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 | |
3820b513 FW |
16 | static bool rcu_rdp_is_offloaded(struct rcu_data *rdp) |
17 | { | |
18 | /* | |
19 | * In order to read the offloaded state of an rdp is a safe | |
20 | * and stable way and prevent from its value to be changed | |
21 | * under us, we must either hold the barrier mutex, the cpu | |
22 | * hotplug lock (read or write) or the nocb lock. Local | |
23 | * non-preemptible reads are also safe. NOCB kthreads and | |
24 | * timers have their own means of synchronization against the | |
25 | * offloaded state updaters. | |
26 | */ | |
27 | RCU_LOCKDEP_WARN( | |
28 | !(lockdep_is_held(&rcu_state.barrier_mutex) || | |
29 | (IS_ENABLED(CONFIG_HOTPLUG_CPU) && lockdep_is_cpus_held()) || | |
30 | rcu_lockdep_is_held_nocb(rdp) || | |
31 | (rdp == this_cpu_ptr(&rcu_data) && | |
32 | !(IS_ENABLED(CONFIG_PREEMPT_COUNT) && preemptible())) || | |
d76e0926 | 33 | rcu_current_is_nocb_kthread(rdp)), |
3820b513 FW |
34 | "Unsafe read of RCU_NOCB offloaded state" |
35 | ); | |
36 | ||
37 | return rcu_segcblist_is_offloaded(&rdp->cblist); | |
38 | } | |
39 | ||
26845c28 PM |
40 | /* |
41 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 42 | * messages about anything out of the ordinary. |
26845c28 PM |
43 | */ |
44 | static void __init rcu_bootup_announce_oddness(void) | |
45 | { | |
ab6f5bd6 | 46 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 47 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
48 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
49 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
a7538352 JP |
50 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n", |
51 | RCU_FANOUT); | |
7fa27001 | 52 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
53 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
54 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
55 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 56 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 57 | pr_info("\tRCU lockdep checking is enabled.\n"); |
8cbd0e38 PM |
58 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) |
59 | pr_info("\tRCU strict (and thus non-scalable) grace periods enabled.\n"); | |
42621697 AG |
60 | if (RCU_NUM_LVLS >= 4) |
61 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 62 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 63 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
64 | RCU_FANOUT_LEAF); |
65 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
a7538352 JP |
66 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", |
67 | rcu_fanout_leaf); | |
cca6f393 | 68 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 69 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b | 70 | #ifdef CONFIG_RCU_BOOST |
a7538352 JP |
71 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", |
72 | kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
17c7798b PM |
73 | #endif |
74 | if (blimit != DEFAULT_RCU_BLIMIT) | |
75 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
76 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
77 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
78 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
79 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
b2b00ddf | 80 | if (qovld != DEFAULT_RCU_QOVLD) |
aa96a93b | 81 | pr_info("\tBoot-time adjustment of callback overload level to %ld.\n", qovld); |
17c7798b PM |
82 | if (jiffies_till_first_fqs != ULONG_MAX) |
83 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
84 | if (jiffies_till_next_fqs != ULONG_MAX) | |
85 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
c06aed0e PM |
86 | if (jiffies_till_sched_qs != ULONG_MAX) |
87 | pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs); | |
17c7798b PM |
88 | if (rcu_kick_kthreads) |
89 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
90 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
91 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 92 | if (gp_preinit_delay) |
17c7798b | 93 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 94 | if (gp_init_delay) |
17c7798b | 95 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 96 | if (gp_cleanup_delay) |
17c7798b | 97 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
48d07c04 SAS |
98 | if (!use_softirq) |
99 | pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n"); | |
17c7798b PM |
100 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) |
101 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 102 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
103 | } |
104 | ||
28f6569a | 105 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 106 | |
63d4c8c9 | 107 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 108 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 109 | |
f41d911f PM |
110 | /* |
111 | * Tell them what RCU they are running. | |
112 | */ | |
0e0fc1c2 | 113 | static void __init rcu_bootup_announce(void) |
f41d911f | 114 | { |
efc151c3 | 115 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 116 | rcu_bootup_announce_oddness(); |
f41d911f PM |
117 | } |
118 | ||
8203d6d0 PM |
119 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
120 | #define RCU_GP_TASKS 0x8 | |
121 | #define RCU_EXP_TASKS 0x4 | |
122 | #define RCU_GP_BLKD 0x2 | |
123 | #define RCU_EXP_BLKD 0x1 | |
124 | ||
125 | /* | |
126 | * Queues a task preempted within an RCU-preempt read-side critical | |
127 | * section into the appropriate location within the ->blkd_tasks list, | |
128 | * depending on the states of any ongoing normal and expedited grace | |
129 | * periods. The ->gp_tasks pointer indicates which element the normal | |
130 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
131 | * indicates which element the expedited grace period is waiting on (again, | |
132 | * NULL if none). If a grace period is waiting on a given element in the | |
133 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
134 | * adding a task to the tail of the list blocks any grace period that is | |
135 | * already waiting on one of the elements. In contrast, adding a task | |
136 | * to the head of the list won't block any grace period that is already | |
137 | * waiting on one of the elements. | |
138 | * | |
139 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
140 | * period wait for a task that is not strictly speaking blocking it. | |
141 | * Given the choice, we needlessly block a normal grace period rather than | |
142 | * blocking an expedited grace period. | |
143 | * | |
144 | * Note that an endless sequence of expedited grace periods still cannot | |
145 | * indefinitely postpone a normal grace period. Eventually, all of the | |
146 | * fixed number of preempted tasks blocking the normal grace period that are | |
147 | * not also blocking the expedited grace period will resume and complete | |
148 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
149 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
150 | * the corresponding expedited grace period will also be the end of the | |
151 | * normal grace period. | |
152 | */ | |
46a5d164 PM |
153 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
154 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
155 | { |
156 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
157 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
158 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
159 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
160 | struct task_struct *t = current; | |
161 | ||
a32e01ee | 162 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 163 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 164 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
165 | /* RCU better not be waiting on newly onlined CPUs! */ |
166 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
167 | rdp->grpmask); | |
ea9b0c8a | 168 | |
8203d6d0 PM |
169 | /* |
170 | * Decide where to queue the newly blocked task. In theory, | |
171 | * this could be an if-statement. In practice, when I tried | |
172 | * that, it was quite messy. | |
173 | */ | |
174 | switch (blkd_state) { | |
175 | case 0: | |
176 | case RCU_EXP_TASKS: | |
177 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
178 | case RCU_GP_TASKS: | |
