3da06 PM |
747 | return; |
748 | } | |
d9a3da06 | 749 | } |
1943c89d PM |
750 | if (ULONG_CMP_LT(snap, ACCESS_ONCE(sync_rcu_preempt_exp_count))) { |
751 | put_online_cpus(); | |
d9a3da06 | 752 | goto unlock_mb_ret; /* Others did our work for us. */ |
1943c89d | 753 | } |
d9a3da06 | 754 | |
12f5f524 | 755 | /* force all RCU readers onto ->blkd_tasks lists. */ |
d9a3da06 PM |
756 | synchronize_sched_expedited(); |
757 | ||
d9a3da06 PM |
758 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ |
759 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
1943c89d | 760 | raw_spin_lock_irqsave(&rnp->lock, flags); |
6303b9c8 | 761 | smp_mb__after_unlock_lock(); |
d9a3da06 | 762 | rnp->expmask = rnp->qsmaskinit; |
1943c89d | 763 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
764 | } |
765 | ||
12f5f524 | 766 | /* Snapshot current state of ->blkd_tasks lists. */ |
d9a3da06 PM |
767 | rcu_for_each_leaf_node(rsp, rnp) |
768 | sync_rcu_preempt_exp_init(rsp, rnp); | |
769 | if (NUM_RCU_NODES > 1) | |
770 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
771 | ||
1943c89d | 772 | put_online_cpus(); |
d9a3da06 | 773 | |
12f5f524 | 774 | /* Wait for snapshotted ->blkd_tasks lists to drain. */ |
d9a3da06 PM |
775 | rnp = rcu_get_root(rsp); |
776 | wait_event(sync_rcu_preempt_exp_wq, | |
777 | sync_rcu_preempt_exp_done(rnp)); | |
778 | ||
779 | /* Clean up and exit. */ | |
780 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
4de376a1 PK |
781 | ACCESS_ONCE(sync_rcu_preempt_exp_count) = |
782 | sync_rcu_preempt_exp_count + 1; | |
d9a3da06 PM |
783 | unlock_mb_ret: |
784 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
785 | mb_ret: | |
786 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
019129d5 PM |
787 | } |
788 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
789 | ||
e74f4c45 PM |
790 | /** |
791 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
792 | * |
793 | * Note that this primitive does not necessarily wait for an RCU grace period | |
794 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
795 | * in the system, then rcu_barrier() is within its rights to return | |
796 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
797 | */ |
798 | void rcu_barrier(void) | |
799 | { | |
037b64ed | 800 | _rcu_barrier(&rcu_preempt_state); |
e74f4c45 PM |
801 | } |
802 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
803 | ||
1eba8f84 | 804 | /* |
6cc68793 | 805 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
806 | */ |
807 | static void __init __rcu_init_preempt(void) | |
808 | { | |
394f99a9 | 809 | rcu_init_one(&rcu_preempt_state, &rcu_preempt_data); |
1eba8f84 PM |
810 | } |
811 | ||
2439b696 PM |
812 | /* |
813 | * Check for a task exiting while in a preemptible-RCU read-side | |
814 | * critical section, clean up if so. No need to issue warnings, | |
815 | * as debug_check_no_locks_held() already does this if lockdep | |
816 | * is enabled. | |
817 | */ | |
818 | void exit_rcu(void) | |
819 | { | |
820 | struct task_struct *t = current; | |
821 | ||
822 | if (likely(list_empty(¤t->rcu_node_entry))) | |
823 | return; | |
824 | t->rcu_read_lock_nesting = 1; | |
825 | barrier(); | |
1d082fd0 | 826 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 PM |
827 | __rcu_read_unlock(); |
828 | } | |
829 | ||
28f6569a | 830 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 831 | |
e534165b | 832 | static struct rcu_state *rcu_state_p = &rcu_sched_state; |
27f4d280 | 833 | |
f41d911f PM |
834 | /* |
835 | * Tell them what RCU they are running. | |
836 | */ | |
0e0fc1c2 | 837 | static void __init rcu_bootup_announce(void) |
f41d911f | 838 | { |
efc151c3 | 839 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 840 | rcu_bootup_announce_oddness(); |
f41d911f PM |
841 | } |
842 | ||
cba6d0d6 PM |
843 | /* |
844 | * Because preemptible RCU does not exist, we never have to check for | |
845 | * CPUs being in quiescent states. | |
846 | */ | |
38200cf2 | 847 | static void rcu_preempt_note_context_switch(void) |
cba6d0d6 PM |
848 | { |
849 | } | |
850 | ||
fc2219d4 | 851 | /* |
6cc68793 | 852 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
853 | * RCU readers. |
854 | */ | |
27f4d280 | 855 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
856 | { |
857 | return 0; | |
858 | } | |
859 | ||
b668c9cf PM |
860 | #ifdef CONFIG_HOTPLUG_CPU |
861 | ||
8af3a5e7 PM |
862 | /* |
863 | * Because there is no preemptible RCU, there can be no readers blocked. | |
864 | */ | |
865 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 866 | { |
8af3a5e7 | 867 | return false; |
b668c9cf PM |
868 | } |
869 | ||
870 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
871 | ||
1ed509a2 | 872 | /* |
6cc68793 | 873 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
874 | * tasks blocked within RCU read-side critical sections. |
875 | */ | |
876 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
877 | { | |
878 | } | |
879 | ||
f41d911f | 880 | /* |
6cc68793 | 881 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
882 | * tasks blocked within RCU read-side critical sections. |
883 | */ | |
9bc8b558 | 884 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 885 | { |
9bc8b558 | 886 | return 0; |
f41d911f PM |
887 | } |
888 | ||
b0e165c0 | 889 | /* |
6cc68793 | 890 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
891 | * so there is no need to check for blocked tasks. So check only for |
892 | * bogus qsmask values. | |
b0e165c0 PM |
893 | */ |
894 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
895 | { | |
49e29126 | 896 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
897 | } |
898 | ||
f41d911f | 899 | /* |
6cc68793 | 900 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
901 | * to check. |
902 | */ | |
86aea0e6 | 903 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
904 | { |
905 | } | |
906 | ||
019129d5 PM |
907 | /* |
908 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
6cc68793 | 909 | * But because preemptible RCU does not exist, map to rcu-sched. |
019129d5 PM |
910 | */ |
911 | void synchronize_rcu_expedited(void) | |
912 | { | |
913 | synchronize_sched_expedited(); | |
914 | } | |
915 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
916 | ||
e74f4c45 | 917 | /* |
6cc68793 | 918 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
919 | * another name for rcu_barrier_sched(). |
920 | */ | |
921 | void rcu_barrier(void) | |
922 | { | |
923 | rcu_barrier_sched(); | |
924 | } | |
925 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
926 | ||
1eba8f84 | 927 | /* |
6cc68793 | 928 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
929 | */ |
930 | static void __init __rcu_init_preempt(void) | |
931 | { | |
932 | } | |
933 | ||
2439b696 PM |
934 | /* |
935 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
936 | * while in preemptible RCU read-side critical sections. | |
937 | */ | |
938 | void exit_rcu(void) | |
939 | { | |
940 | } | |
941 | ||
28f6569a | 942 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 943 | |
27f4d280 PM |
944 | #ifdef CONFIG_RCU_BOOST |
945 | ||
1696a8be | 946 | #include "../locking/rtmutex_common.h" |
27f4d280 | 947 | |
0ea1f2eb PM |
948 | #ifdef CONFIG_RCU_TRACE |
949 | ||
950 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
951 | { | |
96e92021 | 952 | if (!rcu_preempt_has_tasks(rnp)) |
0ea1f2eb PM |
953 | rnp->n_balk_blkd_tasks++; |
954 | else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL) | |
955 | rnp->n_balk_exp_gp_tasks++; | |
956 | else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL) | |
957 | rnp->n_balk_boost_tasks++; | |
958 | else if (rnp->gp_tasks != NULL && rnp->qsmask != 0) | |
959 | rnp->n_balk_notblocked++; | |
960 | else if (rnp->gp_tasks != NULL && | |
a9f4793d | 961 | ULONG_CMP_LT(jiffies, rnp->boost_time)) |
0ea1f2eb PM |
962 | rnp->n_balk_notyet++; |
963 | else | |
964 | rnp->n_balk_nos++; | |
965 | } | |
966 | ||
967 | #else /* #ifdef CONFIG_RCU_TRACE */ | |
968 | ||
969 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
970 | { | |
971 | } | |
972 | ||
973 | #endif /* #else #ifdef CONFIG_RCU_TRACE */ | |
974 | ||
5d01bbd1 TG |
975 | static void rcu_wake_cond(struct task_struct *t, int status) |
976 | { | |
977 | /* | |
978 | * If the thread is yielding, only wake it when this | |
979 | * is invoked from idle | |
980 | */ | |
981 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
982 | wake_up_process(t); | |
983 | } | |
984 | ||
27f4d280 PM |
985 | /* |
986 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
987 | * or ->boost_tasks, advancing the pointer to the next task in the | |
988 | * ->blkd_tasks list. | |
989 | * | |
990 | * Note that irqs must be enabled: boosting the task can block. | |
991 | * Returns 1 if there are more tasks needing to be boosted. | |
992 | */ | |
993 | static int rcu_boost(struct rcu_node *rnp) | |
994 | { | |
995 | unsigned long flags; | |
27f4d280 PM |
996 | struct task_struct *t; |
997 | struct list_head *tb; | |
998 | ||
b08ea27d PM |
999 | if (ACCESS_ONCE(rnp->exp_tasks) == NULL && |
1000 | ACCESS_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1001 | return 0; /* Nothing left to boost. */ |
1002 | ||
1003 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
6303b9c8 | 1004 | smp_mb__after_unlock_lock(); |
27f4d280 PM |
1005 | |
1006 | /* | |
1007 | * Recheck under the lock: all tasks in need of boosting | |
1008 | * might exit their RCU read-side critical sections on their own. | |
1009 | */ | |
1010 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
1011 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Preferentially boost tasks blocking expedited grace periods. | |
1017 | * This cannot starve the normal grace periods because a second | |
1018 | * expedited grace period must boost all blocked tasks, including | |
1019 | * those blocking the pre-existing normal grace period. | |
1020 | */ | |
0ea1f2eb | 1021 | if (rnp->exp_tasks != NULL) { |
27f4d280 | 1022 | tb = rnp->exp_tasks; |
0ea1f2eb PM |
1023 | rnp->n_exp_boosts++; |
1024 | } else { | |
27f4d280 | 1025 | tb = rnp->boost_tasks; |
0ea1f2eb PM |
1026 | rnp->n_normal_boosts++; |
1027 | } | |
1028 | rnp->n_tasks_boosted++; | |
27f4d280 PM |
1029 | |
1030 | /* | |
1031 | * We boost task t by manufacturing an rt_mutex that appears to | |
1032 | * be held by task t. We leave a pointer to that rt_mutex where | |
1033 | * task t can find it, and task t will release the mutex when it | |
1034 | * exits its outermost RCU read-side critical section. Then | |
1035 | * simply acquiring this artificial rt_mutex will boost task | |
1036 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1037 | * | |
1038 | * Note that task t must acquire rnp->lock to remove itself from | |
1039 | * the ->blkd_tasks list, which it will do from exit() if from | |
1040 | * nowhere else. We therefore are guaranteed that task t will | |
1041 | * stay around at least until we drop rnp->lock. Note that | |
1042 | * rnp->lock also resolves races between our priority boosting | |
1043 | * and task t's exiting its outermost RCU read-side critical | |
1044 | * section. | |
1045 | */ | |
1046 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1047 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
27f4d280 | 1048 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
abaa93d9 PM |
1049 | /* Lock only for side effect: boosts task t's priority. */ |
1050 | rt_mutex_lock(&rnp->boost_mtx); | |
1051 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1052 | |
4f89b336 PM |
1053 | return ACCESS_ONCE(rnp->exp_tasks) != NULL || |
1054 | ACCESS_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1055 | } |
1056 | ||
27f4d280 PM |
1057 | /* |
