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); | |
b2b00ddf | 59 | if (qovld != DEFAULT_RCU_QOVLD) |
aa96a93b | 60 | pr_info("\tBoot-time adjustment of callback overload level to %ld.\n", qovld); |
17c7798b PM |
61 | if (jiffies_till_first_fqs != ULONG_MAX) |
62 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
63 | if (jiffies_till_next_fqs != ULONG_MAX) | |
64 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
c06aed0e PM |
65 | if (jiffies_till_sched_qs != ULONG_MAX) |
66 | pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs); | |
17c7798b PM |
67 | if (rcu_kick_kthreads) |
68 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
69 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
70 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 71 | if (gp_preinit_delay) |
17c7798b | 72 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 73 | if (gp_init_delay) |
17c7798b | 74 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 75 | if (gp_cleanup_delay) |
17c7798b | 76 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
48d07c04 SAS |
77 | if (!use_softirq) |
78 | pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n"); | |
17c7798b PM |
79 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) |
80 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 81 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
82 | } |
83 | ||
28f6569a | 84 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 85 | |
63d4c8c9 | 86 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 87 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 88 | |
f41d911f PM |
89 | /* |
90 | * Tell them what RCU they are running. | |
91 | */ | |
0e0fc1c2 | 92 | static void __init rcu_bootup_announce(void) |
f41d911f | 93 | { |
efc151c3 | 94 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 95 | rcu_bootup_announce_oddness(); |
f41d911f PM |
96 | } |
97 | ||
8203d6d0 PM |
98 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
99 | #define RCU_GP_TASKS 0x8 | |
100 | #define RCU_EXP_TASKS 0x4 | |
101 | #define RCU_GP_BLKD 0x2 | |
102 | #define RCU_EXP_BLKD 0x1 | |
103 | ||
104 | /* | |
105 | * Queues a task preempted within an RCU-preempt read-side critical | |
106 | * section into the appropriate location within the ->blkd_tasks list, | |
107 | * depending on the states of any ongoing normal and expedited grace | |
108 | * periods. The ->gp_tasks pointer indicates which element the normal | |
109 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
110 | * indicates which element the expedited grace period is waiting on (again, | |
111 | * NULL if none). If a grace period is waiting on a given element in the | |
112 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
113 | * adding a task to the tail of the list blocks any grace period that is | |
114 | * already waiting on one of the elements. In contrast, adding a task | |
115 | * to the head of the list won't block any grace period that is already | |
116 | * waiting on one of the elements. | |
117 | * | |
118 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
119 | * period wait for a task that is not strictly speaking blocking it. | |
120 | * Given the choice, we needlessly block a normal grace period rather than | |
121 | * blocking an expedited grace period. | |
122 | * | |
123 | * Note that an endless sequence of expedited grace periods still cannot | |
124 | * indefinitely postpone a normal grace period. Eventually, all of the | |
125 | * fixed number of preempted tasks blocking the normal grace period that are | |
126 | * not also blocking the expedited grace period will resume and complete | |
127 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
128 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
129 | * the corresponding expedited grace period will also be the end of the | |
130 | * normal grace period. | |
131 | */ | |
46a5d164 PM |
132 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
133 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
134 | { |
135 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
136 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
137 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
138 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
139 | struct task_struct *t = current; | |
140 | ||
a32e01ee | 141 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 142 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 143 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
144 | /* RCU better not be waiting on newly onlined CPUs! */ |
145 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
146 | rdp->grpmask); | |
ea9b0c8a | 147 | |
8203d6d0 PM |
148 | /* |
149 | * Decide where to queue the newly blocked task. In theory, | |
150 | * this could be an if-statement. In practice, when I tried | |
151 | * that, it was quite messy. | |
152 | */ | |
153 | switch (blkd_state) { | |
154 | case 0: | |
155 | case RCU_EXP_TASKS: | |
156 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
157 | case RCU_GP_TASKS: | |
158 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
159 | ||
160 | /* | |
161 | * Blocking neither GP, or first task blocking the normal | |
162 | * GP but not blocking the already-waiting expedited GP. | |
163 | * Queue at the head of the list to avoid unnecessarily | |
164 | * blocking the already-waiting GPs. | |
165 | */ | |
166 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
167 | break; | |
168 | ||
169 | case RCU_EXP_BLKD: | |
170 | case RCU_GP_BLKD: | |
171 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
172 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
173 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
174 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
175 | ||
176 | /* | |
177 | * First task arriving that blocks either GP, or first task | |
178 | * arriving that blocks the expedited GP (with the normal | |
179 | * GP already waiting), or a task arriving that blocks | |
180 | * both GPs with both GPs already waiting. Queue at the | |
181 | * tail of the list to avoid any GP waiting on any of the | |
182 | * already queued tasks that are not blocking it. | |
183 | */ | |
184 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
185 | break; | |
186 | ||
187 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
188 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
189 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
190 | ||
191 | /* | |
192 | * Second or subsequent task blocking the expedited GP. | |
193 | * The task either does not block the normal GP, or is the | |
194 | * first task blocking the normal GP. Queue just after | |
195 | * the first task blocking the expedited GP. | |
196 | */ | |
197 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
198 | break; | |
199 | ||
200 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
201 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
202 | ||
203 | /* | |
204 | * Second or subsequent task blocking the normal GP. | |
205 | * The task does not block the expedited GP. Queue just | |
206 | * after the first task blocking the normal GP. | |
207 | */ | |
208 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
209 | break; | |
210 | ||
211 | default: | |
212 | ||
213 | /* Yet another exercise in excessive paranoia. */ | |
214 | WARN_ON_ONCE(1); | |
215 | break; | |
216 | } | |
217 | ||
218 | /* | |
219 | * We have now queued the task. If it was the first one to | |
220 | * block either grace period, update the ->gp_tasks and/or | |
221 | * ->exp_tasks pointers, respectively, to reference the newly | |
222 | * blocked tasks. | |
223 | */ | |
4bc8d555 | 224 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
6935c398 | 225 | WRITE_ONCE(rnp->gp_tasks, &t->rcu_node_entry); |
d43a5d32 | 226 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 227 | } |
8203d6d0 | 228 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
314eeb43 | 229 | WRITE_ONCE(rnp->exp_tasks, &t->rcu_node_entry); |
2dee9404 PM |
230 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
231 | !(rnp->qsmask & rdp->grpmask)); | |
232 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
233 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 234 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
235 | |
236 | /* | |
237 | * Report the quiescent state for the expedited GP. This expedited | |
238 | * GP should not be able to end until we report, so there should be | |
239 | * no need to check for a subsequent expedited GP. (Though we are | |
240 | * still in a quiescent state in any case.) | |
241 | */ | |
1bb33644 | 242 | if (blkd_state & RCU_EXP_BLKD && rdp->exp_deferred_qs) |
63d4c8c9 | 243 | rcu_report_exp_rdp(rdp); |
fcc878e4 | 244 | else |
1bb33644 | 245 | WARN_ON_ONCE(rdp->exp_deferred_qs); |
8203d6d0 PM |
246 | } |
247 | ||
f41d911f | 248 | /* |
c7037ff5 PM |
249 | * Record a preemptible-RCU quiescent state for the specified CPU. |
250 | * Note that this does not necessarily mean that the task currently running | |
251 | * on the CPU is in a quiescent state: Instead, it means that the current | |
252 | * grace period need not wait on any RCU read-side critical section that | |
253 | * starts later on this CPU. It also means that if the current task is | |
254 | * in an RCU read-side critical section, it has already added itself to | |
255 | * some leaf rcu_node structure's ->blkd_tasks list. In addition to the | |
256 | * current task, there might be any number of other tasks blocked while | |
257 | * in an RCU read-side critical section. | |
25502a6c | 258 | * |
c7037ff5 | 259 | * Callers to this function must disable preemption. |
f41d911f | 260 | */ |
45975c7d | 261 | static void rcu_qs(void) |
f41d911f | 262 | { |
45975c7d | 263 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n"); |
2280ee5a | 264 | if (__this_cpu_read(rcu_data.cpu_no_qs.s)) { |
284a8c93 | 265 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2280ee5a | 266 | __this_cpu_read(rcu_data.gp_seq), |
284a8c93 | 267 | TPS("cpuqs")); |
2280ee5a | 268 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); |
c98cac60 | 269 | barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */ |
add0d37b | 270 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false); |
284a8c93 | 271 | } |
f41d911f PM |
272 | } |
273 | ||
274 | /* | |
c3422bea PM |
275 | * We have entered the scheduler, and the current task might soon be |
276 | * context-switched away from. If this task is in an RCU read-side | |
277 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
278 | * record that fact, so we enqueue the task on the blkd_tasks list. |
279 | * The task will dequeue itself when it exits the outermost enclosing | |
280 | * RCU read-side critical section. Therefore, the current grace period | |
281 | * cannot be permitted to complete until the blkd_tasks list entries | |
282 | * predating the current grace period drain, in other words, until | |
283 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 284 | * |
46a5d164 | 285 | * Caller must disable interrupts. |
f41d911f | 286 | */ |
45975c7d | 287 | void rcu_note_context_switch(bool preempt) |
f41d911f PM |
288 | { |
289 | struct task_struct *t = current; | |
da1df50d | 290 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
f41d911f PM |
291 | struct rcu_node *rnp; |
292 | ||
45975c7d | 293 | trace_rcu_utilization(TPS("Start context switch")); |
b04db8e1 | 294 | lockdep_assert_irqs_disabled(); |
77339e61 LJ |
295 | WARN_ON_ONCE(!preempt && rcu_preempt_depth() > 0); |
296 | if (rcu_preempt_depth() > 0 && | |
1d082fd0 | 297 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
298 | |
299 | /* Possibly blocking in an RCU read-side critical section. */ | |
f41d911f | 300 | rnp = rdp->mynode; |
46a5d164 | 301 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 302 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 303 | t->rcu_blocked_node = rnp; |
f41d911f PM |
304 | |
305 | /* | |
8203d6d0 PM |
306 | * Verify the CPU's sanity, trace the preemption, and |
307 | * then queue the task as required based on the states | |
308 | * of any ongoing and expedited grace periods. | |
f41d911f | 309 | */ |
0aa04b05 | 310 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 311 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
88d1bead | 312 | trace_rcu_preempt_task(rcu_state.name, |
d4c08f2a PM |
313 | t->pid, |
314 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
315 | ? rnp->gp_seq |
316 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 317 | rcu_preempt_ctxt_queue(rnp, rdp); |
3e310098 PM |
318 | } else { |
319 | rcu_preempt_deferred_qs(t); | |
f41d911f PM |
320 | } |
321 | ||
322 | /* | |
323 | * Either we were not in an RCU read-side critical section to | |
324 | * begin with, or we have now recorded that critical section | |
325 | * globally. Either way, we can now note a quiescent state | |
326 | * for this CPU. Again, if we were in an RCU read-side critical | |
327 | * section, and if that critical section was blocking the current | |
328 | * grace period, then the fact that the task has been enqueued | |
329 | * means that we continue to block the current grace period. | |
330 | */ | |
45975c7d | 331 | rcu_qs(); |
1bb33644 | 332 | if (rdp->exp_deferred_qs) |
63d4c8c9 | 333 | rcu_report_exp_rdp(rdp); |
43766c3e | 334 | rcu_tasks_qs(current, preempt); |
45975c7d | 335 | trace_rcu_utilization(TPS("End context switch")); |
f41d911f | 336 | } |
45975c7d | 337 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
f41d911f | 338 | |
fc2219d4 PM |
339 | /* |
340 | * Check for preempted RCU readers blocking the current grace period | |
