Commit | Line | Data |
---|---|---|
22e40925 | 1 | /* SPDX-License-Identifier: GPL-2.0+ */ |
f41d911f PM |
2 | /* |
3 | * Read-Copy Update mechanism for mutual exclusion (tree-based version) | |
4 | * Internal non-public definitions that provide either classic | |
6cc68793 | 5 | * or preemptible semantics. |
f41d911f | 6 | * |
f41d911f PM |
7 | * Copyright Red Hat, 2009 |
8 | * Copyright IBM Corporation, 2009 | |
9 | * | |
10 | * Author: Ingo Molnar <mingo@elte.hu> | |
22e40925 | 11 | * Paul E. McKenney <paulmck@linux.ibm.com> |
f41d911f PM |
12 | */ |
13 | ||
abaa93d9 | 14 | #include "../locking/rtmutex_common.h" |
5b61b0ba | 15 | |
3fbfbf7a PM |
16 | #ifdef CONFIG_RCU_NOCB_CPU |
17 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
1b0048a4 | 18 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
19 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
20 | ||
26845c28 PM |
21 | /* |
22 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 23 | * messages about anything out of the ordinary. |
26845c28 PM |
24 | */ |
25 | static void __init rcu_bootup_announce_oddness(void) | |
26 | { | |
ab6f5bd6 | 27 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 28 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
29 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
30 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
a7538352 JP |
31 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n", |
32 | RCU_FANOUT); | |
7fa27001 | 33 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
34 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
35 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
36 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 37 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 38 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
39 | if (RCU_NUM_LVLS >= 4) |
40 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 41 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 42 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
43 | RCU_FANOUT_LEAF); |
44 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
a7538352 JP |
45 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", |
46 | rcu_fanout_leaf); | |
cca6f393 | 47 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 48 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b | 49 | #ifdef CONFIG_RCU_BOOST |
a7538352 JP |
50 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", |
51 | kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
17c7798b PM |
52 | #endif |
53 | if (blimit != DEFAULT_RCU_BLIMIT) | |
54 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
55 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
56 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
57 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
58 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
59 | if (jiffies_till_first_fqs != ULONG_MAX) | |
60 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
61 | if (jiffies_till_next_fqs != ULONG_MAX) | |
62 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
c06aed0e PM |
63 | if (jiffies_till_sched_qs != ULONG_MAX) |
64 | pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs); | |
17c7798b PM |
65 | if (rcu_kick_kthreads) |
66 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
67 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
68 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 69 | if (gp_preinit_delay) |
17c7798b | 70 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 71 | if (gp_init_delay) |
17c7798b | 72 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 73 | if (gp_cleanup_delay) |
17c7798b | 74 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
48d07c04 SAS |
75 | if (!use_softirq) |
76 | pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n"); | |
17c7798b PM |
77 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) |
78 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 79 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
80 | } |
81 | ||
28f6569a | 82 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 83 | |
63d4c8c9 | 84 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 85 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 86 | |
f41d911f PM |
87 | /* |
88 | * Tell them what RCU they are running. | |
89 | */ | |
0e0fc1c2 | 90 | static void __init rcu_bootup_announce(void) |
f41d911f | 91 | { |
efc151c3 | 92 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 93 | rcu_bootup_announce_oddness(); |
f41d911f PM |
94 | } |
95 | ||
8203d6d0 PM |
96 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
97 | #define RCU_GP_TASKS 0x8 | |
98 | #define RCU_EXP_TASKS 0x4 | |
99 | #define RCU_GP_BLKD 0x2 | |
100 | #define RCU_EXP_BLKD 0x1 | |
101 | ||
102 | /* | |
103 | * Queues a task preempted within an RCU-preempt read-side critical | |
104 | * section into the appropriate location within the ->blkd_tasks list, | |
105 | * depending on the states of any ongoing normal and expedited grace | |
106 | * periods. The ->gp_tasks pointer indicates which element the normal | |
107 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
108 | * indicates which element the expedited grace period is waiting on (again, | |
109 | * NULL if none). If a grace period is waiting on a given element in the | |
110 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
111 | * adding a task to the tail of the list blocks any grace period that is | |
112 | * already waiting on one of the elements. In contrast, adding a task | |
113 | * to the head of the list won't block any grace period that is already | |
114 | * waiting on one of the elements. | |
115 | * | |
116 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
117 | * period wait for a task that is not strictly speaking blocking it. | |
118 | * Given the choice, we needlessly block a normal grace period rather than | |
119 | * blocking an expedited grace period. | |
120 | * | |
121 | * Note that an endless sequence of expedited grace periods still cannot | |
122 | * indefinitely postpone a normal grace period. Eventually, all of the | |
123 | * fixed number of preempted tasks blocking the normal grace period that are | |
124 | * not also blocking the expedited grace period will resume and complete | |
125 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
126 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
127 | * the corresponding expedited grace period will also be the end of the | |
128 | * normal grace period. | |
129 | */ | |
46a5d164 PM |
130 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
131 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
132 | { |
133 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
134 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
135 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
136 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
137 | struct task_struct *t = current; | |
138 | ||
a32e01ee | 139 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 140 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 141 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
142 | /* RCU better not be waiting on newly onlined CPUs! */ |
143 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
144 | rdp->grpmask); | |
ea9b0c8a | 145 | |
8203d6d0 PM |
146 | /* |
147 | * Decide where to queue the newly blocked task. In theory, | |
148 | * this could be an if-statement. In practice, when I tried | |
149 | * that, it was quite messy. | |
150 | */ | |
151 | switch (blkd_state) { | |
152 | case 0: | |
153 | case RCU_EXP_TASKS: | |
154 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
155 | case RCU_GP_TASKS: | |
156 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
157 | ||
158 | /* | |
159 | * Blocking neither GP, or first task blocking the normal | |
160 | * GP but not blocking the already-waiting expedited GP. | |
161 | * Queue at the head of the list to avoid unnecessarily | |
162 | * blocking the already-waiting GPs. | |
163 | */ | |
164 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
165 | break; | |
166 | ||
167 | case RCU_EXP_BLKD: | |
168 | case RCU_GP_BLKD: | |
169 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
170 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
171 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
172 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
173 | ||
174 | /* | |
175 | * First task arriving that blocks either GP, or first task | |
176 | * arriving that blocks the expedited GP (with the normal | |
177 | * GP already waiting), or a task arriving that blocks | |
178 | * both GPs with both GPs already waiting. Queue at the | |
179 | * tail of the list to avoid any GP waiting on any of the | |
180 | * already queued tasks that are not blocking it. | |
181 | */ | |
182 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
183 | break; | |
184 | ||
185 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
186 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
187 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
188 | ||
189 | /* | |
190 | * Second or subsequent task blocking the expedited GP. | |
191 | * The task either does not block the normal GP, or is the | |
192 | * first task blocking the normal GP. Queue just after | |
193 | * the first task blocking the expedited GP. | |
194 | */ | |
195 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
196 | break; | |
197 | ||
198 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
199 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
200 | ||
201 | /* | |
202 | * Second or subsequent task blocking the normal GP. | |
203 | * The task does not block the expedited GP. Queue just | |
204 | * after the first task blocking the normal GP. | |
205 | */ | |
206 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
207 | break; | |
208 | ||
209 | default: | |
210 | ||
211 | /* Yet another exercise in excessive paranoia. */ | |
212 | WARN_ON_ONCE(1); | |
213 | break; | |
214 | } | |
215 | ||
216 | /* | |
217 | * We have now queued the task. If it was the first one to | |
218 | * block either grace period, update the ->gp_tasks and/or | |
219 | * ->exp_tasks pointers, respectively, to reference the newly | |
220 | * blocked tasks. | |
221 | */ | |
4bc8d555 | 222 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
8203d6d0 | 223 | rnp->gp_tasks = &t->rcu_node_entry; |
d43a5d32 | 224 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 225 | } |
8203d6d0 PM |
226 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
227 | rnp->exp_tasks = &t->rcu_node_entry; | |
2dee9404 PM |
228 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
229 | !(rnp->qsmask & rdp->grpmask)); | |
230 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
231 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 232 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
233 | |
234 | /* | |
235 | * Report the quiescent state for the expedited GP. This expedited | |
236 | * GP should not be able to end until we report, so there should be | |
237 | * no need to check for a subsequent expedited GP. (Though we are | |
238 | * still in a quiescent state in any case.) | |
239 | */ | |
1bb33644 | 240 | if (blkd_state & RCU_EXP_BLKD && rdp->exp_deferred_qs) |
63d4c8c9 | 241 | rcu_report_exp_rdp(rdp); |
fcc878e4 | 242 | else |
1bb33644 | 243 | WARN_ON_ONCE(rdp->exp_deferred_qs); |
8203d6d0 PM |
244 | } |
245 | ||
f41d911f | 246 | /* |
c7037ff5 PM |
247 | * Record a preemptible-RCU quiescent state for the specified CPU. |
248 | * Note that this does not necessarily mean that the task currently running | |
249 | * on the CPU is in a quiescent state: Instead, it means that the current | |
250 | * grace period need not wait on any RCU read-side critical section that | |