1058 | * Priority-boosting kthread. One per leaf rcu_node and one for the | |
1059 | * root rcu_node. | |
1060 | */ | |
1061 | static int rcu_boost_kthread(void *arg) | |
1062 | { | |
1063 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1064 | int spincnt = 0; | |
1065 | int more2boost; | |
1066 | ||
f7f7bac9 | 1067 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1068 | for (;;) { |
d71df90e | 1069 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1070 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1071 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1072 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1073 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1074 | more2boost = rcu_boost(rnp); |
1075 | if (more2boost) | |
1076 | spincnt++; | |
1077 | else | |
1078 | spincnt = 0; | |
1079 | if (spincnt > 10) { | |
5d01bbd1 | 1080 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1081 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1082 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1083 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1084 | spincnt = 0; |
1085 | } | |
1086 | } | |
1217ed1b | 1087 | /* NOTREACHED */ |
f7f7bac9 | 1088 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1089 | return 0; |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * Check to see if it is time to start boosting RCU readers that are | |
1094 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1095 | * kthread to start boosting them. If there is an expedited grace | |
1096 | * period in progress, it is always time to boost. | |
1097 | * | |
b065a853 PM |
1098 | * The caller must hold rnp->lock, which this function releases. |
1099 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1100 | * about it going away. | |
27f4d280 | 1101 | */ |
1217ed1b | 1102 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1103 | __releases(rnp->lock) |
27f4d280 PM |
1104 | { |
1105 | struct task_struct *t; | |
1106 | ||
0ea1f2eb PM |
1107 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
1108 | rnp->n_balk_exp_gp_tasks++; | |
1217ed1b | 1109 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1110 | return; |
0ea1f2eb | 1111 | } |
27f4d280 PM |
1112 | if (rnp->exp_tasks != NULL || |
1113 | (rnp->gp_tasks != NULL && | |
1114 | rnp->boost_tasks == NULL && | |
1115 | rnp->qsmask == 0 && | |
1116 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1117 | if (rnp->exp_tasks == NULL) | |
1118 | rnp->boost_tasks = rnp->gp_tasks; | |
1217ed1b | 1119 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1120 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1121 | if (t) |
1122 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1123 | } else { |
0ea1f2eb | 1124 | rcu_initiate_boost_trace(rnp); |
1217ed1b PM |
1125 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1126 | } | |
27f4d280 PM |
1127 | } |
1128 | ||
a46e0899 PM |
1129 | /* |
1130 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1131 | */ | |
1132 | static void invoke_rcu_callbacks_kthread(void) | |
1133 | { | |
1134 | unsigned long flags; | |
1135 | ||
1136 | local_irq_save(flags); | |
1137 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1138 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1139 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1140 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1141 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1142 | } | |
a46e0899 PM |
1143 | local_irq_restore(flags); |
1144 | } | |
1145 | ||
dff1672d PM |
1146 | /* |
1147 | * Is the current CPU running the RCU-callbacks kthread? | |
1148 | * Caller must have preemption disabled. | |
1149 | */ | |
1150 | static bool rcu_is_callbacks_kthread(void) | |
1151 | { | |
c9d4b0af | 1152 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1153 | } |
1154 | ||
27f4d280 PM |
1155 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1156 | ||
1157 | /* | |
1158 | * Do priority-boost accounting for the start of a new grace period. | |
1159 | */ | |
1160 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1161 | { | |
1162 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1163 | } | |
1164 | ||
27f4d280 PM |
1165 | /* |
1166 | * Create an RCU-boost kthread for the specified node if one does not | |
1167 | * already exist. We only create this kthread for preemptible RCU. | |
1168 | * Returns zero if all is well, a negated errno otherwise. | |
1169 | */ | |
49fb4c62 | 1170 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
5d01bbd1 | 1171 | struct rcu_node *rnp) |
27f4d280 | 1172 | { |
5d01bbd1 | 1173 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1174 | unsigned long flags; |
1175 | struct sched_param sp; | |
1176 | struct task_struct *t; | |
1177 | ||
1178 | if (&rcu_preempt_state != rsp) | |
1179 | return 0; | |
5d01bbd1 TG |
1180 | |
1181 | if (!rcu_scheduler_fully_active || rnp->qsmaskinit == 0) | |
1182 | return 0; | |
1183 | ||
a46e0899 | 1184 | rsp->boost = 1; |
27f4d280 PM |
1185 | if (rnp->boost_kthread_task != NULL) |
1186 | return 0; | |
1187 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1188 | "rcub/%d", rnp_index); |
27f4d280 PM |
1189 | if (IS_ERR(t)) |
1190 | return PTR_ERR(t); | |
1191 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
6303b9c8 | 1192 | smp_mb__after_unlock_lock(); |
27f4d280 PM |
1193 | rnp->boost_kthread_task = t; |
1194 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
21871d7e | 1195 | sp.sched_priority = kthread_prio; |
27f4d280 | 1196 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1197 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1198 | return 0; |
1199 | } | |
1200 | ||
f8b7fc6b PM |
1201 | static void rcu_kthread_do_work(void) |
1202 | { | |
c9d4b0af CL |
1203 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1204 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
f8b7fc6b PM |
1205 | rcu_preempt_do_callbacks(); |
1206 | } | |
1207 | ||
62ab7072 | 1208 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1209 | { |
f8b7fc6b | 1210 | struct sched_param sp; |
f8b7fc6b | 1211 | |
21871d7e | 1212 | sp.sched_priority = kthread_prio; |
62ab7072 | 1213 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1214 | } |
1215 | ||
62ab7072 | 1216 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1217 | { |
62ab7072 | 1218 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1219 | } |
1220 | ||
62ab7072 | 1221 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1222 | { |
c9d4b0af | 1223 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1224 | } |
1225 | ||
1226 | /* | |
1227 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1228 | * RCU softirq used in flavors and configurations of RCU that do not |
1229 | * support RCU priority boosting. | |
f8b7fc6b | 1230 | */ |
62ab7072 | 1231 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1232 | { |
c9d4b0af CL |
1233 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1234 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1235 | int spincnt; |
f8b7fc6b | 1236 | |
62ab7072 | 1237 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1238 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1239 | local_bh_disable(); |
f8b7fc6b | 1240 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1241 | this_cpu_inc(rcu_cpu_kthread_loops); |
1242 | local_irq_disable(); | |
f8b7fc6b PM |
1243 | work = *workp; |
1244 | *workp = 0; | |
62ab7072 | 1245 | local_irq_enable(); |
f8b7fc6b PM |
1246 | if (work) |
1247 | rcu_kthread_do_work(); | |
1248 | local_bh_enable(); | |
62ab7072 | 1249 | if (*workp == 0) { |
f7f7bac9 | 1250 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1251 | *statusp = RCU_KTHREAD_WAITING; |
1252 | return; | |
f8b7fc6b PM |
1253 | } |
1254 | } | |
62ab7072 | 1255 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1256 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1257 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1258 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1259 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1260 | } |
1261 | ||
1262 | /* | |
1263 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1264 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1265 | * held, so the value of rnp->qsmaskinit will be stable. | |
1266 | * | |
1267 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1268 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1269 | * this function allows the kthread to execute on any CPU. | |
1270 | */ | |
5d01bbd1 | 1271 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1272 | { |
5d01bbd1 TG |
1273 | struct task_struct *t = rnp->boost_kthread_task; |
1274 | unsigned long mask = rnp->qsmaskinit; | |
f8b7fc6b PM |
1275 | cpumask_var_t cm; |
1276 | int cpu; | |
f8b7fc6b | 1277 | |
5d01bbd1 | 1278 | if (!t) |
f8b7fc6b | 1279 | return; |
5d01bbd1 | 1280 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1281 | return; |
f8b7fc6b PM |
1282 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) |
1283 | if ((mask & 0x1) && cpu != outgoingcpu) | |
1284 | cpumask_set_cpu(cpu, cm); | |
5d0b0249 | 1285 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1286 | cpumask_setall(cm); |
5d01bbd1 | 1287 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1288 | free_cpumask_var(cm); |
1289 | } | |
1290 | ||
62ab7072 PM |
1291 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1292 | .store = &rcu_cpu_kthread_task, | |
1293 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1294 | .thread_fn = rcu_cpu_kthread, | |
1295 | .thread_comm = "rcuc/%u", | |
1296 | .setup = rcu_cpu_kthread_setup, | |
1297 | .park = rcu_cpu_kthread_park, | |
1298 | }; | |
f8b7fc6b PM |
1299 | |
1300 | /* | |
9386c0b7 | 1301 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1302 | */ |
9386c0b7 | 1303 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1304 | { |
f8b7fc6b | 1305 | struct rcu_node *rnp; |
5d01bbd1 | 1306 | int cpu; |
f8b7fc6b | 1307 | |
62ab7072 | 1308 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1309 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1310 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
3e9f5c70 PM |
1311 | rcu_for_each_leaf_node(rcu_state_p, rnp) |
1312 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); | |
f8b7fc6b | 1313 | } |
f8b7fc6b | 1314 | |
49fb4c62 | 1315 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1316 | { |
e534165b | 1317 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
f8b7fc6b PM |
1318 | struct rcu_node *rnp = rdp->mynode; |
1319 | ||
1320 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1321 | if (rcu_scheduler_fully_active) |
e534165b | 1322 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); |
f8b7fc6b PM |
1323 | } |
1324 | ||
27f4d280 PM |
1325 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1326 | ||
1217ed1b | 1327 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1328 | __releases(rnp->lock) |
27f4d280 | 1329 | { |
1217ed1b | 1330 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 PM |
1331 | } |
1332 | ||
a46e0899 | 1333 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1334 | { |
a46e0899 | 1335 | WARN_ON_ONCE(1); |
27f4d280 PM |
1336 | } |
1337 | ||
dff1672d PM |
1338 | static bool rcu_is_callbacks_kthread(void) |
1339 | { | |
1340 | return false; | |
1341 | } | |
1342 | ||
27f4d280 PM |
1343 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1344 | { | |
1345 | } | |
1346 | ||
5d01bbd1 | 1347 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1348 | { |
1349 | } | |
1350 | ||
9386c0b7 | 1351 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1352 | { |
b0d30417 | 1353 | } |
b0d30417 | 1354 | |
49fb4c62 | 1355 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1356 | { |
1357 | } | |
1358 | ||
27f4d280 PM |