341 | * for the specified rcu_node structure. If the caller needs a reliable | |
342 | * answer, it must hold the rcu_node's ->lock. | |
343 | */ | |
27f4d280 | 344 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 345 | { |
6935c398 | 346 | return READ_ONCE(rnp->gp_tasks) != NULL; |
fc2219d4 PM |
347 | } |
348 | ||
5f5fa7ea | 349 | /* limit value for ->rcu_read_lock_nesting. */ |
5f1a6ef3 PM |
350 | #define RCU_NEST_PMAX (INT_MAX / 2) |
351 | ||
77339e61 LJ |
352 | static void rcu_preempt_read_enter(void) |
353 | { | |
354 | current->rcu_read_lock_nesting++; | |
355 | } | |
356 | ||
5f5fa7ea | 357 | static int rcu_preempt_read_exit(void) |
77339e61 | 358 | { |
5f5fa7ea | 359 | return --current->rcu_read_lock_nesting; |
77339e61 LJ |
360 | } |
361 | ||
362 | static void rcu_preempt_depth_set(int val) | |
363 | { | |
364 | current->rcu_read_lock_nesting = val; | |
365 | } | |
366 | ||
0e5da22e PM |
367 | /* |
368 | * Preemptible RCU implementation for rcu_read_lock(). | |
369 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
370 | * if we block. | |
371 | */ | |
372 | void __rcu_read_lock(void) | |
373 | { | |
77339e61 | 374 | rcu_preempt_read_enter(); |
5f1a6ef3 | 375 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
77339e61 | 376 | WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX); |
0e5da22e PM |
377 | barrier(); /* critical section after entry code. */ |
378 | } | |
379 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
380 | ||
381 | /* | |
382 | * Preemptible RCU implementation for rcu_read_unlock(). | |
383 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
384 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
385 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
386 | * in an RCU read-side critical section and other special cases. | |
387 | */ | |
388 | void __rcu_read_unlock(void) | |
389 | { | |
390 | struct task_struct *t = current; | |
391 | ||
5f5fa7ea | 392 | if (rcu_preempt_read_exit() == 0) { |
0e5da22e | 393 | barrier(); /* critical section before exit code. */ |
0e5da22e PM |
394 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) |
395 | rcu_read_unlock_special(t); | |
0e5da22e | 396 | } |
5f1a6ef3 | 397 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
77339e61 | 398 | int rrln = rcu_preempt_depth(); |
0e5da22e | 399 | |
5f5fa7ea | 400 | WARN_ON_ONCE(rrln < 0 || rrln > RCU_NEST_PMAX); |
0e5da22e | 401 | } |
0e5da22e PM |
402 | } |
403 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
404 | ||
12f5f524 PM |
405 | /* |
406 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
407 | * returning NULL if at the end of the list. | |
408 | */ | |
409 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
410 | struct rcu_node *rnp) | |
411 | { | |
412 | struct list_head *np; | |
413 | ||
414 | np = t->rcu_node_entry.next; | |
415 | if (np == &rnp->blkd_tasks) | |
416 | np = NULL; | |
417 | return np; | |
418 | } | |
419 | ||
8af3a5e7 PM |
420 | /* |
421 | * Return true if the specified rcu_node structure has tasks that were | |
422 | * preempted within an RCU read-side critical section. | |
423 | */ | |
424 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
425 | { | |
426 | return !list_empty(&rnp->blkd_tasks); | |
427 | } | |
428 | ||
b668c9cf | 429 | /* |
3e310098 PM |
430 | * Report deferred quiescent states. The deferral time can |
431 | * be quite short, for example, in the case of the call from | |
432 | * rcu_read_unlock_special(). | |
b668c9cf | 433 | */ |
3e310098 PM |
434 | static void |
435 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 436 | { |
b6a932d1 PM |
437 | bool empty_exp; |
438 | bool empty_norm; | |
439 | bool empty_exp_now; | |
12f5f524 | 440 | struct list_head *np; |
abaa93d9 | 441 | bool drop_boost_mutex = false; |
8203d6d0 | 442 | struct rcu_data *rdp; |
f41d911f | 443 | struct rcu_node *rnp; |
1d082fd0 | 444 | union rcu_special special; |
f41d911f | 445 | |
f41d911f | 446 | /* |
8203d6d0 PM |
447 | * If RCU core is waiting for this CPU to exit its critical section, |
448 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 449 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
450 | */ |
451 | special = t->rcu_read_unlock_special; | |
da1df50d | 452 | rdp = this_cpu_ptr(&rcu_data); |
1bb33644 | 453 | if (!special.s && !rdp->exp_deferred_qs) { |
3e310098 PM |
454 | local_irq_restore(flags); |
455 | return; | |
456 | } | |
3717e1e9 LJ |
457 | t->rcu_read_unlock_special.s = 0; |
458 | if (special.b.need_qs) | |
45975c7d | 459 | rcu_qs(); |
f41d911f | 460 | |
8203d6d0 | 461 | /* |
3e310098 PM |
462 | * Respond to a request by an expedited grace period for a |
463 | * quiescent state from this CPU. Note that requests from | |
464 | * tasks are handled when removing the task from the | |
465 | * blocked-tasks list below. | |
8203d6d0 | 466 | */ |
3717e1e9 | 467 | if (rdp->exp_deferred_qs) |
63d4c8c9 | 468 | rcu_report_exp_rdp(rdp); |
8203d6d0 | 469 | |
f41d911f | 470 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 | 471 | if (special.b.blocked) { |
f41d911f | 472 | |
dd5d19ba | 473 | /* |
0a0ba1c9 | 474 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
475 | * now remains queued on the rcu_node corresponding to the |
476 | * CPU it first blocked on, so there is no longer any need | |
477 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 478 | */ |
8ba9153b PM |
479 | rnp = t->rcu_blocked_node; |
480 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
481 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 482 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 483 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 484 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 485 | (!empty_norm || rnp->qsmask)); |
6c7d7dbf | 486 | empty_exp = sync_rcu_exp_done(rnp); |
d9a3da06 | 487 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 488 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 489 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 490 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 491 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 492 | rnp->gp_seq, t->pid); |
12f5f524 | 493 | if (&t->rcu_node_entry == rnp->gp_tasks) |
6935c398 | 494 | WRITE_ONCE(rnp->gp_tasks, np); |
12f5f524 | 495 | if (&t->rcu_node_entry == rnp->exp_tasks) |
314eeb43 | 496 | WRITE_ONCE(rnp->exp_tasks, np); |
727b705b | 497 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
498 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
499 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 | 500 | if (&t->rcu_node_entry == rnp->boost_tasks) |
5822b812 | 501 | WRITE_ONCE(rnp->boost_tasks, np); |
727b705b | 502 | } |
f41d911f PM |
503 | |
504 | /* | |
505 | * If this was the last task on the current list, and if | |
506 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
507 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
508 | * so we must take a snapshot of the expedited state. | |
f41d911f | 509 | */ |
6c7d7dbf | 510 | empty_exp_now = sync_rcu_exp_done(rnp); |
74e871ac | 511 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 512 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 513 | rnp->gp_seq, |
d4c08f2a PM |
514 | 0, rnp->qsmask, |
515 | rnp->level, | |
516 | rnp->grplo, | |
517 | rnp->grphi, | |
518 | !!rnp->gp_tasks); | |
139ad4da | 519 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 520 | } else { |
67c583a7 | 521 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 522 | } |
d9a3da06 | 523 | |
27f4d280 | 524 | /* Unboost if we were boosted. */ |
727b705b | 525 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 526 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 527 | |
d9a3da06 PM |
528 | /* |
529 | * If this was the last task on the expedited lists, | |
530 | * then we need to report up the rcu_node hierarchy. | |
531 | */ | |
389abd48 | 532 | if (!empty_exp && empty_exp_now) |
63d4c8c9 | 533 | rcu_report_exp_rnp(rnp, true); |
b668c9cf PM |
534 | } else { |
535 | local_irq_restore(flags); | |
f41d911f | 536 | } |
f41d911f PM |
537 | } |
538 | ||
3e310098 PM |
539 | /* |
540 | * Is a deferred quiescent-state pending, and are we also not in | |
541 | * an RCU read-side critical section? It is the caller's responsibility | |
542 | * to ensure it is otherwise safe to report any deferred quiescent | |
543 | * states. The reason for this is that it is safe to report a | |
544 | * quiescent state during context switch even though preemption | |
545 | * is disabled. This function cannot be expected to understand these | |
546 | * nuances, so the caller must handle them. | |
547 | */ | |
548 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
549 | { | |
1bb33644 | 550 | return (__this_cpu_read(rcu_data.exp_deferred_qs) || |
3e310098 | 551 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
5f5fa7ea | 552 | rcu_preempt_depth() == 0; |
3e310098 PM |
553 | } |
554 | ||
555 | /* | |
556 | * Report a deferred quiescent state if needed and safe to do so. | |
557 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
558 | * not being in an RCU read-side critical section. The caller must | |
559 | * evaluate safety in terms of interrupt, softirq, and preemption | |
560 | * disabling. | |
561 | */ | |
562 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
563 | { | |
564 | unsigned long flags; | |
3e310098 PM |
565 | |
566 | if (!rcu_preempt_need_deferred_qs(t)) | |
567 | return; | |
3e310098 PM |
568 | local_irq_save(flags); |
569 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
3e310098 PM |
570 | } |
571 | ||
0864f057 PM |
572 | /* |
573 | * Minimal handler to give the scheduler a chance to re-evaluate. | |
574 | */ | |
575 | static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp) | |
576 | { | |
577 | struct rcu_data *rdp; | |
578 | ||
579 | rdp = container_of(iwp, struct rcu_data, defer_qs_iw); | |
580 | rdp->defer_qs_iw_pending = false; | |
581 | } | |
582 | ||
3e310098 PM |
583 | /* |
584 | * Handle special cases during rcu_read_unlock(), such as needing to | |
585 | * notify RCU core processing or task having blocked during the RCU | |
586 | * read-side critical section. | |
587 | */ | |
588 | static void rcu_read_unlock_special(struct task_struct *t) | |
589 | { | |
590 | unsigned long flags; | |
591 | bool preempt_bh_were_disabled = | |
592 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
593 | bool irqs_were_disabled; | |
594 | ||
595 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
596 | if (in_nmi()) | |
597 | return; | |
598 | ||
599 | local_irq_save(flags); | |
600 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 | 601 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
25102de6 PM |
602 | bool exp; |
603 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
604 | struct rcu_node *rnp = rdp->mynode; | |
605 | ||
e4453d8a PM |
606 | exp = (t->rcu_blocked_node && |
607 | READ_ONCE(t->rcu_blocked_node->exp_tasks)) || | |
608 | (rdp->grpmask & READ_ONCE(rnp->expmask)); | |
23634ebc | 609 | // Need to defer quiescent state until everything is enabled. |
e4453d8a PM |
610 | if (use_softirq && (in_irq() || (exp && !irqs_were_disabled))) { |
611 | // Using softirq, safe to awaken, and either the | |
612 | // wakeup is free or there is an expedited GP. | |
05f41571 PM |
613 | raise_softirq_irqoff(RCU_SOFTIRQ); |
614 | } else { | |
23634ebc | 615 | // Enabling BH or preempt does reschedule, so... |
e4453d8a PM |
616 | // Also if no expediting, slow is OK. |
617 | // Plus nohz_full CPUs eventually get tick enabled. | |
05f41571 PM |
618 | set_tsk_need_resched(current); |
619 | set_preempt_need_resched(); | |
d143b3d1 | 620 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && |
0864f057 PM |
621 | !rdp->defer_qs_iw_pending && exp) { |
622 | // Get scheduler to re-evaluate and call hooks. | |
623 | // If !IRQ_WORK, FQS scan will eventually IPI. | |
624 | init_irq_work(&rdp->defer_qs_iw, | |
625 | rcu_preempt_deferred_qs_handler); | |
626 | rdp->defer_qs_iw_pending = true; | |
627 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); | |
628 | } | |
05f41571 | 629 | } |
3e310098 PM |
630 | local_irq_restore(flags); |
631 | return; | |
632 | } | |
633 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
634 | } | |
635 | ||
b0e165c0 PM |
636 | /* |
637 | * Check that the list of blocked tasks for the newly completed grace | |
638 | * period is in fact empty. It is a serious bug to complete a grace | |
639 | * period that still has RCU readers blocked! This function must be | |
03bd2983 | 640 | * invoked -before- updating this rnp's ->gp_seq. |
12f5f524 PM |
641 | * |
642 | * Also, if there are blocked tasks on the list, they automatically | |
643 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 644 | */ |
81ab59a3 | 645 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 646 | { |
c5ebe66c PM |
647 | struct task_struct *t; |
648 | ||
ea9b0c8a | 649 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
03bd2983 | 650 | raw_lockdep_assert_held_rcu_node(rnp); |
4bc8d555 | 651 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 652 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
653 | if (rcu_preempt_has_tasks(rnp) && |
654 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
6935c398 | 655 | WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next); |
c5ebe66c PM |
656 | t = container_of(rnp->gp_tasks, struct task_struct, |
657 | rcu_node_entry); | |
658 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 659 | rnp->gp_seq, t->pid); |
c5ebe66c | 660 | } |
28ecd580 | 661 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
662 | } |
663 | ||
f41d911f | 664 | /* |
c98cac60 PM |
665 | * Check for a quiescent state from the current CPU, including voluntary |