251 | * starts later on this CPU. It also means that if the current task is | |
252 | * in an RCU read-side critical section, it has already added itself to | |
253 | * some leaf rcu_node structure's ->blkd_tasks list. In addition to the | |
254 | * current task, there might be any number of other tasks blocked while | |
255 | * in an RCU read-side critical section. | |
25502a6c | 256 | * |
c7037ff5 | 257 | * Callers to this function must disable preemption. |
f41d911f | 258 | */ |
45975c7d | 259 | static void rcu_qs(void) |
f41d911f | 260 | { |
45975c7d | 261 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n"); |
2280ee5a | 262 | if (__this_cpu_read(rcu_data.cpu_no_qs.s)) { |
284a8c93 | 263 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2280ee5a | 264 | __this_cpu_read(rcu_data.gp_seq), |
284a8c93 | 265 | TPS("cpuqs")); |
2280ee5a | 266 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); |
c98cac60 | 267 | barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */ |
add0d37b | 268 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false); |
284a8c93 | 269 | } |
f41d911f PM |
270 | } |
271 | ||
272 | /* | |
c3422bea PM |
273 | * We have entered the scheduler, and the current task might soon be |
274 | * context-switched away from. If this task is in an RCU read-side | |
275 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
276 | * record that fact, so we enqueue the task on the blkd_tasks list. |
277 | * The task will dequeue itself when it exits the outermost enclosing | |
278 | * RCU read-side critical section. Therefore, the current grace period | |
279 | * cannot be permitted to complete until the blkd_tasks list entries | |
280 | * predating the current grace period drain, in other words, until | |
281 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 282 | * |
46a5d164 | 283 | * Caller must disable interrupts. |
f41d911f | 284 | */ |
45975c7d | 285 | void rcu_note_context_switch(bool preempt) |
f41d911f PM |
286 | { |
287 | struct task_struct *t = current; | |
da1df50d | 288 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
f41d911f PM |
289 | struct rcu_node *rnp; |
290 | ||
45975c7d | 291 | trace_rcu_utilization(TPS("Start context switch")); |
b04db8e1 | 292 | lockdep_assert_irqs_disabled(); |
5b72f964 | 293 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 294 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 295 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
296 | |
297 | /* Possibly blocking in an RCU read-side critical section. */ | |
f41d911f | 298 | rnp = rdp->mynode; |
46a5d164 | 299 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 300 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 301 | t->rcu_blocked_node = rnp; |
f41d911f PM |
302 | |
303 | /* | |
8203d6d0 PM |
304 | * Verify the CPU's sanity, trace the preemption, and |
305 | * then queue the task as required based on the states | |
306 | * of any ongoing and expedited grace periods. | |
f41d911f | 307 | */ |
0aa04b05 | 308 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 309 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
88d1bead | 310 | trace_rcu_preempt_task(rcu_state.name, |
d4c08f2a PM |
311 | t->pid, |
312 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
313 | ? rnp->gp_seq |
314 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 315 | rcu_preempt_ctxt_queue(rnp, rdp); |
3e310098 PM |
316 | } else { |
317 | rcu_preempt_deferred_qs(t); | |
f41d911f PM |
318 | } |
319 | ||
320 | /* | |
321 | * Either we were not in an RCU read-side critical section to | |
322 | * begin with, or we have now recorded that critical section | |
323 | * globally. Either way, we can now note a quiescent state | |
324 | * for this CPU. Again, if we were in an RCU read-side critical | |
325 | * section, and if that critical section was blocking the current | |
326 | * grace period, then the fact that the task has been enqueued | |
327 | * means that we continue to block the current grace period. | |
328 | */ | |
45975c7d | 329 | rcu_qs(); |
1bb33644 | 330 | if (rdp->exp_deferred_qs) |
63d4c8c9 | 331 | rcu_report_exp_rdp(rdp); |
45975c7d | 332 | trace_rcu_utilization(TPS("End context switch")); |
f41d911f | 333 | } |
45975c7d | 334 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
f41d911f | 335 | |
fc2219d4 PM |
336 | /* |
337 | * Check for preempted RCU readers blocking the current grace period | |
338 | * for the specified rcu_node structure. If the caller needs a reliable | |
339 | * answer, it must hold the rcu_node's ->lock. | |
340 | */ | |
27f4d280 | 341 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 342 | { |
12f5f524 | 343 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
344 | } |
345 | ||
5f1a6ef3 PM |
346 | /* Bias and limit values for ->rcu_read_lock_nesting. */ |
347 | #define RCU_NEST_BIAS INT_MAX | |
348 | #define RCU_NEST_NMAX (-INT_MAX / 2) | |
349 | #define RCU_NEST_PMAX (INT_MAX / 2) | |
350 | ||
0e5da22e PM |
351 | /* |
352 | * Preemptible RCU implementation for rcu_read_lock(). | |
353 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
354 | * if we block. | |
355 | */ | |
356 | void __rcu_read_lock(void) | |
357 | { | |
358 | current->rcu_read_lock_nesting++; | |
5f1a6ef3 PM |
359 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
360 | WARN_ON_ONCE(current->rcu_read_lock_nesting > RCU_NEST_PMAX); | |
0e5da22e PM |
361 | barrier(); /* critical section after entry code. */ |
362 | } | |
363 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
364 | ||
365 | /* | |
366 | * Preemptible RCU implementation for rcu_read_unlock(). | |
367 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
368 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
369 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
370 | * in an RCU read-side critical section and other special cases. | |
371 | */ | |
372 | void __rcu_read_unlock(void) | |
373 | { | |
374 | struct task_struct *t = current; | |
375 | ||
376 | if (t->rcu_read_lock_nesting != 1) { | |
377 | --t->rcu_read_lock_nesting; | |
378 | } else { | |
379 | barrier(); /* critical section before exit code. */ | |
5f1a6ef3 | 380 | t->rcu_read_lock_nesting = -RCU_NEST_BIAS; |
0e5da22e PM |
381 | barrier(); /* assign before ->rcu_read_unlock_special load */ |
382 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) | |
383 | rcu_read_unlock_special(t); | |
384 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
385 | t->rcu_read_lock_nesting = 0; | |
386 | } | |
5f1a6ef3 PM |
387 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
388 | int rrln = t->rcu_read_lock_nesting; | |
0e5da22e | 389 | |
5f1a6ef3 | 390 | WARN_ON_ONCE(rrln < 0 && rrln > RCU_NEST_NMAX); |
0e5da22e | 391 | } |
0e5da22e PM |
392 | } |
393 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
394 | ||
12f5f524 PM |
395 | /* |
396 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
397 | * returning NULL if at the end of the list. | |
398 | */ | |
399 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
400 | struct rcu_node *rnp) | |
401 | { | |
402 | struct list_head *np; | |
403 | ||
404 | np = t->rcu_node_entry.next; | |
405 | if (np == &rnp->blkd_tasks) | |
406 | np = NULL; | |
407 | return np; | |
408 | } | |
409 | ||
8af3a5e7 PM |
410 | /* |
411 | * Return true if the specified rcu_node structure has tasks that were | |
412 | * preempted within an RCU read-side critical section. | |
413 | */ | |
414 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
415 | { | |
416 | return !list_empty(&rnp->blkd_tasks); | |
417 | } | |
418 | ||
b668c9cf | 419 | /* |
3e310098 PM |
420 | * Report deferred quiescent states. The deferral time can |
421 | * be quite short, for example, in the case of the call from | |
422 | * rcu_read_unlock_special(). | |
b668c9cf | 423 | */ |
3e310098 PM |
424 | static void |
425 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 426 | { |
b6a932d1 PM |
427 | bool empty_exp; |
428 | bool empty_norm; | |
429 | bool empty_exp_now; | |
12f5f524 | 430 | struct list_head *np; |
abaa93d9 | 431 | bool drop_boost_mutex = false; |
8203d6d0 | 432 | struct rcu_data *rdp; |
f41d911f | 433 | struct rcu_node *rnp; |
1d082fd0 | 434 | union rcu_special special; |
f41d911f | 435 | |
f41d911f | 436 | /* |
8203d6d0 PM |
437 | * If RCU core is waiting for this CPU to exit its critical section, |
438 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 439 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
440 | */ |
441 | special = t->rcu_read_unlock_special; | |
da1df50d | 442 | rdp = this_cpu_ptr(&rcu_data); |
1bb33644 | 443 | if (!special.s && !rdp->exp_deferred_qs) { |
3e310098 PM |
444 | local_irq_restore(flags); |
445 | return; | |
446 | } | |
23634ebc | 447 | t->rcu_read_unlock_special.b.deferred_qs = false; |
1d082fd0 | 448 | if (special.b.need_qs) { |
45975c7d | 449 | rcu_qs(); |
c0135d07 | 450 | t->rcu_read_unlock_special.b.need_qs = false; |
1bb33644 | 451 | if (!t->rcu_read_unlock_special.s && !rdp->exp_deferred_qs) { |
79a62f95 LJ |
452 | local_irq_restore(flags); |
453 | return; | |
454 | } | |
f41d911f PM |
455 | } |
456 | ||
8203d6d0 | 457 | /* |
3e310098 PM |
458 | * Respond to a request by an expedited grace period for a |
459 | * quiescent state from this CPU. Note that requests from | |
460 | * tasks are handled when removing the task from the | |
461 | * blocked-tasks list below. | |
8203d6d0 | 462 | */ |
1bb33644 | 463 | if (rdp->exp_deferred_qs) { |
63d4c8c9 | 464 | rcu_report_exp_rdp(rdp); |
8203d6d0 PM |
465 | if (!t->rcu_read_unlock_special.s) { |
466 | local_irq_restore(flags); | |
467 | return; | |
468 | } | |
469 | } | |
470 | ||
f41d911f | 471 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 PM |
472 | if (special.b.blocked) { |
473 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 474 | |
dd5d19ba | 475 | /* |
0a0ba1c9 | 476 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
477 | * now remains queued on the rcu_node corresponding to the |
478 | * CPU it first blocked on, so there is no longer any need | |
479 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 480 | */ |
8ba9153b PM |
481 | rnp = t->rcu_blocked_node; |
482 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
483 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 484 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 485 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 486 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 487 | (!empty_norm || rnp->qsmask)); |
8203d6d0 | 488 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 489 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 490 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 491 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 492 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 493 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 494 | rnp->gp_seq, t->pid); |
12f5f524 PM |
495 | if (&t->rcu_node_entry == rnp->gp_tasks) |
496 | rnp->gp_tasks = np; | |
497 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
498 | rnp->exp_tasks = np; | |
727b705b | 499 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
500 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
501 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
502 | if (&t->rcu_node_entry == rnp->boost_tasks) |
503 | rnp->boost_tasks = np; | |
727b705b | 504 | } |
f41d911f PM |
505 | |
506 | /* | |
507 | * If this was the last task on the current list, and if | |
508 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
509 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
510 | * so we must take a snapshot of the expedited state. | |
f41d911f | 511 | */ |