1359 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1360 | ||
8bd93a2c PM |
1361 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1362 | ||
1363 | /* | |
1364 | * Check to see if any future RCU-related work will need to be done | |
1365 | * by the current CPU, even if none need be done immediately, returning | |
1366 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1367 | * an exported member of the RCU API. | |
1368 | * | |
7cb92499 PM |
1369 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1370 | * any flavor of RCU. | |
8bd93a2c | 1371 | */ |
ffa83fb5 | 1372 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
aa6da514 | 1373 | int rcu_needs_cpu(unsigned long *delta_jiffies) |
8bd93a2c | 1374 | { |
aa9b1630 | 1375 | *delta_jiffies = ULONG_MAX; |
aa6da514 | 1376 | return rcu_cpu_has_callbacks(NULL); |
7cb92499 | 1377 | } |
ffa83fb5 | 1378 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
7cb92499 PM |
1379 | |
1380 | /* | |
1381 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1382 | * after it. | |
1383 | */ | |
8fa7845d | 1384 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1385 | { |
1386 | } | |
1387 | ||
aea1b35e | 1388 | /* |
a858af28 | 1389 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1390 | * is nothing. |
1391 | */ | |
198bbf81 | 1392 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1393 | { |
1394 | } | |
1395 | ||
c57afe80 PM |
1396 | /* |
1397 | * Don't bother keeping a running count of the number of RCU callbacks | |
1398 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1399 | */ | |
1400 | static void rcu_idle_count_callbacks_posted(void) | |
1401 | { | |
1402 | } | |
1403 | ||
8bd93a2c PM |
1404 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1405 | ||
f23f7fa1 PM |
1406 | /* |
1407 | * This code is invoked when a CPU goes idle, at which point we want | |
1408 | * to have the CPU do everything required for RCU so that it can enter | |
1409 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1410 | * state machine implemented by rcu_prepare_for_idle() below. | |
1411 | * | |
1412 | * The following three proprocessor symbols control this state machine: | |
1413 | * | |
f23f7fa1 PM |
1414 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1415 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1416 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1417 | * benchmarkers who might otherwise be tempted to set this to a large | |
1418 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1419 | * system. And if you are -that- concerned about energy efficiency, | |
1420 | * just power the system down and be done with it! | |
778d250a PM |
1421 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1422 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1423 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1424 | * |
1425 | * The values below work well in practice. If future workloads require | |
1426 | * adjustment, they can be converted into kernel config parameters, though | |
1427 | * making the state machine smarter might be a better option. | |
1428 | */ | |
e84c48ae | 1429 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1430 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1431 | |
5e44ce35 PM |
1432 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1433 | module_param(rcu_idle_gp_delay, int, 0644); | |
1434 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1435 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1436 | |
d689fe22 | 1437 | extern int tick_nohz_active; |
486e2593 PM |
1438 | |
1439 | /* | |
c229828c PM |
1440 | * Try to advance callbacks for all flavors of RCU on the current CPU, but |
1441 | * only if it has been awhile since the last time we did so. Afterwards, | |
1442 | * if there are any callbacks ready for immediate invocation, return true. | |
486e2593 | 1443 | */ |
f1f399d1 | 1444 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1445 | { |
c0f4dfd4 PM |
1446 | bool cbs_ready = false; |
1447 | struct rcu_data *rdp; | |
c229828c | 1448 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1449 | struct rcu_node *rnp; |
1450 | struct rcu_state *rsp; | |
486e2593 | 1451 | |
c229828c PM |
1452 | /* Exit early if we advanced recently. */ |
1453 | if (jiffies == rdtp->last_advance_all) | |
d0bc90fd | 1454 | return false; |
c229828c PM |
1455 | rdtp->last_advance_all = jiffies; |
1456 | ||
c0f4dfd4 PM |
1457 | for_each_rcu_flavor(rsp) { |
1458 | rdp = this_cpu_ptr(rsp->rda); | |
1459 | rnp = rdp->mynode; | |
486e2593 | 1460 | |
c0f4dfd4 PM |
1461 | /* |
1462 | * Don't bother checking unless a grace period has | |
1463 | * completed since we last checked and there are | |
1464 | * callbacks not yet ready to invoke. | |
1465 | */ | |
e3663b10 PM |
1466 | if ((rdp->completed != rnp->completed || |
1467 | unlikely(ACCESS_ONCE(rdp->gpwrap))) && | |
c0f4dfd4 | 1468 | rdp->nxttail[RCU_DONE_TAIL] != rdp->nxttail[RCU_NEXT_TAIL]) |
470716fc | 1469 | note_gp_changes(rsp, rdp); |
486e2593 | 1470 | |
c0f4dfd4 PM |
1471 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1472 | cbs_ready = true; | |
1473 | } | |
1474 | return cbs_ready; | |
486e2593 PM |
1475 | } |
1476 | ||
aa9b1630 | 1477 | /* |
c0f4dfd4 PM |
1478 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1479 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1480 | * caller to set the timeout based on whether or not there are non-lazy | |
1481 | * callbacks. | |
aa9b1630 | 1482 | * |
c0f4dfd4 | 1483 | * The caller must have disabled interrupts. |
aa9b1630 | 1484 | */ |
ffa83fb5 | 1485 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
aa6da514 | 1486 | int rcu_needs_cpu(unsigned long *dj) |
aa9b1630 | 1487 | { |
aa6da514 | 1488 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
aa9b1630 | 1489 | |
c0f4dfd4 PM |
1490 | /* Snapshot to detect later posting of non-lazy callback. */ |
1491 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1492 | ||
aa9b1630 | 1493 | /* If no callbacks, RCU doesn't need the CPU. */ |
aa6da514 | 1494 | if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) { |
c0f4dfd4 | 1495 | *dj = ULONG_MAX; |
aa9b1630 PM |
1496 | return 0; |
1497 | } | |
c0f4dfd4 PM |
1498 | |
1499 | /* Attempt to advance callbacks. */ | |
1500 | if (rcu_try_advance_all_cbs()) { | |
1501 | /* Some ready to invoke, so initiate later invocation. */ | |
1502 | invoke_rcu_core(); | |
aa9b1630 PM |
1503 | return 1; |
1504 | } | |
c0f4dfd4 PM |
1505 | rdtp->last_accelerate = jiffies; |
1506 | ||
1507 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1508 | if (!rdtp->all_lazy) { |
c0f4dfd4 PM |
1509 | *dj = round_up(rcu_idle_gp_delay + jiffies, |
1510 | rcu_idle_gp_delay) - jiffies; | |
e84c48ae | 1511 | } else { |
c0f4dfd4 | 1512 | *dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1513 | } |
aa9b1630 PM |
1514 | return 0; |
1515 | } | |
ffa83fb5 | 1516 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
aa9b1630 | 1517 | |
21e52e15 | 1518 | /* |
c0f4dfd4 PM |
1519 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1520 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1521 | * The second major task is to check to see if a non-lazy callback has | |
1522 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1523 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1524 | * any recently arrived callbacks. | |
aea1b35e PM |
1525 | * |
1526 | * The caller must have disabled interrupts. | |
8bd93a2c | 1527 | */ |
198bbf81 | 1528 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1529 | { |
f1f399d1 | 1530 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
48a7639c | 1531 | bool needwake; |
c0f4dfd4 | 1532 | struct rcu_data *rdp; |
198bbf81 | 1533 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1534 | struct rcu_node *rnp; |
1535 | struct rcu_state *rsp; | |
9d2ad243 PM |
1536 | int tne; |
1537 | ||
1538 | /* Handle nohz enablement switches conservatively. */ | |
d689fe22 | 1539 | tne = ACCESS_ONCE(tick_nohz_active); |
9d2ad243 | 1540 | if (tne != rdtp->tick_nohz_enabled_snap) { |
aa6da514 | 1541 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 PM |
1542 | invoke_rcu_core(); /* force nohz to see update. */ |
1543 | rdtp->tick_nohz_enabled_snap = tne; | |
1544 | return; | |
1545 | } | |
1546 | if (!tne) | |
1547 | return; | |
f511fc62 | 1548 | |
c0f4dfd4 | 1549 | /* If this is a no-CBs CPU, no callbacks, just return. */ |
198bbf81 | 1550 | if (rcu_is_nocb_cpu(smp_processor_id())) |
9a0c6fef | 1551 | return; |
9a0c6fef | 1552 | |
c57afe80 | 1553 | /* |
c0f4dfd4 PM |
1554 | * If a non-lazy callback arrived at a CPU having only lazy |
1555 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1556 | * idle duration on re-entry to idle. | |
c57afe80 | 1557 | */ |
c0f4dfd4 PM |
1558 | if (rdtp->all_lazy && |
1559 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
c337f8f5 PM |
1560 | rdtp->all_lazy = false; |
1561 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
c0f4dfd4 | 1562 | invoke_rcu_core(); |
c57afe80 PM |
1563 | return; |
1564 | } | |
c57afe80 | 1565 | |
3084f2f8 | 1566 | /* |
c0f4dfd4 PM |
1567 | * If we have not yet accelerated this jiffy, accelerate all |
1568 | * callbacks on this CPU. | |
3084f2f8 | 1569 | */ |
c0f4dfd4 | 1570 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1571 | return; |
c0f4dfd4 PM |
1572 | rdtp->last_accelerate = jiffies; |
1573 | for_each_rcu_flavor(rsp) { | |
198bbf81 | 1574 | rdp = this_cpu_ptr(rsp->rda); |
c0f4dfd4 PM |
1575 | if (!*rdp->nxttail[RCU_DONE_TAIL]) |
1576 | continue; | |
1577 | rnp = rdp->mynode; | |
1578 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
6303b9c8 | 1579 | smp_mb__after_unlock_lock(); |
48a7639c | 1580 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
c0f4dfd4 | 1581 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
48a7639c PM |
1582 | if (needwake) |
1583 | rcu_gp_kthread_wake(rsp); | |
77e38ed3 | 1584 | } |
f1f399d1 | 1585 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
c0f4dfd4 | 1586 | } |
3084f2f8 | 1587 | |
c0f4dfd4 PM |
1588 | /* |
1589 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1590 | * any grace periods that elapsed while the CPU was idle, and if any | |
1591 | * callbacks are now ready to invoke, initiate invocation. | |
1592 | */ | |
8fa7845d | 1593 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1594 | { |
f1f399d1 | 1595 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
8fa7845d | 1596 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1597 | return; |
7a497c96 PM |
1598 | if (rcu_try_advance_all_cbs()) |
1599 | invoke_rcu_core(); | |
f1f399d1 | 1600 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
8bd93a2c PM |
1601 | } |
1602 | ||
c57afe80 | 1603 | /* |
98248a0e PM |
1604 | * Keep a running count of the number of non-lazy callbacks posted |
1605 | * on this CPU. This running counter (which is never decremented) allows | |
1606 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1607 | * posts a callback, even if an equal number of callbacks are invoked. | |
1608 | * Of course, callbacks should only be posted from within a trace event | |
1609 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1610 | */ |
1611 | static void rcu_idle_count_callbacks_posted(void) | |
1612 | { | |
5955f7ee | 1613 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1614 | } |
1615 | ||
b626c1b6 PM |
1616 | /* |
1617 | * Data for flushing lazy RCU callbacks at OOM time. | |
1618 | */ | |
1619 | static atomic_t oom_callback_count; | |
1620 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1621 | ||
1622 | /* | |
1623 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1624 | * wake-up if we are the last one. | |
1625 | */ | |
1626 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1627 | { | |