666 | * context switches for Tasks RCU. When a task blocks, the task is | |
667 | * recorded in the corresponding CPU's rcu_node structure, which is checked | |
668 | * elsewhere, hence this function need only check for quiescent states | |
669 | * related to the current CPU, not to those related to tasks. | |
f41d911f | 670 | */ |
c98cac60 | 671 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f PM |
672 | { |
673 | struct task_struct *t = current; | |
674 | ||
45975c7d PM |
675 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
676 | rcu_note_voluntary_context_switch(current); | |
677 | } | |
77339e61 | 678 | if (rcu_preempt_depth() > 0 || |
3e310098 PM |
679 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { |
680 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
681 | if (rcu_preempt_need_deferred_qs(t)) { |
682 | set_tsk_need_resched(t); | |
683 | set_preempt_need_resched(); | |
684 | } | |
3e310098 PM |
685 | } else if (rcu_preempt_need_deferred_qs(t)) { |
686 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
687 | return; | |
5f5fa7ea | 688 | } else if (!WARN_ON_ONCE(rcu_preempt_depth())) { |
45975c7d | 689 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
690 | return; |
691 | } | |
3e310098 PM |
692 | |
693 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
77339e61 | 694 | if (rcu_preempt_depth() > 0 && |
2280ee5a PM |
695 | __this_cpu_read(rcu_data.core_needs_qs) && |
696 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 697 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 698 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 699 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
700 | } |
701 | ||
2439b696 PM |
702 | /* |
703 | * Check for a task exiting while in a preemptible-RCU read-side | |
884157ce PM |
704 | * critical section, clean up if so. No need to issue warnings, as |
705 | * debug_check_no_locks_held() already does this if lockdep is enabled. | |
706 | * Besides, if this function does anything other than just immediately | |
707 | * return, there was a bug of some sort. Spewing warnings from this | |
708 | * function is like as not to simply obscure important prior warnings. | |
2439b696 PM |
709 | */ |
710 | void exit_rcu(void) | |
711 | { | |
712 | struct task_struct *t = current; | |
713 | ||
884157ce | 714 | if (unlikely(!list_empty(¤t->rcu_node_entry))) { |
77339e61 | 715 | rcu_preempt_depth_set(1); |
884157ce | 716 | barrier(); |
add0d37b | 717 | WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true); |
77339e61 LJ |
718 | } else if (unlikely(rcu_preempt_depth())) { |
719 | rcu_preempt_depth_set(1); | |
884157ce | 720 | } else { |
2439b696 | 721 | return; |
884157ce | 722 | } |
2439b696 | 723 | __rcu_read_unlock(); |
3e310098 | 724 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
725 | } |
726 | ||
4bc8d555 PM |
727 | /* |
728 | * Dump the blocked-tasks state, but limit the list dump to the | |
729 | * specified number of elements. | |
730 | */ | |
57738942 | 731 | static void |
81ab59a3 | 732 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 733 | { |
57738942 | 734 | int cpu; |
4bc8d555 PM |
735 | int i; |
736 | struct list_head *lhp; | |
57738942 PM |
737 | bool onl; |
738 | struct rcu_data *rdp; | |
ff3cee39 | 739 | struct rcu_node *rnp1; |
4bc8d555 | 740 | |
ce11fae8 | 741 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 742 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 743 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
8ff37290 | 744 | (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs); |
ff3cee39 PM |
745 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) |
746 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
747 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf | 748 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
065a6db1 | 749 | __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks), |
314eeb43 | 750 | READ_ONCE(rnp->exp_tasks)); |
77cfc7bf | 751 | pr_info("%s: ->blkd_tasks", __func__); |
4bc8d555 PM |
752 | i = 0; |
753 | list_for_each(lhp, &rnp->blkd_tasks) { | |
754 | pr_cont(" %p", lhp); | |
cd6d17b4 | 755 | if (++i >= ncheck) |
4bc8d555 PM |
756 | break; |
757 | } | |
758 | pr_cont("\n"); | |
57738942 | 759 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 760 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 PM |
761 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); |
762 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
763 | cpu, ".o"[onl], | |
764 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
765 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
766 | } | |
4bc8d555 PM |
767 | } |
768 | ||
28f6569a | 769 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f PM |
770 | |
771 | /* | |
772 | * Tell them what RCU they are running. | |
773 | */ | |
0e0fc1c2 | 774 | static void __init rcu_bootup_announce(void) |
f41d911f | 775 | { |
efc151c3 | 776 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 777 | rcu_bootup_announce_oddness(); |
f41d911f PM |
778 | } |
779 | ||
45975c7d | 780 | /* |
90326f05 | 781 | * Note a quiescent state for PREEMPTION=n. Because we do not need to know |
45975c7d PM |
782 | * how many quiescent states passed, just if there was at least one since |
783 | * the start of the grace period, this just sets a flag. The caller must | |
784 | * have disabled preemption. | |
785 | */ | |
786 | static void rcu_qs(void) | |
d28139c4 | 787 | { |
45975c7d PM |
788 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
789 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
790 | return; | |
791 | trace_rcu_grace_period(TPS("rcu_sched"), | |
792 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
793 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
794 | if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) | |
795 | return; | |
796 | __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false); | |
63d4c8c9 | 797 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
d28139c4 PM |
798 | } |
799 | ||
395a2f09 PM |
800 | /* |
801 | * Register an urgently needed quiescent state. If there is an | |
802 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
803 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
804 | * some cases serve for expedited as well as normal grace periods. | |
805 | * Either way, register a lightweight quiescent state. | |
395a2f09 PM |
806 | */ |
807 | void rcu_all_qs(void) | |
808 | { | |
809 | unsigned long flags; | |
810 | ||
2dba13f0 | 811 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
812 | return; |
813 | preempt_disable(); | |
814 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 815 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
816 | preempt_enable(); |
817 | return; | |
818 | } | |
2dba13f0 | 819 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
2dba13f0 | 820 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
821 | local_irq_save(flags); |
822 | rcu_momentary_dyntick_idle(); | |
823 | local_irq_restore(flags); | |
824 | } | |
7e28c5af | 825 | rcu_qs(); |
395a2f09 PM |
826 | preempt_enable(); |
827 | } | |
828 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
829 | ||
cba6d0d6 | 830 | /* |
90326f05 | 831 | * Note a PREEMPTION=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 832 | */ |
45975c7d | 833 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 834 | { |
45975c7d PM |
835 | trace_rcu_utilization(TPS("Start context switch")); |
836 | rcu_qs(); | |
837 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 838 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 839 | goto out; |
2dba13f0 PM |
840 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
841 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 842 | rcu_momentary_dyntick_idle(); |
43766c3e | 843 | rcu_tasks_qs(current, preempt); |
45975c7d PM |
844 | out: |
845 | trace_rcu_utilization(TPS("End context switch")); | |
cba6d0d6 | 846 | } |
45975c7d | 847 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 848 | |
fc2219d4 | 849 | /* |
6cc68793 | 850 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
851 | * RCU readers. |
852 | */ | |
27f4d280 | 853 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
854 | { |
855 | return 0; | |
856 | } | |
857 | ||
8af3a5e7 PM |
858 | /* |
859 | * Because there is no preemptible RCU, there can be no readers blocked. | |
860 | */ | |
861 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 862 | { |
8af3a5e7 | 863 | return false; |
b668c9cf PM |
864 | } |
865 | ||
3e310098 PM |
866 | /* |
867 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
868 | * states. | |
869 | */ | |
870 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
871 | { | |
872 | return false; | |
873 | } | |
874 | static void rcu_preempt_deferred_qs(struct task_struct *t) { } | |
875 | ||
b0e165c0 | 876 | /* |
6cc68793 | 877 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
878 | * so there is no need to check for blocked tasks. So check only for |
879 | * bogus qsmask values. | |
b0e165c0 | 880 | */ |
81ab59a3 | 881 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 882 | { |
49e29126 | 883 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
884 | } |
885 | ||
f41d911f | 886 | /* |
c98cac60 PM |
887 | * Check to see if this CPU is in a non-context-switch quiescent state, |
888 | * namely user mode and idle loop. | |
f41d911f | 889 | */ |
c98cac60 | 890 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f | 891 | { |
45975c7d | 892 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 893 | |
45975c7d PM |
894 | /* |
895 | * Get here if this CPU took its interrupt from user | |
896 | * mode or from the idle loop, and if this is not a | |
897 | * nested interrupt. In this case, the CPU is in | |
898 | * a quiescent state, so note it. | |
899 | * | |
900 | * No memory barrier is required here because rcu_qs() | |
901 | * references only CPU-local variables that other CPUs | |
902 | * neither access nor modify, at least not while the | |
903 | * corresponding CPU is online. | |
904 | */ | |
905 | ||
906 | rcu_qs(); | |
907 | } | |
e74f4c45 | 908 | } |
e74f4c45 | 909 | |
2439b696 PM |
910 | /* |
911 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
912 | * while in preemptible RCU read-side critical sections. | |
913 | */ | |
914 | void exit_rcu(void) | |
915 | { | |
916 | } | |
917 | ||
4bc8d555 PM |
918 | /* |
919 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
920 | */ | |
57738942 | 921 | static void |
81ab59a3 | 922 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
923 | { |
924 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
925 | } | |
926 | ||
28f6569a | 927 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 928 | |
48d07c04 SAS |
929 | /* |
930 | * If boosting, set rcuc kthreads to realtime priority. | |
931 | */ | |
932 | static void rcu_cpu_kthread_setup(unsigned int cpu) | |
933 | { | |
27f4d280 | 934 | #ifdef CONFIG_RCU_BOOST |
48d07c04 | 935 | struct sched_param sp; |
27f4d280 | 936 | |
48d07c04 SAS |
937 | sp.sched_priority = kthread_prio; |
938 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
939 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
5d01bbd1 TG |
940 | } |
941 | ||
48d07c04 SAS |
942 | #ifdef CONFIG_RCU_BOOST |
943 | ||
27f4d280 PM |
944 | /* |
945 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
946 | * or ->boost_tasks, advancing the pointer to the next task in the | |
947 | * ->blkd_tasks list. | |
948 | * | |
949 | * Note that irqs must be enabled: boosting the task can block. | |
950 | * Returns 1 if there are more tasks needing to be boosted. | |
951 | */ | |
952 | static int rcu_boost(struct rcu_node *rnp) | |
953 | { | |
954 | unsigned long flags; | |
27f4d280 PM |
955 | struct task_struct *t; |
956 | struct list_head *tb; | |
957 | ||
7d0ae808 PM |
958 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
959 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
960 | return 0; /* Nothing left to boost. */ |
961 | ||
2a67e741 | 962 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
963 | |
964 | /* | |
965 | * Recheck under the lock: all tasks in need of boosting | |
966 | * might exit their RCU read-side critical sections on their own. | |
967 | */ | |
968 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 969 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
970 | return 0; |
971 | } | |
972 | ||
973 | /* | |
974 | * Preferentially boost tasks blocking expedited grace periods. | |
975 | * This cannot starve the normal grace periods because a second | |
976 | * expedited grace period must boost all blocked tasks, including | |
977 | * those blocking the pre-existing normal grace period. | |
978 | */ | |
bec06785 | 979 | if (rnp->exp_tasks != NULL) |
27f4d280 | 980 | tb = rnp->exp_tasks; |
bec06785 | 981 | else |
27f4d280 PM |
982 | tb = rnp->boost_tasks; |
983 | ||
984 | /* | |
985 | * We boost task t by manufacturing an rt_mutex that appears to | |
986 | * be held by task t. We leave a pointer to that rt_mutex where | |
987 | * task t can find it, and task t will release the mutex when it | |
988 | * exits its outermost RCU read-side critical section. Then | |
989 | * simply acquiring this artificial rt_mutex will boost task | |
990 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
991 | * | |
992 | * Note that task t must acquire rnp->lock to remove itself from | |
993 | * the ->blkd_tasks list, which it will do from exit() if from | |
994 | * nowhere else. We therefore are guaranteed that task t will | |
995 | * stay around at least until we drop rnp->lock. Note that | |