8203d6d0 | 512 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 513 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 514 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 515 | rnp->gp_seq, |
d4c08f2a PM |
516 | 0, rnp->qsmask, |
517 | rnp->level, | |
518 | rnp->grplo, | |
519 | rnp->grphi, | |
520 | !!rnp->gp_tasks); | |
139ad4da | 521 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 522 | } else { |
67c583a7 | 523 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 524 | } |
d9a3da06 | 525 | |
27f4d280 | 526 | /* Unboost if we were boosted. */ |
727b705b | 527 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 528 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 529 | |
d9a3da06 PM |
530 | /* |
531 | * If this was the last task on the expedited lists, | |
532 | * then we need to report up the rcu_node hierarchy. | |
533 | */ | |
389abd48 | 534 | if (!empty_exp && empty_exp_now) |
63d4c8c9 | 535 | rcu_report_exp_rnp(rnp, true); |
b668c9cf PM |
536 | } else { |
537 | local_irq_restore(flags); | |
f41d911f | 538 | } |
f41d911f PM |
539 | } |
540 | ||
3e310098 PM |
541 | /* |
542 | * Is a deferred quiescent-state pending, and are we also not in | |
543 | * an RCU read-side critical section? It is the caller's responsibility | |
544 | * to ensure it is otherwise safe to report any deferred quiescent | |
545 | * states. The reason for this is that it is safe to report a | |
546 | * quiescent state during context switch even though preemption | |
547 | * is disabled. This function cannot be expected to understand these | |
548 | * nuances, so the caller must handle them. | |
549 | */ | |
550 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
551 | { | |
1bb33644 | 552 | return (__this_cpu_read(rcu_data.exp_deferred_qs) || |
3e310098 | 553 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
27c744e3 | 554 | t->rcu_read_lock_nesting <= 0; |
3e310098 PM |
555 | } |
556 | ||
557 | /* | |
558 | * Report a deferred quiescent state if needed and safe to do so. | |
559 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
560 | * not being in an RCU read-side critical section. The caller must | |
561 | * evaluate safety in terms of interrupt, softirq, and preemption | |
562 | * disabling. | |
563 | */ | |
564 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
565 | { | |
566 | unsigned long flags; | |
567 | bool couldrecurse = t->rcu_read_lock_nesting >= 0; | |
568 | ||
569 | if (!rcu_preempt_need_deferred_qs(t)) | |
570 | return; | |
571 | if (couldrecurse) | |
5f1a6ef3 | 572 | t->rcu_read_lock_nesting -= RCU_NEST_BIAS; |
3e310098 PM |
573 | local_irq_save(flags); |
574 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
575 | if (couldrecurse) | |
5f1a6ef3 | 576 | t->rcu_read_lock_nesting += RCU_NEST_BIAS; |
3e310098 PM |
577 | } |
578 | ||
0864f057 PM |
579 | /* |
580 | * Minimal handler to give the scheduler a chance to re-evaluate. | |
581 | */ | |
582 | static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp) | |
583 | { | |
584 | struct rcu_data *rdp; | |
585 | ||
586 | rdp = container_of(iwp, struct rcu_data, defer_qs_iw); | |
587 | rdp->defer_qs_iw_pending = false; | |
588 | } | |
589 | ||
3e310098 PM |
590 | /* |
591 | * Handle special cases during rcu_read_unlock(), such as needing to | |
592 | * notify RCU core processing or task having blocked during the RCU | |
593 | * read-side critical section. | |
594 | */ | |
595 | static void rcu_read_unlock_special(struct task_struct *t) | |
596 | { | |
597 | unsigned long flags; | |
598 | bool preempt_bh_were_disabled = | |
599 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
600 | bool irqs_were_disabled; | |
601 | ||
602 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
603 | if (in_nmi()) | |
604 | return; | |
605 | ||
606 | local_irq_save(flags); | |
607 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 | 608 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
25102de6 PM |
609 | bool exp; |
610 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
611 | struct rcu_node *rnp = rdp->mynode; | |
612 | ||
23634ebc | 613 | t->rcu_read_unlock_special.b.exp_hint = false; |
25102de6 PM |
614 | exp = (t->rcu_blocked_node && t->rcu_blocked_node->exp_tasks) || |
615 | (rdp->grpmask & rnp->expmask) || | |
616 | tick_nohz_full_cpu(rdp->cpu); | |
23634ebc | 617 | // Need to defer quiescent state until everything is enabled. |
87446b48 PM |
618 | if (irqs_were_disabled && use_softirq && |
619 | (in_interrupt() || | |
620 | (exp && !t->rcu_read_unlock_special.b.deferred_qs))) { | |
23634ebc PM |
621 | // Using softirq, safe to awaken, and we get |
622 | // no help from enabling irqs, unlike bh/preempt. | |
05f41571 PM |
623 | raise_softirq_irqoff(RCU_SOFTIRQ); |
624 | } else { | |
23634ebc | 625 | // Enabling BH or preempt does reschedule, so... |
25102de6 | 626 | // Also if no expediting or NO_HZ_FULL, slow is OK. |
05f41571 PM |
627 | set_tsk_need_resched(current); |
628 | set_preempt_need_resched(); | |
d143b3d1 | 629 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && |
0864f057 PM |
630 | !rdp->defer_qs_iw_pending && exp) { |
631 | // Get scheduler to re-evaluate and call hooks. | |
632 | // If !IRQ_WORK, FQS scan will eventually IPI. | |
633 | init_irq_work(&rdp->defer_qs_iw, | |
634 | rcu_preempt_deferred_qs_handler); | |
635 | rdp->defer_qs_iw_pending = true; | |
636 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); | |
637 | } | |
05f41571 | 638 | } |
23634ebc | 639 | t->rcu_read_unlock_special.b.deferred_qs = true; |
3e310098 PM |
640 | local_irq_restore(flags); |
641 | return; | |
642 | } | |
05f41571 | 643 | WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false); |
3e310098 PM |
644 | rcu_preempt_deferred_qs_irqrestore(t, flags); |
645 | } | |
646 | ||
b0e165c0 PM |
647 | /* |
648 | * Check that the list of blocked tasks for the newly completed grace | |
649 | * period is in fact empty. It is a serious bug to complete a grace | |
650 | * period that still has RCU readers blocked! This function must be | |
ff3bb6f4 | 651 | * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock |
b0e165c0 | 652 | * must be held by the caller. |
12f5f524 PM |
653 | * |
654 | * Also, if there are blocked tasks on the list, they automatically | |
655 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 656 | */ |
81ab59a3 | 657 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 658 | { |
c5ebe66c PM |
659 | struct task_struct *t; |
660 | ||
ea9b0c8a | 661 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
4bc8d555 | 662 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 663 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
664 | if (rcu_preempt_has_tasks(rnp) && |
665 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
12f5f524 | 666 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
667 | t = container_of(rnp->gp_tasks, struct task_struct, |
668 | rcu_node_entry); | |
669 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 670 | rnp->gp_seq, t->pid); |
c5ebe66c | 671 | } |
28ecd580 | 672 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
673 | } |
674 | ||
f41d911f | 675 | /* |
c98cac60 PM |
676 | * Check for a quiescent state from the current CPU, including voluntary |
677 | * context switches for Tasks RCU. When a task blocks, the task is | |
678 | * recorded in the corresponding CPU's rcu_node structure, which is checked | |
679 | * elsewhere, hence this function need only check for quiescent states | |
680 | * related to the current CPU, not to those related to tasks. | |
f41d911f | 681 | */ |
c98cac60 | 682 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f PM |
683 | { |
684 | struct task_struct *t = current; | |
685 | ||
45975c7d PM |
686 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
687 | rcu_note_voluntary_context_switch(current); | |
688 | } | |
3e310098 PM |
689 | if (t->rcu_read_lock_nesting > 0 || |
690 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { | |
691 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
692 | if (rcu_preempt_need_deferred_qs(t)) { |
693 | set_tsk_need_resched(t); | |
694 | set_preempt_need_resched(); | |
695 | } | |
3e310098 PM |
696 | } else if (rcu_preempt_need_deferred_qs(t)) { |
697 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
698 | return; | |
699 | } else if (!t->rcu_read_lock_nesting) { | |
45975c7d | 700 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
701 | return; |
702 | } | |
3e310098 PM |
703 | |
704 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
10f39bb1 | 705 | if (t->rcu_read_lock_nesting > 0 && |
2280ee5a PM |
706 | __this_cpu_read(rcu_data.core_needs_qs) && |
707 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 708 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 709 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 710 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
711 | } |
712 | ||
2439b696 PM |
713 | /* |
714 | * Check for a task exiting while in a preemptible-RCU read-side | |
884157ce PM |
715 | * critical section, clean up if so. No need to issue warnings, as |
716 | * debug_check_no_locks_held() already does this if lockdep is enabled. | |
717 | * Besides, if this function does anything other than just immediately | |
718 | * return, there was a bug of some sort. Spewing warnings from this | |
719 | * function is like as not to simply obscure important prior warnings. | |
2439b696 PM |
720 | */ |
721 | void exit_rcu(void) | |
722 | { | |
723 | struct task_struct *t = current; | |
724 | ||
884157ce PM |
725 | if (unlikely(!list_empty(¤t->rcu_node_entry))) { |
726 | t->rcu_read_lock_nesting = 1; | |
727 | barrier(); | |
add0d37b | 728 | WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true); |
884157ce PM |
729 | } else if (unlikely(t->rcu_read_lock_nesting)) { |
730 | t->rcu_read_lock_nesting = 1; | |
731 | } else { | |
2439b696 | 732 | return; |
884157ce | 733 | } |
2439b696 | 734 | __rcu_read_unlock(); |
3e310098 | 735 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
736 | } |
737 | ||
4bc8d555 PM |
738 | /* |
739 | * Dump the blocked-tasks state, but limit the list dump to the | |
740 | * specified number of elements. | |
741 | */ | |
57738942 | 742 | static void |
81ab59a3 | 743 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 744 | { |
57738942 | 745 | int cpu; |
4bc8d555 PM |
746 | int i; |
747 | struct list_head *lhp; | |
57738942 PM |
748 | bool onl; |
749 | struct rcu_data *rdp; | |
ff3cee39 | 750 | struct rcu_node *rnp1; |
4bc8d555 | 751 | |
ce11fae8 | 752 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 753 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 754 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
ff3cee39 PM |
755 | (long)rnp->gp_seq, (long)rnp->completedqs); |
756 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
757 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
758 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf PM |
759 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
760 | __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks); | |
761 | pr_info("%s: ->blkd_tasks", __func__); | |
4bc8d555 PM |
762 | i = 0; |
763 | list_for_each(lhp, &rnp->blkd_tasks) { | |
764 | pr_cont(" %p", lhp); | |
cd6d17b4 | 765 | if (++i >= ncheck) |
4bc8d555 PM |
766 | break; |
767 | } | |
768 | pr_cont("\n"); | |
57738942 | 769 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 770 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 PM |
771 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); |
772 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
773 | cpu, ".o"[onl], | |
774 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
775 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
776 | } | |
4bc8d555 PM |
777 | } |
778 | ||
28f6569a | 779 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f PM |