179 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
180 | ||
181 | /* | |
182 | * Blocking neither GP, or first task blocking the normal | |
183 | * GP but not blocking the already-waiting expedited GP. | |
184 | * Queue at the head of the list to avoid unnecessarily | |
185 | * blocking the already-waiting GPs. | |
186 | */ | |
187 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
188 | break; | |
189 | ||
190 | case RCU_EXP_BLKD: | |
191 | case RCU_GP_BLKD: | |
192 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
193 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
194 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
195 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
196 | ||
197 | /* | |
198 | * First task arriving that blocks either GP, or first task | |
199 | * arriving that blocks the expedited GP (with the normal | |
200 | * GP already waiting), or a task arriving that blocks | |
201 | * both GPs with both GPs already waiting. Queue at the | |
202 | * tail of the list to avoid any GP waiting on any of the | |
203 | * already queued tasks that are not blocking it. | |
204 | */ | |
205 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
206 | break; | |
207 | ||
208 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
209 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
210 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
211 | ||
212 | /* | |
213 | * Second or subsequent task blocking the expedited GP. | |
214 | * The task either does not block the normal GP, or is the | |
215 | * first task blocking the normal GP. Queue just after | |
216 | * the first task blocking the expedited GP. | |
217 | */ | |
218 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
219 | break; | |
220 | ||
221 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
222 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
223 | ||
224 | /* | |
225 | * Second or subsequent task blocking the normal GP. | |
226 | * The task does not block the expedited GP. Queue just | |
227 | * after the first task blocking the normal GP. | |
228 | */ | |
229 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
230 | break; | |
231 | ||
232 | default: | |
233 | ||
234 | /* Yet another exercise in excessive paranoia. */ | |
235 | WARN_ON_ONCE(1); | |
236 | break; | |
237 | } | |
238 | ||
239 | /* | |
240 | * We have now queued the task. If it was the first one to | |
241 | * block either grace period, update the ->gp_tasks and/or | |
242 | * ->exp_tasks pointers, respectively, to reference the newly | |
243 | * blocked tasks. | |
244 | */ | |
4bc8d555 | 245 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
6935c398 | 246 | WRITE_ONCE(rnp->gp_tasks, &t->rcu_node_entry); |
d43a5d32 | 247 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 248 | } |
8203d6d0 | 249 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
314eeb43 | 250 | WRITE_ONCE(rnp->exp_tasks, &t->rcu_node_entry); |
2dee9404 PM |
251 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
252 | !(rnp->qsmask & rdp->grpmask)); | |
253 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
254 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 255 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
256 | |
257 | /* | |
258 | * Report the quiescent state for the expedited GP. This expedited | |
259 | * GP should not be able to end until we report, so there should be | |
260 | * no need to check for a subsequent expedited GP. (Though we are | |
261 | * still in a quiescent state in any case.) | |
262 | */ | |
1bb33644 | 263 | if (blkd_state & RCU_EXP_BLKD && rdp->exp_deferred_qs) |
63d4c8c9 | 264 | rcu_report_exp_rdp(rdp); |
fcc878e4 | 265 | else |
1bb33644 | 266 | WARN_ON_ONCE(rdp->exp_deferred_qs); |
8203d6d0 PM |
267 | } |
268 | ||
f41d911f | 269 | /* |
c7037ff5 PM |
270 | * Record a preemptible-RCU quiescent state for the specified CPU. |
271 | * Note that this does not necessarily mean that the task currently running | |
272 | * on the CPU is in a quiescent state: Instead, it means that the current | |
273 | * grace period need not wait on any RCU read-side critical section that | |
274 | * starts later on this CPU. It also means that if the current task is | |
275 | * in an RCU read-side critical section, it has already added itself to | |
276 | * some leaf rcu_node structure's ->blkd_tasks list. In addition to the | |
277 | * current task, there might be any number of other tasks blocked while | |
278 | * in an RCU read-side critical section. | |
25502a6c | 279 | * |
c7037ff5 | 280 | * Callers to this function must disable preemption. |
f41d911f | 281 | */ |
45975c7d | 282 | static void rcu_qs(void) |
f41d911f | 283 | { |
45975c7d | 284 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n"); |
2280ee5a | 285 | if (__this_cpu_read(rcu_data.cpu_no_qs.s)) { |
284a8c93 | 286 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2280ee5a | 287 | __this_cpu_read(rcu_data.gp_seq), |
284a8c93 | 288 | TPS("cpuqs")); |
2280ee5a | 289 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); |
c98cac60 | 290 | barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */ |
add0d37b | 291 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false); |
284a8c93 | 292 | } |
f41d911f PM |
293 | } |
294 | ||
295 | /* | |
c3422bea PM |
296 | * We have entered the scheduler, and the current task might soon be |
297 | * context-switched away from. If this task is in an RCU read-side | |
298 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
299 | * record that fact, so we enqueue the task on the blkd_tasks list. |
300 | * The task will dequeue itself when it exits the outermost enclosing | |
301 | * RCU read-side critical section. Therefore, the current grace period | |
302 | * cannot be permitted to complete until the blkd_tasks list entries | |
303 | * predating the current grace period drain, in other words, until | |
304 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 305 | * |
46a5d164 | 306 | * Caller must disable interrupts. |
f41d911f | 307 | */ |
45975c7d | 308 | void rcu_note_context_switch(bool preempt) |
f41d911f PM |
309 | { |
310 | struct task_struct *t = current; | |
da1df50d | 311 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
f41d911f PM |
312 | struct rcu_node *rnp; |
313 | ||
45975c7d | 314 | trace_rcu_utilization(TPS("Start context switch")); |
b04db8e1 | 315 | lockdep_assert_irqs_disabled(); |
521c89b3 | 316 | WARN_ONCE(!preempt && rcu_preempt_depth() > 0, "Voluntary context switch within RCU read-side critical section!"); |
77339e61 | 317 | if (rcu_preempt_depth() > 0 && |
1d082fd0 | 318 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
319 | |
320 | /* Possibly blocking in an RCU read-side critical section. */ | |
f41d911f | 321 | rnp = rdp->mynode; |
46a5d164 | 322 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 323 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 324 | t->rcu_blocked_node = rnp; |
f41d911f PM |
325 | |
326 | /* | |
8203d6d0 PM |
327 | * Verify the CPU's sanity, trace the preemption, and |
328 | * then queue the task as required based on the states | |
329 | * of any ongoing and expedited grace periods. | |
f41d911f | 330 | */ |
0aa04b05 | 331 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 332 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
88d1bead | 333 | trace_rcu_preempt_task(rcu_state.name, |
d4c08f2a PM |
334 | t->pid, |
335 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
336 | ? rnp->gp_seq |
337 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 338 | rcu_preempt_ctxt_queue(rnp, rdp); |
3e310098 PM |
339 | } else { |
340 | rcu_preempt_deferred_qs(t); | |
f41d911f PM |
341 | } |
342 | ||
343 | /* | |
344 | * Either we were not in an RCU read-side critical section to | |
345 | * begin with, or we have now recorded that critical section | |
346 | * globally. Either way, we can now note a quiescent state | |
347 | * for this CPU. Again, if we were in an RCU read-side critical | |
348 | * section, and if that critical section was blocking the current | |
349 | * grace period, then the fact that the task has been enqueued | |
350 | * means that we continue to block the current grace period. | |
351 | */ | |