996 | * rnp->lock also resolves races between our priority boosting | |
997 | * and task t's exiting its outermost RCU read-side critical | |
998 | * section. | |
999 | */ | |
1000 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1001 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1002 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1003 | /* Lock only for side effect: boosts task t's priority. */ |
1004 | rt_mutex_lock(&rnp->boost_mtx); | |
1005 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1006 | |
7d0ae808 PM |
1007 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1008 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1009 | } |
1010 | ||
27f4d280 | 1011 | /* |
bc17ea10 | 1012 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1013 | */ |
1014 | static int rcu_boost_kthread(void *arg) | |
1015 | { | |
1016 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1017 | int spincnt = 0; | |
1018 | int more2boost; | |
1019 | ||
f7f7bac9 | 1020 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1021 | for (;;) { |
3ca3b0e2 | 1022 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING); |
f7f7bac9 | 1023 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
065a6db1 PM |
1024 | rcu_wait(READ_ONCE(rnp->boost_tasks) || |
1025 | READ_ONCE(rnp->exp_tasks)); | |
f7f7bac9 | 1026 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
3ca3b0e2 | 1027 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING); |
27f4d280 PM |
1028 | more2boost = rcu_boost(rnp); |
1029 | if (more2boost) | |
1030 | spincnt++; | |
1031 | else | |
1032 | spincnt = 0; | |
1033 | if (spincnt > 10) { | |
3ca3b0e2 | 1034 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING); |
f7f7bac9 | 1035 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1036 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1037 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1038 | spincnt = 0; |
1039 | } | |
1040 | } | |
1217ed1b | 1041 | /* NOTREACHED */ |
f7f7bac9 | 1042 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1043 | return 0; |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Check to see if it is time to start boosting RCU readers that are | |
1048 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1049 | * kthread to start boosting them. If there is an expedited grace | |
1050 | * period in progress, it is always time to boost. | |
1051 | * | |
b065a853 PM |
1052 | * The caller must hold rnp->lock, which this function releases. |
1053 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1054 | * about it going away. | |
27f4d280 | 1055 | */ |
1217ed1b | 1056 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1057 | __releases(rnp->lock) |
27f4d280 | 1058 | { |
a32e01ee | 1059 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1060 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1061 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1062 | return; |
0ea1f2eb | 1063 | } |
27f4d280 PM |
1064 | if (rnp->exp_tasks != NULL || |
1065 | (rnp->gp_tasks != NULL && | |
1066 | rnp->boost_tasks == NULL && | |
1067 | rnp->qsmask == 0 && | |
7b241311 | 1068 | (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld))) { |
27f4d280 | 1069 | if (rnp->exp_tasks == NULL) |
5822b812 | 1070 | WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks); |
67c583a7 | 1071 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
a2badefa | 1072 | rcu_wake_cond(rnp->boost_kthread_task, |
3ca3b0e2 | 1073 | READ_ONCE(rnp->boost_kthread_status)); |
1217ed1b | 1074 | } else { |
67c583a7 | 1075 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1076 | } |
27f4d280 PM |
1077 | } |
1078 | ||
dff1672d PM |
1079 | /* |
1080 | * Is the current CPU running the RCU-callbacks kthread? | |
1081 | * Caller must have preemption disabled. | |
1082 | */ | |
1083 | static bool rcu_is_callbacks_kthread(void) | |
1084 | { | |
37f62d7c | 1085 | return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current; |
dff1672d PM |
1086 | } |
1087 | ||
27f4d280 PM |
1088 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1089 | ||
1090 | /* | |
1091 | * Do priority-boost accounting for the start of a new grace period. | |
1092 | */ | |
1093 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1094 | { | |
1095 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1096 | } | |
1097 | ||
27f4d280 PM |
1098 | /* |
1099 | * Create an RCU-boost kthread for the specified node if one does not | |
1100 | * already exist. We only create this kthread for preemptible RCU. | |
1101 | * Returns zero if all is well, a negated errno otherwise. | |
1102 | */ | |
3545832f | 1103 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 | 1104 | { |
6dbfdc14 | 1105 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1106 | unsigned long flags; |
1107 | struct sched_param sp; | |
1108 | struct task_struct *t; | |
1109 | ||
6dbfdc14 | 1110 | if (!IS_ENABLED(CONFIG_PREEMPT_RCU)) |
3545832f | 1111 | return; |
5d01bbd1 | 1112 | |
0aa04b05 | 1113 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
3545832f | 1114 | return; |
5d01bbd1 | 1115 | |
6dbfdc14 | 1116 | rcu_state.boost = 1; |
3545832f | 1117 | |
27f4d280 | 1118 | if (rnp->boost_kthread_task != NULL) |
3545832f BP |
1119 | return; |
1120 | ||
27f4d280 | 1121 | t = kthread_create(rcu_boost_kthread, (void *)rnp, |
5b61b0ba | 1122 | "rcub/%d", rnp_index); |
3545832f BP |
1123 | if (WARN_ON_ONCE(IS_ERR(t))) |
1124 | return; | |
1125 | ||
2a67e741 | 1126 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1127 | rnp->boost_kthread_task = t; |
67c583a7 | 1128 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1129 | sp.sched_priority = kthread_prio; |
27f4d280 | 1130 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1131 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1132 | } |
1133 | ||
f8b7fc6b PM |
1134 | /* |
1135 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1136 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1137 | * held, so the value of rnp->qsmaskinit will be stable. | |
1138 | * | |
1139 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1140 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1141 | * this function allows the kthread to execute on any CPU. | |
1142 | */ | |
5d01bbd1 | 1143 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1144 | { |
5d01bbd1 | 1145 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1146 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1147 | cpumask_var_t cm; |
1148 | int cpu; | |
f8b7fc6b | 1149 | |
5d01bbd1 | 1150 | if (!t) |
f8b7fc6b | 1151 | return; |
5d01bbd1 | 1152 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1153 | return; |
bc75e999 MR |
1154 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1155 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1156 | cpu != outgoingcpu) | |
f8b7fc6b | 1157 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1158 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1159 | cpumask_setall(cm); |
5d01bbd1 | 1160 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1161 | free_cpumask_var(cm); |
1162 | } | |
1163 | ||
f8b7fc6b | 1164 | /* |
9386c0b7 | 1165 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1166 | */ |
9386c0b7 | 1167 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1168 | { |
f8b7fc6b PM |
1169 | struct rcu_node *rnp; |
1170 | ||
aedf4ba9 | 1171 | rcu_for_each_leaf_node(rnp) |
3545832f | 1172 | rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b | 1173 | } |
f8b7fc6b | 1174 | |
49fb4c62 | 1175 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1176 | { |
da1df50d | 1177 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
f8b7fc6b PM |
1178 | struct rcu_node *rnp = rdp->mynode; |
1179 | ||
1180 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1181 | if (rcu_scheduler_fully_active) |
3545832f | 1182 | rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b PM |
1183 | } |
1184 | ||
27f4d280 PM |
1185 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1186 | ||
1217ed1b | 1187 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1188 | __releases(rnp->lock) |
27f4d280 | 1189 | { |
67c583a7 | 1190 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1191 | } |
1192 | ||
dff1672d PM |
1193 | static bool rcu_is_callbacks_kthread(void) |
1194 | { | |
1195 | return false; | |
1196 | } | |
1197 | ||
27f4d280 PM |
1198 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1199 | { | |
1200 | } | |
1201 | ||
5d01bbd1 | 1202 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1203 | { |
1204 | } | |
1205 | ||
9386c0b7 | 1206 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1207 | { |
b0d30417 | 1208 | } |
b0d30417 | 1209 | |
49fb4c62 | 1210 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1211 | { |
1212 | } | |
1213 | ||
27f4d280 PM |
1214 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1215 | ||
8bd93a2c PM |
1216 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1217 | ||
1218 | /* | |
0bd55c69 PM |
1219 | * Check to see if any future non-offloaded RCU-related work will need |
1220 | * to be done by the current CPU, even if none need be done immediately, | |
1221 | * returning 1 if so. This function is part of the RCU implementation; | |
1222 | * it is -not- an exported member of the RCU API. | |
8bd93a2c | 1223 | * |
0ae86a27 PM |
1224 | * Because we not have RCU_FAST_NO_HZ, just check whether or not this |
1225 | * CPU has RCU callbacks queued. | |
8bd93a2c | 1226 | */ |
c1ad348b | 1227 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1228 | { |
c1ad348b | 1229 | *nextevt = KTIME_MAX; |
0bd55c69 PM |
1230 | return !rcu_segcblist_empty(&this_cpu_ptr(&rcu_data)->cblist) && |
1231 | !rcu_segcblist_is_offloaded(&this_cpu_ptr(&rcu_data)->cblist); | |
7cb92499 PM |
1232 | } |
1233 | ||
1234 | /* | |
1235 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1236 | * after it. | |
1237 | */ | |
8fa7845d | 1238 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1239 | { |
1240 | } | |
1241 | ||
aea1b35e | 1242 | /* |
a858af28 | 1243 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1244 | * is nothing. |
1245 | */ | |
198bbf81 | 1246 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1247 | { |
1248 | } | |
1249 | ||
8bd93a2c PM |
1250 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1251 | ||
f23f7fa1 PM |
1252 | /* |
1253 | * This code is invoked when a CPU goes idle, at which point we want | |
1254 | * to have the CPU do everything required for RCU so that it can enter | |
77a40f97 | 1255 | * the energy-efficient dyntick-idle mode. |
f23f7fa1 | 1256 | * |
77a40f97 | 1257 | * The following preprocessor symbol controls this: |
f23f7fa1 | 1258 | * |
f23f7fa1 PM |
1259 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1260 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1261 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1262 | * benchmarkers who might otherwise be tempted to set this to a large | |
1263 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1264 | * system. And if you are -that- concerned about energy efficiency, | |
1265 | * just power the system down and be done with it! | |
1266 | * | |
77a40f97 | 1267 | * The value below works well in practice. If future workloads require |
f23f7fa1 PM |
1268 | * adjustment, they can be converted into kernel config parameters, though |
1269 | * making the state machine smarter might be a better option. | |
1270 | */ | |
e84c48ae | 1271 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
f23f7fa1 | 1272 | |
5e44ce35 PM |
1273 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1274 | module_param(rcu_idle_gp_delay, int, 0644); | |
486e2593 | 1275 | |
486e2593 | 1276 | /* |
0ae86a27 PM |
1277 | * Try to advance callbacks on the current CPU, but only if it has been |
1278 | * awhile since the last time we did so. Afterwards, if there are any | |
1279 | * callbacks ready for immediate invocation, return true. | |
486e2593 | 1280 | */ |
f1f399d1 | 1281 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1282 | { |
c0f4dfd4 | 1283 | bool cbs_ready = false; |
5998a75a | 1284 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1285 | struct rcu_node *rnp; |
486e2593 | 1286 | |
c229828c | 1287 | /* Exit early if we advanced recently. */ |
5998a75a | 1288 | if (jiffies == rdp->last_advance_all) |
d0bc90fd | 1289 | return false; |
5998a75a | 1290 | rdp->last_advance_all = jiffies; |
c229828c | 1291 | |
b97d23c5 | 1292 | rnp = rdp->mynode; |
486e2593 | 1293 | |
b97d23c5 PM |
1294 | /* |
1295 | * Don't bother checking unless a grace period has | |
1296 | * completed since we last checked and there are | |
1297 | * callbacks not yet ready to invoke. | |
1298 | */ | |
1299 | if ((rcu_seq_completed_gp(rdp->gp_seq, | |
1300 | rcu_seq_current(&rnp->gp_seq)) || | |
1301 | unlikely(READ_ONCE(rdp->gpwrap))) && | |
1302 | rcu_segcblist_pend_cbs(&rdp->cblist)) | |
1303 | note_gp_changes(rdp); | |
1304 | ||
1305 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
1306 | cbs_ready = true; | |
c0f4dfd4 | 1307 | return cbs_ready; |
486e2593 PM |
1308 | } |
1309 | ||
aa9b1630 | 1310 | /* |
c0f4dfd4 PM |
1311 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1312 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
77a40f97 | 1313 | * caller about what to set the timeout. |
aa9b1630 | 1314 | * |
c0f4dfd4 | 1315 | * The caller must have disabled interrupts. |
aa9b1630 | 1316 | */ |
c1ad348b | 1317 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1318 | { |
5998a75a | 1319 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c1ad348b | 1320 | unsigned long dj; |
aa9b1630 | 1321 | |
b04db8e1 | 1322 | lockdep_assert_irqs_disabled(); |
3382adbc | 1323 | |
0bd55c69 PM |