780 | |
781 | /* | |
782 | * Tell them what RCU they are running. | |
783 | */ | |
0e0fc1c2 | 784 | static void __init rcu_bootup_announce(void) |
f41d911f | 785 | { |
efc151c3 | 786 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 787 | rcu_bootup_announce_oddness(); |
f41d911f PM |
788 | } |
789 | ||
45975c7d PM |
790 | /* |
791 | * Note a quiescent state for PREEMPT=n. Because we do not need to know | |
792 | * how many quiescent states passed, just if there was at least one since | |
793 | * the start of the grace period, this just sets a flag. The caller must | |
794 | * have disabled preemption. | |
795 | */ | |
796 | static void rcu_qs(void) | |
d28139c4 | 797 | { |
45975c7d PM |
798 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
799 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
800 | return; | |
801 | trace_rcu_grace_period(TPS("rcu_sched"), | |
802 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
803 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
804 | if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) | |
805 | return; | |
806 | __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false); | |
63d4c8c9 | 807 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
d28139c4 PM |
808 | } |
809 | ||
395a2f09 PM |
810 | /* |
811 | * Register an urgently needed quiescent state. If there is an | |
812 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
813 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
814 | * some cases serve for expedited as well as normal grace periods. | |
815 | * Either way, register a lightweight quiescent state. | |
395a2f09 PM |
816 | */ |
817 | void rcu_all_qs(void) | |
818 | { | |
819 | unsigned long flags; | |
820 | ||
2dba13f0 | 821 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
822 | return; |
823 | preempt_disable(); | |
824 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 825 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
826 | preempt_enable(); |
827 | return; | |
828 | } | |
2dba13f0 | 829 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
2dba13f0 | 830 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
831 | local_irq_save(flags); |
832 | rcu_momentary_dyntick_idle(); | |
833 | local_irq_restore(flags); | |
834 | } | |
7e28c5af | 835 | rcu_qs(); |
395a2f09 PM |
836 | preempt_enable(); |
837 | } | |
838 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
839 | ||
cba6d0d6 | 840 | /* |
45975c7d | 841 | * Note a PREEMPT=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 842 | */ |
45975c7d | 843 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 844 | { |
45975c7d PM |
845 | trace_rcu_utilization(TPS("Start context switch")); |
846 | rcu_qs(); | |
847 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 848 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 849 | goto out; |
2dba13f0 PM |
850 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
851 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 852 | rcu_momentary_dyntick_idle(); |
45975c7d PM |
853 | if (!preempt) |
854 | rcu_tasks_qs(current); | |
855 | out: | |
856 | trace_rcu_utilization(TPS("End context switch")); | |
cba6d0d6 | 857 | } |
45975c7d | 858 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 859 | |
fc2219d4 | 860 | /* |
6cc68793 | 861 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
862 | * RCU readers. |
863 | */ | |
27f4d280 | 864 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
865 | { |
866 | return 0; | |
867 | } | |
868 | ||
8af3a5e7 PM |
869 | /* |
870 | * Because there is no preemptible RCU, there can be no readers blocked. | |
871 | */ | |
872 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 873 | { |
8af3a5e7 | 874 | return false; |
b668c9cf PM |
875 | } |
876 | ||
3e310098 PM |
877 | /* |
878 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
879 | * states. | |
880 | */ | |
881 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
882 | { | |
883 | return false; | |
884 | } | |
885 | static void rcu_preempt_deferred_qs(struct task_struct *t) { } | |
886 | ||
b0e165c0 | 887 | /* |
6cc68793 | 888 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
889 | * so there is no need to check for blocked tasks. So check only for |
890 | * bogus qsmask values. | |
b0e165c0 | 891 | */ |
81ab59a3 | 892 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 893 | { |
49e29126 | 894 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
895 | } |
896 | ||
f41d911f | 897 | /* |
c98cac60 PM |
898 | * Check to see if this CPU is in a non-context-switch quiescent state, |
899 | * namely user mode and idle loop. | |
f41d911f | 900 | */ |
c98cac60 | 901 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f | 902 | { |
45975c7d | 903 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 904 | |
45975c7d PM |
905 | /* |
906 | * Get here if this CPU took its interrupt from user | |
907 | * mode or from the idle loop, and if this is not a | |
908 | * nested interrupt. In this case, the CPU is in | |
909 | * a quiescent state, so note it. | |
910 | * | |
911 | * No memory barrier is required here because rcu_qs() | |
912 | * references only CPU-local variables that other CPUs | |
913 | * neither access nor modify, at least not while the | |
914 | * corresponding CPU is online. | |
915 | */ | |
916 | ||
917 | rcu_qs(); | |
918 | } | |
e74f4c45 | 919 | } |
e74f4c45 | 920 | |
2439b696 PM |
921 | /* |
922 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
923 | * while in preemptible RCU read-side critical sections. | |
924 | */ | |
925 | void exit_rcu(void) | |
926 | { | |
927 | } | |
928 | ||
4bc8d555 PM |
929 | /* |
930 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
931 | */ | |
57738942 | 932 | static void |
81ab59a3 | 933 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
934 | { |
935 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
936 | } | |
937 | ||
28f6569a | 938 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 939 | |
48d07c04 SAS |
940 | /* |
941 | * If boosting, set rcuc kthreads to realtime priority. | |
942 | */ | |
943 | static void rcu_cpu_kthread_setup(unsigned int cpu) | |
944 | { | |
27f4d280 | 945 | #ifdef CONFIG_RCU_BOOST |
48d07c04 | 946 | struct sched_param sp; |
27f4d280 | 947 | |
48d07c04 SAS |
948 | sp.sched_priority = kthread_prio; |
949 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
950 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
5d01bbd1 TG |
951 | } |
952 | ||
48d07c04 SAS |
953 | #ifdef CONFIG_RCU_BOOST |
954 | ||
27f4d280 PM |
955 | /* |
956 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
957 | * or ->boost_tasks, advancing the pointer to the next task in the | |
958 | * ->blkd_tasks list. | |
959 | * | |
960 | * Note that irqs must be enabled: boosting the task can block. | |
961 | * Returns 1 if there are more tasks needing to be boosted. | |
962 | */ | |
963 | static int rcu_boost(struct rcu_node *rnp) | |
964 | { | |
965 | unsigned long flags; | |
27f4d280 PM |
966 | struct task_struct *t; |
967 | struct list_head *tb; | |
968 | ||
7d0ae808 PM |
969 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
970 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
971 | return 0; /* Nothing left to boost. */ |
972 | ||
2a67e741 | 973 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
974 | |
975 | /* | |
976 | * Recheck under the lock: all tasks in need of boosting | |
977 | * might exit their RCU read-side critical sections on their own. | |
978 | */ | |
979 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 980 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
981 | return 0; |
982 | } | |
983 | ||
984 | /* | |
985 | * Preferentially boost tasks blocking expedited grace periods. | |
986 | * This cannot starve the normal grace periods because a second | |
987 | * expedited grace period must boost all blocked tasks, including | |
988 | * those blocking the pre-existing normal grace period. | |
989 | */ | |
bec06785 | 990 | if (rnp->exp_tasks != NULL) |
27f4d280 | 991 | tb = rnp->exp_tasks; |
bec06785 | 992 | else |
27f4d280 PM |
993 | tb = rnp->boost_tasks; |
994 | ||
995 | /* | |
996 | * We boost task t by manufacturing an rt_mutex that appears to | |
997 | * be held by task t. We leave a pointer to that rt_mutex where | |
998 | * task t can find it, and task t will release the mutex when it | |
999 | * exits its outermost RCU read-side critical section. Then | |
1000 | * simply acquiring this artificial rt_mutex will boost task | |
1001 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1002 | * | |
1003 | * Note that task t must acquire rnp->lock to remove itself from | |
1004 | * the ->blkd_tasks list, which it will do from exit() if from | |
1005 | * nowhere else. We therefore are guaranteed that task t will | |
1006 | * stay around at least until we drop rnp->lock. Note that | |
1007 | * rnp->lock also resolves races between our priority boosting | |
1008 | * and task t's exiting its outermost RCU read-side critical | |
1009 | * section. | |
1010 | */ | |
1011 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1012 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1013 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1014 | /* Lock only for side effect: boosts task t's priority. */ |
1015 | rt_mutex_lock(&rnp->boost_mtx); | |
1016 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1017 | |
7d0ae808 PM |
1018 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1019 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1020 | } |
1021 | ||
27f4d280 | 1022 | /* |
bc17ea10 | 1023 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1024 | */ |
1025 | static int rcu_boost_kthread(void *arg) | |
1026 | { | |
1027 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1028 | int spincnt = 0; | |
1029 | int more2boost; | |
1030 | ||
f7f7bac9 | 1031 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1032 | for (;;) { |
d71df90e | 1033 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1034 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1035 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1036 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1037 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1038 | more2boost = rcu_boost(rnp); |
1039 | if (more2boost) | |
1040 | spincnt++; | |
1041 | else | |
1042 | spincnt = 0; | |
1043 | if (spincnt > 10) { | |
5d01bbd1 | 1044 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1045 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1046 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1047 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1048 | spincnt = 0; |
1049 | } | |
1050 | } | |
1217ed1b | 1051 | /* NOTREACHED */ |
f7f7bac9 | 1052 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1053 | return 0; |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Check to see if it is time to start boosting RCU readers that are | |
1058 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1059 | * kthread to start boosting them. If there is an expedited grace | |
1060 | * period in progress, it is always time to boost. | |
1061 | * | |
b065a853 PM |
1062 | * The caller must hold rnp->lock, which this function releases. |
1063 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1064 | * about it going away. | |
27f4d280 | 1065 | */ |
1217ed1b | 1066 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1067 | __releases(rnp->lock) |
27f4d280 | 1068 | { |
a32e01ee | 1069 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1070 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1071 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1072 | return; |
0ea1f2eb | 1073 | } |
27f4d280 PM |
1074 | if (rnp->exp_tasks != NULL || |