45975c7d | 352 | rcu_qs(); |
1bb33644 | 353 | if (rdp->exp_deferred_qs) |
63d4c8c9 | 354 | rcu_report_exp_rdp(rdp); |
43766c3e | 355 | rcu_tasks_qs(current, preempt); |
45975c7d | 356 | trace_rcu_utilization(TPS("End context switch")); |
f41d911f | 357 | } |
45975c7d | 358 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
f41d911f | 359 | |
fc2219d4 PM |
360 | /* |
361 | * Check for preempted RCU readers blocking the current grace period | |
362 | * for the specified rcu_node structure. If the caller needs a reliable | |
363 | * answer, it must hold the rcu_node's ->lock. | |
364 | */ | |
27f4d280 | 365 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 366 | { |
6935c398 | 367 | return READ_ONCE(rnp->gp_tasks) != NULL; |
fc2219d4 PM |
368 | } |
369 | ||
5f5fa7ea | 370 | /* limit value for ->rcu_read_lock_nesting. */ |
5f1a6ef3 PM |
371 | #define RCU_NEST_PMAX (INT_MAX / 2) |
372 | ||
77339e61 LJ |
373 | static void rcu_preempt_read_enter(void) |
374 | { | |
5fcb3a5f | 375 | WRITE_ONCE(current->rcu_read_lock_nesting, READ_ONCE(current->rcu_read_lock_nesting) + 1); |
77339e61 LJ |
376 | } |
377 | ||
5f5fa7ea | 378 | static int rcu_preempt_read_exit(void) |
77339e61 | 379 | { |
5fcb3a5f PM |
380 | int ret = READ_ONCE(current->rcu_read_lock_nesting) - 1; |
381 | ||
382 | WRITE_ONCE(current->rcu_read_lock_nesting, ret); | |
383 | return ret; | |
77339e61 LJ |
384 | } |
385 | ||
386 | static void rcu_preempt_depth_set(int val) | |
387 | { | |
5fcb3a5f | 388 | WRITE_ONCE(current->rcu_read_lock_nesting, val); |
77339e61 LJ |
389 | } |
390 | ||
0e5da22e PM |
391 | /* |
392 | * Preemptible RCU implementation for rcu_read_lock(). | |
393 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
394 | * if we block. | |
395 | */ | |
396 | void __rcu_read_lock(void) | |
397 | { | |
77339e61 | 398 | rcu_preempt_read_enter(); |
5f1a6ef3 | 399 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
77339e61 | 400 | WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX); |
f19920e4 PM |
401 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread) |
402 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true); | |
0e5da22e PM |
403 | barrier(); /* critical section after entry code. */ |
404 | } | |
405 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
406 | ||
407 | /* | |
408 | * Preemptible RCU implementation for rcu_read_unlock(). | |
409 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
410 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
411 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
412 | * in an RCU read-side critical section and other special cases. | |
413 | */ | |
414 | void __rcu_read_unlock(void) | |
415 | { | |
416 | struct task_struct *t = current; | |
417 | ||
7e937220 | 418 | barrier(); // critical section before exit code. |
5f5fa7ea | 419 | if (rcu_preempt_read_exit() == 0) { |
7e937220 | 420 | barrier(); // critical-section exit before .s check. |
0e5da22e PM |
421 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) |
422 | rcu_read_unlock_special(t); | |
0e5da22e | 423 | } |
5f1a6ef3 | 424 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
77339e61 | 425 | int rrln = rcu_preempt_depth(); |
0e5da22e | 426 | |
5f5fa7ea | 427 | WARN_ON_ONCE(rrln < 0 || rrln > RCU_NEST_PMAX); |
0e5da22e | 428 | } |
0e5da22e PM |
429 | } |
430 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
431 | ||
12f5f524 PM |
432 | /* |
433 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
434 | * returning NULL if at the end of the list. | |
435 | */ | |
436 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
437 | struct rcu_node *rnp) | |
438 | { | |
439 | struct list_head *np; | |
440 | ||
441 | np = t->rcu_node_entry.next; | |
442 | if (np == &rnp->blkd_tasks) | |
443 | np = NULL; | |
444 | return np; | |
445 | } | |
446 | ||
8af3a5e7 PM |
447 | /* |
448 | * Return true if the specified rcu_node structure has tasks that were | |
449 | * preempted within an RCU read-side critical section. | |
450 | */ | |
451 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
452 | { | |
453 | return !list_empty(&rnp->blkd_tasks); | |
454 | } | |
455 | ||
b668c9cf | 456 | /* |
3e310098 PM |
457 | * Report deferred quiescent states. The deferral time can |
458 | * be quite short, for example, in the case of the call from | |
459 | * rcu_read_unlock_special(). | |
b668c9cf | 460 | */ |
3e310098 PM |
461 | static void |
462 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 463 | { |
b6a932d1 PM |
464 | bool empty_exp; |
465 | bool empty_norm; | |
466 | bool empty_exp_now; | |
12f5f524 | 467 | struct list_head *np; |
abaa93d9 | 468 | bool drop_boost_mutex = false; |
8203d6d0 | 469 | struct rcu_data *rdp; |
f41d911f | 470 | struct rcu_node *rnp; |
1d082fd0 | 471 | union rcu_special special; |
f41d911f | 472 | |
f41d911f | 473 | /* |
8203d6d0 PM |
474 | * If RCU core is waiting for this CPU to exit its critical section, |
475 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 476 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
477 | */ |
478 | special = t->rcu_read_unlock_special; | |
da1df50d | 479 | rdp = this_cpu_ptr(&rcu_data); |
1bb33644 | 480 | if (!special.s && !rdp->exp_deferred_qs) { |
3e310098 PM |
481 | local_irq_restore(flags); |
482 | return; | |
483 | } | |
3717e1e9 | 484 | t->rcu_read_unlock_special.s = 0; |
44bad5b3 | 485 | if (special.b.need_qs) { |
3d29aaf1 | 486 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) { |
cfeac397 | 487 | rcu_report_qs_rdp(rdp); |
3d29aaf1 PM |
488 | udelay(rcu_unlock_delay); |
489 | } else { | |
44bad5b3 | 490 | rcu_qs(); |
3d29aaf1 | 491 | } |
44bad5b3 | 492 | } |
f41d911f | 493 | |
8203d6d0 | 494 | /* |
3e310098 PM |
495 | * Respond to a request by an expedited grace period for a |
496 | * quiescent state from this CPU. Note that requests from | |
497 | * tasks are handled when removing the task from the | |
498 | * blocked-tasks list below. | |
8203d6d0 | 499 | */ |
3717e1e9 | 500 | if (rdp->exp_deferred_qs) |
63d4c8c9 | 501 | rcu_report_exp_rdp(rdp); |
8203d6d0 | 502 | |
f41d911f | 503 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 | 504 | if (special.b.blocked) { |
f41d911f | 505 | |
dd5d19ba | 506 | /* |
0a0ba1c9 | 507 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
508 | * now remains queued on the rcu_node corresponding to the |
509 | * CPU it first blocked on, so there is no longer any need | |
510 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 511 | */ |
8ba9153b PM |
512 | rnp = t->rcu_blocked_node; |
513 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
514 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 515 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 516 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 517 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 518 | (!empty_norm || rnp->qsmask)); |
6c7d7dbf | 519 | empty_exp = sync_rcu_exp_done(rnp); |
d9a3da06 | 520 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 521 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 522 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 523 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 524 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 525 | rnp->gp_seq, t->pid); |
12f5f524 | 526 | if (&t->rcu_node_entry == rnp->gp_tasks) |
6935c398 | 527 | WRITE_ONCE(rnp->gp_tasks, np); |
12f5f524 | 528 | if (&t->rcu_node_entry == rnp->exp_tasks) |
314eeb43 | 529 | WRITE_ONCE(rnp->exp_tasks, np); |
727b705b | 530 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b | 531 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
830e6acc | 532 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx.rtmutex) == t; |
2dee9404 | 533 | if (&t->rcu_node_entry == rnp->boost_tasks) |
5822b812 | 534 | WRITE_ONCE(rnp->boost_tasks, np); |
727b705b | 535 | } |
f41d911f PM |
536 | |
537 | /* | |
538 | * If this was the last task on the current list, and if | |
539 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
540 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
541 | * so we must take a snapshot of the expedited state. | |
f41d911f | 542 | */ |
6c7d7dbf | 543 | empty_exp_now = sync_rcu_exp_done(rnp); |
