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 | /* | |
17ea3718 | 19 | * In order to read the offloaded state of an rdp in a safe |
3820b513 FW |
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 | 53 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
c4a09ff7 | 54 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 55 | pr_info("\tRCU lockdep checking is enabled.\n"); |
8cbd0e38 | 56 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) |
17ea3718 | 57 | pr_info("\tRCU strict (and thus non-scalable) grace periods are enabled.\n"); |
42621697 AG |
58 | if (RCU_NUM_LVLS >= 4) |
59 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 60 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 61 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
62 | RCU_FANOUT_LEAF); |
63 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
a7538352 JP |
64 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", |
65 | rcu_fanout_leaf); | |
cca6f393 | 66 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 67 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b | 68 | #ifdef CONFIG_RCU_BOOST |
a7538352 JP |
69 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", |
70 | kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
17c7798b PM |
71 | #endif |
72 | if (blimit != DEFAULT_RCU_BLIMIT) | |
73 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
74 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
75 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
76 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
77 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
b2b00ddf | 78 | if (qovld != DEFAULT_RCU_QOVLD) |
aa96a93b | 79 | pr_info("\tBoot-time adjustment of callback overload level to %ld.\n", qovld); |
17c7798b PM |
80 | if (jiffies_till_first_fqs != ULONG_MAX) |
81 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
82 | if (jiffies_till_next_fqs != ULONG_MAX) | |
83 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
c06aed0e PM |
84 | if (jiffies_till_sched_qs != ULONG_MAX) |
85 | pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs); | |
17c7798b PM |
86 | if (rcu_kick_kthreads) |
87 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
88 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
17ea3718 | 89 | pr_info("\tRCU callback double-/use-after-free debug is enabled.\n"); |
90040c9e | 90 | if (gp_preinit_delay) |
17c7798b | 91 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 92 | if (gp_init_delay) |
17c7798b | 93 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 94 | if (gp_cleanup_delay) |
17ea3718 | 95 | pr_info("\tRCU debug GP cleanup slowdown %d jiffies.\n", gp_cleanup_delay); |
48d07c04 SAS |
96 | if (!use_softirq) |
97 | pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n"); | |
17c7798b PM |
98 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) |
99 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 100 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
101 | } |
102 | ||
28f6569a | 103 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 104 | |
63d4c8c9 | 105 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 106 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 107 | |
f41d911f PM |
108 | /* |
109 | * Tell them what RCU they are running. | |
110 | */ | |
0e0fc1c2 | 111 | static void __init rcu_bootup_announce(void) |
f41d911f | 112 | { |
efc151c3 | 113 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 114 | rcu_bootup_announce_oddness(); |
f41d911f PM |
115 | } |
116 | ||
8203d6d0 PM |
117 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
118 | #define RCU_GP_TASKS 0x8 | |
119 | #define RCU_EXP_TASKS 0x4 | |
120 | #define RCU_GP_BLKD 0x2 | |
121 | #define RCU_EXP_BLKD 0x1 | |
122 | ||
123 | /* | |
124 | * Queues a task preempted within an RCU-preempt read-side critical | |
125 | * section into the appropriate location within the ->blkd_tasks list, | |
126 | * depending on the states of any ongoing normal and expedited grace | |
127 | * periods. The ->gp_tasks pointer indicates which element the normal | |
128 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
129 | * indicates which element the expedited grace period is waiting on (again, | |
130 | * NULL if none). If a grace period is waiting on a given element in the | |
131 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
132 | * adding a task to the tail of the list blocks any grace period that is | |
133 | * already waiting on one of the elements. In contrast, adding a task | |
134 | * to the head of the list won't block any grace period that is already | |
135 | * waiting on one of the elements. | |
136 | * | |
137 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
138 | * period wait for a task that is not strictly speaking blocking it. | |
139 | * Given the choice, we needlessly block a normal grace period rather than | |
140 | * blocking an expedited grace period. | |
141 | * | |
142 | * Note that an endless sequence of expedited grace periods still cannot | |
143 | * indefinitely postpone a normal grace period. Eventually, all of the | |
144 | * fixed number of preempted tasks blocking the normal grace period that are | |