1075 | (rnp->gp_tasks != NULL && | |
1076 | rnp->boost_tasks == NULL && | |
1077 | rnp->qsmask == 0 && | |
1078 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1079 | if (rnp->exp_tasks == NULL) | |
1080 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1081 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
a2badefa PM |
1082 | rcu_wake_cond(rnp->boost_kthread_task, |
1083 | rnp->boost_kthread_status); | |
1217ed1b | 1084 | } else { |
67c583a7 | 1085 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1086 | } |
27f4d280 PM |
1087 | } |
1088 | ||
dff1672d PM |
1089 | /* |
1090 | * Is the current CPU running the RCU-callbacks kthread? | |
1091 | * Caller must have preemption disabled. | |
1092 | */ | |
1093 | static bool rcu_is_callbacks_kthread(void) | |
1094 | { | |
37f62d7c | 1095 | return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current; |
dff1672d PM |
1096 | } |
1097 | ||
27f4d280 PM |
1098 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1099 | ||
1100 | /* | |
1101 | * Do priority-boost accounting for the start of a new grace period. | |
1102 | */ | |
1103 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1104 | { | |
1105 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1106 | } | |
1107 | ||
27f4d280 PM |
1108 | /* |
1109 | * Create an RCU-boost kthread for the specified node if one does not | |
1110 | * already exist. We only create this kthread for preemptible RCU. | |
1111 | * Returns zero if all is well, a negated errno otherwise. | |
1112 | */ | |
3545832f | 1113 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 | 1114 | { |
6dbfdc14 | 1115 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1116 | unsigned long flags; |
1117 | struct sched_param sp; | |
1118 | struct task_struct *t; | |
1119 | ||
6dbfdc14 | 1120 | if (!IS_ENABLED(CONFIG_PREEMPT_RCU)) |
3545832f | 1121 | return; |
5d01bbd1 | 1122 | |
0aa04b05 | 1123 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
3545832f | 1124 | return; |
5d01bbd1 | 1125 | |
6dbfdc14 | 1126 | rcu_state.boost = 1; |
3545832f | 1127 | |
27f4d280 | 1128 | if (rnp->boost_kthread_task != NULL) |
3545832f BP |
1129 | return; |
1130 | ||
27f4d280 | 1131 | t = kthread_create(rcu_boost_kthread, (void *)rnp, |
5b61b0ba | 1132 | "rcub/%d", rnp_index); |
3545832f BP |
1133 | if (WARN_ON_ONCE(IS_ERR(t))) |
1134 | return; | |
1135 | ||
2a67e741 | 1136 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1137 | rnp->boost_kthread_task = t; |
67c583a7 | 1138 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1139 | sp.sched_priority = kthread_prio; |
27f4d280 | 1140 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1141 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1142 | } |
1143 | ||
f8b7fc6b PM |
1144 | /* |
1145 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1146 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1147 | * held, so the value of rnp->qsmaskinit will be stable. | |
1148 | * | |
1149 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1150 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1151 | * this function allows the kthread to execute on any CPU. | |
1152 | */ | |
5d01bbd1 | 1153 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1154 | { |
5d01bbd1 | 1155 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1156 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1157 | cpumask_var_t cm; |
1158 | int cpu; | |
f8b7fc6b | 1159 | |
5d01bbd1 | 1160 | if (!t) |
f8b7fc6b | 1161 | return; |
5d01bbd1 | 1162 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1163 | return; |
bc75e999 MR |
1164 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1165 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1166 | cpu != outgoingcpu) | |
f8b7fc6b | 1167 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1168 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1169 | cpumask_setall(cm); |
5d01bbd1 | 1170 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1171 | free_cpumask_var(cm); |
1172 | } | |
1173 | ||
f8b7fc6b | 1174 | /* |
9386c0b7 | 1175 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1176 | */ |
9386c0b7 | 1177 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1178 | { |
f8b7fc6b PM |
1179 | struct rcu_node *rnp; |
1180 | ||
aedf4ba9 | 1181 | rcu_for_each_leaf_node(rnp) |
3545832f | 1182 | rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b | 1183 | } |
f8b7fc6b | 1184 | |
49fb4c62 | 1185 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1186 | { |
da1df50d | 1187 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
f8b7fc6b PM |
1188 | struct rcu_node *rnp = rdp->mynode; |
1189 | ||
1190 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1191 | if (rcu_scheduler_fully_active) |
3545832f | 1192 | rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b PM |
1193 | } |
1194 | ||
27f4d280 PM |
1195 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1196 | ||
1217ed1b | 1197 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1198 | __releases(rnp->lock) |
27f4d280 | 1199 | { |
67c583a7 | 1200 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1201 | } |
1202 | ||
dff1672d PM |
1203 | static bool rcu_is_callbacks_kthread(void) |
1204 | { | |
1205 | return false; | |
1206 | } | |
1207 | ||
27f4d280 PM |
1208 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1209 | { | |
1210 | } | |
1211 | ||
5d01bbd1 | 1212 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1213 | { |
1214 | } | |
1215 | ||
9386c0b7 | 1216 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1217 | { |
b0d30417 | 1218 | } |
b0d30417 | 1219 | |
49fb4c62 | 1220 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1221 | { |
1222 | } | |
1223 | ||
27f4d280 PM |
1224 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1225 | ||
8bd93a2c PM |
1226 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1227 | ||
1228 | /* | |
0bd55c69 PM |
1229 | * Check to see if any future non-offloaded RCU-related work will need |
1230 | * to be done by the current CPU, even if none need be done immediately, | |
1231 | * returning 1 if so. This function is part of the RCU implementation; | |
1232 | * it is -not- an exported member of the RCU API. | |
8bd93a2c | 1233 | * |
0ae86a27 PM |
1234 | * Because we not have RCU_FAST_NO_HZ, just check whether or not this |
1235 | * CPU has RCU callbacks queued. | |
8bd93a2c | 1236 | */ |
c1ad348b | 1237 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1238 | { |
c1ad348b | 1239 | *nextevt = KTIME_MAX; |
0bd55c69 PM |
1240 | return !rcu_segcblist_empty(&this_cpu_ptr(&rcu_data)->cblist) && |
1241 | !rcu_segcblist_is_offloaded(&this_cpu_ptr(&rcu_data)->cblist); | |
7cb92499 PM |
1242 | } |
1243 | ||
1244 | /* | |
1245 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1246 | * after it. | |
1247 | */ | |
8fa7845d | 1248 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1249 | { |
1250 | } | |
1251 | ||
aea1b35e | 1252 | /* |
a858af28 | 1253 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1254 | * is nothing. |
1255 | */ | |
198bbf81 | 1256 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1257 | { |
1258 | } | |
1259 | ||
8bd93a2c PM |
1260 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1261 | ||
f23f7fa1 PM |
1262 | /* |
1263 | * This code is invoked when a CPU goes idle, at which point we want | |
1264 | * to have the CPU do everything required for RCU so that it can enter | |
1265 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1266 | * state machine implemented by rcu_prepare_for_idle() below. | |
1267 | * | |
1268 | * The following three proprocessor symbols control this state machine: | |
1269 | * | |
f23f7fa1 PM |
1270 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1271 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1272 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1273 | * benchmarkers who might otherwise be tempted to set this to a large | |
1274 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1275 | * system. And if you are -that- concerned about energy efficiency, | |
1276 | * just power the system down and be done with it! | |
778d250a PM |
1277 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1278 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1279 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1280 | * |
1281 | * The values below work well in practice. If future workloads require | |
1282 | * adjustment, they can be converted into kernel config parameters, though | |
1283 | * making the state machine smarter might be a better option. | |
1284 | */ | |
e84c48ae | 1285 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1286 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1287 | |
5e44ce35 PM |
1288 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1289 | module_param(rcu_idle_gp_delay, int, 0644); | |
1290 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1291 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1292 | |
486e2593 | 1293 | /* |
0ae86a27 PM |
1294 | * Try to advance callbacks on the current CPU, but only if it has been |
1295 | * awhile since the last time we did so. Afterwards, if there are any | |
1296 | * callbacks ready for immediate invocation, return true. | |
486e2593 | 1297 | */ |
f1f399d1 | 1298 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1299 | { |
c0f4dfd4 | 1300 | bool cbs_ready = false; |
5998a75a | 1301 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1302 | struct rcu_node *rnp; |
486e2593 | 1303 | |
c229828c | 1304 | /* Exit early if we advanced recently. */ |
5998a75a | 1305 | if (jiffies == rdp->last_advance_all) |
d0bc90fd | 1306 | return false; |
5998a75a | 1307 | rdp->last_advance_all = jiffies; |
c229828c | 1308 | |
b97d23c5 | 1309 | rnp = rdp->mynode; |
486e2593 | 1310 | |
b97d23c5 PM |
1311 | /* |
1312 | * Don't bother checking unless a grace period has | |
1313 | * completed since we last checked and there are | |
1314 | * callbacks not yet ready to invoke. | |
1315 | */ | |
1316 | if ((rcu_seq_completed_gp(rdp->gp_seq, | |
1317 | rcu_seq_current(&rnp->gp_seq)) || | |
1318 | unlikely(READ_ONCE(rdp->gpwrap))) && | |
1319 | rcu_segcblist_pend_cbs(&rdp->cblist)) | |
1320 | note_gp_changes(rdp); | |
1321 | ||
1322 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
1323 | cbs_ready = true; | |
c0f4dfd4 | 1324 | return cbs_ready; |
486e2593 PM |
1325 | } |
1326 | ||
aa9b1630 | 1327 | /* |
c0f4dfd4 PM |
1328 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1329 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1330 | * caller to set the timeout based on whether or not there are non-lazy | |
1331 | * callbacks. | |
aa9b1630 | 1332 | * |
c0f4dfd4 | 1333 | * The caller must have disabled interrupts. |
aa9b1630 | 1334 | */ |
c1ad348b | 1335 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1336 | { |
5998a75a | 1337 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c1ad348b | 1338 | unsigned long dj; |
aa9b1630 | 1339 | |
b04db8e1 | 1340 | lockdep_assert_irqs_disabled(); |
3382adbc | 1341 | |
0bd55c69 PM |
1342 | /* If no non-offloaded callbacks, RCU doesn't need the CPU. */ |
1343 | if (rcu_segcblist_empty(&rdp->cblist) || | |
1344 | rcu_segcblist_is_offloaded(&this_cpu_ptr(&rcu_data)->cblist)) { | |
c1ad348b | 1345 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1346 | return 0; |
1347 | } | |
c0f4dfd4 PM |
1348 | |
1349 | /* Attempt to advance callbacks. */ | |
1350 | if (rcu_try_advance_all_cbs()) { | |
1351 | /* Some ready to invoke, so initiate later invocation. */ | |
1352 | invoke_rcu_core(); | |
aa9b1630 PM |
1353 | return 1; |
1354 | } | |
5998a75a | 1355 | rdp->last_accelerate = jiffies; |
c0f4dfd4 PM |
1356 | |
1357 | /* Request timer delay depending on laziness, and round. */ | |
260e1e4f PM |