1628 | if (atomic_dec_and_test(&oom_callback_count)) | |
1629 | wake_up(&oom_callback_wq); | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1634 | * least one lazy callback. This will unnecessarily post callbacks | |
1635 | * to CPUs that already have a non-lazy callback at the end of their | |
1636 | * callback list, but this is an infrequent operation, so accept some | |
1637 | * extra overhead to keep things simple. | |
1638 | */ | |
1639 | static void rcu_oom_notify_cpu(void *unused) | |
1640 | { | |
1641 | struct rcu_state *rsp; | |
1642 | struct rcu_data *rdp; | |
1643 | ||
1644 | for_each_rcu_flavor(rsp) { | |
fa07a58f | 1645 | rdp = raw_cpu_ptr(rsp->rda); |
b626c1b6 PM |
1646 | if (rdp->qlen_lazy != 0) { |
1647 | atomic_inc(&oom_callback_count); | |
1648 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1649 | } | |
1650 | } | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1655 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1656 | * ensuring that they free up the corresponding memory in a timely manner. | |
1657 | * Because an uncertain amount of memory will be freed in some uncertain | |
1658 | * timeframe, we do not claim to have freed anything. | |
1659 | */ | |
1660 | static int rcu_oom_notify(struct notifier_block *self, | |
1661 | unsigned long notused, void *nfreed) | |
1662 | { | |
1663 | int cpu; | |
1664 | ||
1665 | /* Wait for callbacks from earlier instance to complete. */ | |
1666 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
78e4bc34 | 1667 | smp_mb(); /* Ensure callback reuse happens after callback invocation. */ |
b626c1b6 PM |
1668 | |
1669 | /* | |
1670 | * Prevent premature wakeup: ensure that all increments happen | |
1671 | * before there is a chance of the counter reaching zero. | |
1672 | */ | |
1673 | atomic_set(&oom_callback_count, 1); | |
1674 | ||
1675 | get_online_cpus(); | |
1676 | for_each_online_cpu(cpu) { | |
1677 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
bde6c3aa | 1678 | cond_resched_rcu_qs(); |
b626c1b6 PM |
1679 | } |
1680 | put_online_cpus(); | |
1681 | ||
1682 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1683 | atomic_dec(&oom_callback_count); | |
1684 | ||
1685 | return NOTIFY_OK; | |
1686 | } | |
1687 | ||
1688 | static struct notifier_block rcu_oom_nb = { | |
1689 | .notifier_call = rcu_oom_notify | |
1690 | }; | |
1691 | ||
1692 | static int __init rcu_register_oom_notifier(void) | |
1693 | { | |
1694 | register_oom_notifier(&rcu_oom_nb); | |
1695 | return 0; | |
1696 | } | |
1697 | early_initcall(rcu_register_oom_notifier); | |
1698 | ||
8bd93a2c | 1699 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 PM |
1700 | |
1701 | #ifdef CONFIG_RCU_CPU_STALL_INFO | |
1702 | ||
1703 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
1704 | ||
1705 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1706 | { | |
5955f7ee | 1707 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1708 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1709 | |
c0f4dfd4 PM |
1710 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1711 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1712 | ulong2long(nlpd), | |
1713 | rdtp->all_lazy ? 'L' : '.', | |
1714 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1715 | } |
1716 | ||
1717 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1718 | ||
1719 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1720 | { | |
1c17e4d4 | 1721 | *cp = '\0'; |
a858af28 PM |
1722 | } |
1723 | ||
1724 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1725 | ||
1726 | /* Initiate the stall-info list. */ | |
1727 | static void print_cpu_stall_info_begin(void) | |
1728 | { | |
efc151c3 | 1729 | pr_cont("\n"); |
a858af28 PM |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * Print out diagnostic information for the specified stalled CPU. | |
1734 | * | |
1735 | * If the specified CPU is aware of the current RCU grace period | |
1736 | * (flavor specified by rsp), then print the number of scheduling | |
1737 | * clock interrupts the CPU has taken during the time that it has | |
1738 | * been aware. Otherwise, print the number of RCU grace periods | |
1739 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1740 | * aware of the previous grace period. | |
1741 | * | |
1742 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1743 | */ | |
1744 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1745 | { | |
1746 | char fast_no_hz[72]; | |
1747 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1748 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1749 | char *ticks_title; | |
1750 | unsigned long ticks_value; | |
1751 | ||
1752 | if (rsp->gpnum == rdp->gpnum) { | |
1753 | ticks_title = "ticks this GP"; | |
1754 | ticks_value = rdp->ticks_this_gp; | |
1755 | } else { | |
1756 | ticks_title = "GPs behind"; | |
1757 | ticks_value = rsp->gpnum - rdp->gpnum; | |
1758 | } | |
1759 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
fc908ed3 | 1760 | pr_err("\t%d: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n", |
a858af28 PM |
1761 | cpu, ticks_value, ticks_title, |
1762 | atomic_read(&rdtp->dynticks) & 0xfff, | |
1763 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, | |
6231069b | 1764 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
fc908ed3 | 1765 | ACCESS_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart, |
a858af28 PM |
1766 | fast_no_hz); |
1767 | } | |
1768 | ||
1769 | /* Terminate the stall-info list. */ | |
1770 | static void print_cpu_stall_info_end(void) | |
1771 | { | |
efc151c3 | 1772 | pr_err("\t"); |
a858af28 PM |
1773 | } |
1774 | ||
1775 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1776 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1777 | { | |
1778 | rdp->ticks_this_gp = 0; | |
6231069b | 1779 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1780 | } |
1781 | ||
1782 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1783 | static void increment_cpu_stall_ticks(void) | |
1784 | { | |
115f7a7c PM |
1785 | struct rcu_state *rsp; |
1786 | ||
1787 | for_each_rcu_flavor(rsp) | |
fa07a58f | 1788 | raw_cpu_inc(rsp->rda->ticks_this_gp); |
a858af28 PM |
1789 | } |
1790 | ||
1791 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
1792 | ||
1793 | static void print_cpu_stall_info_begin(void) | |
1794 | { | |
efc151c3 | 1795 | pr_cont(" {"); |
a858af28 PM |
1796 | } |
1797 | ||
1798 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1799 | { | |
efc151c3 | 1800 | pr_cont(" %d", cpu); |
a858af28 PM |
1801 | } |
1802 | ||
1803 | static void print_cpu_stall_info_end(void) | |
1804 | { | |
efc151c3 | 1805 | pr_cont("} "); |
a858af28 PM |
1806 | } |
1807 | ||
1808 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1809 | { | |
1810 | } | |
1811 | ||
1812 | static void increment_cpu_stall_ticks(void) | |
1813 | { | |
1814 | } | |
1815 | ||
1816 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
3fbfbf7a PM |
1817 | |
1818 | #ifdef CONFIG_RCU_NOCB_CPU | |
1819 | ||
1820 | /* | |
1821 | * Offload callback processing from the boot-time-specified set of CPUs | |
1822 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1823 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1824 | * waits for a grace period to elapse, and invokes the callbacks. | |
1825 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1826 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1827 | * has been specified, in which case each kthread actively polls its | |
1828 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1829 | * reduce RCU's overhead on that CPU.) | |
1830 | * | |
1831 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1832 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1833 | * running CPU-bound user-mode computations. | |
1834 | * | |
1835 | * Offloading of callback processing could also in theory be used as | |
1836 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1837 | * queued are more aggressive about entering dyntick-idle mode. | |
1838 | */ | |
1839 | ||
1840 | ||
1841 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1842 | static int __init rcu_nocb_setup(char *str) | |
1843 | { | |
1844 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
1845 | have_rcu_nocb_mask = true; | |
1846 | cpulist_parse(str, rcu_nocb_mask); | |
1847 | return 1; | |
1848 | } | |
1849 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1850 | ||
1b0048a4 PG |
1851 | static int __init parse_rcu_nocb_poll(char *arg) |
1852 | { | |
1853 | rcu_nocb_poll = 1; | |
1854 | return 0; | |
1855 | } | |
1856 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1857 | ||
dae6e64d | 1858 | /* |
0446be48 PM |
1859 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1860 | * grace period. | |
dae6e64d | 1861 | */ |
0446be48 | 1862 | static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
dae6e64d | 1863 | { |
0446be48 | 1864 | wake_up_all(&rnp->nocb_gp_wq[rnp->completed & 0x1]); |
dae6e64d PM |
1865 | } |
1866 | ||
1867 | /* | |
8b425aa8 | 1868 | * Set the root rcu_node structure's ->need_future_gp field |
dae6e64d PM |
1869 | * based on the sum of those of all rcu_node structures. This does |
1870 | * double-count the root rcu_node structure's requests, but this | |
1871 | * is necessary to handle the possibility of a rcu_nocb_kthread() | |
1872 | * having awakened during the time that the rcu_node structures | |
1873 | * were being updated for the end of the previous grace period. | |
34ed6246 | 1874 | */ |
dae6e64d PM |
1875 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
1876 | { | |
8b425aa8 | 1877 | rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq; |
dae6e64d PM |
1878 | } |
1879 | ||
1880 | static void rcu_init_one_nocb(struct rcu_node *rnp) | |
34ed6246 | 1881 | { |
dae6e64d PM |
1882 | init_waitqueue_head(&rnp->nocb_gp_wq[0]); |
1883 | init_waitqueue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1884 | } |
1885 | ||
2f33b512 | 1886 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
24342c96 | 1887 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1888 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a PM |
1889 | { |
1890 | if (have_rcu_nocb_mask) | |
1891 | return cpumask_test_cpu(cpu, rcu_nocb_mask); | |
1892 | return false; | |
1893 | } | |
2f33b512 | 1894 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
3fbfbf7a | 1895 | |
fbce7497 PM |
1896 | /* |
1897 | * Kick the leader kthread for this NOCB group. | |
1898 | */ | |
1899 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1900 | { | |
1901 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1902 | ||
1903 | if (!ACCESS_ONCE(rdp_leader->nocb_kthread)) | |
1904 | return; | |
11ed7f93 | 1905 | if (ACCESS_ONCE(rdp_leader->nocb_leader_sleep) || force) { |
39953dfd | 1906 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
11ed7f93 | 1907 | ACCESS_ONCE(rdp_leader->nocb_leader_sleep) = false; |
fbce7497 PM |
1908 | wake_up(&rdp_leader->nocb_wq); |
1909 | } | |
1910 | } | |
1911 | ||
d7e29933 PM |
1912 | /* |
1913 | * Does the specified CPU need an RCU callback for the specified flavor | |
1914 | * of rcu_barrier()? | |
1915 | */ | |
1916 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) | |
1917 | { | |
1918 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
41050a00 PM |
1919 | unsigned long ret; |
1920 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1921 | struct rcu_head *rhp; |
41050a00 | 1922 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1923 | |
41050a00 PM |
1924 | /* |
1925 | * Check count of all no-CBs callbacks awaiting invocation. | |
1926 | * There needs to be a barrier before this function is called, | |
1927 | * but associated with a prior determination that no more | |