145 | * not also blocking the expedited grace period will resume and complete | |
146 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
147 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
148 | * the corresponding expedited grace period will also be the end of the | |
149 | * normal grace period. | |
150 | */ | |
46a5d164 PM |
151 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
152 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
153 | { |
154 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
155 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
156 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
157 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
158 | struct task_struct *t = current; | |
159 | ||
a32e01ee | 160 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 161 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 162 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
163 | /* RCU better not be waiting on newly onlined CPUs! */ |
164 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
165 | rdp->grpmask); | |
ea9b0c8a | 166 | |
8203d6d0 PM |
167 | /* |
168 | * Decide where to queue the newly blocked task. In theory, | |
169 | * this could be an if-statement. In practice, when I tried | |
170 | * that, it was quite messy. | |
171 | */ | |
172 | switch (blkd_state) { | |
173 | case 0: | |
174 | case RCU_EXP_TASKS: | |
175 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
176 | case RCU_GP_TASKS: | |
177 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
178 | ||
179 | /* | |
180 | * Blocking neither GP, or first task blocking the normal | |
181 | * GP but not blocking the already-waiting expedited GP. | |
182 | * Queue at the head of the list to avoid unnecessarily | |
183 | * blocking the already-waiting GPs. | |
184 | */ | |
185 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
186 | break; | |
187 | ||
188 | case RCU_EXP_BLKD: | |
189 | case RCU_GP_BLKD: | |
190 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
191 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
192 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
193 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
194 | ||
195 | /* | |
196 | * First task arriving that blocks either GP, or first task | |
197 | * arriving that blocks the expedited GP (with the normal | |
198 | * GP already waiting), or a task arriving that blocks | |
199 | * both GPs with both GPs already waiting. Queue at the | |
200 | * tail of the list to avoid any GP waiting on any of the | |
201 | * already queued tasks that are not blocking it. | |
202 | */ | |
203 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
204 | break; | |
205 | ||
206 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
207 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
208 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
209 | ||
210 | /* | |
211 | * Second or subsequent task blocking the expedited GP. | |
212 | * The task either does not block the normal GP, or is the | |
213 | * first task blocking the normal GP. Queue just after | |
214 | * the first task blocking the expedited GP. | |
215 | */ | |
216 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
217 | break; | |
218 | ||
219 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
220 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
221 | ||
222 | /* | |
223 | * Second or subsequent task blocking the normal GP. | |
224 | * The task does not block the expedited GP. Queue just | |
225 | * after the first task blocking the normal GP. | |
226 | */ | |
227 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
228 | break; | |
229 | ||
230 | default: | |
231 | ||
232 | /* Yet another exercise in excessive paranoia. */ | |
233 | WARN_ON_ONCE(1); | |
234 | break; | |
235 | } | |
236 | ||
237 | /* | |
238 | * We have now queued the task. If it was the first one to | |
239 | * block either grace period, update the ->gp_tasks and/or | |
240 | * ->exp_tasks pointers, respectively, to reference the newly | |
241 | * blocked tasks. | |
242 | */ | |
4bc8d555 | 243 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
6935c398 | 244 | WRITE_ONCE(rnp->gp_tasks, &t->rcu_node_entry); |
d43a5d32 | 245 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 246 | } |
8203d6d0 | 247 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
314eeb43 | 248 | WRITE_ONCE(rnp->exp_tasks, &t->rcu_node_entry); |
2dee9404 PM |
249 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
250 | !(rnp->qsmask & rdp->grpmask)); | |
251 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
252 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 253 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
254 | |
255 | /* | |
256 | * Report the quiescent state for the expedited GP. This expedited | |
257 | * GP should not be able to end until we report, so there should be | |
258 | * no need to check for a subsequent expedited GP. (Though we are | |
259 | * still in a quiescent state in any case.) | |
260 | */ | |
6120b72e | 261 | if (blkd_state & RCU_EXP_BLKD && rdp->cpu_no_qs.b.exp) |