74e871ac | 544 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 545 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 546 | rnp->gp_seq, |
d4c08f2a PM |
547 | 0, rnp->qsmask, |
548 | rnp->level, | |
549 | rnp->grplo, | |
550 | rnp->grphi, | |
551 | !!rnp->gp_tasks); | |
139ad4da | 552 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 553 | } else { |
67c583a7 | 554 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 555 | } |
d9a3da06 | 556 | |
27f4d280 | 557 | /* Unboost if we were boosted. */ |
727b705b | 558 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
830e6acc | 559 | rt_mutex_futex_unlock(&rnp->boost_mtx.rtmutex); |
27f4d280 | 560 | |
d9a3da06 PM |
561 | /* |
562 | * If this was the last task on the expedited lists, | |
563 | * then we need to report up the rcu_node hierarchy. | |
564 | */ | |
389abd48 | 565 | if (!empty_exp && empty_exp_now) |
63d4c8c9 | 566 | rcu_report_exp_rnp(rnp, true); |
b668c9cf PM |
567 | } else { |
568 | local_irq_restore(flags); | |
f41d911f | 569 | } |
f41d911f PM |
570 | } |
571 | ||
3e310098 PM |
572 | /* |
573 | * Is a deferred quiescent-state pending, and are we also not in | |
574 | * an RCU read-side critical section? It is the caller's responsibility | |
575 | * to ensure it is otherwise safe to report any deferred quiescent | |
576 | * states. The reason for this is that it is safe to report a | |
577 | * quiescent state during context switch even though preemption | |
578 | * is disabled. This function cannot be expected to understand these | |
579 | * nuances, so the caller must handle them. | |
580 | */ | |
581 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
582 | { | |
1bb33644 | 583 | return (__this_cpu_read(rcu_data.exp_deferred_qs) || |
3e310098 | 584 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
5f5fa7ea | 585 | rcu_preempt_depth() == 0; |
3e310098 PM |
586 | } |
587 | ||
588 | /* | |
589 | * Report a deferred quiescent state if needed and safe to do so. | |
590 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
591 | * not being in an RCU read-side critical section. The caller must | |
592 | * evaluate safety in terms of interrupt, softirq, and preemption | |
593 | * disabling. | |
594 | */ | |
595 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
596 | { | |
597 | unsigned long flags; | |
3e310098 PM |
598 | |
599 | if (!rcu_preempt_need_deferred_qs(t)) | |
600 | return; | |
3e310098 PM |
601 | local_irq_save(flags); |
602 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
3e310098 PM |
603 | } |
604 | ||
0864f057 PM |
605 | /* |
606 | * Minimal handler to give the scheduler a chance to re-evaluate. | |
607 | */ | |
608 | static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp) | |
609 | { | |
610 | struct rcu_data *rdp; | |
611 | ||
612 | rdp = container_of(iwp, struct rcu_data, defer_qs_iw); | |
613 | rdp->defer_qs_iw_pending = false; | |
614 | } | |
615 | ||
3e310098 PM |
616 | /* |
617 | * Handle special cases during rcu_read_unlock(), such as needing to | |
618 | * notify RCU core processing or task having blocked during the RCU | |
619 | * read-side critical section. | |
620 | */ | |
621 | static void rcu_read_unlock_special(struct task_struct *t) | |
622 | { | |
623 | unsigned long flags; | |
39bbfc62 | 624 | bool irqs_were_disabled; |
3e310098 PM |
625 | bool preempt_bh_were_disabled = |
626 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
3e310098 PM |
627 | |
628 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
629 | if (in_nmi()) | |
630 | return; | |
631 | ||
632 | local_irq_save(flags); | |
633 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 | 634 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
39bbfc62 | 635 | bool expboost; // Expedited GP in flight or possible boosting. |
25102de6 PM |
636 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
637 | struct rcu_node *rnp = rdp->mynode; | |
638 | ||
39bbfc62 PM |
639 | expboost = (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks)) || |
640 | (rdp->grpmask & READ_ONCE(rnp->expmask)) || | |
7308e024 | 641 | IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) || |
39bbfc62 PM |
642 | (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && |
643 | t->rcu_blocked_node); | |
23634ebc | 644 | // Need to defer quiescent state until everything is enabled. |
39bbfc62 | 645 | if (use_softirq && (in_irq() || (expboost && !irqs_were_disabled))) { |
e4453d8a | 646 | // Using softirq, safe to awaken, and either the |
39bbfc62 PM |
647 | // wakeup is free or there is either an expedited |
648 | // GP in flight or a potential need to deboost. | |
05f41571 PM |
649 | raise_softirq_irqoff(RCU_SOFTIRQ); |
650 | } else { | |
23634ebc | 651 | // Enabling BH or preempt does reschedule, so... |
39bbfc62 PM |
652 | // Also if no expediting and no possible deboosting, |
653 | // slow is OK. Plus nohz_full CPUs eventually get | |
654 | // tick enabled. | |
05f41571 PM |
655 | set_tsk_need_resched(current); |
656 | set_preempt_need_resched(); | |
d143b3d1 | 657 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && |
39bbfc62 | 658 | expboost && !rdp->defer_qs_iw_pending && cpu_online(rdp->cpu)) { |
0864f057 PM |
659 | // Get scheduler to re-evaluate and call hooks. |
660 | // If !IRQ_WORK, FQS scan will eventually IPI. | |
39bbfc62 | 661 | init_irq_work(&rdp->defer_qs_iw, rcu_preempt_deferred_qs_handler); |
0864f057 PM |
662 | rdp->defer_qs_iw_pending = true; |
663 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); | |
664 | } | |
05f41571 | 665 | } |
3e310098 PM |
666 | local_irq_restore(flags); |
667 | return; | |
668 | } | |
669 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
670 | } | |
671 | ||
b0e165c0 PM |
672 | /* |
673 | * Check that the list of blocked tasks for the newly completed grace | |
674 | * period is in fact empty. It is a serious bug to complete a grace | |
675 | * period that still has RCU readers blocked! This function must be | |
03bd2983 | 676 | * invoked -before- updating this rnp's ->gp_seq. |
12f5f524 PM |
677 | * |
678 | * Also, if there are blocked tasks on the list, they automatically | |
679 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 680 | */ |
81ab59a3 | 681 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 682 | { |
c5ebe66c PM |
683 | struct task_struct *t; |
684 | ||
ea9b0c8a | 685 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
03bd2983 | 686 | raw_lockdep_assert_held_rcu_node(rnp); |
4bc8d555 | 687 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 688 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
689 | if (rcu_preempt_has_tasks(rnp) && |
690 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
6935c398 | 691 | WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next); |
c5ebe66c PM |
692 | t = container_of(rnp->gp_tasks, struct task_struct, |
693 | rcu_node_entry); | |
694 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 695 | rnp->gp_seq, t->pid); |
c5ebe66c | 696 | } |
28ecd580 | 697 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
698 | } |
699 | ||
f41d911f | 700 | /* |
c98cac60 PM |
701 | * Check for a quiescent state from the current CPU, including voluntary |
702 | * context switches for Tasks RCU. When a task blocks, the task is | |
703 | * recorded in the corresponding CPU's rcu_node structure, which is checked | |
704 | * elsewhere, hence this function need only check for quiescent states | |
705 | * related to the current CPU, not to those related to tasks. | |
f41d911f | 706 | */ |
c98cac60 | 707 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f PM |
708 | { |
709 | struct task_struct *t = current; | |
710 | ||
a649d25d | 711 | lockdep_assert_irqs_disabled(); |
45975c7d PM |
712 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
713 | rcu_note_voluntary_context_switch(current); | |
714 | } | |
77339e61 | 715 | if (rcu_preempt_depth() > 0 || |
3e310098 PM |
716 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { |
717 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
718 | if (rcu_preempt_need_deferred_qs(t)) { |
719 | set_tsk_need_resched(t); | |
720 | set_preempt_need_resched(); | |
721 | } | |
3e310098 PM |
722 | } else if (rcu_preempt_need_deferred_qs(t)) { |
723 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
724 | return; | |