1928 | * callbacks would be posted. In the worst case, the first | |
1929 | * barrier in _rcu_barrier() suffices (but the caller cannot | |
1930 | * necessarily rely on this, not a substitute for the caller | |
1931 | * getting the concurrency design right!). There must also be | |
1932 | * a barrier between the following load an posting of a callback | |
1933 | * (if a callback is in fact needed). This is associated with an | |
1934 | * atomic_inc() in the caller. | |
1935 | */ | |
1936 | ret = atomic_long_read(&rdp->nocb_q_count); | |
d7e29933 | 1937 | |
41050a00 | 1938 | #ifdef CONFIG_PROVE_RCU |
d7e29933 PM |
1939 | rhp = ACCESS_ONCE(rdp->nocb_head); |
1940 | if (!rhp) | |
1941 | rhp = ACCESS_ONCE(rdp->nocb_gp_head); | |
1942 | if (!rhp) | |
1943 | rhp = ACCESS_ONCE(rdp->nocb_follower_head); | |
1944 | ||
1945 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
1946 | if (!ACCESS_ONCE(rdp->nocb_kthread) && rhp) { | |
1947 | /* RCU callback enqueued before CPU first came online??? */ | |
1948 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1949 | cpu, rhp->func); | |
1950 | WARN_ON_ONCE(1); | |
1951 | } | |
41050a00 | 1952 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1953 | |
41050a00 | 1954 | return !!ret; |
d7e29933 PM |
1955 | } |
1956 | ||
3fbfbf7a PM |
1957 | /* |
1958 | * Enqueue the specified string of rcu_head structures onto the specified | |
1959 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
1960 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
1961 | * counts are supplied by rhcount and rhcount_lazy. | |
1962 | * | |
1963 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1964 | */ | |
1965 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
1966 | struct rcu_head *rhp, | |
1967 | struct rcu_head **rhtp, | |
96d3fd0d PM |
1968 | int rhcount, int rhcount_lazy, |
1969 | unsigned long flags) | |
3fbfbf7a PM |
1970 | { |
1971 | int len; | |
1972 | struct rcu_head **old_rhpp; | |
1973 | struct task_struct *t; | |
1974 | ||
1975 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
1976 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
1977 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a PM |
1978 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
1979 | ACCESS_ONCE(*old_rhpp) = rhp; | |
3fbfbf7a | 1980 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 1981 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
1982 | |
1983 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
1984 | t = ACCESS_ONCE(rdp->nocb_kthread); | |
25e03a74 | 1985 | if (rcu_nocb_poll || !t) { |
9261dd0d PM |
1986 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1987 | TPS("WakeNotPoll")); | |
3fbfbf7a | 1988 | return; |
9261dd0d | 1989 | } |
3fbfbf7a PM |
1990 | len = atomic_long_read(&rdp->nocb_q_count); |
1991 | if (old_rhpp == &rdp->nocb_head) { | |
96d3fd0d | 1992 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
1993 | /* ... if queue was empty ... */ |
1994 | wake_nocb_leader(rdp, false); | |
96d3fd0d PM |
1995 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1996 | TPS("WakeEmpty")); | |
1997 | } else { | |
9fdd3bc9 | 1998 | rdp->nocb_defer_wakeup = RCU_NOGP_WAKE; |
96d3fd0d PM |
1999 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2000 | TPS("WakeEmptyIsDeferred")); | |
2001 | } | |
3fbfbf7a PM |
2002 | rdp->qlen_last_fqs_check = 0; |
2003 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 2004 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
2005 | if (!irqs_disabled_flags(flags)) { |
2006 | wake_nocb_leader(rdp, true); | |
2007 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2008 | TPS("WakeOvf")); | |
2009 | } else { | |
2010 | rdp->nocb_defer_wakeup = RCU_NOGP_WAKE_FORCE; | |
2011 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2012 | TPS("WakeOvfIsDeferred")); | |
2013 | } | |
3fbfbf7a | 2014 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d PM |
2015 | } else { |
2016 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot")); | |
3fbfbf7a PM |
2017 | } |
2018 | return; | |
2019 | } | |
2020 | ||
2021 | /* | |
2022 | * This is a helper for __call_rcu(), which invokes this when the normal | |
2023 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
2024 | * function returns failure back to __call_rcu(), which can complain | |
2025 | * appropriately. | |
2026 | * | |
2027 | * Otherwise, this function queues the callback where the corresponding | |
2028 | * "rcuo" kthread can find it. | |
2029 | */ | |
2030 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 2031 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
2032 | { |
2033 | ||
d1e43fa5 | 2034 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 2035 | return false; |
96d3fd0d | 2036 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 PM |
2037 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
2038 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
2039 | (unsigned long)rhp->func, | |
756cbf6b PM |
2040 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2041 | -atomic_long_read(&rdp->nocb_q_count)); | |
21e7a608 PM |
2042 | else |
2043 | trace_rcu_callback(rdp->rsp->name, rhp, | |
756cbf6b PM |
2044 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2045 | -atomic_long_read(&rdp->nocb_q_count)); | |
1772947b PM |
2046 | |
2047 | /* | |
2048 | * If called from an extended quiescent state with interrupts | |
2049 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2050 | * deferred-wakeup check to function. | |
2051 | */ | |
2052 | if (irqs_disabled_flags(flags) && | |
2053 | !rcu_is_watching() && | |
2054 | cpu_online(smp_processor_id())) | |
2055 | invoke_rcu_core(); | |
2056 | ||
c271d3a9 | 2057 | return true; |
3fbfbf7a PM |
2058 | } |
2059 | ||
2060 | /* | |
2061 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2062 | * not a no-CBs CPU. | |
2063 | */ | |
2064 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
96d3fd0d PM |
2065 | struct rcu_data *rdp, |
2066 | unsigned long flags) | |
3fbfbf7a PM |
2067 | { |
2068 | long ql = rsp->qlen; | |
2069 | long qll = rsp->qlen_lazy; | |
2070 | ||
2071 | /* If this is not a no-CBs CPU, tell the caller to do it the old way. */ | |
d1e43fa5 | 2072 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
0a9e1e11 | 2073 | return false; |
3fbfbf7a PM |
2074 | rsp->qlen = 0; |
2075 | rsp->qlen_lazy = 0; | |
2076 | ||
2077 | /* First, enqueue the donelist, if any. This preserves CB ordering. */ | |
2078 | if (rsp->orphan_donelist != NULL) { | |
2079 | __call_rcu_nocb_enqueue(rdp, rsp->orphan_donelist, | |
96d3fd0d | 2080 | rsp->orphan_donetail, ql, qll, flags); |
3fbfbf7a PM |
2081 | ql = qll = 0; |
2082 | rsp->orphan_donelist = NULL; | |
2083 | rsp->orphan_donetail = &rsp->orphan_donelist; | |
2084 | } | |
2085 | if (rsp->orphan_nxtlist != NULL) { | |
2086 | __call_rcu_nocb_enqueue(rdp, rsp->orphan_nxtlist, | |
96d3fd0d | 2087 | rsp->orphan_nxttail, ql, qll, flags); |
3fbfbf7a PM |
2088 | ql = qll = 0; |
2089 | rsp->orphan_nxtlist = NULL; | |
2090 | rsp->orphan_nxttail = &rsp->orphan_nxtlist; | |
2091 | } | |
0a9e1e11 | 2092 | return true; |
3fbfbf7a PM |
2093 | } |
2094 | ||
2095 | /* | |
34ed6246 PM |
2096 | * If necessary, kick off a new grace period, and either way wait |
2097 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2098 | */ |
34ed6246 | 2099 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2100 | { |
34ed6246 | 2101 | unsigned long c; |
dae6e64d | 2102 | bool d; |
34ed6246 | 2103 | unsigned long flags; |
48a7639c | 2104 | bool needwake; |
34ed6246 PM |
2105 | struct rcu_node *rnp = rdp->mynode; |
2106 | ||
2107 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
6303b9c8 | 2108 | smp_mb__after_unlock_lock(); |
48a7639c | 2109 | needwake = rcu_start_future_gp(rnp, rdp, &c); |
0446be48 | 2110 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
48a7639c PM |
2111 | if (needwake) |
2112 | rcu_gp_kthread_wake(rdp->rsp); | |
3fbfbf7a PM |
2113 | |
2114 | /* | |
34ed6246 PM |
2115 | * Wait for the grace period. Do so interruptibly to avoid messing |
2116 | * up the load average. | |
3fbfbf7a | 2117 | */ |
f7f7bac9 | 2118 | trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2119 | for (;;) { |
dae6e64d PM |
2120 | wait_event_interruptible( |
2121 | rnp->nocb_gp_wq[c & 0x1], | |
2122 | (d = ULONG_CMP_GE(ACCESS_ONCE(rnp->completed), c))); | |
2123 | if (likely(d)) | |
34ed6246 | 2124 | break; |
73a860cd | 2125 | WARN_ON(signal_pending(current)); |
f7f7bac9 | 2126 | trace_rcu_future_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2127 | } |
f7f7bac9 | 2128 | trace_rcu_future_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2129 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2130 | } |
2131 | ||
fbce7497 PM |
2132 | /* |
2133 | * Leaders come here to wait for additional callbacks to show up. | |
2134 | * This function does not return until callbacks appear. | |
2135 | */ | |
2136 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2137 | { | |
2138 | bool firsttime = true; | |
2139 | bool gotcbs; | |
2140 | struct rcu_data *rdp; | |
2141 | struct rcu_head **tail; | |
2142 | ||
2143 | wait_again: | |
2144 | ||
2145 | /* Wait for callbacks to appear. */ | |
2146 | if (!rcu_nocb_poll) { | |
2147 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, "Sleep"); | |
2148 | wait_event_interruptible(my_rdp->nocb_wq, | |
11ed7f93 | 2149 | !ACCESS_ONCE(my_rdp->nocb_leader_sleep)); |
fbce7497 PM |
2150 | /* Memory barrier handled by smp_mb() calls below and repoll. */ |
2151 | } else if (firsttime) { | |
2152 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
2153 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, "Poll"); | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Each pass through the following loop checks a follower for CBs. | |
2158 | * We are our own first follower. Any CBs found are moved to | |
2159 | * nocb_gp_head, where they await a grace period. | |
2160 | */ | |
2161 | gotcbs = false; | |
2162 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
2163 | rdp->nocb_gp_head = ACCESS_ONCE(rdp->nocb_head); | |
2164 | if (!rdp->nocb_gp_head) | |
2165 | continue; /* No CBs here, try next follower. */ | |
2166 | ||
2167 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
2168 | ACCESS_ONCE(rdp->nocb_head) = NULL; | |
2169 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); | |
fbce7497 PM |
2170 | gotcbs = true; |
2171 | } | |
2172 | ||
2173 | /* | |
2174 | * If there were no callbacks, sleep a bit, rescan after a | |
2175 | * memory barrier, and go retry. | |
2176 | */ | |
2177 | if (unlikely(!gotcbs)) { | |
2178 | if (!rcu_nocb_poll) | |
2179 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, | |
2180 | "WokeEmpty"); | |
73a860cd | 2181 | WARN_ON(signal_pending(current)); |
fbce7497 PM |
2182 | schedule_timeout_interruptible(1); |
2183 | ||
2184 | /* Rescan in case we were a victim of memory ordering. */ | |
11ed7f93 PK |
2185 | my_rdp->nocb_leader_sleep = true; |
2186 | smp_mb(); /* Ensure _sleep true before scan. */ | |
fbce7497 PM |
2187 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) |
2188 | if (ACCESS_ONCE(rdp->nocb_head)) { | |
2189 | /* Found CB, so short-circuit next wait. */ | |
11ed7f93 | 2190 | my_rdp->nocb_leader_sleep = false; |
fbce7497 PM |
2191 | break; |
2192 | } | |
2193 | goto wait_again; | |
2194 | } | |
2195 | ||
2196 | /* Wait for one grace period. */ | |
2197 | rcu_nocb_wait_gp(my_rdp); | |
2198 | ||
2199 | /* | |
11ed7f93 PK |
2200 | * We left ->nocb_leader_sleep unset to reduce cache thrashing. |
2201 | * We set it now, but recheck for new callbacks while | |
fbce7497 PM |
2202 | * traversing our follower list. |
2203 | */ | |
11ed7f93 PK |
2204 | my_rdp->nocb_leader_sleep = true; |