63d4c8c9 | 262 | rcu_report_exp_rdp(rdp); |
fcc878e4 | 263 | else |
6120b72e | 264 | WARN_ON_ONCE(rdp->cpu_no_qs.b.exp); |
8203d6d0 PM |
265 | } |
266 | ||
f41d911f | 267 | /* |
c7037ff5 PM |
268 | * Record a preemptible-RCU quiescent state for the specified CPU. |
269 | * Note that this does not necessarily mean that the task currently running | |
270 | * on the CPU is in a quiescent state: Instead, it means that the current | |
271 | * grace period need not wait on any RCU read-side critical section that | |
272 | * starts later on this CPU. It also means that if the current task is | |
273 | * in an RCU read-side critical section, it has already added itself to | |
274 | * some leaf rcu_node structure's ->blkd_tasks list. In addition to the | |
275 | * current task, there might be any number of other tasks blocked while | |
276 | * in an RCU read-side critical section. | |
25502a6c | 277 | * |
a4382659 FW |
278 | * Unlike non-preemptible-RCU, quiescent state reports for expedited |
279 | * grace periods are handled separately via deferred quiescent states | |
280 | * and context switch events. | |
281 | * | |
c7037ff5 | 282 | * Callers to this function must disable preemption. |
f41d911f | 283 | */ |
45975c7d | 284 | static void rcu_qs(void) |
f41d911f | 285 | { |
45975c7d | 286 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n"); |
a4382659 | 287 | if (__this_cpu_read(rcu_data.cpu_no_qs.b.norm)) { |
284a8c93 | 288 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2280ee5a | 289 | __this_cpu_read(rcu_data.gp_seq), |
284a8c93 | 290 | TPS("cpuqs")); |
2280ee5a | 291 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); |
c98cac60 | 292 | barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */ |
add0d37b | 293 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false); |
284a8c93 | 294 | } |
f41d911f PM |
295 | } |
296 | ||
297 | /* | |
c3422bea PM |
298 | * We have entered the scheduler, and the current task might soon be |
299 | * context-switched away from. If this task is in an RCU read-side | |
300 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
301 | * record that fact, so we enqueue the task on the blkd_tasks list. |
302 | * The task will dequeue itself when it exits the outermost enclosing | |
303 | * RCU read-side critical section. Therefore, the current grace period | |
304 | * cannot be permitted to complete until the blkd_tasks list entries | |
305 | * predating the current grace period drain, in other words, until | |
306 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 307 | * |
46a5d164 | 308 | * Caller must disable interrupts. |
f41d911f | 309 | */ |
45975c7d | 310 | void rcu_note_context_switch(bool preempt) |
f41d911f PM |
311 | { |
312 | struct task_struct *t = current; | |
da1df50d | 313 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
f41d911f PM |
314 | struct rcu_node *rnp; |
315 | ||
45975c7d | 316 | trace_rcu_utilization(TPS("Start context switch")); |
b04db8e1 | 317 | lockdep_assert_irqs_disabled(); |
521c89b3 | 318 | WARN_ONCE(!preempt && rcu_preempt_depth() > 0, "Voluntary context switch within RCU read-side critical section!"); |
77339e61 | 319 | if (rcu_preempt_depth() > 0 && |
1d082fd0 | 320 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
321 | |
322 | /* Possibly blocking in an RCU read-side critical section. */ | |
f41d911f | 323 | rnp = rdp->mynode; |
46a5d164 | 324 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 325 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 326 | t->rcu_blocked_node = rnp; |
f41d911f PM |
327 | |
328 | /* | |
8203d6d0 PM |
329 | * Verify the CPU's sanity, trace the preemption, and |
330 | * then queue the task as required based on the states | |
331 | * of any ongoing and expedited grace periods. | |
f41d911f | 332 | */ |
5ae0f1b5 | 333 | WARN_ON_ONCE(!rcu_rdp_cpu_online(rdp)); |
e7d8842e | 334 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
88d1bead | 335 | trace_rcu_preempt_task(rcu_state.name, |
d4c08f2a PM |
336 | t->pid, |
337 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
338 | ? rnp->gp_seq |
339 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 340 | rcu_preempt_ctxt_queue(rnp, rdp); |
3e310098 PM |
341 | } else { |
342 | rcu_preempt_deferred_qs(t); | |
f41d911f PM |
343 | } |
344 | ||
345 | /* | |
346 | * Either we were not in an RCU read-side critical section to | |
347 | * begin with, or we have now recorded that critical section | |
348 | * globally. Either way, we can now note a quiescent state | |
349 | * for this CPU. Again, if we were in an RCU read-side critical | |
350 | * section, and if that critical section was blocking the current | |
351 | * grace period, then the fact that the task has been enqueued | |
352 | * means that we continue to block the current grace period. | |
353 | */ | |
45975c7d | 354 | rcu_qs(); |
6120b72e | 355 | if (rdp->cpu_no_qs.b.exp) |
63d4c8c9 | 356 | rcu_report_exp_rdp(rdp); |
43766c3e | 357 | rcu_tasks_qs(current, preempt); |
45975c7d | 358 | trace_rcu_utilization(TPS("End context switch")); |
f41d911f | 359 | } |
45975c7d | 360 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