1324 | /* If no non-offloaded callbacks, RCU doesn't need the CPU. */ |
1325 | if (rcu_segcblist_empty(&rdp->cblist) || | |
1326 | rcu_segcblist_is_offloaded(&this_cpu_ptr(&rcu_data)->cblist)) { | |
c1ad348b | 1327 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1328 | return 0; |
1329 | } | |
c0f4dfd4 PM |
1330 | |
1331 | /* Attempt to advance callbacks. */ | |
1332 | if (rcu_try_advance_all_cbs()) { | |
1333 | /* Some ready to invoke, so initiate later invocation. */ | |
1334 | invoke_rcu_core(); | |
aa9b1630 PM |
1335 | return 1; |
1336 | } | |
5998a75a | 1337 | rdp->last_accelerate = jiffies; |
c0f4dfd4 | 1338 | |
77a40f97 JFG |
1339 | /* Request timer and round. */ |
1340 | dj = round_up(rcu_idle_gp_delay + jiffies, rcu_idle_gp_delay) - jiffies; | |
1341 | ||
c1ad348b | 1342 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1343 | return 0; |
1344 | } | |
1345 | ||
21e52e15 | 1346 | /* |
77a40f97 JFG |
1347 | * Prepare a CPU for idle from an RCU perspective. The first major task is to |
1348 | * sense whether nohz mode has been enabled or disabled via sysfs. The second | |
1349 | * major task is to accelerate (that is, assign grace-period numbers to) any | |
1350 | * recently arrived callbacks. | |
aea1b35e PM |
1351 | * |
1352 | * The caller must have disabled interrupts. | |
8bd93a2c | 1353 | */ |
198bbf81 | 1354 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1355 | { |
48a7639c | 1356 | bool needwake; |
0fd79e75 | 1357 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1358 | struct rcu_node *rnp; |
9d2ad243 PM |
1359 | int tne; |
1360 | ||
b04db8e1 | 1361 | lockdep_assert_irqs_disabled(); |
ce5215c1 | 1362 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) |
3382adbc PM |
1363 | return; |
1364 | ||
9d2ad243 | 1365 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1366 | tne = READ_ONCE(tick_nohz_active); |
0fd79e75 | 1367 | if (tne != rdp->tick_nohz_enabled_snap) { |
260e1e4f | 1368 | if (!rcu_segcblist_empty(&rdp->cblist)) |
9d2ad243 | 1369 | invoke_rcu_core(); /* force nohz to see update. */ |
0fd79e75 | 1370 | rdp->tick_nohz_enabled_snap = tne; |
9d2ad243 PM |
1371 | return; |
1372 | } | |
1373 | if (!tne) | |
1374 | return; | |
f511fc62 | 1375 | |
3084f2f8 | 1376 | /* |
c0f4dfd4 PM |
1377 | * If we have not yet accelerated this jiffy, accelerate all |
1378 | * callbacks on this CPU. | |
3084f2f8 | 1379 | */ |
5998a75a | 1380 | if (rdp->last_accelerate == jiffies) |
aea1b35e | 1381 | return; |
5998a75a | 1382 | rdp->last_accelerate = jiffies; |
b97d23c5 | 1383 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
c0f4dfd4 | 1384 | rnp = rdp->mynode; |
2a67e741 | 1385 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
02f50142 | 1386 | needwake = rcu_accelerate_cbs(rnp, rdp); |
67c583a7 | 1387 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c | 1388 | if (needwake) |
532c00c9 | 1389 | rcu_gp_kthread_wake(); |
77e38ed3 | 1390 | } |
c0f4dfd4 | 1391 | } |
3084f2f8 | 1392 | |
c0f4dfd4 PM |
1393 | /* |
1394 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1395 | * any grace periods that elapsed while the CPU was idle, and if any | |
1396 | * callbacks are now ready to invoke, initiate invocation. | |
1397 | */ | |
8fa7845d | 1398 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1399 | { |
ce5215c1 PM |
1400 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
1401 | ||
b04db8e1 | 1402 | lockdep_assert_irqs_disabled(); |
ce5215c1 | 1403 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) |
aea1b35e | 1404 | return; |
7a497c96 PM |
1405 | if (rcu_try_advance_all_cbs()) |
1406 | invoke_rcu_core(); | |
8bd93a2c PM |
1407 | } |
1408 | ||
1409 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ | |
a858af28 | 1410 | |
3fbfbf7a PM |
1411 | #ifdef CONFIG_RCU_NOCB_CPU |
1412 | ||
1413 | /* | |
1414 | * Offload callback processing from the boot-time-specified set of CPUs | |
a9fefdb2 PM |
1415 | * specified by rcu_nocb_mask. For the CPUs in the set, there are kthreads |
1416 | * created that pull the callbacks from the corresponding CPU, wait for | |
1417 | * a grace period to elapse, and invoke the callbacks. These kthreads | |
6484fe54 PM |
1418 | * are organized into GP kthreads, which manage incoming callbacks, wait for |
1419 | * grace periods, and awaken CB kthreads, and the CB kthreads, which only | |
1420 | * invoke callbacks. Each GP kthread invokes its own CBs. The no-CBs CPUs | |
1421 | * do a wake_up() on their GP kthread when they insert a callback into any | |
a9fefdb2 PM |
1422 | * empty list, unless the rcu_nocb_poll boot parameter has been specified, |
1423 | * in which case each kthread actively polls its CPU. (Which isn't so great | |
1424 | * for energy efficiency, but which does reduce RCU's overhead on that CPU.) | |
3fbfbf7a PM |
1425 | * |
1426 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1427 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1428 | * running CPU-bound user-mode computations. | |
1429 | * | |
a9fefdb2 PM |
1430 | * Offloading of callbacks can also be used as an energy-efficiency |
1431 | * measure because CPUs with no RCU callbacks queued are more aggressive | |
1432 | * about entering dyntick-idle mode. | |
3fbfbf7a PM |
1433 | */ |
1434 | ||
1435 | ||
497e4260 PM |
1436 | /* |
1437 | * Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. | |
1438 | * The string after the "rcu_nocbs=" is either "all" for all CPUs, or a | |
1439 | * comma-separated list of CPUs and/or CPU ranges. If an invalid list is | |
1440 | * given, a warning is emitted and all CPUs are offloaded. | |
1441 | */ | |
3fbfbf7a PM |
1442 | static int __init rcu_nocb_setup(char *str) |
1443 | { | |
1444 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
da8739f2 PM |
1445 | if (!strcasecmp(str, "all")) |
1446 | cpumask_setall(rcu_nocb_mask); | |
1447 | else | |
497e4260 PM |
1448 | if (cpulist_parse(str, rcu_nocb_mask)) { |
1449 | pr_warn("rcu_nocbs= bad CPU range, all CPUs set\n"); | |
1450 | cpumask_setall(rcu_nocb_mask); | |
1451 | } | |
3fbfbf7a PM |
1452 | return 1; |
1453 | } | |
1454 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1455 | ||
1b0048a4 PG |
1456 | static int __init parse_rcu_nocb_poll(char *arg) |
1457 | { | |
5455a7f6 | 1458 | rcu_nocb_poll = true; |
1b0048a4 PG |
1459 | return 0; |
1460 | } | |
1461 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1462 | ||
5d6742b3 | 1463 | /* |
d1b222c6 PM |
1464 | * Don't bother bypassing ->cblist if the call_rcu() rate is low. |
1465 | * After all, the main point of bypassing is to avoid lock contention | |
1466 | * on ->nocb_lock, which only can happen at high call_rcu() rates. | |
5d6742b3 | 1467 | */ |
d1b222c6 PM |
1468 | int nocb_nobypass_lim_per_jiffy = 16 * 1000 / HZ; |
1469 | module_param(nocb_nobypass_lim_per_jiffy, int, 0); | |
1470 | ||
1471 | /* | |
1472 | * Acquire the specified rcu_data structure's ->nocb_bypass_lock. If the | |
1473 | * lock isn't immediately available, increment ->nocb_lock_contended to | |
1474 | * flag the contention. | |
1475 | */ | |
1476 | static void rcu_nocb_bypass_lock(struct rcu_data *rdp) | |
9ced4548 | 1477 | __acquires(&rdp->nocb_bypass_lock) |
5d6742b3 | 1478 | { |
81c0b3d7 | 1479 | lockdep_assert_irqs_disabled(); |
d1b222c6 | 1480 | if (raw_spin_trylock(&rdp->nocb_bypass_lock)) |
81c0b3d7 PM |
1481 | return; |
1482 | atomic_inc(&rdp->nocb_lock_contended); | |
6aacd88d | 1483 | WARN_ON_ONCE(smp_processor_id() != rdp->cpu); |
81c0b3d7 | 1484 | smp_mb__after_atomic(); /* atomic_inc() before lock. */ |
d1b222c6 | 1485 | raw_spin_lock(&rdp->nocb_bypass_lock); |
81c0b3d7 PM |
1486 | smp_mb__before_atomic(); /* atomic_dec() after lock. */ |
1487 | atomic_dec(&rdp->nocb_lock_contended); | |
1488 | } | |
1489 | ||
1490 | /* | |
1491 | * Spinwait until the specified rcu_data structure's ->nocb_lock is | |
1492 | * not contended. Please note that this is extremely special-purpose, | |
1493 | * relying on the fact that at most two kthreads and one CPU contend for | |
1494 | * this lock, and also that the two kthreads are guaranteed to have frequent | |
1495 | * grace-period-duration time intervals between successive acquisitions | |
1496 | * of the lock. This allows us to use an extremely simple throttling | |
1497 | * mechanism, and further to apply it only to the CPU doing floods of | |
1498 | * call_rcu() invocations. Don't try this at home! | |
1499 | */ | |
1500 | static void rcu_nocb_wait_contended(struct rcu_data *rdp) | |
1501 | { | |
6aacd88d PM |
1502 | WARN_ON_ONCE(smp_processor_id() != rdp->cpu); |
1503 | while (WARN_ON_ONCE(atomic_read(&rdp->nocb_lock_contended))) | |
81c0b3d7 | 1504 | cpu_relax(); |
5d6742b3 PM |
1505 | } |
1506 | ||
d1b222c6 PM |
1507 | /* |
1508 | * Conditionally acquire the specified rcu_data structure's | |
1509 | * ->nocb_bypass_lock. | |
1510 | */ | |
1511 | static bool rcu_nocb_bypass_trylock(struct rcu_data *rdp) | |
1512 | { | |
1513 | lockdep_assert_irqs_disabled(); | |
1514 | return raw_spin_trylock(&rdp->nocb_bypass_lock); | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Release the specified rcu_data structure's ->nocb_bypass_lock. | |
1519 | */ | |
1520 | static void rcu_nocb_bypass_unlock(struct rcu_data *rdp) | |
92c0b889 | 1521 | __releases(&rdp->nocb_bypass_lock) |
d1b222c6 PM |
1522 | { |
1523 | lockdep_assert_irqs_disabled(); | |
1524 | raw_spin_unlock(&rdp->nocb_bypass_lock); | |
1525 | } | |
1526 | ||
1527 | /* | |
1528 | * Acquire the specified rcu_data structure's ->nocb_lock, but only | |
1529 | * if it corresponds to a no-CBs CPU. | |
1530 | */ | |
1531 | static void rcu_nocb_lock(struct rcu_data *rdp) | |
1532 | { | |
1533 | lockdep_assert_irqs_disabled(); | |
1534 | if (!rcu_segcblist_is_offloaded(&rdp->cblist)) | |
1535 | return; | |
1536 | raw_spin_lock(&rdp->nocb_lock); | |
1537 | } | |
1538 | ||
5d6742b3 PM |
1539 | /* |
1540 | * Release the specified rcu_data structure's ->nocb_lock, but only | |
1541 | * if it corresponds to a no-CBs CPU. | |
1542 | */ | |
1543 | static void rcu_nocb_unlock(struct rcu_data *rdp) | |
1544 | { | |
1545 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) { | |
1546 | lockdep_assert_irqs_disabled(); | |
1547 | raw_spin_unlock(&rdp->nocb_lock); | |
1548 | } | |
1549 | } | |
1550 | ||
1551 | /* | |
1552 | * Release the specified rcu_data structure's ->nocb_lock and restore | |
1553 | * interrupts, but only if it corresponds to a no-CBs CPU. | |
1554 | */ | |
1555 | static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp, | |
1556 | unsigned long flags) | |
1557 | { | |
1558 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) { | |
1559 | lockdep_assert_irqs_disabled(); | |
1560 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
1561 | } else { | |
1562 | local_irq_restore(flags); | |
1563 | } | |
1564 | } | |
1565 | ||
d1b222c6 PM |
1566 | /* Lockdep check that ->cblist may be safely accessed. */ |
1567 | static void rcu_lockdep_assert_cblist_protected(struct rcu_data *rdp) | |
1568 | { | |
1569 | lockdep_assert_irqs_disabled(); | |
13817dd5 | 1570 | if (rcu_segcblist_is_offloaded(&rdp->cblist)) |
d1b222c6 PM |
1571 | lockdep_assert_held(&rdp->nocb_lock); |
1572 | } | |
1573 | ||
dae6e64d | 1574 | /* |
0446be48 PM |
1575 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1576 | * grace period. | |
dae6e64d | 1577 | */ |
abedf8e2 | 1578 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1579 | { |
abedf8e2 | 1580 | swake_up_all(sq); |
dae6e64d PM |
1581 | } |
1582 | ||
abedf8e2 | 1583 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1584 | { |
e0da2374 | 1585 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1586 | } |
1587 | ||
dae6e64d | 1588 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1589 | { |
abedf8e2 PG |
1590 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1591 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1592 | } |
1593 | ||
24342c96 | 1594 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1595 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1596 | { |
84b12b75 | 1597 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1598 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1599 | return false; | |
1600 | } | |
1601 | ||
fbce7497 | 1602 | /* |
6484fe54 | 1603 | * Kick the GP kthread for this NOCB group. Caller holds ->nocb_lock |
8be6e1b1 | 1604 | * and this function releases it. |
fbce7497 | 1605 | */ |
5d6742b3 | 1606 | static void wake_nocb_gp(struct rcu_data *rdp, bool force, |
5f675ba6 | 1607 | unsigned long flags) |
8be6e1b1 | 1608 | __releases(rdp->nocb_lock) |
fbce7497 | 1609 | { |
d1b222c6 | 1610 | bool needwake = false; |
5f675ba6 | 1611 | struct rcu_data *rdp_gp = rdp->nocb_gp_rdp; |
fbce7497 | 1612 | |
8be6e1b1 | 1613 | lockdep_assert_held(&rdp->nocb_lock); |
5f675ba6 | 1614 | if (!READ_ONCE(rdp_gp->nocb_gp_kthread)) { |
d1b222c6 PM |
1615 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
1616 | TPS("AlreadyAwake")); | |
81c0b3d7 | 1617 | rcu_nocb_unlock_irqrestore(rdp, flags); |
fbce7497 | 1618 | return; |
8be6e1b1 | 1619 | } |
d1b222c6 PM |
1620 | del_timer(&rdp->nocb_timer); |
1621 | rcu_nocb_unlock_irqrestore(rdp, flags); | |
1622 | raw_spin_lock_irqsave(&rdp_gp->nocb_gp_lock, flags); | |
1623 | if (force || READ_ONCE(rdp_gp->nocb_gp_sleep)) { | |
5d6742b3 | 1624 | WRITE_ONCE(rdp_gp->nocb_gp_sleep, false); |
d1b222c6 PM |
1625 | needwake = true; |
1626 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DoWake")); | |
fbce7497 | 1627 | } |
d1b222c6 PM |
1628 | raw_spin_unlock_irqrestore(&rdp_gp->nocb_gp_lock, flags); |
1629 | if (needwake) | |
1630 | wake_up_process(rdp_gp->nocb_gp_kthread); | |
fbce7497 PM |
1631 | } |
1632 | ||
8be6e1b1 | 1633 | /* |
6484fe54 PM |