1358 | rdp->all_lazy = !rcu_segcblist_n_nonlazy_cbs(&rdp->cblist); |
1359 | if (rdp->all_lazy) { | |
1360 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; | |
1361 | } else { | |
c1ad348b | 1362 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1363 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1364 | } |
c1ad348b | 1365 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1366 | return 0; |
1367 | } | |
1368 | ||
21e52e15 | 1369 | /* |
c0f4dfd4 PM |
1370 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1371 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1372 | * The second major task is to check to see if a non-lazy callback has | |
1373 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1374 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1375 | * any recently arrived callbacks. | |
aea1b35e PM |
1376 | * |
1377 | * The caller must have disabled interrupts. | |
8bd93a2c | 1378 | */ |
198bbf81 | 1379 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1380 | { |
48a7639c | 1381 | bool needwake; |
0fd79e75 | 1382 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1383 | struct rcu_node *rnp; |
9d2ad243 PM |
1384 | int tne; |
1385 | ||
b04db8e1 | 1386 | lockdep_assert_irqs_disabled(); |
ce5215c1 | 1387 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) |
3382adbc PM |
1388 | return; |
1389 | ||
9d2ad243 | 1390 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1391 | tne = READ_ONCE(tick_nohz_active); |
0fd79e75 | 1392 | if (tne != rdp->tick_nohz_enabled_snap) { |
260e1e4f | 1393 | if (!rcu_segcblist_empty(&rdp->cblist)) |
9d2ad243 | 1394 | invoke_rcu_core(); /* force nohz to see update. */ |
0fd79e75 | 1395 | rdp->tick_nohz_enabled_snap = tne; |
9d2ad243 PM |
1396 | return; |
1397 | } | |
1398 | if (!tne) | |
1399 | return; | |
f511fc62 | 1400 | |
c57afe80 | 1401 | /* |
c0f4dfd4 PM |
1402 | * If a non-lazy callback arrived at a CPU having only lazy |
1403 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1404 | * idle duration on re-entry to idle. | |
c57afe80 | 1405 | */ |
260e1e4f | 1406 | if (rdp->all_lazy && rcu_segcblist_n_nonlazy_cbs(&rdp->cblist)) { |
c458a89e | 1407 | rdp->all_lazy = false; |
c0f4dfd4 | 1408 | invoke_rcu_core(); |
c57afe80 PM |
1409 | return; |
1410 | } | |
c57afe80 | 1411 | |
3084f2f8 | 1412 | /* |
c0f4dfd4 PM |
1413 | * If we have not yet accelerated this jiffy, accelerate all |
1414 | * callbacks on this CPU. | |
3084f2f8 | 1415 | */ |
5998a75a | 1416 | if (rdp->last_accelerate == jiffies) |
aea1b35e | 1417 | return; |
5998a75a | 1418 | rdp->last_accelerate = jiffies; |
b97d23c5 | 1419 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
c0f4dfd4 | 1420 | rnp = rdp->mynode; |
2a67e741 | 1421 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
02f50142 | 1422 | needwake = rcu_accelerate_cbs(rnp, rdp); |
67c583a7 | 1423 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c | 1424 | if (needwake) |
532c00c9 | 1425 | rcu_gp_kthread_wake(); |
77e38ed3 | 1426 | } |
c0f4dfd4 | 1427 | } |
3084f2f8 | 1428 | |
c0f4dfd4 PM |
1429 | /* |
1430 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1431 | * any grace periods that elapsed while the CPU was idle, and if any | |
1432 | * callbacks are now ready to invoke, initiate invocation. | |
1433 | */ | |
8fa7845d | 1434 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1435 | { |
ce5215c1 PM |
1436 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
1437 | ||
b04db8e1 | 1438 | lockdep_assert_irqs_disabled(); |
ce5215c1 | 1439 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) |
aea1b35e | 1440 | return; |
7a497c96 PM |
1441 | if (rcu_try_advance_all_cbs()) |
1442 | invoke_rcu_core(); | |
8bd93a2c PM |
1443 | } |
1444 | ||
1445 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ | |
a858af28 | 1446 | |
3fbfbf7a PM |
1447 | #ifdef CONFIG_RCU_NOCB_CPU |
1448 | ||
1449 | /* | |
1450 | * Offload callback processing from the boot-time-specified set of CPUs | |
a9fefdb2 PM |
1451 | * specified by rcu_nocb_mask. For the CPUs in the set, there are kthreads |
1452 | * created that pull the callbacks from the corresponding CPU, wait for | |
1453 | * a grace period to elapse, and invoke the callbacks. These kthreads | |
6484fe54 PM |
1454 | * are organized into GP kthreads, which manage incoming callbacks, wait for |
1455 | * grace periods, and awaken CB kthreads, and the CB kthreads, which only | |
1456 | * invoke callbacks. Each GP kthread invokes its own CBs. The no-CBs CPUs | |
1457 | * do a wake_up() on their GP kthread when they insert a callback into any | |
a9fefdb2 PM |
1458 | * empty list, unless the rcu_nocb_poll boot parameter has been specified, |
1459 | * in which case each kthread actively polls its CPU. (Which isn't so great | |
1460 | * for energy efficiency, but which does reduce RCU's overhead on that CPU.) | |
3fbfbf7a PM |
1461 | * |
1462 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1463 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1464 | * running CPU-bound user-mode computations. | |
1465 | * | |
a9fefdb2 PM |
1466 | * Offloading of callbacks can also be used as an energy-efficiency |
1467 | * measure because CPUs with no RCU callbacks queued are more aggressive | |
1468 | * about entering dyntick-idle mode. | |
3fbfbf7a PM |
1469 | */ |
1470 | ||
1471 | ||
497e4260 PM |
1472 | /* |
1473 | * Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. | |
1474 | * The string after the "rcu_nocbs=" is either "all" for all CPUs, or a | |
1475 | * comma-separated list of CPUs and/or CPU ranges. If an invalid list is | |
1476 | * given, a warning is emitted and all CPUs are offloaded. | |
1477 | */ | |
3fbfbf7a PM |
1478 | static int __init rcu_nocb_setup(char *str) |
1479 | { | |
1480 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
da8739f2 PM |
1481 | if (!strcasecmp(str, "all")) |
1482 | cpumask_setall(rcu_nocb_mask); | |
1483 | else | |
497e4260 PM |
1484 | if (cpulist_parse(str, rcu_nocb_mask)) { |
1485 | pr_warn("rcu_nocbs= bad CPU range, all CPUs set\n"); | |
1486 | cpumask_setall(rcu_nocb_mask); | |
1487 | } | |
3fbfbf7a PM |
1488 | return 1; |
1489 | } | |
1490 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1491 | ||
1b0048a4 PG |
1492 | static int __init parse_rcu_nocb_poll(char *arg) |
1493 | { | |
5455a7f6 | 1494 | rcu_nocb_poll = true; |
1b0048a4 PG |
1495 | return 0; |
1496 | } | |
1497 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1498 | ||
5d6742b3 PM |
1499 | /* |
1500 | * Acquire the specified rcu_data structure's ->nocb_lock, but only | |
81c0b3d7 PM |
1501 | * if it corresponds to a no-CBs CPU. If the lock isn't immediately |
1502 | * available, increment ->nocb_lock_contended to flag the contention. | |
5d6742b3 PM |
1503 | */ |
1504 | static void rcu_nocb_lock(struct rcu_data *rdp) | |
1505 | { | |
81c0b3d7 PM |
1506 | lockdep_assert_irqs_disabled(); |
1507 | if (!rcu_segcblist_is_offloaded(&rdp->cblist) || | |
1508 | raw_spin_trylock(&rdp->nocb_lock)) | |
1509 | return; | |
1510 | atomic_inc(&rdp->nocb_lock_contended); | |
1511 | smp_mb__after_atomic(); /* atomic_inc() before lock. */ | |
1512 | raw_spin_lock(&rdp->nocb_lock); | |
1513 | smp_mb__before_atomic(); /* atomic_dec() after lock. */ | |
1514 | atomic_dec(&rdp->nocb_lock_contended); | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Spinwait until the specified rcu_data structure's ->nocb_lock is | |
1519 | * not contended. Please note that this is extremely special-purpose, | |
1520 | * relying on the fact that at most two kthreads and one CPU contend for | |
1521 | * this lock, and also that the two kthreads are guaranteed to have frequent | |
1522 | * grace-period-duration time intervals between successive acquisitions | |
1523 | * of the lock. This allows us to use an extremely simple throttling | |
1524 | * mechanism, and further to apply it only to the CPU doing floods of | |
1525 | * call_rcu() invocations. Don't try this at home! | |
1526 | */ | |
1527 | static void rcu_nocb_wait_contended(struct rcu_data *rdp) | |
1528 | { | |
1529 | while (atomic_read(&rdp->nocb_lock_contended)) | |
1530 | cpu_relax(); | |
5d6742b3 PM |
1531 | } |
1532 | ||
1533 | /* | |
1534 | * Release the specified rcu_data structure's ->nocb_lock, but only | |
1535 | * if it corresponds to a no-CBs CPU. | |
1536 | */ | |
1537 | static void rcu_nocb_unlock(struct rcu_data *rdp) | |
1538 | { | |
1539 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) { | |
1540 | lockdep_assert_irqs_disabled(); | |
1541 | raw_spin_unlock(&rdp->nocb_lock); | |
1542 | } | |
1543 | } | |
1544 | ||
1545 | /* | |
1546 | * Release the specified rcu_data structure's ->nocb_lock and restore | |
1547 | * interrupts, but only if it corresponds to a no-CBs CPU. | |
1548 | */ | |
1549 | static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp, | |
1550 | unsigned long flags) | |
1551 | { | |
1552 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) { | |
1553 | lockdep_assert_irqs_disabled(); | |
1554 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
1555 | } else { | |
1556 | local_irq_restore(flags); | |
1557 | } | |
1558 | } | |
1559 | ||
dae6e64d | 1560 | /* |
0446be48 PM |
1561 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1562 | * grace period. | |
dae6e64d | 1563 | */ |
abedf8e2 | 1564 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1565 | { |
abedf8e2 | 1566 | swake_up_all(sq); |
dae6e64d PM |
1567 | } |
1568 | ||
abedf8e2 | 1569 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1570 | { |
e0da2374 | 1571 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1572 | } |
1573 | ||
dae6e64d | 1574 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1575 | { |
abedf8e2 PG |
1576 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1577 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1578 | } |
1579 | ||
24342c96 | 1580 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1581 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1582 | { |
84b12b75 | 1583 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1584 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1585 | return false; | |
1586 | } | |
1587 | ||
fbce7497 | 1588 | /* |
6484fe54 | 1589 | * Kick the GP kthread for this NOCB group. Caller holds ->nocb_lock |
8be6e1b1 | 1590 | * and this function releases it. |
fbce7497 | 1591 | */ |
5d6742b3 | 1592 | static void wake_nocb_gp(struct rcu_data *rdp, bool force, |
5f675ba6 | 1593 | unsigned long flags) |
8be6e1b1 | 1594 | __releases(rdp->nocb_lock) |
fbce7497 | 1595 | { |
5f675ba6 | 1596 | struct rcu_data *rdp_gp = rdp->nocb_gp_rdp; |
fbce7497 | 1597 | |
8be6e1b1 | 1598 | lockdep_assert_held(&rdp->nocb_lock); |
5f675ba6 | 1599 | if (!READ_ONCE(rdp_gp->nocb_gp_kthread)) { |
81c0b3d7 | 1600 | rcu_nocb_unlock_irqrestore(rdp, flags); |
fbce7497 | 1601 | return; |
8be6e1b1 | 1602 | } |
5d6742b3 | 1603 | if (READ_ONCE(rdp_gp->nocb_gp_sleep) || force) { |
8be6e1b1 | 1604 | del_timer(&rdp->nocb_timer); |
81c0b3d7 | 1605 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 | 1606 | smp_mb(); /* enqueue before ->nocb_gp_sleep. */ |
81c0b3d7 | 1607 | rcu_nocb_lock_irqsave(rdp_gp, flags); |
5d6742b3 | 1608 | WRITE_ONCE(rdp_gp->nocb_gp_sleep, false); |
81c0b3d7 | 1609 | rcu_nocb_unlock_irqrestore(rdp_gp, flags); |
5d6742b3 | 1610 | wake_up_process(rdp_gp->nocb_gp_kthread); |
8be6e1b1 | 1611 | } else { |
81c0b3d7 | 1612 | rcu_nocb_unlock_irqrestore(rdp, flags); |
fbce7497 PM |
1613 | } |
1614 | } | |
1615 | ||
8be6e1b1 | 1616 | /* |
6484fe54 PM |
1617 | * Arrange to wake the GP kthread for this NOCB group at some future |
1618 | * time when it is safe to do so. | |
8be6e1b1 | 1619 | */ |
0d52a665 PM |
1620 | static void wake_nocb_gp_defer(struct rcu_data *rdp, int waketype, |
1621 | const char *reason) | |