f41d911f | 361 | |
fc2219d4 PM |
362 | /* |
363 | * Check for preempted RCU readers blocking the current grace period | |
364 | * for the specified rcu_node structure. If the caller needs a reliable | |
365 | * answer, it must hold the rcu_node's ->lock. | |
366 | */ | |
27f4d280 | 367 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 368 | { |
6935c398 | 369 | return READ_ONCE(rnp->gp_tasks) != NULL; |
fc2219d4 PM |
370 | } |
371 | ||
5f5fa7ea | 372 | /* limit value for ->rcu_read_lock_nesting. */ |
5f1a6ef3 PM |
373 | #define RCU_NEST_PMAX (INT_MAX / 2) |
374 | ||
77339e61 LJ |
375 | static void rcu_preempt_read_enter(void) |
376 | { | |
5fcb3a5f | 377 | WRITE_ONCE(current->rcu_read_lock_nesting, READ_ONCE(current->rcu_read_lock_nesting) + 1); |
77339e61 LJ |
378 | } |
379 | ||
5f5fa7ea | 380 | static int rcu_preempt_read_exit(void) |
77339e61 | 381 | { |
5fcb3a5f PM |
382 | int ret = READ_ONCE(current->rcu_read_lock_nesting) - 1; |
383 | ||
384 | WRITE_ONCE(current->rcu_read_lock_nesting, ret); | |
385 | return ret; | |
77339e61 LJ |
386 | } |
387 | ||
388 | static void rcu_preempt_depth_set(int val) | |
389 | { | |
5fcb3a5f | 390 | WRITE_ONCE(current->rcu_read_lock_nesting, val); |
77339e61 LJ |
391 | } |
392 | ||
0e5da22e PM |
393 | /* |
394 | * Preemptible RCU implementation for rcu_read_lock(). | |
395 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
396 | * if we block. | |
397 | */ | |
398 | void __rcu_read_lock(void) | |
399 | { | |
77339e61 | 400 | rcu_preempt_read_enter(); |
5f1a6ef3 | 401 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
77339e61 | 402 | WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX); |
f19920e4 PM |
403 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread) |
404 | WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true); | |
0e5da22e PM |
405 | barrier(); /* critical section after entry code. */ |
406 | } | |
407 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
408 | ||
409 | /* | |
410 | * Preemptible RCU implementation for rcu_read_unlock(). | |
411 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
412 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
413 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
414 | * in an RCU read-side critical section and other special cases. | |
415 | */ | |
416 | void __rcu_read_unlock(void) | |
417 | { | |
418 | struct task_struct *t = current; | |
419 | ||
7e937220 | 420 | barrier(); // critical section before exit code. |
5f5fa7ea | 421 | if (rcu_preempt_read_exit() == 0) { |
7e937220 | 422 | barrier(); // critical-section exit before .s check. |
0e5da22e PM |
423 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) |
424 | rcu_read_unlock_special(t); | |
0e5da22e | 425 | } |
5f1a6ef3 | 426 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
77339e61 | 427 | int rrln = rcu_preempt_depth(); |
0e5da22e | 428 | |
5f5fa7ea | 429 | WARN_ON_ONCE(rrln < 0 || rrln > RCU_NEST_PMAX); |
0e5da22e | 430 | } |
0e5da22e PM |
431 | } |
432 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
433 | ||
12f5f524 PM |
434 | /* |
435 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
436 | * returning NULL if at the end of the list. | |
437 | */ | |
438 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
439 | struct rcu_node *rnp) | |
440 | { | |
441 | struct list_head *np; | |
442 | ||
443 | np = t->rcu_node_entry.next; | |
444 | if (np == &rnp->blkd_tasks) | |
445 | np = NULL; | |
446 | return np; | |
447 | } | |
448 | ||
8af3a5e7 PM |
449 | /* |
450 | * Return true if the specified rcu_node structure has tasks that were | |
451 | * preempted within an RCU read-side critical section. | |
452 | */ | |
453 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
454 | { | |
455 | return !list_empty(&rnp->blkd_tasks); | |
456 | } | |
457 | ||
b668c9cf | 458 | /* |
3e310098 PM |
459 | * Report deferred quiescent states. The deferral time can |
460 | * be quite short, for example, in the case of the call from | |
461 | * rcu_read_unlock_special(). | |
b668c9cf | 462 | */ |
3e310098 PM |
463 | static void |
464 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 465 | { |
b6a932d1 PM |
466 | bool empty_exp; |
467 | bool empty_norm; | |
468 | bool empty_exp_now; | |
12f5f524 | 469 | struct list_head *np; |
abaa93d9 | 470 | bool drop_boost_mutex = false; |
8203d6d0 | 471 | struct rcu_data *rdp; |
f41d911f | 472 | struct rcu_node *rnp; |
1d082fd0 | 473 | union rcu_special special; |
f41d911f | 474 | |
f41d911f | 475 | /* |
8203d6d0 PM |
476 | * If RCU core is waiting for this CPU to exit its critical section, |
477 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 478 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
479 | */ |
480 | special = t->rcu_read_unlock_special; | |
da1df50d | 481 | rdp = this_cpu_ptr(&rcu_data); |
6120b72e | 482 | if (!special.s && !rdp->cpu_no_qs.b.exp) { |
3e310098 PM |
483 | local_irq_restore(flags); |
484 | return; | |
485 | } | |
3717e1e9 | 486 | t->rcu_read_unlock_special.s = 0; |
44bad5b3 | 487 | if (special.b.need_qs) { |
3d29aaf1 | 488 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) { |