5f5fa7ea | 725 | } else if (!WARN_ON_ONCE(rcu_preempt_depth())) { |
45975c7d | 726 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
727 | return; |
728 | } | |
3e310098 PM |
729 | |
730 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
77339e61 | 731 | if (rcu_preempt_depth() > 0 && |
2280ee5a PM |
732 | __this_cpu_read(rcu_data.core_needs_qs) && |
733 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 734 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 735 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 736 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
737 | } |
738 | ||
2439b696 PM |
739 | /* |
740 | * Check for a task exiting while in a preemptible-RCU read-side | |
884157ce PM |
741 | * critical section, clean up if so. No need to issue warnings, as |
742 | * debug_check_no_locks_held() already does this if lockdep is enabled. | |
743 | * Besides, if this function does anything other than just immediately | |
744 | * return, there was a bug of some sort. Spewing warnings from this | |
745 | * function is like as not to simply obscure important prior warnings. | |
2439b696 PM |
746 | */ |
747 | void exit_rcu(void) | |
748 | { | |
749 | struct task_struct *t = current; | |
750 | ||
884157ce | 751 | if (unlikely(!list_empty(¤t->rcu_node_entry))) { |
77339e61 | 752 | rcu_preempt_depth_set(1); |
884157ce | 753 | barrier(); |
add0d37b | 754 | WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true); |
77339e61 LJ |
755 | } else if (unlikely(rcu_preempt_depth())) { |
756 | rcu_preempt_depth_set(1); | |
884157ce | 757 | } else { |
2439b696 | 758 | return; |
884157ce | 759 | } |
2439b696 | 760 | __rcu_read_unlock(); |
3e310098 | 761 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
762 | } |
763 | ||
4bc8d555 PM |
764 | /* |
765 | * Dump the blocked-tasks state, but limit the list dump to the | |
766 | * specified number of elements. | |
767 | */ | |
57738942 | 768 | static void |
81ab59a3 | 769 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 770 | { |
57738942 | 771 | int cpu; |
4bc8d555 PM |
772 | int i; |
773 | struct list_head *lhp; | |
57738942 PM |
774 | bool onl; |
775 | struct rcu_data *rdp; | |
ff3cee39 | 776 | struct rcu_node *rnp1; |
4bc8d555 | 777 | |
ce11fae8 | 778 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 779 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 780 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
8ff37290 | 781 | (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs); |
ff3cee39 PM |
782 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) |
783 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
784 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf | 785 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
065a6db1 | 786 | __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks), |
314eeb43 | 787 | READ_ONCE(rnp->exp_tasks)); |
77cfc7bf | 788 | pr_info("%s: ->blkd_tasks", __func__); |
4bc8d555 PM |
789 | i = 0; |
790 | list_for_each(lhp, &rnp->blkd_tasks) { | |
791 | pr_cont(" %p", lhp); | |
cd6d17b4 | 792 | if (++i >= ncheck) |
4bc8d555 PM |
793 | break; |
794 | } | |
795 | pr_cont("\n"); | |
57738942 | 796 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 797 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 PM |
798 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); |
799 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
800 | cpu, ".o"[onl], | |
801 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
802 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
803 | } | |
4bc8d555 PM |
804 | } |
805 | ||
28f6569a | 806 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 807 | |
aa40c138 PM |
808 | /* |
809 | * If strict grace periods are enabled, and if the calling | |
810 | * __rcu_read_unlock() marks the beginning of a quiescent state, immediately | |
811 | * report that quiescent state and, if requested, spin for a bit. | |
812 | */ | |
813 | void rcu_read_unlock_strict(void) | |
814 | { | |
815 | struct rcu_data *rdp; | |
816 | ||
925da92b | 817 | if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread) |
aa40c138 PM |
818 | return; |
819 | rdp = this_cpu_ptr(&rcu_data); | |
cfeac397 | 820 | rcu_report_qs_rdp(rdp); |
aa40c138 PM |
821 | udelay(rcu_unlock_delay); |
822 | } | |
823 | EXPORT_SYMBOL_GPL(rcu_read_unlock_strict); | |
824 | ||
f41d911f PM |
825 | /* |
826 | * Tell them what RCU they are running. | |
827 | */ | |
0e0fc1c2 | 828 | static void __init rcu_bootup_announce(void) |
f41d911f | 829 | { |
efc151c3 | 830 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 831 | rcu_bootup_announce_oddness(); |
f41d911f PM |
832 | } |
833 | ||
45975c7d | 834 | /* |
90326f05 | 835 | * Note a quiescent state for PREEMPTION=n. Because we do not need to know |
45975c7d PM |
836 | * how many quiescent states passed, just if there was at least one since |
837 | * the start of the grace period, this just sets a flag. The caller must | |
838 | * have disabled preemption. | |
839 | */ | |
840 | static void rcu_qs(void) | |
d28139c4 | 841 | { |
45975c7d PM |
842 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
843 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
844 | return; | |
845 | trace_rcu_grace_period(TPS("rcu_sched"), | |
846 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
847 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
848 | if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) | |
849 | return; | |
850 | __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false); | |
63d4c8c9 | 851 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
d28139c4 PM |
852 | } |
853 | ||
395a2f09 PM |
854 | /* |
855 | * Register an urgently needed quiescent state. If there is an | |
856 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
857 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
858 | * some cases serve for expedited as well as normal grace periods. | |
859 | * Either way, register a lightweight quiescent state. | |
395a2f09 PM |
860 | */ |
861 | void rcu_all_qs(void) | |
862 | { | |
863 | unsigned long flags; | |
864 | ||
2dba13f0 | 865 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
866 | return; |
867 | preempt_disable(); | |
868 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 869 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
870 | preempt_enable(); |
871 | return; | |
872 | } | |
2dba13f0 | 873 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
2dba13f0 | 874 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
875 | local_irq_save(flags); |
876 | rcu_momentary_dyntick_idle(); | |
877 | local_irq_restore(flags); | |
878 | } | |
7e28c5af | 879 | rcu_qs(); |
395a2f09 PM |
880 | preempt_enable(); |
881 | } | |
882 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
883 | ||
cba6d0d6 | 884 | /* |
90326f05 | 885 | * Note a PREEMPTION=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 886 | */ |
45975c7d | 887 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 888 | { |
45975c7d PM |
889 | trace_rcu_utilization(TPS("Start context switch")); |
890 | rcu_qs(); | |
891 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 892 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 893 | goto out; |
2dba13f0 PM |
894 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
895 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 896 | rcu_momentary_dyntick_idle(); |
43766c3e | 897 | rcu_tasks_qs(current, preempt); |
45975c7d PM |
898 | out: |
899 | trace_rcu_utilization(TPS("End context switch")); | |
cba6d0d6 | 900 | } |
45975c7d | 901 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 902 | |
fc2219d4 | 903 | /* |
6cc68793 | 904 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
905 | * RCU readers. |
906 | */ | |
27f4d280 | 907 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
908 | { |
909 | return 0; | |
910 | } | |
911 | ||
8af3a5e7 PM |
912 | /* |
913 | * Because there is no preemptible RCU, there can be no readers blocked. | |
914 | */ | |
915 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 916 | { |
8af3a5e7 | 917 | return false; |
b668c9cf PM |
918 | } |
919 | ||
3e310098 PM |
920 | /* |
921 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
922 | * states. | |