8be6e1b1 | 1622 | { |
8be6e1b1 PM |
1623 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) |
1624 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
383e1332 PM |
1625 | if (rdp->nocb_defer_wakeup < waketype) |
1626 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
88d1bead | 1627 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, reason); |
d7e29933 PM |
1628 | } |
1629 | ||
3fbfbf7a | 1630 | /* |
5d6742b3 PM |
1631 | * Awaken the no-CBs grace-period kthead if needed, either due to it |
1632 | * legitimately being asleep or due to overload conditions. | |
3fbfbf7a PM |
1633 | * |
1634 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1635 | */ | |
5d6742b3 PM |
1636 | static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone, |
1637 | unsigned long flags) | |
1638 | __releases(rdp->nocb_lock) | |
3fbfbf7a | 1639 | { |
ce0a825e | 1640 | long len; |
3fbfbf7a PM |
1641 | struct task_struct *t; |
1642 | ||
5d6742b3 | 1643 | // If we are being polled or there is no kthread, just leave. |
12f54c3a | 1644 | t = READ_ONCE(rdp->nocb_gp_kthread); |
25e03a74 | 1645 | if (rcu_nocb_poll || !t) { |
88d1bead | 1646 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9261dd0d | 1647 | TPS("WakeNotPoll")); |
5d6742b3 | 1648 | rcu_nocb_unlock_irqrestore(rdp, flags); |
3fbfbf7a | 1649 | return; |
9261dd0d | 1650 | } |
5d6742b3 PM |
1651 | // Need to actually to a wakeup. |
1652 | len = rcu_segcblist_n_cbs(&rdp->cblist); | |
1653 | if (was_alldone) { | |
aeeacd9d | 1654 | rdp->qlen_last_fqs_check = len; |
96d3fd0d | 1655 | if (!irqs_disabled_flags(flags)) { |
fbce7497 | 1656 | /* ... if queue was empty ... */ |
5d6742b3 | 1657 | wake_nocb_gp(rdp, false, flags); |
88d1bead | 1658 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
96d3fd0d PM |
1659 | TPS("WakeEmpty")); |
1660 | } else { | |
0d52a665 PM |
1661 | wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE, |
1662 | TPS("WakeEmptyIsDeferred")); | |
5d6742b3 | 1663 | rcu_nocb_unlock_irqrestore(rdp, flags); |
96d3fd0d | 1664 | } |
3fbfbf7a | 1665 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { |
fbce7497 | 1666 | /* ... or if many callbacks queued. */ |
aeeacd9d | 1667 | rdp->qlen_last_fqs_check = len; |
7f36ef82 PM |
1668 | if (!rdp->nocb_cb_sleep && |
1669 | rcu_segcblist_ready_cbs(&rdp->cblist)) { | |
1670 | // Already going full tilt, so don't try to rewake. | |
6608c3a0 | 1671 | } else if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
7f36ef82 | 1672 | rcu_advance_cbs_nowake(rdp->mynode, rdp); |
9fdd3bc9 | 1673 | } else { |
0d52a665 PM |
1674 | wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE_FORCE, |
1675 | TPS("WakeOvfIsDeferred")); | |
9fdd3bc9 | 1676 | } |
81c0b3d7 | 1677 | rcu_nocb_unlock_irqrestore(rdp, flags); |
9261dd0d | 1678 | } else { |
88d1bead | 1679 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WakeNot")); |
5d6742b3 | 1680 | rcu_nocb_unlock_irqrestore(rdp, flags); |
3fbfbf7a | 1681 | } |
81c0b3d7 | 1682 | if (!irqs_disabled_flags(flags)) { |
5d6742b3 | 1683 | lockdep_assert_irqs_enabled(); |
81c0b3d7 PM |
1684 | rcu_nocb_wait_contended(rdp); |
1685 | } | |
3fbfbf7a PM |
1686 | return; |
1687 | } | |
1688 | ||
1689 | /* | |
5d6742b3 PM |
1690 | * No-CBs GP kthreads come here to wait for additional callbacks to show up |
1691 | * or for grace periods to end. | |
fbce7497 | 1692 | */ |
12f54c3a | 1693 | static void nocb_gp_wait(struct rcu_data *my_rdp) |
fbce7497 | 1694 | { |
5d6742b3 PM |
1695 | int __maybe_unused cpu = my_rdp->cpu; |
1696 | unsigned long cur_gp_seq; | |
8be6e1b1 | 1697 | unsigned long flags; |
fbce7497 | 1698 | bool gotcbs; |
969974e5 | 1699 | bool needwait_gp = false; // This prevents actual uninitialized use. |
5d6742b3 PM |
1700 | bool needwake; |
1701 | bool needwake_gp; | |
fbce7497 | 1702 | struct rcu_data *rdp; |
5d6742b3 | 1703 | struct rcu_node *rnp; |
969974e5 | 1704 | unsigned long wait_gp_seq = 0; // Suppress "use uninitialized" warning. |
fbce7497 PM |
1705 | |
1706 | /* | |
5d6742b3 PM |
1707 | * Each pass through the following loop checks for CBs and for the |
1708 | * nearest grace period (if any) to wait for next. The CB kthreads | |
1709 | * and the global grace-period kthread are awakened if needed. | |
fbce7497 | 1710 | */ |
58bf6f77 | 1711 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_cb_rdp) { |
5d6742b3 PM |
1712 | if (rcu_segcblist_empty(&rdp->cblist)) |
1713 | continue; /* No callbacks here, try next. */ | |
1714 | rnp = rdp->mynode; | |
81c0b3d7 | 1715 | rcu_nocb_lock_irqsave(rdp, flags); |
5d6742b3 PM |
1716 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
1717 | del_timer(&my_rdp->nocb_timer); | |
1718 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
1719 | needwake_gp = rcu_advance_cbs(rnp, rdp); | |
1720 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ | |
1721 | // Need to wait on some grace period? | |
1722 | if (rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq)) { | |
1723 | if (!needwait_gp || | |
1724 | ULONG_CMP_LT(cur_gp_seq, wait_gp_seq)) | |
1725 | wait_gp_seq = cur_gp_seq; | |
1726 | needwait_gp = true; | |
8be6e1b1 | 1727 | } |
5d6742b3 PM |
1728 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) { |
1729 | needwake = rdp->nocb_cb_sleep; | |
1730 | WRITE_ONCE(rdp->nocb_cb_sleep, false); | |
1731 | smp_mb(); /* CB invocation -after- GP end. */ | |
1732 | } else { | |
1733 | needwake = false; | |
8be6e1b1 | 1734 | } |
81c0b3d7 | 1735 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 | 1736 | if (needwake) { |
12f54c3a | 1737 | swake_up_one(&rdp->nocb_cb_wq); |
5d6742b3 | 1738 | gotcbs = true; |
fbce7497 | 1739 | } |
5d6742b3 PM |
1740 | if (needwake_gp) |
1741 | rcu_gp_kthread_wake(); | |
1742 | } | |
1743 | ||
1744 | if (rcu_nocb_poll) { | |
1745 | /* Polling, so trace if first poll in the series. */ | |
1746 | if (gotcbs) | |
1747 | trace_rcu_nocb_wake(rcu_state.name, cpu, TPS("Poll")); | |
1748 | schedule_timeout_interruptible(1); | |
1749 | } else if (!needwait_gp) { | |
1750 | /* Wait for callbacks to appear. */ | |
1751 | trace_rcu_nocb_wake(rcu_state.name, cpu, TPS("Sleep")); | |
1752 | swait_event_interruptible_exclusive(my_rdp->nocb_gp_wq, | |
1753 | !READ_ONCE(my_rdp->nocb_gp_sleep)); | |
1754 | } else { | |
1755 | rnp = my_rdp->mynode; | |
1756 | trace_rcu_this_gp(rnp, my_rdp, wait_gp_seq, TPS("StartWait")); | |
1757 | swait_event_interruptible_exclusive( | |
1758 | rnp->nocb_gp_wq[rcu_seq_ctr(wait_gp_seq) & 0x1], | |
1759 | rcu_seq_done(&rnp->gp_seq, wait_gp_seq) || | |
1760 | !READ_ONCE(my_rdp->nocb_gp_sleep)); | |
1761 | trace_rcu_this_gp(rnp, my_rdp, wait_gp_seq, TPS("EndWait")); | |
1762 | } | |
1763 | if (!rcu_nocb_poll) { | |
81c0b3d7 | 1764 | rcu_nocb_lock_irqsave(my_rdp, flags); |
5d6742b3 | 1765 | WRITE_ONCE(my_rdp->nocb_gp_sleep, true); |
81c0b3d7 | 1766 | rcu_nocb_unlock_irqrestore(my_rdp, flags); |
fbce7497 | 1767 | } |
5d6742b3 | 1768 | WARN_ON(signal_pending(current)); |
12f54c3a | 1769 | } |
fbce7497 | 1770 | |
12f54c3a PM |
1771 | /* |
1772 | * No-CBs grace-period-wait kthread. There is one of these per group | |
1773 | * of CPUs, but only once at least one CPU in that group has come online | |
1774 | * at least once since boot. This kthread checks for newly posted | |
1775 | * callbacks from any of the CPUs it is responsible for, waits for a | |
1776 | * grace period, then awakens all of the rcu_nocb_cb_kthread() instances | |
1777 | * that then have callback-invocation work to do. | |
1778 | */ | |
1779 | static int rcu_nocb_gp_kthread(void *arg) | |
1780 | { | |
1781 | struct rcu_data *rdp = arg; | |
1782 | ||
5d6742b3 | 1783 | for (;;) { |
12f54c3a | 1784 | nocb_gp_wait(rdp); |
5d6742b3 PM |
1785 | cond_resched_tasks_rcu_qs(); |
1786 | } | |
12f54c3a | 1787 | return 0; |
fbce7497 PM |
1788 | } |
1789 | ||
1790 | /* | |
5d6742b3 PM |
1791 | * Invoke any ready callbacks from the corresponding no-CBs CPU, |
1792 | * then, if there are no more, wait for more to appear. | |
fbce7497 | 1793 | */ |
5d6742b3 | 1794 | static void nocb_cb_wait(struct rcu_data *rdp) |
fbce7497 | 1795 | { |
5d6742b3 PM |
1796 | unsigned long flags; |
1797 | bool needwake_gp = false; | |
1798 | struct rcu_node *rnp = rdp->mynode; | |
1799 | ||
1800 | local_irq_save(flags); | |
1801 | rcu_momentary_dyntick_idle(); | |
1802 | local_irq_restore(flags); | |
1803 | local_bh_disable(); | |
1804 | rcu_do_batch(rdp); | |
1805 | local_bh_enable(); | |
1806 | lockdep_assert_irqs_enabled(); | |
81c0b3d7 | 1807 | rcu_nocb_lock_irqsave(rdp, flags); |
5d6742b3 PM |
1808 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
1809 | needwake_gp = rcu_advance_cbs(rdp->mynode, rdp); | |
1810 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ | |
1811 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) { | |
81c0b3d7 | 1812 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 PM |
1813 | if (needwake_gp) |
1814 | rcu_gp_kthread_wake(); | |
1815 | return; | |
1816 | } | |
1817 | ||
f7c9a9b6 | 1818 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("CBSleep")); |
5d6742b3 | 1819 | WRITE_ONCE(rdp->nocb_cb_sleep, true); |
81c0b3d7 | 1820 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 PM |
1821 | if (needwake_gp) |
1822 | rcu_gp_kthread_wake(); | |
12f54c3a | 1823 | swait_event_interruptible_exclusive(rdp->nocb_cb_wq, |
5d6742b3 PM |
1824 | !READ_ONCE(rdp->nocb_cb_sleep)); |
1825 | if (!smp_load_acquire(&rdp->nocb_cb_sleep)) { /* VVV */ | |
1826 | /* ^^^ Ensure CB invocation follows _sleep test. */ | |
1827 | return; | |
fbce7497 | 1828 | } |
12f54c3a PM |
1829 | WARN_ON(signal_pending(current)); |
1830 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty")); | |
fbce7497 PM |
1831 | } |
1832 | ||
3fbfbf7a | 1833 | /* |
5d6742b3 PM |
1834 | * Per-rcu_data kthread, but only for no-CBs CPUs. Repeatedly invoke |
1835 | * nocb_cb_wait() to do the dirty work. | |
3fbfbf7a | 1836 | */ |
12f54c3a | 1837 | static int rcu_nocb_cb_kthread(void *arg) |
3fbfbf7a | 1838 | { |
3fbfbf7a PM |
1839 | struct rcu_data *rdp = arg; |
1840 | ||
5d6742b3 PM |
1841 | // Each pass through this loop does one callback batch, and, |
1842 | // if there are no more ready callbacks, waits for them. | |
3fbfbf7a | 1843 | for (;;) { |
5d6742b3 PM |
1844 | nocb_cb_wait(rdp); |
1845 | cond_resched_tasks_rcu_qs(); | |
3fbfbf7a PM |
1846 | } |
1847 | return 0; | |
1848 | } | |
1849 | ||
96d3fd0d | 1850 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 1851 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 1852 | { |
7d0ae808 | 1853 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
1854 | } |
1855 | ||
1856 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 1857 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 1858 | { |
8be6e1b1 | 1859 | unsigned long flags; |
9fdd3bc9 PM |
1860 | int ndw; |
1861 | ||
81c0b3d7 | 1862 | rcu_nocb_lock_irqsave(rdp, flags); |
8be6e1b1 | 1863 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { |
81c0b3d7 | 1864 | rcu_nocb_unlock_irqrestore(rdp, flags); |
96d3fd0d | 1865 | return; |
8be6e1b1 | 1866 | } |
7d0ae808 | 1867 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 1868 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
5d6742b3 | 1869 | wake_nocb_gp(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
88d1bead | 1870 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
1871 | } |
1872 | ||
8be6e1b1 | 1873 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 1874 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 1875 | { |
fd30b717 KC |
1876 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
1877 | ||
1878 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
1879 | } |
1880 | ||
1881 | /* | |
1882 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