1634 | * Arrange to wake the GP kthread for this NOCB group at some future |
1635 | * time when it is safe to do so. | |
8be6e1b1 | 1636 | */ |
0d52a665 PM |
1637 | static void wake_nocb_gp_defer(struct rcu_data *rdp, int waketype, |
1638 | const char *reason) | |
8be6e1b1 | 1639 | { |
8be6e1b1 PM |
1640 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) |
1641 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
383e1332 PM |
1642 | if (rdp->nocb_defer_wakeup < waketype) |
1643 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
88d1bead | 1644 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, reason); |
d7e29933 PM |
1645 | } |
1646 | ||
d1b222c6 PM |
1647 | /* |
1648 | * Flush the ->nocb_bypass queue into ->cblist, enqueuing rhp if non-NULL. | |
1649 | * However, if there is a callback to be enqueued and if ->nocb_bypass | |
1650 | * proves to be initially empty, just return false because the no-CB GP | |
1651 | * kthread may need to be awakened in this case. | |
1652 | * | |
1653 | * Note that this function always returns true if rhp is NULL. | |
1654 | */ | |
1655 | static bool rcu_nocb_do_flush_bypass(struct rcu_data *rdp, struct rcu_head *rhp, | |
1656 | unsigned long j) | |
1657 | { | |
1658 | struct rcu_cblist rcl; | |
1659 | ||
1660 | WARN_ON_ONCE(!rcu_segcblist_is_offloaded(&rdp->cblist)); | |
1661 | rcu_lockdep_assert_cblist_protected(rdp); | |
1662 | lockdep_assert_held(&rdp->nocb_bypass_lock); | |
1663 | if (rhp && !rcu_cblist_n_cbs(&rdp->nocb_bypass)) { | |
1664 | raw_spin_unlock(&rdp->nocb_bypass_lock); | |
1665 | return false; | |
1666 | } | |
1667 | /* Note: ->cblist.len already accounts for ->nocb_bypass contents. */ | |
1668 | if (rhp) | |
1669 | rcu_segcblist_inc_len(&rdp->cblist); /* Must precede enqueue. */ | |
1670 | rcu_cblist_flush_enqueue(&rcl, &rdp->nocb_bypass, rhp); | |
1671 | rcu_segcblist_insert_pend_cbs(&rdp->cblist, &rcl); | |
1672 | WRITE_ONCE(rdp->nocb_bypass_first, j); | |
1673 | rcu_nocb_bypass_unlock(rdp); | |
1674 | return true; | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * Flush the ->nocb_bypass queue into ->cblist, enqueuing rhp if non-NULL. | |
1679 | * However, if there is a callback to be enqueued and if ->nocb_bypass | |
1680 | * proves to be initially empty, just return false because the no-CB GP | |
1681 | * kthread may need to be awakened in this case. | |
1682 | * | |
1683 | * Note that this function always returns true if rhp is NULL. | |
1684 | */ | |
1685 | static bool rcu_nocb_flush_bypass(struct rcu_data *rdp, struct rcu_head *rhp, | |
1686 | unsigned long j) | |
1687 | { | |
1688 | if (!rcu_segcblist_is_offloaded(&rdp->cblist)) | |
1689 | return true; | |
1690 | rcu_lockdep_assert_cblist_protected(rdp); | |
1691 | rcu_nocb_bypass_lock(rdp); | |
1692 | return rcu_nocb_do_flush_bypass(rdp, rhp, j); | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * If the ->nocb_bypass_lock is immediately available, flush the | |
1697 | * ->nocb_bypass queue into ->cblist. | |
1698 | */ | |
1699 | static void rcu_nocb_try_flush_bypass(struct rcu_data *rdp, unsigned long j) | |
1700 | { | |
1701 | rcu_lockdep_assert_cblist_protected(rdp); | |
1702 | if (!rcu_segcblist_is_offloaded(&rdp->cblist) || | |
1703 | !rcu_nocb_bypass_trylock(rdp)) | |
1704 | return; | |
1705 | WARN_ON_ONCE(!rcu_nocb_do_flush_bypass(rdp, NULL, j)); | |
1706 | } | |
1707 | ||
1708 | /* | |
1709 | * See whether it is appropriate to use the ->nocb_bypass list in order | |
1710 | * to control contention on ->nocb_lock. A limited number of direct | |
1711 | * enqueues are permitted into ->cblist per jiffy. If ->nocb_bypass | |
1712 | * is non-empty, further callbacks must be placed into ->nocb_bypass, | |
1713 | * otherwise rcu_barrier() breaks. Use rcu_nocb_flush_bypass() to switch | |
1714 | * back to direct use of ->cblist. However, ->nocb_bypass should not be | |
1715 | * used if ->cblist is empty, because otherwise callbacks can be stranded | |
1716 | * on ->nocb_bypass because we cannot count on the current CPU ever again | |
1717 | * invoking call_rcu(). The general rule is that if ->nocb_bypass is | |
1718 | * non-empty, the corresponding no-CBs grace-period kthread must not be | |
1719 | * in an indefinite sleep state. | |
1720 | * | |
1721 | * Finally, it is not permitted to use the bypass during early boot, | |
1722 | * as doing so would confuse the auto-initialization code. Besides | |
1723 | * which, there is no point in worrying about lock contention while | |
1724 | * there is only one CPU in operation. | |
1725 | */ | |
1726 | static bool rcu_nocb_try_bypass(struct rcu_data *rdp, struct rcu_head *rhp, | |
1727 | bool *was_alldone, unsigned long flags) | |
1728 | { | |
1729 | unsigned long c; | |
1730 | unsigned long cur_gp_seq; | |
1731 | unsigned long j = jiffies; | |
1732 | long ncbs = rcu_cblist_n_cbs(&rdp->nocb_bypass); | |
1733 | ||
1734 | if (!rcu_segcblist_is_offloaded(&rdp->cblist)) { | |
1735 | *was_alldone = !rcu_segcblist_pend_cbs(&rdp->cblist); | |
1736 | return false; /* Not offloaded, no bypassing. */ | |
1737 | } | |
1738 | lockdep_assert_irqs_disabled(); | |
1739 | ||
1740 | // Don't use ->nocb_bypass during early boot. | |
1741 | if (rcu_scheduler_active != RCU_SCHEDULER_RUNNING) { | |
1742 | rcu_nocb_lock(rdp); | |
1743 | WARN_ON_ONCE(rcu_cblist_n_cbs(&rdp->nocb_bypass)); | |
1744 | *was_alldone = !rcu_segcblist_pend_cbs(&rdp->cblist); | |
1745 | return false; | |
1746 | } | |
1747 | ||
1748 | // If we have advanced to a new jiffy, reset counts to allow | |
1749 | // moving back from ->nocb_bypass to ->cblist. | |
1750 | if (j == rdp->nocb_nobypass_last) { | |
1751 | c = rdp->nocb_nobypass_count + 1; | |
1752 | } else { | |
1753 | WRITE_ONCE(rdp->nocb_nobypass_last, j); | |
1754 | c = rdp->nocb_nobypass_count - nocb_nobypass_lim_per_jiffy; | |
1755 | if (ULONG_CMP_LT(rdp->nocb_nobypass_count, | |
1756 | nocb_nobypass_lim_per_jiffy)) | |
1757 | c = 0; | |
1758 | else if (c > nocb_nobypass_lim_per_jiffy) | |
1759 | c = nocb_nobypass_lim_per_jiffy; | |
1760 | } | |
1761 | WRITE_ONCE(rdp->nocb_nobypass_count, c); | |
1762 | ||
1763 | // If there hasn't yet been all that many ->cblist enqueues | |
1764 | // this jiffy, tell the caller to enqueue onto ->cblist. But flush | |
1765 | // ->nocb_bypass first. | |
1766 | if (rdp->nocb_nobypass_count < nocb_nobypass_lim_per_jiffy) { | |
1767 | rcu_nocb_lock(rdp); | |
1768 | *was_alldone = !rcu_segcblist_pend_cbs(&rdp->cblist); | |
1769 | if (*was_alldone) | |
1770 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, | |
1771 | TPS("FirstQ")); | |
1772 | WARN_ON_ONCE(!rcu_nocb_flush_bypass(rdp, NULL, j)); | |
1773 | WARN_ON_ONCE(rcu_cblist_n_cbs(&rdp->nocb_bypass)); | |
1774 | return false; // Caller must enqueue the callback. | |
1775 | } | |
1776 | ||
1777 | // If ->nocb_bypass has been used too long or is too full, | |
1778 | // flush ->nocb_bypass to ->cblist. | |
1779 | if ((ncbs && j != READ_ONCE(rdp->nocb_bypass_first)) || | |
1780 | ncbs >= qhimark) { | |
1781 | rcu_nocb_lock(rdp); | |
1782 | if (!rcu_nocb_flush_bypass(rdp, rhp, j)) { | |
1783 | *was_alldone = !rcu_segcblist_pend_cbs(&rdp->cblist); | |
1784 | if (*was_alldone) | |
1785 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, | |
1786 | TPS("FirstQ")); | |
1787 | WARN_ON_ONCE(rcu_cblist_n_cbs(&rdp->nocb_bypass)); | |
1788 | return false; // Caller must enqueue the callback. | |
1789 | } | |
1790 | if (j != rdp->nocb_gp_adv_time && | |
1791 | rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq) && | |
1792 | rcu_seq_done(&rdp->mynode->gp_seq, cur_gp_seq)) { | |
1793 | rcu_advance_cbs_nowake(rdp->mynode, rdp); | |
1794 | rdp->nocb_gp_adv_time = j; | |
1795 | } | |
1796 | rcu_nocb_unlock_irqrestore(rdp, flags); | |
1797 | return true; // Callback already enqueued. | |
1798 | } | |
1799 | ||
1800 | // We need to use the bypass. | |
1801 | rcu_nocb_wait_contended(rdp); | |
1802 | rcu_nocb_bypass_lock(rdp); | |
1803 | ncbs = rcu_cblist_n_cbs(&rdp->nocb_bypass); | |
1804 | rcu_segcblist_inc_len(&rdp->cblist); /* Must precede enqueue. */ | |
1805 | rcu_cblist_enqueue(&rdp->nocb_bypass, rhp); | |
1806 | if (!ncbs) { | |
1807 | WRITE_ONCE(rdp->nocb_bypass_first, j); | |
1808 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("FirstBQ")); | |
1809 | } | |
1810 | rcu_nocb_bypass_unlock(rdp); | |
1811 | smp_mb(); /* Order enqueue before wake. */ | |
1812 | if (ncbs) { | |
1813 | local_irq_restore(flags); | |
1814 | } else { | |
1815 | // No-CBs GP kthread might be indefinitely asleep, if so, wake. | |
1816 | rcu_nocb_lock(rdp); // Rare during call_rcu() flood. | |
1817 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) { | |
1818 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, | |
1819 | TPS("FirstBQwake")); | |
1820 | __call_rcu_nocb_wake(rdp, true, flags); | |
1821 | } else { | |
1822 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, | |
1823 | TPS("FirstBQnoWake")); | |
1824 | rcu_nocb_unlock_irqrestore(rdp, flags); | |
1825 | } | |
1826 | } | |
1827 | return true; // Callback already enqueued. | |
1828 | } | |
1829 | ||
3fbfbf7a | 1830 | /* |
5d6742b3 PM |
1831 | * Awaken the no-CBs grace-period kthead if needed, either due to it |
1832 | * legitimately being asleep or due to overload conditions. | |
3fbfbf7a PM |
1833 | * |
1834 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1835 | */ | |
5d6742b3 PM |
1836 | static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone, |
1837 | unsigned long flags) | |
1838 | __releases(rdp->nocb_lock) | |
3fbfbf7a | 1839 | { |
296181d7 PM |
1840 | unsigned long cur_gp_seq; |
1841 | unsigned long j; | |
ce0a825e | 1842 | long len; |
3fbfbf7a PM |
1843 | struct task_struct *t; |
1844 | ||
5d6742b3 | 1845 | // If we are being polled or there is no kthread, just leave. |
12f54c3a | 1846 | t = READ_ONCE(rdp->nocb_gp_kthread); |
25e03a74 | 1847 | if (rcu_nocb_poll || !t) { |
88d1bead | 1848 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9261dd0d | 1849 | TPS("WakeNotPoll")); |
5d6742b3 | 1850 | rcu_nocb_unlock_irqrestore(rdp, flags); |
3fbfbf7a | 1851 | return; |
9261dd0d | 1852 | } |
5d6742b3 PM |
1853 | // Need to actually to a wakeup. |
1854 | len = rcu_segcblist_n_cbs(&rdp->cblist); | |
1855 | if (was_alldone) { | |
aeeacd9d | 1856 | rdp->qlen_last_fqs_check = len; |
96d3fd0d | 1857 | if (!irqs_disabled_flags(flags)) { |
fbce7497 | 1858 | /* ... if queue was empty ... */ |
5d6742b3 | 1859 | wake_nocb_gp(rdp, false, flags); |
88d1bead | 1860 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
96d3fd0d PM |
1861 | TPS("WakeEmpty")); |
1862 | } else { | |
0d52a665 PM |
1863 | wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE, |
1864 | TPS("WakeEmptyIsDeferred")); | |
5d6742b3 | 1865 | rcu_nocb_unlock_irqrestore(rdp, flags); |
96d3fd0d | 1866 | } |
3fbfbf7a | 1867 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { |
fbce7497 | 1868 | /* ... or if many callbacks queued. */ |
aeeacd9d | 1869 | rdp->qlen_last_fqs_check = len; |
296181d7 PM |
1870 | j = jiffies; |
1871 | if (j != rdp->nocb_gp_adv_time && | |
1872 | rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq) && | |
1873 | rcu_seq_done(&rdp->mynode->gp_seq, cur_gp_seq)) { | |
faca5c25 | 1874 | rcu_advance_cbs_nowake(rdp->mynode, rdp); |
296181d7 PM |
1875 | rdp->nocb_gp_adv_time = j; |
1876 | } | |
f48fe4c5 PM |
1877 | smp_mb(); /* Enqueue before timer_pending(). */ |
1878 | if ((rdp->nocb_cb_sleep || | |
1879 | !rcu_segcblist_ready_cbs(&rdp->cblist)) && | |
1880 | !timer_pending(&rdp->nocb_bypass_timer)) | |
273f0340 PM |
1881 | wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE_FORCE, |
1882 | TPS("WakeOvfIsDeferred")); | |
273f0340 | 1883 | rcu_nocb_unlock_irqrestore(rdp, flags); |
9261dd0d | 1884 | } else { |
88d1bead | 1885 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WakeNot")); |
5d6742b3 | 1886 | rcu_nocb_unlock_irqrestore(rdp, flags); |
3fbfbf7a PM |
1887 | } |
1888 | return; | |
1889 | } | |
1890 | ||
d1b222c6 PM |
1891 | /* Wake up the no-CBs GP kthread to flush ->nocb_bypass. */ |
1892 | static void do_nocb_bypass_wakeup_timer(struct timer_list *t) | |
1893 | { | |
1894 | unsigned long flags; | |
1895 | struct rcu_data *rdp = from_timer(rdp, t, nocb_bypass_timer); | |
1896 | ||
1897 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("Timer")); | |
1898 | rcu_nocb_lock_irqsave(rdp, flags); | |
f48fe4c5 | 1899 | smp_mb__after_spinlock(); /* Timer expire before wakeup. */ |
d1b222c6 PM |
1900 | __call_rcu_nocb_wake(rdp, true, flags); |
1901 | } | |
1902 | ||
3fbfbf7a | 1903 | /* |
5d6742b3 PM |
1904 | * No-CBs GP kthreads come here to wait for additional callbacks to show up |
1905 | * or for grace periods to end. | |
fbce7497 | 1906 | */ |
12f54c3a | 1907 | static void nocb_gp_wait(struct rcu_data *my_rdp) |
fbce7497 | 1908 | { |
d1b222c6 PM |
1909 | bool bypass = false; |
1910 | long bypass_ncbs; | |
5d6742b3 PM |
1911 | int __maybe_unused cpu = my_rdp->cpu; |
1912 | unsigned long cur_gp_seq; | |
8be6e1b1 | 1913 | unsigned long flags; |
b8889c9c | 1914 | bool gotcbs = false; |
d1b222c6 | 1915 | unsigned long j = jiffies; |
969974e5 | 1916 | bool needwait_gp = false; // This prevents actual uninitialized use. |
5d6742b3 PM |
1917 | bool needwake; |
1918 | bool needwake_gp; | |
fbce7497 | 1919 | struct rcu_data *rdp; |
5d6742b3 | 1920 | struct rcu_node *rnp; |
969974e5 | 1921 | unsigned long wait_gp_seq = 0; // Suppress "use uninitialized" warning. |
3d05031a | 1922 | bool wasempty = false; |
fbce7497 PM |
1923 | |
1924 | /* | |
5d6742b3 PM |
1925 | * Each pass through the following loop checks for CBs and for the |
1926 | * nearest grace period (if any) to wait for next. The CB kthreads | |