923 | */ | |
924 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
925 | { | |
926 | return false; | |
927 | } | |
928 | static void rcu_preempt_deferred_qs(struct task_struct *t) { } | |
929 | ||
b0e165c0 | 930 | /* |
6cc68793 | 931 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
932 | * so there is no need to check for blocked tasks. So check only for |
933 | * bogus qsmask values. | |
b0e165c0 | 934 | */ |
81ab59a3 | 935 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 936 | { |
49e29126 | 937 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
938 | } |
939 | ||
f41d911f | 940 | /* |
c98cac60 PM |
941 | * Check to see if this CPU is in a non-context-switch quiescent state, |
942 | * namely user mode and idle loop. | |
f41d911f | 943 | */ |
c98cac60 | 944 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f | 945 | { |
45975c7d | 946 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 947 | |
45975c7d PM |
948 | /* |
949 | * Get here if this CPU took its interrupt from user | |
950 | * mode or from the idle loop, and if this is not a | |
951 | * nested interrupt. In this case, the CPU is in | |
952 | * a quiescent state, so note it. | |
953 | * | |
954 | * No memory barrier is required here because rcu_qs() | |
955 | * references only CPU-local variables that other CPUs | |
956 | * neither access nor modify, at least not while the | |
957 | * corresponding CPU is online. | |
958 | */ | |
959 | ||
960 | rcu_qs(); | |
961 | } | |
e74f4c45 | 962 | } |
e74f4c45 | 963 | |
2439b696 PM |
964 | /* |
965 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
966 | * while in preemptible RCU read-side critical sections. | |
967 | */ | |
968 | void exit_rcu(void) | |
969 | { | |
970 | } | |
971 | ||
4bc8d555 PM |
972 | /* |
973 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
974 | */ | |
57738942 | 975 | static void |
81ab59a3 | 976 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
977 | { |
978 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
979 | } | |
980 | ||
28f6569a | 981 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 982 | |
48d07c04 SAS |
983 | /* |
984 | * If boosting, set rcuc kthreads to realtime priority. | |
985 | */ | |
986 | static void rcu_cpu_kthread_setup(unsigned int cpu) | |
987 | { | |
27f4d280 | 988 | #ifdef CONFIG_RCU_BOOST |
48d07c04 | 989 | struct sched_param sp; |
27f4d280 | 990 | |
48d07c04 SAS |
991 | sp.sched_priority = kthread_prio; |
992 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
993 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
5d01bbd1 TG |
994 | } |
995 | ||
48d07c04 SAS |
996 | #ifdef CONFIG_RCU_BOOST |
997 | ||
27f4d280 PM |
998 | /* |
999 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1000 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1001 | * ->blkd_tasks list. | |
1002 | * | |
1003 | * Note that irqs must be enabled: boosting the task can block. | |
1004 | * Returns 1 if there are more tasks needing to be boosted. | |
1005 | */ | |
1006 | static int rcu_boost(struct rcu_node *rnp) | |
1007 | { | |
1008 | unsigned long flags; | |
27f4d280 PM |
1009 | struct task_struct *t; |
1010 | struct list_head *tb; | |
1011 | ||
7d0ae808 PM |
1012 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1013 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1014 | return 0; /* Nothing left to boost. */ |
1015 | ||
2a67e741 | 1016 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1017 | |
1018 | /* | |
1019 | * Recheck under the lock: all tasks in need of boosting | |
1020 | * might exit their RCU read-side critical sections on their own. | |
1021 | */ | |
1022 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1023 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1024 | return 0; |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Preferentially boost tasks blocking expedited grace periods. | |
1029 | * This cannot starve the normal grace periods because a second | |
1030 | * expedited grace period must boost all blocked tasks, including | |
1031 | * those blocking the pre-existing normal grace period. | |
1032 | */ | |
bec06785 | 1033 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1034 | tb = rnp->exp_tasks; |
bec06785 | 1035 | else |
27f4d280 PM |
1036 | tb = rnp->boost_tasks; |
1037 | ||
1038 | /* | |
1039 | * We boost task t by manufacturing an rt_mutex that appears to | |
1040 | * be held by task t. We leave a pointer to that rt_mutex where | |
1041 | * task t can find it, and task t will release the mutex when it | |
1042 | * exits its outermost RCU read-side critical section. Then | |
1043 | * simply acquiring this artificial rt_mutex will boost task | |
1044 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1045 | * | |
1046 | * Note that task t must acquire rnp->lock to remove itself from | |
1047 | * the ->blkd_tasks list, which it will do from exit() if from | |
1048 | * nowhere else. We therefore are guaranteed that task t will | |
1049 | * stay around at least until we drop rnp->lock. Note that | |
1050 | * rnp->lock also resolves races between our priority boosting | |
1051 | * and task t's exiting its outermost RCU read-side critical | |
1052 | * section. | |
1053 | */ | |
1054 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
830e6acc | 1055 | rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t); |
67c583a7 | 1056 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1057 | /* Lock only for side effect: boosts task t's priority. */ |
1058 | rt_mutex_lock(&rnp->boost_mtx); | |
1059 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
396eba65 | 1060 | rnp->n_boosts++; |
27f4d280 | 1061 | |
7d0ae808 PM |
1062 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1063 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1064 | } |
1065 | ||
27f4d280 | 1066 | /* |
bc17ea10 | 1067 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1068 | */ |
1069 | static int rcu_boost_kthread(void *arg) | |
1070 | { | |
1071 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1072 | int spincnt = 0; | |
1073 | int more2boost; | |
1074 | ||
f7f7bac9 | 1075 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1076 | for (;;) { |
3ca3b0e2 | 1077 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING); |
f7f7bac9 | 1078 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
065a6db1 PM |
1079 | rcu_wait(READ_ONCE(rnp->boost_tasks) || |
1080 | READ_ONCE(rnp->exp_tasks)); | |
f7f7bac9 | 1081 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
3ca3b0e2 | 1082 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING); |
27f4d280 PM |
1083 | more2boost = rcu_boost(rnp); |
1084 | if (more2boost) | |
1085 | spincnt++; | |
1086 | else | |
1087 | spincnt = 0; | |
1088 | if (spincnt > 10) { | |
3ca3b0e2 | 1089 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING); |
f7f7bac9 | 1090 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
a9352f72 | 1091 | schedule_timeout_idle(2); |
f7f7bac9 | 1092 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1093 | spincnt = 0; |
1094 | } | |
1095 | } | |
1217ed1b | 1096 | /* NOTREACHED */ |
f7f7bac9 | 1097 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1098 | return 0; |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Check to see if it is time to start boosting RCU readers that are | |
1103 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1104 | * kthread to start boosting them. If there is an expedited grace | |
1105 | * period in progress, it is always time to boost. | |
1106 | * | |
b065a853 PM |
1107 | * The caller must hold rnp->lock, which this function releases. |
1108 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1109 | * about it going away. | |
27f4d280 | 1110 | */ |
1217ed1b | 1111 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1112 | __releases(rnp->lock) |
27f4d280 | 1113 | { |
a32e01ee | 1114 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1115 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1116 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1117 | return; |
0ea1f2eb | 1118 | } |
27f4d280 PM |
1119 | if (rnp->exp_tasks != NULL || |
1120 | (rnp->gp_tasks != NULL && | |
1121 | rnp->boost_tasks == NULL && | |
1122 | rnp->qsmask == 0 && | |