2205 | smp_mb(); /* Ensure _sleep true before scan of ->nocb_head. */ | |
fbce7497 PM |
2206 | |
2207 | /* Each pass through the following loop wakes a follower, if needed. */ | |
2208 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
2209 | if (ACCESS_ONCE(rdp->nocb_head)) | |
11ed7f93 | 2210 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
fbce7497 PM |
2211 | if (!rdp->nocb_gp_head) |
2212 | continue; /* No CBs, so no need to wake follower. */ | |
2213 | ||
2214 | /* Append callbacks to follower's "done" list. */ | |
2215 | tail = xchg(&rdp->nocb_follower_tail, rdp->nocb_gp_tail); | |
2216 | *tail = rdp->nocb_gp_head; | |
c847f142 | 2217 | smp_mb__after_atomic(); /* Store *tail before wakeup. */ |
fbce7497 PM |
2218 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
2219 | /* | |
2220 | * List was empty, wake up the follower. | |
2221 | * Memory barriers supplied by atomic_long_add(). | |
2222 | */ | |
2223 | wake_up(&rdp->nocb_wq); | |
2224 | } | |
2225 | } | |
2226 | ||
2227 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2228 | if (!my_rdp->nocb_follower_head) | |
2229 | goto wait_again; | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * Followers come here to wait for additional callbacks to show up. | |
2234 | * This function does not return until callbacks appear. | |
2235 | */ | |
2236 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2237 | { | |
2238 | bool firsttime = true; | |
2239 | ||
2240 | for (;;) { | |
2241 | if (!rcu_nocb_poll) { | |
2242 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2243 | "FollowerSleep"); | |
2244 | wait_event_interruptible(rdp->nocb_wq, | |
2245 | ACCESS_ONCE(rdp->nocb_follower_head)); | |
2246 | } else if (firsttime) { | |
2247 | /* Don't drown trace log with "Poll"! */ | |
2248 | firsttime = false; | |
2249 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, "Poll"); | |
2250 | } | |
2251 | if (smp_load_acquire(&rdp->nocb_follower_head)) { | |
2252 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2253 | return; | |
2254 | } | |
2255 | if (!rcu_nocb_poll) | |
2256 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2257 | "WokeEmpty"); | |
73a860cd | 2258 | WARN_ON(signal_pending(current)); |
fbce7497 PM |
2259 | schedule_timeout_interruptible(1); |
2260 | } | |
2261 | } | |
2262 | ||
3fbfbf7a PM |
2263 | /* |
2264 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2265 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2266 | * an optional leader-follower relationship so that the grace-period | |
2267 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2268 | */ |
2269 | static int rcu_nocb_kthread(void *arg) | |
2270 | { | |
2271 | int c, cl; | |
2272 | struct rcu_head *list; | |
2273 | struct rcu_head *next; | |
2274 | struct rcu_head **tail; | |
2275 | struct rcu_data *rdp = arg; | |
2276 | ||
2277 | /* Each pass through this loop invokes one batch of callbacks */ | |
2278 | for (;;) { | |
fbce7497 PM |
2279 | /* Wait for callbacks. */ |
2280 | if (rdp->nocb_leader == rdp) | |
2281 | nocb_leader_wait(rdp); | |
2282 | else | |
2283 | nocb_follower_wait(rdp); | |
2284 | ||
2285 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
2286 | list = ACCESS_ONCE(rdp->nocb_follower_head); | |
2287 | BUG_ON(!list); | |
2288 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, "WokeNonEmpty"); | |
2289 | ACCESS_ONCE(rdp->nocb_follower_head) = NULL; | |
2290 | tail = xchg(&rdp->nocb_follower_tail, &rdp->nocb_follower_head); | |
3fbfbf7a PM |
2291 | |
2292 | /* Each pass through the following loop invokes a callback. */ | |
41050a00 PM |
2293 | trace_rcu_batch_start(rdp->rsp->name, |
2294 | atomic_long_read(&rdp->nocb_q_count_lazy), | |
2295 | atomic_long_read(&rdp->nocb_q_count), -1); | |
3fbfbf7a PM |
2296 | c = cl = 0; |
2297 | while (list) { | |
2298 | next = list->next; | |
2299 | /* Wait for enqueuing to complete, if needed. */ | |
2300 | while (next == NULL && &list->next != tail) { | |
69a79bb1 PM |
2301 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2302 | TPS("WaitQueue")); | |
3fbfbf7a | 2303 | schedule_timeout_interruptible(1); |
69a79bb1 PM |
2304 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2305 | TPS("WokeQueue")); | |
3fbfbf7a PM |
2306 | next = list->next; |
2307 | } | |
2308 | debug_rcu_head_unqueue(list); | |
2309 | local_bh_disable(); | |
2310 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2311 | cl++; | |
2312 | c++; | |
2313 | local_bh_enable(); | |
2314 | list = next; | |
2315 | } | |
2316 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
41050a00 PM |
2317 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2318 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2319 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
c635a4e1 | 2320 | rdp->n_nocbs_invoked += c; |
3fbfbf7a PM |
2321 | } |
2322 | return 0; | |
2323 | } | |
2324 | ||
96d3fd0d | 2325 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2326 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2327 | { |
2328 | return ACCESS_ONCE(rdp->nocb_defer_wakeup); | |
2329 | } | |
2330 | ||
2331 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
2332 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2333 | { | |
9fdd3bc9 PM |
2334 | int ndw; |
2335 | ||
96d3fd0d PM |
2336 | if (!rcu_nocb_need_deferred_wakeup(rdp)) |
2337 | return; | |
9fdd3bc9 PM |
2338 | ndw = ACCESS_ONCE(rdp->nocb_defer_wakeup); |
2339 | ACCESS_ONCE(rdp->nocb_defer_wakeup) = RCU_NOGP_WAKE_NOT; | |
2340 | wake_nocb_leader(rdp, ndw == RCU_NOGP_WAKE_FORCE); | |
2341 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake")); | |
96d3fd0d PM |
2342 | } |
2343 | ||
f4579fc5 PM |
2344 | void __init rcu_init_nohz(void) |
2345 | { | |
2346 | int cpu; | |
2347 | bool need_rcu_nocb_mask = true; | |
2348 | struct rcu_state *rsp; | |
2349 | ||
2350 | #ifdef CONFIG_RCU_NOCB_CPU_NONE | |
2351 | need_rcu_nocb_mask = false; | |
2352 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_NONE */ | |
2353 | ||
2354 | #if defined(CONFIG_NO_HZ_FULL) | |
2355 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2356 | need_rcu_nocb_mask = true; | |
2357 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2358 | ||
2359 | if (!have_rcu_nocb_mask && need_rcu_nocb_mask) { | |
949cccdb PK |
2360 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2361 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2362 | return; | |
2363 | } | |
f4579fc5 PM |
2364 | have_rcu_nocb_mask = true; |
2365 | } | |
2366 | if (!have_rcu_nocb_mask) | |
2367 | return; | |
2368 | ||
2369 | #ifdef CONFIG_RCU_NOCB_CPU_ZERO | |
2370 | pr_info("\tOffload RCU callbacks from CPU 0\n"); | |
2371 | cpumask_set_cpu(0, rcu_nocb_mask); | |
2372 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ZERO */ | |
2373 | #ifdef CONFIG_RCU_NOCB_CPU_ALL | |
2374 | pr_info("\tOffload RCU callbacks from all CPUs\n"); | |
2375 | cpumask_copy(rcu_nocb_mask, cpu_possible_mask); | |
2376 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ALL */ | |
2377 | #if defined(CONFIG_NO_HZ_FULL) | |
2378 | if (tick_nohz_full_running) | |
2379 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2380 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2381 | ||
2382 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
2383 | pr_info("\tNote: kernel parameter 'rcu_nocbs=' contains nonexistent CPUs.\n"); | |
2384 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, | |
2385 | rcu_nocb_mask); | |
2386 | } | |
ad853b48 TH |
2387 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", |
2388 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2389 | if (rcu_nocb_poll) |
2390 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2391 | ||
2392 | for_each_rcu_flavor(rsp) { | |
2393 | for_each_cpu(cpu, rcu_nocb_mask) { | |
2394 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
2395 | ||
2396 | /* | |
2397 | * If there are early callbacks, they will need | |
2398 | * to be moved to the nocb lists. | |
2399 | */ | |
2400 | WARN_ON_ONCE(rdp->nxttail[RCU_NEXT_TAIL] != | |
2401 | &rdp->nxtlist && | |
2402 | rdp->nxttail[RCU_NEXT_TAIL] != NULL); | |
2403 | init_nocb_callback_list(rdp); | |
2404 | } | |
35ce7f29 | 2405 | rcu_organize_nocb_kthreads(rsp); |
f4579fc5 | 2406 | } |
96d3fd0d PM |
2407 | } |
2408 | ||
3fbfbf7a PM |
2409 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2410 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2411 | { | |
2412 | rdp->nocb_tail = &rdp->nocb_head; | |
2413 | init_waitqueue_head(&rdp->nocb_wq); | |
fbce7497 | 2414 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
3fbfbf7a PM |
2415 | } |
2416 | ||
35ce7f29 PM |
2417 | /* |
2418 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2419 | * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are | |
2420 | * brought online out of order, this can require re-organizing the | |
2421 | * leader-follower relationships. | |
2422 | */ | |
2423 | static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu) | |
2424 | { | |
2425 | struct rcu_data *rdp; | |
2426 | struct rcu_data *rdp_last; | |
2427 | struct rcu_data *rdp_old_leader; | |
2428 | struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu); | |
2429 | struct task_struct *t; | |
2430 | ||
2431 | /* | |
2432 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2433 | * then nothing to do. | |
2434 | */ | |
2435 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2436 | return; | |
2437 | ||
2438 | /* If we didn't spawn the leader first, reorganize! */ | |
2439 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2440 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2441 | rdp_last = NULL; | |
2442 | rdp = rdp_old_leader; | |
2443 | do { | |
2444 | rdp->nocb_leader = rdp_spawn; | |
2445 | if (rdp_last && rdp != rdp_spawn) | |
2446 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2447 | if (rdp == rdp_spawn) { |
2448 | rdp = rdp->nocb_next_follower; | |
2449 | } else { | |
2450 | rdp_last = rdp; | |
2451 | rdp = rdp->nocb_next_follower; | |
2452 | rdp_last->nocb_next_follower = NULL; | |
2453 | } | |
35ce7f29 PM |
2454 | } while (rdp); |
2455 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2456 | } | |
2457 | ||
2458 | /* Spawn the kthread for this CPU and RCU flavor. */ | |
2459 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, | |
2460 | "rcuo%c/%d", rsp->abbr, cpu); | |
2461 | BUG_ON(IS_ERR(t)); | |
2462 | ACCESS_ONCE(rdp_spawn->nocb_kthread) = t; | |
2463 | } | |
2464 | ||
2465 | /* | |
2466 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2467 | * rcuo kthreads, spawn them. | |
2468 | */ | |
2469 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2470 | { | |
2471 | struct rcu_state *rsp; | |
2472 | ||
2473 | if (rcu_scheduler_fully_active) | |
2474 | for_each_rcu_flavor(rsp) | |
2475 | rcu_spawn_one_nocb_kthread(rsp, cpu); | |
2476 | } | |
2477 | ||
2478 | /* | |
2479 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2480 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2481 | * non-boot CPUs come online -- if this changes, we will need to add | |
2482 | * some mutual exclusion. | |
2483 | */ | |
2484 | static void __init rcu_spawn_nocb_kthreads(void) | |
2485 | { | |
2486 | int cpu; | |
2487 | ||
2488 | for_each_online_cpu(cpu) | |
2489 | rcu_spawn_all_nocb_kthreads(cpu); | |
2490 | } | |
2491 | ||
fbce7497 PM |
2492 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2493 | static int rcu_nocb_leader_stride = -1; | |
2494 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2495 | ||
2496 | /* | |
35ce7f29 | 2497 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2498 | */ |
35ce7f29 | 2499 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp) |
3fbfbf7a PM |
2500 | { |
2501 | int cpu; | |
fbce7497 PM |
2502 | int ls = rcu_nocb_leader_stride; |