1883 | * This means we do an inexact common-case check. Note that if | |
1884 | * we miss, ->nocb_timer will eventually clean things up. | |
1885 | */ | |
1886 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
1887 | { | |
1888 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
1889 | do_nocb_deferred_wakeup_common(rdp); | |
1890 | } | |
1891 | ||
f4579fc5 PM |
1892 | void __init rcu_init_nohz(void) |
1893 | { | |
1894 | int cpu; | |
ef126206 | 1895 | bool need_rcu_nocb_mask = false; |
e83e73f5 | 1896 | struct rcu_data *rdp; |
f4579fc5 | 1897 | |
f4579fc5 PM |
1898 | #if defined(CONFIG_NO_HZ_FULL) |
1899 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
1900 | need_rcu_nocb_mask = true; | |
1901 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
1902 | ||
84b12b75 | 1903 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
1904 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
1905 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
1906 | return; | |
1907 | } | |
f4579fc5 | 1908 | } |
84b12b75 | 1909 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
1910 | return; |
1911 | ||
f4579fc5 PM |
1912 | #if defined(CONFIG_NO_HZ_FULL) |
1913 | if (tick_nohz_full_running) | |
1914 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
1915 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
1916 | ||
1917 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 1918 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
1919 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
1920 | rcu_nocb_mask); | |
1921 | } | |
3016611e PM |
1922 | if (cpumask_empty(rcu_nocb_mask)) |
1923 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
1924 | else | |
1925 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
1926 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
1927 | if (rcu_nocb_poll) |
1928 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
1929 | ||
e83e73f5 PM |
1930 | for_each_cpu(cpu, rcu_nocb_mask) { |
1931 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
1932 | if (rcu_segcblist_empty(&rdp->cblist)) | |
1933 | rcu_segcblist_init(&rdp->cblist); | |
1934 | rcu_segcblist_offload(&rdp->cblist); | |
1935 | } | |
b97d23c5 | 1936 | rcu_organize_nocb_kthreads(); |
96d3fd0d PM |
1937 | } |
1938 | ||
3fbfbf7a PM |
1939 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
1940 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
1941 | { | |
12f54c3a PM |
1942 | init_swait_queue_head(&rdp->nocb_cb_wq); |
1943 | init_swait_queue_head(&rdp->nocb_gp_wq); | |
8be6e1b1 | 1944 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 1945 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
1946 | } |
1947 | ||
35ce7f29 PM |
1948 | /* |
1949 | * If the specified CPU is a no-CBs CPU that does not already have its | |
12f54c3a PM |
1950 | * rcuo CB kthread, spawn it. Additionally, if the rcuo GP kthread |
1951 | * for this CPU's group has not yet been created, spawn it as well. | |
35ce7f29 | 1952 | */ |
4580b054 | 1953 | static void rcu_spawn_one_nocb_kthread(int cpu) |
35ce7f29 | 1954 | { |
12f54c3a PM |
1955 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
1956 | struct rcu_data *rdp_gp; | |
35ce7f29 PM |
1957 | struct task_struct *t; |
1958 | ||
1959 | /* | |
1960 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
1961 | * then nothing to do. | |
1962 | */ | |
12f54c3a | 1963 | if (!rcu_is_nocb_cpu(cpu) || rdp->nocb_cb_kthread) |
35ce7f29 PM |
1964 | return; |
1965 | ||
6484fe54 | 1966 | /* If we didn't spawn the GP kthread first, reorganize! */ |
12f54c3a PM |
1967 | rdp_gp = rdp->nocb_gp_rdp; |
1968 | if (!rdp_gp->nocb_gp_kthread) { | |
1969 | t = kthread_run(rcu_nocb_gp_kthread, rdp_gp, | |
1970 | "rcuog/%d", rdp_gp->cpu); | |
1971 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo GP kthread, OOM is now expected behavior\n", __func__)) | |
1972 | return; | |
1973 | WRITE_ONCE(rdp_gp->nocb_gp_kthread, t); | |
35ce7f29 PM |
1974 | } |
1975 | ||
0ae86a27 | 1976 | /* Spawn the kthread for this CPU. */ |
12f54c3a | 1977 | t = kthread_run(rcu_nocb_cb_kthread, rdp, |
4580b054 | 1978 | "rcuo%c/%d", rcu_state.abbr, cpu); |
12f54c3a | 1979 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo CB kthread, OOM is now expected behavior\n", __func__)) |
9213784b | 1980 | return; |
12f54c3a PM |
1981 | WRITE_ONCE(rdp->nocb_cb_kthread, t); |
1982 | WRITE_ONCE(rdp->nocb_gp_kthread, rdp_gp->nocb_gp_kthread); | |
35ce7f29 PM |
1983 | } |
1984 | ||
1985 | /* | |
1986 | * If the specified CPU is a no-CBs CPU that does not already have its | |
ad368d15 | 1987 | * rcuo kthread, spawn it. |
35ce7f29 | 1988 | */ |
ad368d15 | 1989 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 | 1990 | { |
35ce7f29 | 1991 | if (rcu_scheduler_fully_active) |
b97d23c5 | 1992 | rcu_spawn_one_nocb_kthread(cpu); |
35ce7f29 PM |
1993 | } |
1994 | ||
1995 | /* | |
1996 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
1997 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
1998 | * non-boot CPUs come online -- if this changes, we will need to add | |
1999 | * some mutual exclusion. | |
2000 | */ | |
2001 | static void __init rcu_spawn_nocb_kthreads(void) | |
2002 | { | |
2003 | int cpu; | |
2004 | ||
2005 | for_each_online_cpu(cpu) | |
ad368d15 | 2006 | rcu_spawn_cpu_nocb_kthread(cpu); |
35ce7f29 PM |
2007 | } |
2008 | ||
6484fe54 | 2009 | /* How many CB CPU IDs per GP kthread? Default of -1 for sqrt(nr_cpu_ids). */ |
f7c612b0 PM |
2010 | static int rcu_nocb_gp_stride = -1; |
2011 | module_param(rcu_nocb_gp_stride, int, 0444); | |
fbce7497 PM |
2012 | |
2013 | /* | |
6484fe54 | 2014 | * Initialize GP-CB relationships for all no-CBs CPU. |
fbce7497 | 2015 | */ |
4580b054 | 2016 | static void __init rcu_organize_nocb_kthreads(void) |
3fbfbf7a PM |
2017 | { |
2018 | int cpu; | |
18cd8c93 | 2019 | bool firsttime = true; |
f7c612b0 | 2020 | int ls = rcu_nocb_gp_stride; |
6484fe54 | 2021 | int nl = 0; /* Next GP kthread. */ |
3fbfbf7a | 2022 | struct rcu_data *rdp; |
0bdc33da | 2023 | struct rcu_data *rdp_gp = NULL; /* Suppress misguided gcc warn. */ |
fbce7497 | 2024 | struct rcu_data *rdp_prev = NULL; |
3fbfbf7a | 2025 | |
84b12b75 | 2026 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2027 | return; |
fbce7497 | 2028 | if (ls == -1) { |
9fcb09bd | 2029 | ls = nr_cpu_ids / int_sqrt(nr_cpu_ids); |
f7c612b0 | 2030 | rcu_nocb_gp_stride = ls; |
fbce7497 PM |
2031 | } |
2032 | ||
2033 | /* | |
9831ce3b PM |
2034 | * Each pass through this loop sets up one rcu_data structure. |
2035 | * Should the corresponding CPU come online in the future, then | |
2036 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2037 | */ |
3fbfbf7a | 2038 | for_each_cpu(cpu, rcu_nocb_mask) { |
da1df50d | 2039 | rdp = per_cpu_ptr(&rcu_data, cpu); |
fbce7497 | 2040 | if (rdp->cpu >= nl) { |
6484fe54 | 2041 | /* New GP kthread, set up for CBs & next GP. */ |
fbce7497 | 2042 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; |
58bf6f77 | 2043 | rdp->nocb_gp_rdp = rdp; |
0bdc33da | 2044 | rdp_gp = rdp; |
18cd8c93 PM |
2045 | if (!firsttime && dump_tree) |
2046 | pr_cont("\n"); | |
2047 | firsttime = false; | |
2048 | pr_alert("%s: No-CB GP kthread CPU %d:", __func__, cpu); | |
fbce7497 | 2049 | } else { |
6484fe54 | 2050 | /* Another CB kthread, link to previous GP kthread. */ |
0bdc33da | 2051 | rdp->nocb_gp_rdp = rdp_gp; |
58bf6f77 | 2052 | rdp_prev->nocb_next_cb_rdp = rdp; |
18cd8c93 | 2053 | pr_alert(" %d", cpu); |
fbce7497 PM |
2054 | } |
2055 | rdp_prev = rdp; | |
3fbfbf7a PM |
2056 | } |
2057 | } | |
2058 | ||
5ab7ab83 PM |
2059 | /* |
2060 | * Bind the current task to the offloaded CPUs. If there are no offloaded | |
2061 | * CPUs, leave the task unbound. Splat if the bind attempt fails. | |
2062 | */ | |
2063 | void rcu_bind_current_to_nocb(void) | |
2064 | { | |
2065 | if (cpumask_available(rcu_nocb_mask) && cpumask_weight(rcu_nocb_mask)) | |
2066 | WARN_ON(sched_setaffinity(current->pid, rcu_nocb_mask)); | |
2067 | } | |
2068 | EXPORT_SYMBOL_GPL(rcu_bind_current_to_nocb); | |
2069 | ||
34ed6246 PM |
2070 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2071 | ||
5d6742b3 PM |
2072 | /* No ->nocb_lock to acquire. */ |
2073 | static void rcu_nocb_lock(struct rcu_data *rdp) | |
d7e29933 | 2074 | { |
5d6742b3 PM |
2075 | } |
2076 | ||
2077 | /* No ->nocb_lock to release. */ | |
2078 | static void rcu_nocb_unlock(struct rcu_data *rdp) | |
2079 | { | |
2080 | } | |
2081 | ||
2082 | /* No ->nocb_lock to release. */ | |
2083 | static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp, | |
2084 | unsigned long flags) | |
2085 | { | |
2086 | local_irq_restore(flags); | |
d7e29933 PM |
2087 | } |
2088 | ||
abedf8e2 | 2089 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2090 | { |
3fbfbf7a PM |
2091 | } |
2092 | ||
abedf8e2 | 2093 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2094 | { |
2095 | return NULL; | |
2096 | } | |
2097 | ||
dae6e64d PM |
2098 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2099 | { | |
2100 | } | |
3fbfbf7a | 2101 | |
5d6742b3 PM |
2102 | static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_empty, |
2103 | unsigned long flags) | |
3fbfbf7a | 2104 | { |
5d6742b3 | 2105 | WARN_ON_ONCE(1); /* Should be dead code! */ |
3fbfbf7a PM |
2106 | } |
2107 | ||
3fbfbf7a PM |
2108 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2109 | { | |
2110 | } | |
2111 | ||
9fdd3bc9 | 2112 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2113 | { |
2114 | return false; | |
2115 | } | |
2116 | ||
2117 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2118 | { | |
2119 | } | |
2120 | ||
ad368d15 | 2121 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 PM |
2122 | { |
2123 | } | |
2124 | ||
2125 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2126 | { |
2127 | } | |
2128 | ||
3fbfbf7a | 2129 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ |
65d798f0 | 2130 | |
a096932f PM |
2131 | /* |
2132 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2133 | * grace-period kthread will do force_quiescent_state() processing? | |
2134 | * The idea is to avoid waking up RCU core processing on such a | |
2135 | * CPU unless the grace period has extended for too long. | |
2136 | * | |
2137 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2138 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f | 2139 | */ |
4580b054 | 2140 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
2141 | { |
2142 | #ifdef CONFIG_NO_HZ_FULL | |
2143 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 2144 | (!rcu_gp_in_progress() || |
4580b054 | 2145 | ULONG_CMP_LT(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 2146 | return true; |
a096932f | 2147 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2148 | return false; |
a096932f | 2149 | } |
5057f55e PM |
2150 | |
2151 | /* | |
265f5f28 | 2152 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2153 | */ |
2154 | static void rcu_bind_gp_kthread(void) | |
2155 | { | |
c0f489d2 | 2156 | if (!tick_nohz_full_enabled()) |
5057f55e | 2157 | return; |
de201559 | 2158 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2159 | } |
176f8f7a PM |
2160 | |
2161 | /* Record the current task on dyntick-idle entry. */ | |
2162 | static void rcu_dynticks_task_enter(void) | |
2163 | { | |
2164 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2165 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2166 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2167 | } | |
2168 | ||
2169 | /* Record no current task on dyntick-idle exit. */ | |
2170 | static void rcu_dynticks_task_exit(void) | |
2171 | { | |
2172 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2173 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2174 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2175 | } |