7b241311 | 1123 | (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld))) { |
27f4d280 | 1124 | if (rnp->exp_tasks == NULL) |
5822b812 | 1125 | WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks); |
67c583a7 | 1126 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
a2badefa | 1127 | rcu_wake_cond(rnp->boost_kthread_task, |
3ca3b0e2 | 1128 | READ_ONCE(rnp->boost_kthread_status)); |
1217ed1b | 1129 | } else { |
67c583a7 | 1130 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1131 | } |
27f4d280 PM |
1132 | } |
1133 | ||
dff1672d PM |
1134 | /* |
1135 | * Is the current CPU running the RCU-callbacks kthread? | |
1136 | * Caller must have preemption disabled. | |
1137 | */ | |
1138 | static bool rcu_is_callbacks_kthread(void) | |
1139 | { | |
37f62d7c | 1140 | return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current; |
dff1672d PM |
1141 | } |
1142 | ||
27f4d280 PM |
1143 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1144 | ||
1145 | /* | |
1146 | * Do priority-boost accounting for the start of a new grace period. | |
1147 | */ | |
1148 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1149 | { | |
1150 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1151 | } | |
1152 | ||
27f4d280 PM |
1153 | /* |
1154 | * Create an RCU-boost kthread for the specified node if one does not | |
1155 | * already exist. We only create this kthread for preemptible RCU. | |
1156 | * Returns zero if all is well, a negated errno otherwise. | |
1157 | */ | |
3545832f | 1158 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 PM |
1159 | { |
1160 | unsigned long flags; | |
3ef5a1c3 | 1161 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1162 | struct sched_param sp; |
1163 | struct task_struct *t; | |
1164 | ||
3ef5a1c3 | 1165 | if (rnp->boost_kthread_task || !rcu_scheduler_fully_active) |
3545832f | 1166 | return; |
5d01bbd1 | 1167 | |
6dbfdc14 | 1168 | rcu_state.boost = 1; |
3545832f | 1169 | |
27f4d280 | 1170 | t = kthread_create(rcu_boost_kthread, (void *)rnp, |
5b61b0ba | 1171 | "rcub/%d", rnp_index); |
3545832f BP |
1172 | if (WARN_ON_ONCE(IS_ERR(t))) |
1173 | return; | |
1174 | ||
2a67e741 | 1175 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1176 | rnp->boost_kthread_task = t; |
67c583a7 | 1177 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1178 | sp.sched_priority = kthread_prio; |
27f4d280 | 1179 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1180 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1181 | } |
1182 | ||
f8b7fc6b PM |
1183 | /* |
1184 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1185 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1186 | * held, so the value of rnp->qsmaskinit will be stable. | |
1187 | * | |
1188 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1189 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1190 | * this function allows the kthread to execute on any CPU. | |
1191 | */ | |
5d01bbd1 | 1192 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1193 | { |
5d01bbd1 | 1194 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1195 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1196 | cpumask_var_t cm; |
1197 | int cpu; | |
f8b7fc6b | 1198 | |
5d01bbd1 | 1199 | if (!t) |
f8b7fc6b | 1200 | return; |
5d01bbd1 | 1201 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1202 | return; |
bc75e999 MR |
1203 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1204 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1205 | cpu != outgoingcpu) | |
f8b7fc6b | 1206 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1207 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1208 | cpumask_setall(cm); |
5d01bbd1 | 1209 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1210 | free_cpumask_var(cm); |
1211 | } | |
1212 | ||
f8b7fc6b | 1213 | /* |
9386c0b7 | 1214 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1215 | */ |
9386c0b7 | 1216 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1217 | { |
f8b7fc6b PM |
1218 | struct rcu_node *rnp; |
1219 | ||
aedf4ba9 | 1220 | rcu_for_each_leaf_node(rnp) |
3ef5a1c3 PM |
1221 | if (rcu_rnp_online_cpus(rnp)) |
1222 | rcu_spawn_one_boost_kthread(rnp); | |
f8b7fc6b PM |
1223 | } |
1224 | ||
27f4d280 PM |
1225 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1226 | ||
1217ed1b | 1227 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1228 | __releases(rnp->lock) |
27f4d280 | 1229 | { |
67c583a7 | 1230 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1231 | } |
1232 | ||
dff1672d PM |
1233 | static bool rcu_is_callbacks_kthread(void) |
1234 | { | |
1235 | return false; | |
1236 | } | |
1237 | ||
27f4d280 PM |
1238 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1239 | { | |
1240 | } | |
1241 | ||
3ef5a1c3 | 1242 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
f8b7fc6b PM |
1243 | { |
1244 | } | |
1245 | ||
3ef5a1c3 | 1246 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
b0d30417 | 1247 | { |
b0d30417 | 1248 | } |
b0d30417 | 1249 | |
3ef5a1c3 | 1250 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b PM |
1251 | { |
1252 | } | |
1253 | ||
27f4d280 PM |
1254 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1255 | ||
8bd93a2c PM |
1256 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1257 | ||
1258 | /* | |
0bd55c69 PM |
1259 | * Check to see if any future non-offloaded RCU-related work will need |
1260 | * to be done by the current CPU, even if none need be done immediately, | |
1261 | * returning 1 if so. This function is part of the RCU implementation; | |
1262 | * it is -not- an exported member of the RCU API. | |
8bd93a2c | 1263 | * |
0ae86a27 PM |
1264 | * Because we not have RCU_FAST_NO_HZ, just check whether or not this |
1265 | * CPU has RCU callbacks queued. | |
8bd93a2c | 1266 | */ |
c1ad348b | 1267 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1268 | { |
c1ad348b | 1269 | *nextevt = KTIME_MAX; |
0bd55c69 | 1270 | return !rcu_segcblist_empty(&this_cpu_ptr(&rcu_data)->cblist) && |
3820b513 | 1271 | !rcu_rdp_is_offloaded(this_cpu_ptr(&rcu_data)); |
7cb92499 PM |
1272 | } |
1273 | ||
1274 | /* | |
1275 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1276 | * after it. | |
1277 | */ | |
8fa7845d | 1278 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1279 | { |
1280 | } | |
1281 | ||
aea1b35e | 1282 | /* |
a858af28 | 1283 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1284 | * is nothing. |
1285 | */ | |
198bbf81 | 1286 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1287 | { |
1288 | } | |
1289 | ||
8bd93a2c PM |
1290 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1291 | ||
f23f7fa1 PM |
1292 | /* |
1293 | * This code is invoked when a CPU goes idle, at which point we want | |
1294 | * to have the CPU do everything required for RCU so that it can enter | |
77a40f97 | 1295 | * the energy-efficient dyntick-idle mode. |
f23f7fa1 | 1296 | * |
77a40f97 | 1297 | * The following preprocessor symbol controls this: |
f23f7fa1 | 1298 | * |
f23f7fa1 PM |
1299 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1300 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1301 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1302 | * benchmarkers who might otherwise be tempted to set this to a large | |
1303 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1304 | * system. And if you are -that- concerned about energy efficiency, | |
1305 | * just power the system down and be done with it! | |
1306 | * | |
77a40f97 | 1307 | * The value below works well in practice. If future workloads require |
f23f7fa1 PM |
1308 | * adjustment, they can be converted into kernel config parameters, though |
1309 | * making the state machine smarter might be a better option. | |
1310 | */ | |
e84c48ae | 1311 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
f23f7fa1 | 1312 | |
5e44ce35 PM |
1313 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1314 | module_param(rcu_idle_gp_delay, int, 0644); | |
486e2593 | 1315 | |
486e2593 | 1316 | /* |
0ae86a27 PM |
1317 | * Try to advance callbacks on the current CPU, but only if it has been |
1318 | * awhile since the last time we did so. Afterwards, if there are any | |
1319 | * callbacks ready for immediate invocation, return true. | |
486e2593 | 1320 | */ |