1927 | * and the global grace-period kthread are awakened if needed. | |
fbce7497 | 1928 | */ |
58bf6f77 | 1929 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_cb_rdp) { |
d1b222c6 PM |
1930 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("Check")); |
1931 | rcu_nocb_lock_irqsave(rdp, flags); | |
1932 | bypass_ncbs = rcu_cblist_n_cbs(&rdp->nocb_bypass); | |
1933 | if (bypass_ncbs && | |
1934 | (time_after(j, READ_ONCE(rdp->nocb_bypass_first) + 1) || | |
1935 | bypass_ncbs > 2 * qhimark)) { | |
1936 | // Bypass full or old, so flush it. | |
1937 | (void)rcu_nocb_try_flush_bypass(rdp, j); | |
1938 | bypass_ncbs = rcu_cblist_n_cbs(&rdp->nocb_bypass); | |
1939 | } else if (!bypass_ncbs && rcu_segcblist_empty(&rdp->cblist)) { | |
1940 | rcu_nocb_unlock_irqrestore(rdp, flags); | |
5d6742b3 | 1941 | continue; /* No callbacks here, try next. */ |
d1b222c6 PM |
1942 | } |
1943 | if (bypass_ncbs) { | |
1944 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, | |
1945 | TPS("Bypass")); | |
1946 | bypass = true; | |
1947 | } | |
5d6742b3 | 1948 | rnp = rdp->mynode; |
d1b222c6 PM |
1949 | if (bypass) { // Avoid race with first bypass CB. |
1950 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, | |
1951 | RCU_NOCB_WAKE_NOT); | |
1952 | del_timer(&my_rdp->nocb_timer); | |
1953 | } | |
1954 | // Advance callbacks if helpful and low contention. | |
1955 | needwake_gp = false; | |
1956 | if (!rcu_segcblist_restempty(&rdp->cblist, | |
1957 | RCU_NEXT_READY_TAIL) || | |
1958 | (rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq) && | |
1959 | rcu_seq_done(&rnp->gp_seq, cur_gp_seq))) { | |
1960 | raw_spin_lock_rcu_node(rnp); /* irqs disabled. */ | |
1961 | needwake_gp = rcu_advance_cbs(rnp, rdp); | |
3d05031a PM |
1962 | wasempty = rcu_segcblist_restempty(&rdp->cblist, |
1963 | RCU_NEXT_READY_TAIL); | |
d1b222c6 PM |
1964 | raw_spin_unlock_rcu_node(rnp); /* irqs disabled. */ |
1965 | } | |
5d6742b3 | 1966 | // Need to wait on some grace period? |
3d05031a PM |
1967 | WARN_ON_ONCE(wasempty && |
1968 | !rcu_segcblist_restempty(&rdp->cblist, | |
d1b222c6 | 1969 | RCU_NEXT_READY_TAIL)); |
5d6742b3 PM |
1970 | if (rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq)) { |
1971 | if (!needwait_gp || | |
1972 | ULONG_CMP_LT(cur_gp_seq, wait_gp_seq)) | |
1973 | wait_gp_seq = cur_gp_seq; | |
1974 | needwait_gp = true; | |
d1b222c6 PM |
1975 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
1976 | TPS("NeedWaitGP")); | |
8be6e1b1 | 1977 | } |
5d6742b3 PM |
1978 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) { |
1979 | needwake = rdp->nocb_cb_sleep; | |
1980 | WRITE_ONCE(rdp->nocb_cb_sleep, false); | |
1981 | smp_mb(); /* CB invocation -after- GP end. */ | |
1982 | } else { | |
1983 | needwake = false; | |
8be6e1b1 | 1984 | } |
81c0b3d7 | 1985 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 | 1986 | if (needwake) { |
12f54c3a | 1987 | swake_up_one(&rdp->nocb_cb_wq); |
5d6742b3 | 1988 | gotcbs = true; |
fbce7497 | 1989 | } |
5d6742b3 PM |
1990 | if (needwake_gp) |
1991 | rcu_gp_kthread_wake(); | |
1992 | } | |
1993 | ||
f7a81b12 PM |
1994 | my_rdp->nocb_gp_bypass = bypass; |
1995 | my_rdp->nocb_gp_gp = needwait_gp; | |
1996 | my_rdp->nocb_gp_seq = needwait_gp ? wait_gp_seq : 0; | |
d1b222c6 PM |
1997 | if (bypass && !rcu_nocb_poll) { |
1998 | // At least one child with non-empty ->nocb_bypass, so set | |
1999 | // timer in order to avoid stranding its callbacks. | |
2000 | raw_spin_lock_irqsave(&my_rdp->nocb_gp_lock, flags); | |
2001 | mod_timer(&my_rdp->nocb_bypass_timer, j + 2); | |
2002 | raw_spin_unlock_irqrestore(&my_rdp->nocb_gp_lock, flags); | |
2003 | } | |
5d6742b3 PM |
2004 | if (rcu_nocb_poll) { |
2005 | /* Polling, so trace if first poll in the series. */ | |
2006 | if (gotcbs) | |
2007 | trace_rcu_nocb_wake(rcu_state.name, cpu, TPS("Poll")); | |
2008 | schedule_timeout_interruptible(1); | |
2009 | } else if (!needwait_gp) { | |
2010 | /* Wait for callbacks to appear. */ | |
2011 | trace_rcu_nocb_wake(rcu_state.name, cpu, TPS("Sleep")); | |
2012 | swait_event_interruptible_exclusive(my_rdp->nocb_gp_wq, | |
2013 | !READ_ONCE(my_rdp->nocb_gp_sleep)); | |
d1b222c6 | 2014 | trace_rcu_nocb_wake(rcu_state.name, cpu, TPS("EndSleep")); |
5d6742b3 PM |
2015 | } else { |
2016 | rnp = my_rdp->mynode; | |
2017 | trace_rcu_this_gp(rnp, my_rdp, wait_gp_seq, TPS("StartWait")); | |
2018 | swait_event_interruptible_exclusive( | |
2019 | rnp->nocb_gp_wq[rcu_seq_ctr(wait_gp_seq) & 0x1], | |
2020 | rcu_seq_done(&rnp->gp_seq, wait_gp_seq) || | |
2021 | !READ_ONCE(my_rdp->nocb_gp_sleep)); | |
2022 | trace_rcu_this_gp(rnp, my_rdp, wait_gp_seq, TPS("EndWait")); | |
2023 | } | |
2024 | if (!rcu_nocb_poll) { | |
4fd8c5f1 | 2025 | raw_spin_lock_irqsave(&my_rdp->nocb_gp_lock, flags); |
d1b222c6 PM |
2026 | if (bypass) |
2027 | del_timer(&my_rdp->nocb_bypass_timer); | |
5d6742b3 | 2028 | WRITE_ONCE(my_rdp->nocb_gp_sleep, true); |
4fd8c5f1 | 2029 | raw_spin_unlock_irqrestore(&my_rdp->nocb_gp_lock, flags); |
fbce7497 | 2030 | } |
f7a81b12 | 2031 | my_rdp->nocb_gp_seq = -1; |
5d6742b3 | 2032 | WARN_ON(signal_pending(current)); |
12f54c3a | 2033 | } |
fbce7497 | 2034 | |
12f54c3a PM |
2035 | /* |
2036 | * No-CBs grace-period-wait kthread. There is one of these per group | |
2037 | * of CPUs, but only once at least one CPU in that group has come online | |
2038 | * at least once since boot. This kthread checks for newly posted | |
2039 | * callbacks from any of the CPUs it is responsible for, waits for a | |
2040 | * grace period, then awakens all of the rcu_nocb_cb_kthread() instances | |
2041 | * that then have callback-invocation work to do. | |
2042 | */ | |
2043 | static int rcu_nocb_gp_kthread(void *arg) | |
2044 | { | |
2045 | struct rcu_data *rdp = arg; | |
2046 | ||
5d6742b3 | 2047 | for (;;) { |
f7a81b12 | 2048 | WRITE_ONCE(rdp->nocb_gp_loops, rdp->nocb_gp_loops + 1); |
12f54c3a | 2049 | nocb_gp_wait(rdp); |
5d6742b3 PM |
2050 | cond_resched_tasks_rcu_qs(); |
2051 | } | |
12f54c3a | 2052 | return 0; |
fbce7497 PM |
2053 | } |
2054 | ||
2055 | /* | |
5d6742b3 PM |
2056 | * Invoke any ready callbacks from the corresponding no-CBs CPU, |
2057 | * then, if there are no more, wait for more to appear. | |
fbce7497 | 2058 | */ |
5d6742b3 | 2059 | static void nocb_cb_wait(struct rcu_data *rdp) |
fbce7497 | 2060 | { |
1d5a81c1 | 2061 | unsigned long cur_gp_seq; |
5d6742b3 PM |
2062 | unsigned long flags; |
2063 | bool needwake_gp = false; | |
2064 | struct rcu_node *rnp = rdp->mynode; | |
2065 | ||
2066 | local_irq_save(flags); | |
2067 | rcu_momentary_dyntick_idle(); | |
2068 | local_irq_restore(flags); | |
2069 | local_bh_disable(); | |
2070 | rcu_do_batch(rdp); | |
2071 | local_bh_enable(); | |
2072 | lockdep_assert_irqs_enabled(); | |
81c0b3d7 | 2073 | rcu_nocb_lock_irqsave(rdp, flags); |
1d5a81c1 PM |
2074 | if (rcu_segcblist_nextgp(&rdp->cblist, &cur_gp_seq) && |
2075 | rcu_seq_done(&rnp->gp_seq, cur_gp_seq) && | |
2076 | raw_spin_trylock_rcu_node(rnp)) { /* irqs already disabled. */ | |
523bddd5 PM |
2077 | needwake_gp = rcu_advance_cbs(rdp->mynode, rdp); |
2078 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ | |
2079 | } | |
5d6742b3 | 2080 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) { |
81c0b3d7 | 2081 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 PM |
2082 | if (needwake_gp) |
2083 | rcu_gp_kthread_wake(); | |
2084 | return; | |
2085 | } | |
2086 | ||
f7c9a9b6 | 2087 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("CBSleep")); |
5d6742b3 | 2088 | WRITE_ONCE(rdp->nocb_cb_sleep, true); |
81c0b3d7 | 2089 | rcu_nocb_unlock_irqrestore(rdp, flags); |
5d6742b3 PM |
2090 | if (needwake_gp) |
2091 | rcu_gp_kthread_wake(); | |
12f54c3a | 2092 | swait_event_interruptible_exclusive(rdp->nocb_cb_wq, |
5d6742b3 PM |
2093 | !READ_ONCE(rdp->nocb_cb_sleep)); |
2094 | if (!smp_load_acquire(&rdp->nocb_cb_sleep)) { /* VVV */ | |
2095 | /* ^^^ Ensure CB invocation follows _sleep test. */ | |
2096 | return; | |
fbce7497 | 2097 | } |
12f54c3a PM |
2098 | WARN_ON(signal_pending(current)); |
2099 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty")); | |
fbce7497 PM |
2100 | } |
2101 | ||
3fbfbf7a | 2102 | /* |
5d6742b3 PM |
2103 | * Per-rcu_data kthread, but only for no-CBs CPUs. Repeatedly invoke |
2104 | * nocb_cb_wait() to do the dirty work. | |
3fbfbf7a | 2105 | */ |
12f54c3a | 2106 | static int rcu_nocb_cb_kthread(void *arg) |
3fbfbf7a | 2107 | { |
3fbfbf7a PM |
2108 | struct rcu_data *rdp = arg; |
2109 | ||
5d6742b3 PM |
2110 | // Each pass through this loop does one callback batch, and, |
2111 | // if there are no more ready callbacks, waits for them. | |
3fbfbf7a | 2112 | for (;;) { |
5d6742b3 PM |
2113 | nocb_cb_wait(rdp); |
2114 | cond_resched_tasks_rcu_qs(); | |
3fbfbf7a PM |
2115 | } |
2116 | return 0; | |
2117 | } | |
2118 | ||
96d3fd0d | 2119 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2120 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2121 | { |
7d0ae808 | 2122 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2123 | } |
2124 | ||
2125 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2126 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2127 | { |
8be6e1b1 | 2128 | unsigned long flags; |
9fdd3bc9 PM |
2129 | int ndw; |
2130 | ||
81c0b3d7 | 2131 | rcu_nocb_lock_irqsave(rdp, flags); |
8be6e1b1 | 2132 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { |
81c0b3d7 | 2133 | rcu_nocb_unlock_irqrestore(rdp, flags); |
96d3fd0d | 2134 | return; |
8be6e1b1 | 2135 | } |
7d0ae808 | 2136 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2137 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
5d6742b3 | 2138 | wake_nocb_gp(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
88d1bead | 2139 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2140 | } |
2141 | ||
8be6e1b1 | 2142 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2143 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2144 | { |
fd30b717 KC |
2145 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2146 | ||
2147 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2148 | } |
2149 | ||
2150 | /* | |
2151 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2152 | * This means we do an inexact common-case check. Note that if | |
2153 | * we miss, ->nocb_timer will eventually clean things up. | |
2154 | */ | |
2155 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2156 | { | |
2157 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2158 | do_nocb_deferred_wakeup_common(rdp); | |
2159 | } | |
2160 | ||
f4579fc5 PM |
2161 | void __init rcu_init_nohz(void) |
2162 | { | |
2163 | int cpu; | |
ef126206 | 2164 | bool need_rcu_nocb_mask = false; |
e83e73f5 | 2165 | struct rcu_data *rdp; |
f4579fc5 | 2166 | |
f4579fc5 PM |
2167 | #if defined(CONFIG_NO_HZ_FULL) |
2168 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2169 | need_rcu_nocb_mask = true; | |
2170 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2171 | ||
84b12b75 | 2172 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
2173 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2174 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2175 | return; | |
2176 | } | |
f4579fc5 | 2177 | } |
84b12b75 | 2178 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
2179 | return; |
2180 | ||
f4579fc5 PM |
2181 | #if defined(CONFIG_NO_HZ_FULL) |
2182 | if (tick_nohz_full_running) | |
2183 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2184 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2185 | ||
2186 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 2187 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
2188 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
2189 | rcu_nocb_mask); | |
2190 | } | |
3016611e PM |
2191 | if (cpumask_empty(rcu_nocb_mask)) |
2192 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
2193 | else | |
2194 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
2195 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2196 | if (rcu_nocb_poll) |
2197 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2198 | ||
e83e73f5 PM |
2199 | for_each_cpu(cpu, rcu_nocb_mask) { |
2200 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
2201 | if (rcu_segcblist_empty(&rdp->cblist)) | |
2202 | rcu_segcblist_init(&rdp->cblist); | |
2203 | rcu_segcblist_offload(&rdp->cblist); | |
2204 | } | |
b97d23c5 | 2205 | rcu_organize_nocb_kthreads(); |
96d3fd0d PM |
2206 | } |
2207 | ||
3fbfbf7a PM |
2208 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2209 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2210 | { | |
12f54c3a PM |
2211 | init_swait_queue_head(&rdp->nocb_cb_wq); |
2212 | init_swait_queue_head(&rdp->nocb_gp_wq); | |
8be6e1b1 | 2213 | raw_spin_lock_init(&rdp->nocb_lock); |
d1b222c6 | 2214 | raw_spin_lock_init(&rdp->nocb_bypass_lock); |
4fd8c5f1 | 2215 | raw_spin_lock_init(&rdp->nocb_gp_lock); |
fd30b717 | 2216 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
d1b222c6 PM |
2217 | timer_setup(&rdp->nocb_bypass_timer, do_nocb_bypass_wakeup_timer, 0); |
2218 | rcu_cblist_init(&rdp->nocb_bypass); | |
3fbfbf7a PM |
2219 | } |
2220 | ||
35ce7f29 PM |
2221 | /* |
2222 | * If the specified CPU is a no-CBs CPU that does not already have its | |
12f54c3a PM |
2223 | * rcuo CB kthread, spawn it. Additionally, if the rcuo GP kthread |
2224 | * for this CPU's group has not yet been created, spawn it as well. | |
35ce7f29 | 2225 | */ |
4580b054 | 2226 | static void rcu_spawn_one_nocb_kthread(int cpu) |
35ce7f29 | 2227 | { |
12f54c3a PM |
2228 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
2229 | struct rcu_data *rdp_gp; | |
35ce7f29 PM |