90d2efe7 | 489 | rdp->cpu_no_qs.b.norm = false; |
cfeac397 | 490 | rcu_report_qs_rdp(rdp); |
3d29aaf1 PM |
491 | udelay(rcu_unlock_delay); |
492 | } else { | |
44bad5b3 | 493 | rcu_qs(); |
3d29aaf1 | 494 | } |
44bad5b3 | 495 | } |
f41d911f | 496 | |
8203d6d0 | 497 | /* |
3e310098 PM |
498 | * Respond to a request by an expedited grace period for a |
499 | * quiescent state from this CPU. Note that requests from | |
500 | * tasks are handled when removing the task from the | |
501 | * blocked-tasks list below. | |
8203d6d0 | 502 | */ |
6120b72e | 503 | if (rdp->cpu_no_qs.b.exp) |
63d4c8c9 | 504 | rcu_report_exp_rdp(rdp); |
8203d6d0 | 505 | |
f41d911f | 506 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 | 507 | if (special.b.blocked) { |
f41d911f | 508 | |
dd5d19ba | 509 | /* |
0a0ba1c9 | 510 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
511 | * now remains queued on the rcu_node corresponding to the |
512 | * CPU it first blocked on, so there is no longer any need | |
513 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 514 | */ |
8ba9153b PM |
515 | rnp = t->rcu_blocked_node; |
516 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
517 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 518 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 519 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 520 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 521 | (!empty_norm || rnp->qsmask)); |
6c7d7dbf | 522 | empty_exp = sync_rcu_exp_done(rnp); |
d9a3da06 | 523 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 524 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 525 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 526 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 527 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 528 | rnp->gp_seq, t->pid); |
12f5f524 | 529 | if (&t->rcu_node_entry == rnp->gp_tasks) |
6935c398 | 530 | WRITE_ONCE(rnp->gp_tasks, np); |
12f5f524 | 531 | if (&t->rcu_node_entry == rnp->exp_tasks) |
314eeb43 | 532 | WRITE_ONCE(rnp->exp_tasks, np); |
727b705b | 533 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b | 534 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
830e6acc | 535 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx.rtmutex) == t; |
2dee9404 | 536 | if (&t->rcu_node_entry == rnp->boost_tasks) |
5822b812 | 537 | WRITE_ONCE(rnp->boost_tasks, np); |
727b705b | 538 | } |
f41d911f PM |
539 | |
540 | /* | |
541 | * If this was the last task on the current list, and if | |
542 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
543 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
544 | * so we must take a snapshot of the expedited state. | |
f41d911f | 545 | */ |
6c7d7dbf | 546 | empty_exp_now = sync_rcu_exp_done(rnp); |
74e871ac | 547 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 548 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 549 | rnp->gp_seq, |
d4c08f2a PM |
550 | 0, rnp->qsmask, |
551 | rnp->level, | |
552 | rnp->grplo, | |
553 | rnp->grphi, | |
554 | !!rnp->gp_tasks); | |
139ad4da | 555 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 556 | } else { |
67c583a7 | 557 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 558 | } |
d9a3da06 PM |
559 | |
560 | /* | |
561 | * If this was the last task on the expedited lists, | |
562 | * then we need to report up the rcu_node hierarchy. | |
563 | */ | |
389abd48 | 564 | if (!empty_exp && empty_exp_now) |
63d4c8c9 | 565 | rcu_report_exp_rnp(rnp, true); |
10c53578 PM |
566 | |
567 | /* Unboost if we were boosted. */ | |
568 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) | |
569 | rt_mutex_futex_unlock(&rnp->boost_mtx.rtmutex); | |
b668c9cf PM |
570 | } else { |
571 | local_irq_restore(flags); | |
f41d911f | 572 | } |
f41d911f PM |
573 | } |
574 | ||
3e310098 PM |
575 | /* |
576 | * Is a deferred quiescent-state pending, and are we also not in | |
577 | * an RCU read-side critical section? It is the caller's responsibility | |
578 | * to ensure it is otherwise safe to report any deferred quiescent | |
579 | * states. The reason for this is that it is safe to report a | |
580 | * quiescent state during context switch even though preemption | |
581 | * is disabled. This function cannot be expected to understand these | |
582 | * nuances, so the caller must handle them. | |
583 | */ | |
584 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
585 | { | |
6120b72e | 586 | return (__this_cpu_read(rcu_data.cpu_no_qs.b.exp) || |
3e310098 | 587 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
5f5fa7ea | 588 | rcu_preempt_depth() == 0; |
3e310098 PM |
589 | } |
590 | ||
591 | /* | |
592 | * Report a deferred quiescent state if needed and safe to do so. | |
593 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
594 | * not being in an RCU read-side critical section. The caller must | |
595 | * evaluate safety in terms of interrupt, softirq, and preemption | |
596 | * disabling. | |
597 | */ | |
598 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
599 | { | |
600 | unsigned long flags; | |
3e310098 PM |
601 | |
602 | if (!rcu_preempt_need_deferred_qs(t)) | |
603 | return; | |
3e310098 PM |