f1f399d1 | 1321 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1322 | { |
c0f4dfd4 | 1323 | bool cbs_ready = false; |
5998a75a | 1324 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1325 | struct rcu_node *rnp; |
486e2593 | 1326 | |
c229828c | 1327 | /* Exit early if we advanced recently. */ |
5998a75a | 1328 | if (jiffies == rdp->last_advance_all) |
d0bc90fd | 1329 | return false; |
5998a75a | 1330 | rdp->last_advance_all = jiffies; |
c229828c | 1331 | |
b97d23c5 | 1332 | rnp = rdp->mynode; |
486e2593 | 1333 | |
b97d23c5 PM |
1334 | /* |
1335 | * Don't bother checking unless a grace period has | |
1336 | * completed since we last checked and there are | |
1337 | * callbacks not yet ready to invoke. | |
1338 | */ | |
1339 | if ((rcu_seq_completed_gp(rdp->gp_seq, | |
1340 | rcu_seq_current(&rnp->gp_seq)) || | |
1341 | unlikely(READ_ONCE(rdp->gpwrap))) && | |
1342 | rcu_segcblist_pend_cbs(&rdp->cblist)) | |
1343 | note_gp_changes(rdp); | |
1344 | ||
1345 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
1346 | cbs_ready = true; | |
c0f4dfd4 | 1347 | return cbs_ready; |
486e2593 PM |
1348 | } |
1349 | ||
aa9b1630 | 1350 | /* |
c0f4dfd4 PM |
1351 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1352 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
77a40f97 | 1353 | * caller about what to set the timeout. |
aa9b1630 | 1354 | * |
c0f4dfd4 | 1355 | * The caller must have disabled interrupts. |
aa9b1630 | 1356 | */ |
c1ad348b | 1357 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1358 | { |
5998a75a | 1359 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c1ad348b | 1360 | unsigned long dj; |
aa9b1630 | 1361 | |
b04db8e1 | 1362 | lockdep_assert_irqs_disabled(); |
3382adbc | 1363 | |
0bd55c69 PM |
1364 | /* If no non-offloaded callbacks, RCU doesn't need the CPU. */ |
1365 | if (rcu_segcblist_empty(&rdp->cblist) || | |
3820b513 | 1366 | rcu_rdp_is_offloaded(rdp)) { |
c1ad348b | 1367 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1368 | return 0; |
1369 | } | |
c0f4dfd4 PM |
1370 | |
1371 | /* Attempt to advance callbacks. */ | |
1372 | if (rcu_try_advance_all_cbs()) { | |
1373 | /* Some ready to invoke, so initiate later invocation. */ | |
1374 | invoke_rcu_core(); | |
aa9b1630 PM |
1375 | return 1; |
1376 | } | |
5998a75a | 1377 | rdp->last_accelerate = jiffies; |
c0f4dfd4 | 1378 | |
77a40f97 JFG |
1379 | /* Request timer and round. */ |
1380 | dj = round_up(rcu_idle_gp_delay + jiffies, rcu_idle_gp_delay) - jiffies; | |
1381 | ||
c1ad348b | 1382 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1383 | return 0; |
1384 | } | |
1385 | ||
21e52e15 | 1386 | /* |
77a40f97 JFG |
1387 | * Prepare a CPU for idle from an RCU perspective. The first major task is to |
1388 | * sense whether nohz mode has been enabled or disabled via sysfs. The second | |
1389 | * major task is to accelerate (that is, assign grace-period numbers to) any | |
1390 | * recently arrived callbacks. | |
aea1b35e PM |
1391 | * |
1392 | * The caller must have disabled interrupts. | |
8bd93a2c | 1393 | */ |
198bbf81 | 1394 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1395 | { |
48a7639c | 1396 | bool needwake; |
0fd79e75 | 1397 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1398 | struct rcu_node *rnp; |
9d2ad243 PM |
1399 | int tne; |
1400 | ||
b04db8e1 | 1401 | lockdep_assert_irqs_disabled(); |
3820b513 | 1402 | if (rcu_rdp_is_offloaded(rdp)) |
3382adbc PM |
1403 | return; |
1404 | ||
9d2ad243 | 1405 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1406 | tne = READ_ONCE(tick_nohz_active); |
0fd79e75 | 1407 | if (tne != rdp->tick_nohz_enabled_snap) { |
260e1e4f | 1408 | if (!rcu_segcblist_empty(&rdp->cblist)) |
9d2ad243 | 1409 | invoke_rcu_core(); /* force nohz to see update. */ |
0fd79e75 | 1410 | rdp->tick_nohz_enabled_snap = tne; |
9d2ad243 PM |
1411 | return; |
1412 | } | |
1413 | if (!tne) | |
1414 | return; | |
f511fc62 | 1415 | |
3084f2f8 | 1416 | /* |
c0f4dfd4 PM |
1417 | * If we have not yet accelerated this jiffy, accelerate all |
1418 | * callbacks on this CPU. | |
3084f2f8 | 1419 | */ |
5998a75a | 1420 | if (rdp->last_accelerate == jiffies) |
aea1b35e | 1421 | return; |
5998a75a | 1422 | rdp->last_accelerate = jiffies; |
b97d23c5 | 1423 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
c0f4dfd4 | 1424 | rnp = rdp->mynode; |
2a67e741 | 1425 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
02f50142 | 1426 | needwake = rcu_accelerate_cbs(rnp, rdp); |
67c583a7 | 1427 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c | 1428 | if (needwake) |
532c00c9 | 1429 | rcu_gp_kthread_wake(); |
77e38ed3 | 1430 | } |
c0f4dfd4 | 1431 | } |
3084f2f8 | 1432 | |
c0f4dfd4 PM |
1433 | /* |
1434 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1435 | * any grace periods that elapsed while the CPU was idle, and if any | |
1436 | * callbacks are now ready to invoke, initiate invocation. | |
1437 | */ | |
8fa7845d | 1438 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1439 | { |
ce5215c1 PM |
1440 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
1441 | ||
b04db8e1 | 1442 | lockdep_assert_irqs_disabled(); |
3820b513 | 1443 | if (rcu_rdp_is_offloaded(rdp)) |
aea1b35e | 1444 | return; |
7a497c96 PM |
1445 | if (rcu_try_advance_all_cbs()) |
1446 | invoke_rcu_core(); | |
8bd93a2c PM |
1447 | } |
1448 | ||
1449 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ | |
a858af28 | 1450 | |
a096932f PM |
1451 | /* |
1452 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
1453 | * grace-period kthread will do force_quiescent_state() processing? | |
1454 | * The idea is to avoid waking up RCU core processing on such a | |
1455 | * CPU unless the grace period has extended for too long. | |
1456 | * | |
1457 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 1458 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f | 1459 | */ |
4580b054 | 1460 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
1461 | { |
1462 | #ifdef CONFIG_NO_HZ_FULL | |
1463 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 1464 | (!rcu_gp_in_progress() || |
e2f3ccfa | 1465 | time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 1466 | return true; |
a096932f | 1467 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 1468 | return false; |
a096932f | 1469 | } |
5057f55e PM |
1470 | |
1471 | /* | |
265f5f28 | 1472 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
1473 | */ |
1474 | static void rcu_bind_gp_kthread(void) | |
1475 | { | |
c0f489d2 | 1476 | if (!tick_nohz_full_enabled()) |
5057f55e | 1477 | return; |
de201559 | 1478 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 1479 | } |
176f8f7a PM |
1480 | |
1481 | /* Record the current task on dyntick-idle entry. */ | |
ff5c4f5c | 1482 | static void noinstr rcu_dynticks_task_enter(void) |
176f8f7a PM |
1483 | { |
1484 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 1485 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
1486 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
1487 | } | |
1488 | ||
1489 | /* Record no current task on dyntick-idle exit. */ | |
ff5c4f5c | 1490 | static void noinstr rcu_dynticks_task_exit(void) |
176f8f7a PM |
1491 | { |
1492 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 1493 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
1494 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
1495 | } | |
7d0c9c50 PM |
1496 | |
1497 | /* Turn on heavyweight RCU tasks trace readers on idle/user entry. */ | |
1498 | static void rcu_dynticks_task_trace_enter(void) | |
1499 | { | |
fed31a4d | 1500 | #ifdef CONFIG_TASKS_TRACE_RCU |
7d0c9c50 PM |
1501 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
1502 | current->trc_reader_special.b.need_mb = true; | |
fed31a4d | 1503 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
7d0c9c50 PM |
1504 | } |
1505 | ||
1506 | /* Turn off heavyweight RCU tasks trace readers on idle/user exit. */ | |
1507 | static void rcu_dynticks_task_trace_exit(void) | |
1508 | { | |
fed31a4d | 1509 | #ifdef CONFIG_TASKS_TRACE_RCU |
7d0c9c50 PM |
1510 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
1511 | current->trc_reader_special.b.need_mb = false; | |
fed31a4d | 1512 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
7d0c9c50 | 1513 | } |