2503 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2504 | struct rcu_data *rdp; |
fbce7497 PM |
2505 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2506 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2507 | |
22c2f669 | 2508 | if (!have_rcu_nocb_mask) |
3fbfbf7a | 2509 | return; |
fbce7497 PM |
2510 | if (ls == -1) { |
2511 | ls = int_sqrt(nr_cpu_ids); | |
2512 | rcu_nocb_leader_stride = ls; | |
2513 | } | |
2514 | ||
2515 | /* | |
2516 | * Each pass through this loop sets up one rcu_data structure and | |
2517 | * spawns one rcu_nocb_kthread(). | |
2518 | */ | |
3fbfbf7a PM |
2519 | for_each_cpu(cpu, rcu_nocb_mask) { |
2520 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
fbce7497 PM |
2521 | if (rdp->cpu >= nl) { |
2522 | /* New leader, set up for followers & next leader. */ | |
2523 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2524 | rdp->nocb_leader = rdp; | |
2525 | rdp_leader = rdp; | |
2526 | } else { | |
2527 | /* Another follower, link to previous leader. */ | |
2528 | rdp->nocb_leader = rdp_leader; | |
2529 | rdp_prev->nocb_next_follower = rdp; | |
2530 | } | |
2531 | rdp_prev = rdp; | |
3fbfbf7a PM |
2532 | } |
2533 | } | |
2534 | ||
2535 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2536 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2537 | { |
22c2f669 | 2538 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2539 | return false; |
22c2f669 | 2540 | |
3fbfbf7a | 2541 | rdp->nxttail[RCU_NEXT_TAIL] = NULL; |
34ed6246 | 2542 | return true; |
3fbfbf7a PM |
2543 | } |
2544 | ||
34ed6246 PM |
2545 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2546 | ||
d7e29933 PM |
2547 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) |
2548 | { | |
2549 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2550 | return false; | |
2551 | } | |
2552 | ||
0446be48 | 2553 | static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
3fbfbf7a | 2554 | { |
3fbfbf7a PM |
2555 | } |
2556 | ||
dae6e64d PM |
2557 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
2558 | { | |
2559 | } | |
2560 | ||
2561 | static void rcu_init_one_nocb(struct rcu_node *rnp) | |
2562 | { | |
2563 | } | |
3fbfbf7a | 2564 | |
3fbfbf7a | 2565 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2566 | bool lazy, unsigned long flags) |
3fbfbf7a | 2567 | { |
4afc7e26 | 2568 | return false; |
3fbfbf7a PM |
2569 | } |
2570 | ||
2571 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
96d3fd0d PM |
2572 | struct rcu_data *rdp, |
2573 | unsigned long flags) | |
3fbfbf7a | 2574 | { |
f4aa84ba | 2575 | return false; |
3fbfbf7a PM |
2576 | } |
2577 | ||
3fbfbf7a PM |
2578 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2579 | { | |
2580 | } | |
2581 | ||
9fdd3bc9 | 2582 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2583 | { |
2584 | return false; | |
2585 | } | |
2586 | ||
2587 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2588 | { | |
2589 | } | |
2590 | ||
35ce7f29 PM |
2591 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2592 | { | |
2593 | } | |
2594 | ||
2595 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2596 | { |
2597 | } | |
2598 | ||
34ed6246 | 2599 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2600 | { |
34ed6246 | 2601 | return false; |
3fbfbf7a PM |
2602 | } |
2603 | ||
2604 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2605 | |
2606 | /* | |
2607 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2608 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2609 | * off. RCU will be paying attention to this CPU because it is in the | |
2610 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2611 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2612 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2613 | * period and has not be idle from an RCU perspective, kick it. | |
2614 | */ | |
4a81e832 | 2615 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu) |
65d798f0 PM |
2616 | { |
2617 | #ifdef CONFIG_NO_HZ_FULL | |
2618 | if (tick_nohz_full_cpu(cpu)) | |
2619 | smp_send_reschedule(cpu); | |
2620 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2621 | } | |
2333210b PM |
2622 | |
2623 | ||
2624 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
2625 | ||
0edd1b17 | 2626 | static int full_sysidle_state; /* Current system-idle state. */ |
d4bd54fb PM |
2627 | #define RCU_SYSIDLE_NOT 0 /* Some CPU is not idle. */ |
2628 | #define RCU_SYSIDLE_SHORT 1 /* All CPUs idle for brief period. */ | |
2629 | #define RCU_SYSIDLE_LONG 2 /* All CPUs idle for long enough. */ | |
2630 | #define RCU_SYSIDLE_FULL 3 /* All CPUs idle, ready for sysidle. */ | |
2631 | #define RCU_SYSIDLE_FULL_NOTED 4 /* Actually entered sysidle state. */ | |
2632 | ||
eb348b89 PM |
2633 | /* |
2634 | * Invoked to note exit from irq or task transition to idle. Note that | |
2635 | * usermode execution does -not- count as idle here! After all, we want | |
2636 | * to detect full-system idle states, not RCU quiescent states and grace | |
2637 | * periods. The caller must have disabled interrupts. | |
2638 | */ | |
28ced795 | 2639 | static void rcu_sysidle_enter(int irq) |
eb348b89 PM |
2640 | { |
2641 | unsigned long j; | |
28ced795 | 2642 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
eb348b89 | 2643 | |
663e1310 PM |
2644 | /* If there are no nohz_full= CPUs, no need to track this. */ |
2645 | if (!tick_nohz_full_enabled()) | |
2646 | return; | |
2647 | ||
eb348b89 PM |
2648 | /* Adjust nesting, check for fully idle. */ |
2649 | if (irq) { | |
2650 | rdtp->dynticks_idle_nesting--; | |
2651 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting < 0); | |
2652 | if (rdtp->dynticks_idle_nesting != 0) | |
2653 | return; /* Still not fully idle. */ | |
2654 | } else { | |
2655 | if ((rdtp->dynticks_idle_nesting & DYNTICK_TASK_NEST_MASK) == | |
2656 | DYNTICK_TASK_NEST_VALUE) { | |
2657 | rdtp->dynticks_idle_nesting = 0; | |
2658 | } else { | |
2659 | rdtp->dynticks_idle_nesting -= DYNTICK_TASK_NEST_VALUE; | |
2660 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting < 0); | |
2661 | return; /* Still not fully idle. */ | |
2662 | } | |
2663 | } | |
2664 | ||
2665 | /* Record start of fully idle period. */ | |
2666 | j = jiffies; | |
2667 | ACCESS_ONCE(rdtp->dynticks_idle_jiffies) = j; | |
4e857c58 | 2668 | smp_mb__before_atomic(); |
eb348b89 | 2669 | atomic_inc(&rdtp->dynticks_idle); |
4e857c58 | 2670 | smp_mb__after_atomic(); |
eb348b89 PM |
2671 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks_idle) & 0x1); |
2672 | } | |
2673 | ||
0edd1b17 PM |
2674 | /* |
2675 | * Unconditionally force exit from full system-idle state. This is | |
2676 | * invoked when a normal CPU exits idle, but must be called separately | |
2677 | * for the timekeeping CPU (tick_do_timer_cpu). The reason for this | |
2678 | * is that the timekeeping CPU is permitted to take scheduling-clock | |
2679 | * interrupts while the system is in system-idle state, and of course | |
2680 | * rcu_sysidle_exit() has no way of distinguishing a scheduling-clock | |
2681 | * interrupt from any other type of interrupt. | |
2682 | */ | |
2683 | void rcu_sysidle_force_exit(void) | |
2684 | { | |
2685 | int oldstate = ACCESS_ONCE(full_sysidle_state); | |
2686 | int newoldstate; | |
2687 | ||
2688 | /* | |
2689 | * Each pass through the following loop attempts to exit full | |
2690 | * system-idle state. If contention proves to be a problem, | |
2691 | * a trylock-based contention tree could be used here. | |
2692 | */ | |
2693 | while (oldstate > RCU_SYSIDLE_SHORT) { | |
2694 | newoldstate = cmpxchg(&full_sysidle_state, | |
2695 | oldstate, RCU_SYSIDLE_NOT); | |
2696 | if (oldstate == newoldstate && | |
2697 | oldstate == RCU_SYSIDLE_FULL_NOTED) { | |
2698 | rcu_kick_nohz_cpu(tick_do_timer_cpu); | |
2699 | return; /* We cleared it, done! */ | |
2700 | } | |
2701 | oldstate = newoldstate; | |
2702 | } | |
2703 | smp_mb(); /* Order initial oldstate fetch vs. later non-idle work. */ | |
2704 | } | |
2705 | ||
eb348b89 PM |
2706 | /* |
2707 | * Invoked to note entry to irq or task transition from idle. Note that | |
2708 | * usermode execution does -not- count as idle here! The caller must | |
2709 | * have disabled interrupts. | |
2710 | */ | |
28ced795 | 2711 | static void rcu_sysidle_exit(int irq) |
eb348b89 | 2712 | { |
28ced795 CL |
2713 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
2714 | ||
663e1310 PM |
2715 | /* If there are no nohz_full= CPUs, no need to track this. */ |
2716 | if (!tick_nohz_full_enabled()) | |
2717 | return; | |
2718 | ||
eb348b89 PM |
2719 | /* Adjust nesting, check for already non-idle. */ |
2720 | if (irq) { | |
2721 | rdtp->dynticks_idle_nesting++; | |
2722 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting <= 0); | |
2723 | if (rdtp->dynticks_idle_nesting != 1) | |
2724 | return; /* Already non-idle. */ | |
2725 | } else { | |
2726 | /* | |
2727 | * Allow for irq misnesting. Yes, it really is possible | |
2728 | * to enter an irq handler then never leave it, and maybe | |
2729 | * also vice versa. Handle both possibilities. | |
2730 | */ | |
2731 | if (rdtp->dynticks_idle_nesting & DYNTICK_TASK_NEST_MASK) { | |
2732 | rdtp->dynticks_idle_nesting += DYNTICK_TASK_NEST_VALUE; | |
2733 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting <= 0); | |
2734 | return; /* Already non-idle. */ | |
2735 | } else { | |
2736 | rdtp->dynticks_idle_nesting = DYNTICK_TASK_EXIT_IDLE; | |
2737 | } | |
2738 | } | |
2739 | ||
2740 | /* Record end of idle period. */ | |
4e857c58 | 2741 | smp_mb__before_atomic(); |
eb348b89 | 2742 | atomic_inc(&rdtp->dynticks_idle); |
4e857c58 | 2743 | smp_mb__after_atomic(); |
eb348b89 | 2744 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks_idle) & 0x1)); |
0edd1b17 PM |
2745 | |
2746 | /* | |
2747 | * If we are the timekeeping CPU, we are permitted to be non-idle | |
2748 | * during a system-idle state. This must be the case, because | |
2749 | * the timekeeping CPU has to take scheduling-clock interrupts | |
2750 | * during the time that the system is transitioning to full | |
2751 | * system-idle state. This means that the timekeeping CPU must | |
2752 | * invoke rcu_sysidle_force_exit() directly if it does anything | |
2753 | * more than take a scheduling-clock interrupt. | |
2754 | */ | |
2755 | if (smp_processor_id() == tick_do_timer_cpu) | |
2756 | return; | |
2757 | ||
2758 | /* Update system-idle state: We are clearly no longer fully idle! */ | |
2759 | rcu_sysidle_force_exit(); | |
2760 | } | |
2761 | ||
2762 | /* | |
2763 | * Check to see if the current CPU is idle. Note that usermode execution | |
2764 | * does not count as idle. The caller must have disabled interrupts. | |
2765 | */ | |
2766 | static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, | |
2767 | unsigned long *maxj) | |
2768 | { | |
2769 | int cur; | |
2770 | unsigned long j; | |
2771 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
2772 | ||
663e1310 PM |
2773 | /* If there are no nohz_full= CPUs, don't check system-wide idleness. */ |
2774 | if (!tick_nohz_full_enabled()) | |
2775 | return; | |
2776 | ||
0edd1b17 PM |
2777 | /* |
2778 | * If some other CPU has already reported non-idle, if this is | |
2779 | * not the flavor of RCU that tracks sysidle state, or if this | |
2780 | * is an offline or the timekeeping CPU, nothing to do. | |
2781 | */ | |
417e8d26 | 2782 | if (!*isidle || rdp->rsp != rcu_state_p || |
0edd1b17 PM |
2783 | cpu_is_offline(rdp->cpu) || rdp->cpu == tick_do_timer_cpu) |
2784 | return; | |
eb75767b PM |
2785 | if (rcu_gp_in_progress(rdp->rsp)) |
2786 | WARN_ON_ONCE(smp_processor_id() != tick_do_timer_cpu); | |
0edd1b17 PM |
2787 | |