604 | local_irq_save(flags); |
605 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
3e310098 PM |
606 | } |
607 | ||
0864f057 PM |
608 | /* |
609 | * Minimal handler to give the scheduler a chance to re-evaluate. | |
610 | */ | |
611 | static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp) | |
612 | { | |
613 | struct rcu_data *rdp; | |
614 | ||
615 | rdp = container_of(iwp, struct rcu_data, defer_qs_iw); | |
616 | rdp->defer_qs_iw_pending = false; | |
617 | } | |
618 | ||
3e310098 PM |
619 | /* |
620 | * Handle special cases during rcu_read_unlock(), such as needing to | |
621 | * notify RCU core processing or task having blocked during the RCU | |
622 | * read-side critical section. | |
623 | */ | |
624 | static void rcu_read_unlock_special(struct task_struct *t) | |
625 | { | |
626 | unsigned long flags; | |
39bbfc62 | 627 | bool irqs_were_disabled; |
3e310098 PM |
628 | bool preempt_bh_were_disabled = |
629 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
3e310098 PM |
630 | |
631 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
632 | if (in_nmi()) | |
633 | return; | |
634 | ||
635 | local_irq_save(flags); | |
636 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 | 637 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
39bbfc62 | 638 | bool expboost; // Expedited GP in flight or possible boosting. |
25102de6 PM |
639 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
640 | struct rcu_node *rnp = rdp->mynode; | |
641 | ||
39bbfc62 PM |
642 | expboost = (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks)) || |
643 | (rdp->grpmask & READ_ONCE(rnp->expmask)) || | |
7308e024 | 644 | IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) || |
39bbfc62 PM |
645 | (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled && |
646 | t->rcu_blocked_node); | |
23634ebc | 647 | // Need to defer quiescent state until everything is enabled. |
2407a64f | 648 | if (use_softirq && (in_hardirq() || (expboost && !irqs_were_disabled))) { |
e4453d8a | 649 | // Using softirq, safe to awaken, and either the |
39bbfc62 PM |
650 | // wakeup is free or there is either an expedited |
651 | // GP in flight or a potential need to deboost. | |
05f41571 PM |
652 | raise_softirq_irqoff(RCU_SOFTIRQ); |
653 | } else { | |
23634ebc | 654 | // Enabling BH or preempt does reschedule, so... |
39bbfc62 PM |
655 | // Also if no expediting and no possible deboosting, |
656 | // slow is OK. Plus nohz_full CPUs eventually get | |
657 | // tick enabled. | |
05f41571 PM |
658 | set_tsk_need_resched(current); |
659 | set_preempt_need_resched(); | |
d143b3d1 | 660 | if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled && |
39bbfc62 | 661 | expboost && !rdp->defer_qs_iw_pending && cpu_online(rdp->cpu)) { |
0864f057 PM |
662 | // Get scheduler to re-evaluate and call hooks. |
663 | // If !IRQ_WORK, FQS scan will eventually IPI. | |
39bbfc62 | 664 | init_irq_work(&rdp->defer_qs_iw, rcu_preempt_deferred_qs_handler); |
0864f057 PM |
665 | rdp->defer_qs_iw_pending = true; |
666 | irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu); | |
667 | } | |
05f41571 | 668 | } |
3e310098 PM |
669 | local_irq_restore(flags); |
670 | return; | |
671 | } | |
672 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
673 | } | |
674 | ||
b0e165c0 PM |
675 | /* |
676 | * Check that the list of blocked tasks for the newly completed grace | |
677 | * period is in fact empty. It is a serious bug to complete a grace | |
678 | * period that still has RCU readers blocked! This function must be | |
03bd2983 | 679 | * invoked -before- updating this rnp's ->gp_seq. |
12f5f524 PM |
680 | * |
681 | * Also, if there are blocked tasks on the list, they automatically | |
682 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 683 | */ |
81ab59a3 | 684 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 685 | { |
c5ebe66c PM |
686 | struct task_struct *t; |
687 | ||
ea9b0c8a | 688 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
03bd2983 | 689 | raw_lockdep_assert_held_rcu_node(rnp); |
4bc8d555 | 690 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 691 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
692 | if (rcu_preempt_has_tasks(rnp) && |
693 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
6935c398 | 694 | WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next); |
c5ebe66c PM |
695 | t = container_of(rnp->gp_tasks, struct task_struct, |
696 | rcu_node_entry); | |
697 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 698 | rnp->gp_seq, t->pid); |
c5ebe66c | 699 | } |
28ecd580 | 700 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
701 | } |
702 | ||
f41d911f | 703 | /* |
c98cac60 PM |
704 | * Check for a quiescent state from the current CPU, including voluntary |
705 | * context switches for Tasks RCU. When a task blocks, the task is | |
706 | * recorded in the corresponding CPU's rcu_node structure, which is checked | |
707 | * elsewhere, hence this function need only check for quiescent states | |
708 | * related to the current CPU, not to those related to tasks. | |
f41d911f | 709 | */ |
c98cac60 | 710 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f PM |