2230 | struct task_struct *t; |
2231 | ||
2232 | /* | |
2233 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2234 | * then nothing to do. | |
2235 | */ | |
12f54c3a | 2236 | if (!rcu_is_nocb_cpu(cpu) || rdp->nocb_cb_kthread) |
35ce7f29 PM |
2237 | return; |
2238 | ||
6484fe54 | 2239 | /* If we didn't spawn the GP kthread first, reorganize! */ |
12f54c3a PM |
2240 | rdp_gp = rdp->nocb_gp_rdp; |
2241 | if (!rdp_gp->nocb_gp_kthread) { | |
2242 | t = kthread_run(rcu_nocb_gp_kthread, rdp_gp, | |
2243 | "rcuog/%d", rdp_gp->cpu); | |
2244 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo GP kthread, OOM is now expected behavior\n", __func__)) | |
2245 | return; | |
2246 | WRITE_ONCE(rdp_gp->nocb_gp_kthread, t); | |
35ce7f29 PM |
2247 | } |
2248 | ||
0ae86a27 | 2249 | /* Spawn the kthread for this CPU. */ |
12f54c3a | 2250 | t = kthread_run(rcu_nocb_cb_kthread, rdp, |
4580b054 | 2251 | "rcuo%c/%d", rcu_state.abbr, cpu); |
12f54c3a | 2252 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo CB kthread, OOM is now expected behavior\n", __func__)) |
9213784b | 2253 | return; |
12f54c3a PM |
2254 | WRITE_ONCE(rdp->nocb_cb_kthread, t); |
2255 | WRITE_ONCE(rdp->nocb_gp_kthread, rdp_gp->nocb_gp_kthread); | |
35ce7f29 PM |
2256 | } |
2257 | ||
2258 | /* | |
2259 | * If the specified CPU is a no-CBs CPU that does not already have its | |
ad368d15 | 2260 | * rcuo kthread, spawn it. |
35ce7f29 | 2261 | */ |
ad368d15 | 2262 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 | 2263 | { |
35ce7f29 | 2264 | if (rcu_scheduler_fully_active) |
b97d23c5 | 2265 | rcu_spawn_one_nocb_kthread(cpu); |
35ce7f29 PM |
2266 | } |
2267 | ||
2268 | /* | |
2269 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2270 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2271 | * non-boot CPUs come online -- if this changes, we will need to add | |
2272 | * some mutual exclusion. | |
2273 | */ | |
2274 | static void __init rcu_spawn_nocb_kthreads(void) | |
2275 | { | |
2276 | int cpu; | |
2277 | ||
2278 | for_each_online_cpu(cpu) | |
ad368d15 | 2279 | rcu_spawn_cpu_nocb_kthread(cpu); |
35ce7f29 PM |
2280 | } |
2281 | ||
6484fe54 | 2282 | /* How many CB CPU IDs per GP kthread? Default of -1 for sqrt(nr_cpu_ids). */ |
f7c612b0 PM |
2283 | static int rcu_nocb_gp_stride = -1; |
2284 | module_param(rcu_nocb_gp_stride, int, 0444); | |
fbce7497 PM |
2285 | |
2286 | /* | |
6484fe54 | 2287 | * Initialize GP-CB relationships for all no-CBs CPU. |
fbce7497 | 2288 | */ |
4580b054 | 2289 | static void __init rcu_organize_nocb_kthreads(void) |
3fbfbf7a PM |
2290 | { |
2291 | int cpu; | |
18cd8c93 | 2292 | bool firsttime = true; |
610dea36 SR |
2293 | bool gotnocbs = false; |
2294 | bool gotnocbscbs = true; | |
f7c612b0 | 2295 | int ls = rcu_nocb_gp_stride; |
6484fe54 | 2296 | int nl = 0; /* Next GP kthread. */ |
3fbfbf7a | 2297 | struct rcu_data *rdp; |
0bdc33da | 2298 | struct rcu_data *rdp_gp = NULL; /* Suppress misguided gcc warn. */ |
fbce7497 | 2299 | struct rcu_data *rdp_prev = NULL; |
3fbfbf7a | 2300 | |
84b12b75 | 2301 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2302 | return; |
fbce7497 | 2303 | if (ls == -1) { |
9fcb09bd | 2304 | ls = nr_cpu_ids / int_sqrt(nr_cpu_ids); |
f7c612b0 | 2305 | rcu_nocb_gp_stride = ls; |
fbce7497 PM |
2306 | } |
2307 | ||
2308 | /* | |
9831ce3b PM |
2309 | * Each pass through this loop sets up one rcu_data structure. |
2310 | * Should the corresponding CPU come online in the future, then | |
2311 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2312 | */ |
3fbfbf7a | 2313 | for_each_cpu(cpu, rcu_nocb_mask) { |
da1df50d | 2314 | rdp = per_cpu_ptr(&rcu_data, cpu); |
fbce7497 | 2315 | if (rdp->cpu >= nl) { |
6484fe54 | 2316 | /* New GP kthread, set up for CBs & next GP. */ |
610dea36 | 2317 | gotnocbs = true; |
fbce7497 | 2318 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; |
58bf6f77 | 2319 | rdp->nocb_gp_rdp = rdp; |
0bdc33da | 2320 | rdp_gp = rdp; |
610dea36 SR |
2321 | if (dump_tree) { |
2322 | if (!firsttime) | |
2323 | pr_cont("%s\n", gotnocbscbs | |
2324 | ? "" : " (self only)"); | |
2325 | gotnocbscbs = false; | |
2326 | firsttime = false; | |
2327 | pr_alert("%s: No-CB GP kthread CPU %d:", | |
2328 | __func__, cpu); | |
2329 | } | |
fbce7497 | 2330 | } else { |
6484fe54 | 2331 | /* Another CB kthread, link to previous GP kthread. */ |
610dea36 | 2332 | gotnocbscbs = true; |
0bdc33da | 2333 | rdp->nocb_gp_rdp = rdp_gp; |
58bf6f77 | 2334 | rdp_prev->nocb_next_cb_rdp = rdp; |
610dea36 SR |
2335 | if (dump_tree) |
2336 | pr_cont(" %d", cpu); | |
fbce7497 PM |
2337 | } |
2338 | rdp_prev = rdp; | |
3fbfbf7a | 2339 | } |
610dea36 SR |
2340 | if (gotnocbs && dump_tree) |
2341 | pr_cont("%s\n", gotnocbscbs ? "" : " (self only)"); | |
3fbfbf7a PM |
2342 | } |
2343 | ||
5ab7ab83 PM |
2344 | /* |
2345 | * Bind the current task to the offloaded CPUs. If there are no offloaded | |
2346 | * CPUs, leave the task unbound. Splat if the bind attempt fails. | |
2347 | */ | |
2348 | void rcu_bind_current_to_nocb(void) | |
2349 | { | |
2350 | if (cpumask_available(rcu_nocb_mask) && cpumask_weight(rcu_nocb_mask)) | |
2351 | WARN_ON(sched_setaffinity(current->pid, rcu_nocb_mask)); | |
2352 | } | |
2353 | EXPORT_SYMBOL_GPL(rcu_bind_current_to_nocb); | |
2354 | ||
f7a81b12 PM |
2355 | /* |
2356 | * Dump out nocb grace-period kthread state for the specified rcu_data | |
2357 | * structure. | |
2358 | */ | |
2359 | static void show_rcu_nocb_gp_state(struct rcu_data *rdp) | |
2360 | { | |
2361 | struct rcu_node *rnp = rdp->mynode; | |
2362 | ||
2363 | pr_info("nocb GP %d %c%c%c%c%c%c %c[%c%c] %c%c:%ld rnp %d:%d %lu\n", | |
2364 | rdp->cpu, | |
2365 | "kK"[!!rdp->nocb_gp_kthread], | |
2366 | "lL"[raw_spin_is_locked(&rdp->nocb_gp_lock)], | |
2367 | "dD"[!!rdp->nocb_defer_wakeup], | |
2368 | "tT"[timer_pending(&rdp->nocb_timer)], | |
2369 | "bB"[timer_pending(&rdp->nocb_bypass_timer)], | |
2370 | "sS"[!!rdp->nocb_gp_sleep], | |
2371 | ".W"[swait_active(&rdp->nocb_gp_wq)], | |
2372 | ".W"[swait_active(&rnp->nocb_gp_wq[0])], | |
2373 | ".W"[swait_active(&rnp->nocb_gp_wq[1])], | |
2374 | ".B"[!!rdp->nocb_gp_bypass], | |
2375 | ".G"[!!rdp->nocb_gp_gp], | |
2376 | (long)rdp->nocb_gp_seq, | |
2377 | rnp->grplo, rnp->grphi, READ_ONCE(rdp->nocb_gp_loops)); | |
2378 | } | |
2379 | ||
2380 | /* Dump out nocb kthread state for the specified rcu_data structure. */ | |
2381 | static void show_rcu_nocb_state(struct rcu_data *rdp) | |
2382 | { | |
2383 | struct rcu_segcblist *rsclp = &rdp->cblist; | |
2384 | bool waslocked; | |
2385 | bool wastimer; | |
2386 | bool wassleep; | |
2387 | ||
2388 | if (rdp->nocb_gp_rdp == rdp) | |
2389 | show_rcu_nocb_gp_state(rdp); | |
2390 | ||
2391 | pr_info(" CB %d->%d %c%c%c%c%c%c F%ld L%ld C%d %c%c%c%c%c q%ld\n", | |
2392 | rdp->cpu, rdp->nocb_gp_rdp->cpu, | |
2393 | "kK"[!!rdp->nocb_cb_kthread], | |
2394 | "bB"[raw_spin_is_locked(&rdp->nocb_bypass_lock)], | |
2395 | "cC"[!!atomic_read(&rdp->nocb_lock_contended)], | |
2396 | "lL"[raw_spin_is_locked(&rdp->nocb_lock)], | |
2397 | "sS"[!!rdp->nocb_cb_sleep], | |
2398 | ".W"[swait_active(&rdp->nocb_cb_wq)], | |
2399 | jiffies - rdp->nocb_bypass_first, | |
2400 | jiffies - rdp->nocb_nobypass_last, | |
2401 | rdp->nocb_nobypass_count, | |
2402 | ".D"[rcu_segcblist_ready_cbs(rsclp)], | |
2403 | ".W"[!rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL)], | |
2404 | ".R"[!rcu_segcblist_restempty(rsclp, RCU_WAIT_TAIL)], | |
2405 | ".N"[!rcu_segcblist_restempty(rsclp, RCU_NEXT_READY_TAIL)], | |
2406 | ".B"[!!rcu_cblist_n_cbs(&rdp->nocb_bypass)], | |
2407 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2408 | ||
2409 | /* It is OK for GP kthreads to have GP state. */ | |
2410 | if (rdp->nocb_gp_rdp == rdp) | |
2411 | return; | |
2412 | ||
2413 | waslocked = raw_spin_is_locked(&rdp->nocb_gp_lock); | |
2414 | wastimer = timer_pending(&rdp->nocb_timer); | |
2415 | wassleep = swait_active(&rdp->nocb_gp_wq); | |
2416 | if (!rdp->nocb_defer_wakeup && !rdp->nocb_gp_sleep && | |
2417 | !waslocked && !wastimer && !wassleep) | |
2418 | return; /* Nothing untowards. */ | |
2419 | ||
2420 | pr_info(" !!! %c%c%c%c %c\n", | |
2421 | "lL"[waslocked], | |
2422 | "dD"[!!rdp->nocb_defer_wakeup], | |
2423 | "tT"[wastimer], | |
2424 | "sS"[!!rdp->nocb_gp_sleep], | |
2425 | ".W"[wassleep]); | |
2426 | } | |
2427 | ||
34ed6246 PM |
2428 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2429 | ||
5d6742b3 PM |
2430 | /* No ->nocb_lock to acquire. */ |
2431 | static void rcu_nocb_lock(struct rcu_data *rdp) | |
d7e29933 | 2432 | { |
5d6742b3 PM |
2433 | } |
2434 | ||
2435 | /* No ->nocb_lock to release. */ | |
2436 | static void rcu_nocb_unlock(struct rcu_data *rdp) | |
2437 | { | |
2438 | } | |
2439 | ||
2440 | /* No ->nocb_lock to release. */ | |
2441 | static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp, | |
2442 | unsigned long flags) | |
2443 | { | |
2444 | local_irq_restore(flags); | |
d7e29933 PM |
2445 | } |
2446 | ||
d1b222c6 PM |
2447 | /* Lockdep check that ->cblist may be safely accessed. */ |
2448 | static void rcu_lockdep_assert_cblist_protected(struct rcu_data *rdp) | |
2449 | { | |
2450 | lockdep_assert_irqs_disabled(); | |
2451 | } | |
2452 | ||
abedf8e2 | 2453 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2454 | { |
3fbfbf7a PM |
2455 | } |
2456 | ||
abedf8e2 | 2457 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2458 | { |
2459 | return NULL; | |
2460 | } | |
2461 | ||
dae6e64d PM |
2462 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2463 | { | |
2464 | } | |
3fbfbf7a | 2465 | |
d1b222c6 PM |
2466 | static bool rcu_nocb_flush_bypass(struct rcu_data *rdp, struct rcu_head *rhp, |
2467 | unsigned long j) | |
2468 | { | |
2469 | return true; | |
2470 | } | |
2471 | ||
2472 | static bool rcu_nocb_try_bypass(struct rcu_data *rdp, struct rcu_head *rhp, | |
2473 | bool *was_alldone, unsigned long flags) | |
2474 | { | |
2475 | return false; | |
2476 | } | |
2477 | ||
5d6742b3 PM |
2478 | static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_empty, |
2479 | unsigned long flags) | |
3fbfbf7a | 2480 | { |
5d6742b3 | 2481 | WARN_ON_ONCE(1); /* Should be dead code! */ |
3fbfbf7a PM |
2482 | } |
2483 | ||
3fbfbf7a PM |
2484 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2485 | { | |
2486 | } | |
2487 | ||
9fdd3bc9 | 2488 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2489 | { |
2490 | return false; | |
2491 | } | |
2492 | ||
2493 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2494 | { | |
2495 | } | |
2496 | ||
ad368d15 | 2497 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 PM |
2498 | { |
2499 | } | |
2500 | ||
2501 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2502 | { |
2503 | } | |
2504 | ||
f7a81b12 PM |
2505 | static void show_rcu_nocb_state(struct rcu_data *rdp) |
2506 | { | |
2507 | } | |
2508 | ||
3fbfbf7a | 2509 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ |
65d798f0 | 2510 | |
a096932f PM |
2511 | /* |
2512 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2513 | * grace-period kthread will do force_quiescent_state() processing? | |
2514 | * The idea is to avoid waking up RCU core processing on such a | |
2515 | * CPU unless the grace period has extended for too long. | |
2516 | * | |
2517 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2518 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f | 2519 | */ |
4580b054 | 2520 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
2521 | { |
2522 | #ifdef CONFIG_NO_HZ_FULL | |
2523 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 2524 | (!rcu_gp_in_progress() || |
e2f3ccfa | 2525 | time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 2526 | return true; |
a096932f | 2527 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2528 | return false; |
a096932f | 2529 | } |
5057f55e PM |
2530 | |
2531 | /* | |
265f5f28 | 2532 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2533 | */ |
2534 | static void rcu_bind_gp_kthread(void) | |
2535 | { | |
c0f489d2 | 2536 | if (!tick_nohz_full_enabled()) |
5057f55e | 2537 | return; |
de201559 | 2538 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2539 | } |
176f8f7a PM |
2540 | |
2541 | /* Record the current task on dyntick-idle entry. */ | |
ff5c4f5c | 2542 | static void noinstr rcu_dynticks_task_enter(void) |
176f8f7a PM |
2543 | { |
2544 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2545 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2546 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2547 | } | |
2548 | ||
2549 | /* Record no current task on dyntick-idle exit. */ | |
ff5c4f5c | 2550 | static void noinstr rcu_dynticks_task_exit(void) |
176f8f7a PM |
2551 | { |
2552 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2553 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2554 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2555 | } | |
7d0c9c50 PM |
2556 | |
2557 | /* Turn on heavyweight RCU tasks trace readers on idle/user entry. */ | |
2558 | static void rcu_dynticks_task_trace_enter(void) | |
2559 | { | |
2560 | #ifdef CONFIG_TASKS_RCU_TRACE | |
2561 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) | |
2562 | current->trc_reader_special.b.need_mb = true; | |
2563 | #endif /* #ifdef CONFIG_TASKS_RCU_TRACE */ | |
2564 | } | |
2565 | ||
2566 | /* Turn off heavyweight RCU tasks trace readers on idle/user exit. */ | |
2567 | static void rcu_dynticks_task_trace_exit(void) | |
2568 | { | |
2569 | #ifdef CONFIG_TASKS_RCU_TRACE | |
2570 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) | |
2571 | current->trc_reader_special.b.need_mb = false; | |
2572 | #endif /* #ifdef CONFIG_TASKS_RCU_TRACE */ | |
2573 | } |