2788 | /* Pick up current idle and NMI-nesting counter and check. */ | |
2789 | cur = atomic_read(&rdtp->dynticks_idle); | |
2790 | if (cur & 0x1) { | |
2791 | *isidle = false; /* We are not idle! */ | |
2792 | return; | |
2793 | } | |
2794 | smp_mb(); /* Read counters before timestamps. */ | |
2795 | ||
2796 | /* Pick up timestamps. */ | |
2797 | j = ACCESS_ONCE(rdtp->dynticks_idle_jiffies); | |
2798 | /* If this CPU entered idle more recently, update maxj timestamp. */ | |
2799 | if (ULONG_CMP_LT(*maxj, j)) | |
2800 | *maxj = j; | |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * Is this the flavor of RCU that is handling full-system idle? | |
2805 | */ | |
2806 | static bool is_sysidle_rcu_state(struct rcu_state *rsp) | |
2807 | { | |
417e8d26 | 2808 | return rsp == rcu_state_p; |
0edd1b17 PM |
2809 | } |
2810 | ||
2811 | /* | |
2812 | * Return a delay in jiffies based on the number of CPUs, rcu_node | |
2813 | * leaf fanout, and jiffies tick rate. The idea is to allow larger | |
2814 | * systems more time to transition to full-idle state in order to | |
2815 | * avoid the cache thrashing that otherwise occur on the state variable. | |
2816 | * Really small systems (less than a couple of tens of CPUs) should | |
2817 | * instead use a single global atomically incremented counter, and later | |
2818 | * versions of this will automatically reconfigure themselves accordingly. | |
2819 | */ | |
2820 | static unsigned long rcu_sysidle_delay(void) | |
2821 | { | |
2822 | if (nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) | |
2823 | return 0; | |
2824 | return DIV_ROUND_UP(nr_cpu_ids * HZ, rcu_fanout_leaf * 1000); | |
2825 | } | |
2826 | ||
2827 | /* | |
2828 | * Advance the full-system-idle state. This is invoked when all of | |
2829 | * the non-timekeeping CPUs are idle. | |
2830 | */ | |
2831 | static void rcu_sysidle(unsigned long j) | |
2832 | { | |
2833 | /* Check the current state. */ | |
2834 | switch (ACCESS_ONCE(full_sysidle_state)) { | |
2835 | case RCU_SYSIDLE_NOT: | |
2836 | ||
2837 | /* First time all are idle, so note a short idle period. */ | |
2838 | ACCESS_ONCE(full_sysidle_state) = RCU_SYSIDLE_SHORT; | |
2839 | break; | |
2840 | ||
2841 | case RCU_SYSIDLE_SHORT: | |
2842 | ||
2843 | /* | |
2844 | * Idle for a bit, time to advance to next state? | |
2845 | * cmpxchg failure means race with non-idle, let them win. | |
2846 | */ | |
2847 | if (ULONG_CMP_GE(jiffies, j + rcu_sysidle_delay())) | |
2848 | (void)cmpxchg(&full_sysidle_state, | |
2849 | RCU_SYSIDLE_SHORT, RCU_SYSIDLE_LONG); | |
2850 | break; | |
2851 | ||
2852 | case RCU_SYSIDLE_LONG: | |
2853 | ||
2854 | /* | |
2855 | * Do an additional check pass before advancing to full. | |
2856 | * cmpxchg failure means race with non-idle, let them win. | |
2857 | */ | |
2858 | if (ULONG_CMP_GE(jiffies, j + rcu_sysidle_delay())) | |
2859 | (void)cmpxchg(&full_sysidle_state, | |
2860 | RCU_SYSIDLE_LONG, RCU_SYSIDLE_FULL); | |
2861 | break; | |
2862 | ||
2863 | default: | |
2864 | break; | |
2865 | } | |
2866 | } | |
2867 | ||
2868 | /* | |
2869 | * Found a non-idle non-timekeeping CPU, so kick the system-idle state | |
2870 | * back to the beginning. | |
2871 | */ | |
2872 | static void rcu_sysidle_cancel(void) | |
2873 | { | |
2874 | smp_mb(); | |
becb41bf PM |
2875 | if (full_sysidle_state > RCU_SYSIDLE_SHORT) |
2876 | ACCESS_ONCE(full_sysidle_state) = RCU_SYSIDLE_NOT; | |
0edd1b17 PM |
2877 | } |
2878 | ||
2879 | /* | |
2880 | * Update the sysidle state based on the results of a force-quiescent-state | |
2881 | * scan of the CPUs' dyntick-idle state. | |
2882 | */ | |
2883 | static void rcu_sysidle_report(struct rcu_state *rsp, int isidle, | |
2884 | unsigned long maxj, bool gpkt) | |
2885 | { | |
417e8d26 | 2886 | if (rsp != rcu_state_p) |
0edd1b17 PM |
2887 | return; /* Wrong flavor, ignore. */ |
2888 | if (gpkt && nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) | |
2889 | return; /* Running state machine from timekeeping CPU. */ | |
2890 | if (isidle) | |
2891 | rcu_sysidle(maxj); /* More idle! */ | |
2892 | else | |
2893 | rcu_sysidle_cancel(); /* Idle is over. */ | |
2894 | } | |
2895 | ||
2896 | /* | |
2897 | * Wrapper for rcu_sysidle_report() when called from the grace-period | |
2898 | * kthread's context. | |
2899 | */ | |
2900 | static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, | |
2901 | unsigned long maxj) | |
2902 | { | |
663e1310 PM |
2903 | /* If there are no nohz_full= CPUs, no need to track this. */ |
2904 | if (!tick_nohz_full_enabled()) | |
2905 | return; | |
2906 | ||
0edd1b17 PM |
2907 | rcu_sysidle_report(rsp, isidle, maxj, true); |
2908 | } | |
2909 | ||
2910 | /* Callback and function for forcing an RCU grace period. */ | |
2911 | struct rcu_sysidle_head { | |
2912 | struct rcu_head rh; | |
2913 | int inuse; | |
2914 | }; | |
2915 | ||
2916 | static void rcu_sysidle_cb(struct rcu_head *rhp) | |
2917 | { | |
2918 | struct rcu_sysidle_head *rshp; | |
2919 | ||
2920 | /* | |
2921 | * The following memory barrier is needed to replace the | |
2922 | * memory barriers that would normally be in the memory | |
2923 | * allocator. | |
2924 | */ | |
2925 | smp_mb(); /* grace period precedes setting inuse. */ | |
2926 | ||
2927 | rshp = container_of(rhp, struct rcu_sysidle_head, rh); | |
2928 | ACCESS_ONCE(rshp->inuse) = 0; | |
2929 | } | |
2930 | ||
2931 | /* | |
2932 | * Check to see if the system is fully idle, other than the timekeeping CPU. | |
663e1310 PM |
2933 | * The caller must have disabled interrupts. This is not intended to be |
2934 | * called unless tick_nohz_full_enabled(). | |
0edd1b17 PM |
2935 | */ |
2936 | bool rcu_sys_is_idle(void) | |
2937 | { | |
2938 | static struct rcu_sysidle_head rsh; | |
2939 | int rss = ACCESS_ONCE(full_sysidle_state); | |
2940 | ||
2941 | if (WARN_ON_ONCE(smp_processor_id() != tick_do_timer_cpu)) | |
2942 | return false; | |
2943 | ||
2944 | /* Handle small-system case by doing a full scan of CPUs. */ | |
2945 | if (nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) { | |
2946 | int oldrss = rss - 1; | |
2947 | ||
2948 | /* | |
2949 | * One pass to advance to each state up to _FULL. | |
2950 | * Give up if any pass fails to advance the state. | |
2951 | */ | |
2952 | while (rss < RCU_SYSIDLE_FULL && oldrss < rss) { | |
2953 | int cpu; | |
2954 | bool isidle = true; | |
2955 | unsigned long maxj = jiffies - ULONG_MAX / 4; | |
2956 | struct rcu_data *rdp; | |
2957 | ||
2958 | /* Scan all the CPUs looking for nonidle CPUs. */ | |
2959 | for_each_possible_cpu(cpu) { | |
417e8d26 | 2960 | rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
0edd1b17 PM |
2961 | rcu_sysidle_check_cpu(rdp, &isidle, &maxj); |
2962 | if (!isidle) | |
2963 | break; | |
2964 | } | |
417e8d26 | 2965 | rcu_sysidle_report(rcu_state_p, isidle, maxj, false); |
0edd1b17 PM |
2966 | oldrss = rss; |
2967 | rss = ACCESS_ONCE(full_sysidle_state); | |
2968 | } | |
2969 | } | |
2970 | ||
2971 | /* If this is the first observation of an idle period, record it. */ | |
2972 | if (rss == RCU_SYSIDLE_FULL) { | |
2973 | rss = cmpxchg(&full_sysidle_state, | |
2974 | RCU_SYSIDLE_FULL, RCU_SYSIDLE_FULL_NOTED); | |
2975 | return rss == RCU_SYSIDLE_FULL; | |
2976 | } | |
2977 | ||
2978 | smp_mb(); /* ensure rss load happens before later caller actions. */ | |
2979 | ||
2980 | /* If already fully idle, tell the caller (in case of races). */ | |
2981 | if (rss == RCU_SYSIDLE_FULL_NOTED) | |
2982 | return true; | |
2983 | ||
2984 | /* | |
2985 | * If we aren't there yet, and a grace period is not in flight, | |
2986 | * initiate a grace period. Either way, tell the caller that | |
2987 | * we are not there yet. We use an xchg() rather than an assignment | |
2988 | * to make up for the memory barriers that would otherwise be | |
2989 | * provided by the memory allocator. | |
2990 | */ | |
2991 | if (nr_cpu_ids > CONFIG_NO_HZ_FULL_SYSIDLE_SMALL && | |
417e8d26 | 2992 | !rcu_gp_in_progress(rcu_state_p) && |
0edd1b17 PM |
2993 | !rsh.inuse && xchg(&rsh.inuse, 1) == 0) |
2994 | call_rcu(&rsh.rh, rcu_sysidle_cb); | |
2995 | return false; | |
eb348b89 PM |
2996 | } |
2997 | ||
2333210b PM |
2998 | /* |
2999 | * Initialize dynticks sysidle state for CPUs coming online. | |
3000 | */ | |
3001 | static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp) | |
3002 | { | |
3003 | rdtp->dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE; | |
3004 | } | |
3005 | ||
3006 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
3007 | ||
28ced795 | 3008 | static void rcu_sysidle_enter(int irq) |
eb348b89 PM |
3009 | { |
3010 | } | |
3011 | ||
28ced795 | 3012 | static void rcu_sysidle_exit(int irq) |
eb348b89 PM |
3013 | { |
3014 | } | |
3015 | ||
0edd1b17 PM |
3016 | static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, |
3017 | unsigned long *maxj) | |
3018 | { | |
3019 | } | |
3020 | ||
3021 | static bool is_sysidle_rcu_state(struct rcu_state *rsp) | |
3022 | { | |
3023 | return false; | |
3024 | } | |
3025 | ||
3026 | static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, | |
3027 | unsigned long maxj) | |
3028 | { | |
3029 | } | |
3030 | ||
2333210b PM |
3031 | static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp) |
3032 | { | |
3033 | } | |
3034 | ||
3035 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
a096932f PM |
3036 | |
3037 | /* | |
3038 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
3039 | * grace-period kthread will do force_quiescent_state() processing? | |
3040 | * The idea is to avoid waking up RCU core processing on such a | |
3041 | * CPU unless the grace period has extended for too long. | |
3042 | * | |
3043 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 3044 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f PM |
3045 | */ |
3046 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp) | |
3047 | { | |
3048 | #ifdef CONFIG_NO_HZ_FULL | |
3049 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
3050 | (!rcu_gp_in_progress(rsp) || | |
3051 | ULONG_CMP_LT(jiffies, ACCESS_ONCE(rsp->gp_start) + HZ))) | |
3052 | return 1; | |
3053 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
3054 | return 0; | |
3055 | } | |
5057f55e PM |
3056 | |
3057 | /* | |
3058 | * Bind the grace-period kthread for the sysidle flavor of RCU to the | |
3059 | * timekeeping CPU. | |
3060 | */ | |
3061 | static void rcu_bind_gp_kthread(void) | |
3062 | { | |
c0f489d2 | 3063 | int __maybe_unused cpu; |
5057f55e | 3064 | |
c0f489d2 | 3065 | if (!tick_nohz_full_enabled()) |
5057f55e | 3066 | return; |
c0f489d2 PM |
3067 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE |
3068 | cpu = tick_do_timer_cpu; | |
3069 | if (cpu >= 0 && cpu < nr_cpu_ids && raw_smp_processor_id() != cpu) | |
5057f55e | 3070 | set_cpus_allowed_ptr(current, cpumask_of(cpu)); |
c0f489d2 PM |
3071 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ |
3072 | if (!is_housekeeping_cpu(raw_smp_processor_id())) | |
3073 | housekeeping_affine(current); | |
3074 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
5057f55e | 3075 | } |
176f8f7a PM |
3076 | |
3077 | /* Record the current task on dyntick-idle entry. */ | |
3078 | static void rcu_dynticks_task_enter(void) | |
3079 | { | |
3080 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
3081 | ACCESS_ONCE(current->rcu_tasks_idle_cpu) = smp_processor_id(); | |
3082 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ | |
3083 | } | |
3084 | ||
3085 | /* Record no current task on dyntick-idle exit. */ | |
3086 | static void rcu_dynticks_task_exit(void) | |
3087 | { | |
3088 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
3089 | ACCESS_ONCE(current->rcu_tasks_idle_cpu) = -1; | |
3090 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ | |
3091 | } |