711 | { |
712 | struct task_struct *t = current; | |
713 | ||
a649d25d | 714 | lockdep_assert_irqs_disabled(); |
45975c7d PM |
715 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
716 | rcu_note_voluntary_context_switch(current); | |
717 | } | |
77339e61 | 718 | if (rcu_preempt_depth() > 0 || |
3e310098 PM |
719 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { |
720 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
721 | if (rcu_preempt_need_deferred_qs(t)) { |
722 | set_tsk_need_resched(t); | |
723 | set_preempt_need_resched(); | |
724 | } | |
3e310098 PM |
725 | } else if (rcu_preempt_need_deferred_qs(t)) { |
726 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
727 | return; | |
5f5fa7ea | 728 | } else if (!WARN_ON_ONCE(rcu_preempt_depth())) { |
45975c7d | 729 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
730 | return; |
731 | } | |
3e310098 PM |
732 | |
733 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
77339e61 | 734 | if (rcu_preempt_depth() > 0 && |
2280ee5a PM |
735 | __this_cpu_read(rcu_data.core_needs_qs) && |
736 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 737 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 738 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 739 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
740 | } |
741 | ||
2439b696 PM |
742 | /* |
743 | * Check for a task exiting while in a preemptible-RCU read-side | |
884157ce PM |
744 | * critical section, clean up if so. No need to issue warnings, as |
745 | * debug_check_no_locks_held() already does this if lockdep is enabled. | |
746 | * Besides, if this function does anything other than just immediately | |
747 | * return, there was a bug of some sort. Spewing warnings from this | |
748 | * function is like as not to simply obscure important prior warnings. | |
2439b696 PM |
749 | */ |
750 | void exit_rcu(void) | |
751 | { | |
752 | struct task_struct *t = current; | |
753 | ||
884157ce | 754 | if (unlikely(!list_empty(¤t->rcu_node_entry))) { |
77339e61 | 755 | rcu_preempt_depth_set(1); |
884157ce | 756 | barrier(); |
add0d37b | 757 | WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true); |
77339e61 LJ |
758 | } else if (unlikely(rcu_preempt_depth())) { |
759 | rcu_preempt_depth_set(1); | |
884157ce | 760 | } else { |
2439b696 | 761 | return; |
884157ce | 762 | } |
2439b696 | 763 | __rcu_read_unlock(); |
3e310098 | 764 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
765 | } |
766 | ||
4bc8d555 PM |
767 | /* |
768 | * Dump the blocked-tasks state, but limit the list dump to the | |
769 | * specified number of elements. | |
770 | */ | |
57738942 | 771 | static void |
81ab59a3 | 772 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 773 | { |
57738942 | 774 | int cpu; |
4bc8d555 PM |
775 | int i; |
776 | struct list_head *lhp; | |
57738942 | 777 | struct rcu_data *rdp; |
ff3cee39 | 778 | struct rcu_node *rnp1; |
4bc8d555 | 779 | |
ce11fae8 | 780 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 781 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 782 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
8ff37290 | 783 | (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs); |
ff3cee39 PM |
784 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) |
785 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
786 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf | 787 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
065a6db1 | 788 | __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks), |
314eeb43 | 789 | READ_ONCE(rnp->exp_tasks)); |
77cfc7bf | 790 | pr_info("%s: ->blkd_tasks", __func__); |
4bc8d555 PM |
791 | i = 0; |
792 | list_for_each(lhp, &rnp->blkd_tasks) { | |
793 | pr_cont(" %p", lhp); | |
cd6d17b4 | 794 | if (++i >= ncheck) |
4bc8d555 PM |
795 | break; |
796 | } | |
797 | pr_cont("\n"); | |
57738942 | 798 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 799 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 | 800 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", |
5ae0f1b5 | 801 | cpu, ".o"[rcu_rdp_cpu_online(rdp)], |
57738942 PM |
802 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, |
803 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
804 | } | |
4bc8d555 PM |
805 | } |
806 | ||
28f6569a | 807 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 808 | |
aa40c138 PM |
809 | /* |
810 | * If strict grace periods are enabled, and if the calling | |
811 | * __rcu_read_unlock() marks the beginning of a quiescent state, immediately | |
812 | * report that quiescent state and, if requested, spin for a bit. | |
813 | */ | |
814 | void rcu_read_unlock_strict(void) | |
815 | { | |
816 | struct rcu_data *rdp; | |
817 | ||
925da92b | 818 | if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread) |
aa40c138 PM |
819 | return; |
820 | rdp = this_cpu_ptr(&rcu_data); | |
cfeac397 | 821 | rcu_report_qs_rdp(rdp); |
aa40c138 PM |
822 | udelay(rcu_unlock_delay); |
823 | } | |
824 | EXPORT_SYMBOL_GPL(rcu_read_unlock_strict); | |
825 | ||
f41d911f PM |
826 | /* |
827 | * Tell them what RCU they are running. | |
828 | */ | |
0e0fc1c2 | 829 | static void __init rcu_bootup_announce(void) |
f41d911f | 830 | { |
efc151c3 | 831 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 832 | rcu_bootup_announce_oddness(); |
f41d911f PM |
833 | } |
834 | ||
45975c7d | 835 | /* |
90326f05 | 836 | * Note a quiescent state for PREEMPTION=n. Because we do not need to know |
45975c7d PM |
837 | * how many quiescent states passed, just if there was at least one since |
838 | * the start of the grace period, this just sets a flag. The caller must | |
839 | * have disabled preemption. | |
840 | */ | |
841 | static void rcu_qs(void) | |
d28139c4 | 842 | { |
45975c7d PM |
843 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
844 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
845 | return; | |
846 | trace_rcu_grace_period(TPS("rcu_sched"), | |
847 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
848 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
6e16b0f7 FW |
849 | if (__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) |
850 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); | |
d28139c4 PM |
851 | } |
852 | ||
395a2f09 PM |
853 | /* |
854 | * Register an urgently needed quiescent state. If there is an | |
855 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
856 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
857 | * some cases serve for expedited as well as normal grace periods. | |
858 | * Either way, register a lightweight quiescent state. | |
395a2f09 PM |
859 | */ |
860 | void rcu_all_qs(void) | |
861 | { | |
862 | unsigned long flags; | |
863 | ||
2dba13f0 | 864 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
865 | return; |
866 | preempt_disable(); | |
867 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 868 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
869 | preempt_enable(); |
870 | return; | |
871 | } | |
2dba13f0 | 872 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
2dba13f0 | 873 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
874 | local_irq_save(flags); |
875 | rcu_momentary_dyntick_idle(); | |
876 | local_irq_restore(flags); | |
877 | } | |
7e28c5af | 878 | rcu_qs(); |
395a2f09 PM |
879 | preempt_enable(); |
880 | } | |
881 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
882 | ||
cba6d0d6 | 883 | /* |
90326f05 | 884 | * Note a PREEMPTION=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 885 | */ |
45975c7d | 886 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 887 | { |
45975c7d PM |
888 | trace_rcu_utilization(TPS("Start context switch")); |
889 | rcu_qs(); | |
890 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 891 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 892 | goto out; |
2dba13f0 PM |
893 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
894 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 895 | rcu_momentary_dyntick_idle(); |
43766c3e | 896 | rcu_tasks_qs(current, preempt); |
45975c7d PM |
897 | out: |
898 | trace_rcu_utilization(TPS("End context switch")); | |
cba6d0d6 | 899 | } |
45975c7d | 900 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 901 | |
fc2219d4 | 902 | /* |
6cc68793 | 903 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
904 | * RCU readers. |
905 | */ | |
27f4d280 | 906 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
907 | { |
908 | return 0; | |
909 | } | |
910 | ||
8af3a5e7 PM |
911 | /* |
912 | * Because there is no preemptible RCU, there can be no readers blocked. | |
913 | */ | |
914 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 915 | { |
8af3a5e7 | 916 | return false; |
b668c9cf PM |
917 | } |
918 | ||
3e310098 PM |
919 | /* |
920 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
921 | * states. | |
922 | */ | |
923 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
924 | { | |
925 | return false; | |
926 | } | |
790da248 PM |
927 | |
928 | // Except that we do need to respond to a request by an expedited grace | |
929 | // period for a quiescent state from this CPU. Note that requests from | |
930 | // tasks are handled when removing the task from the blocked-tasks list | |
931 | // below. | |
932 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
933 | { | |
934 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
935 | ||
936 | if (rdp->cpu_no_qs.b.exp) | |
937 | rcu_report_exp_rdp(rdp); | |
938 | } | |
3e310098 | 939 | |
b0e165c0 | 940 | /* |
6cc68793 | 941 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
942 | * so there is no need to check for blocked tasks. So check only for |
943 | * bogus qsmask values. | |
b0e165c0 | 944 | */ |
81ab59a3 | 945 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 946 | { |
49e29126 | 947 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
948 | } |
949 | ||
f41d911f | 950 | /* |
c98cac60 PM |
951 | * Check to see if this CPU is in a non-context-switch quiescent state, |
952 | * namely user mode and idle loop. | |
f41d911f | 953 | */ |
c98cac60 | 954 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f | 955 | { |
45975c7d | 956 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 957 | |
45975c7d PM |
958 | /* |
959 | * Get here if this CPU took its interrupt from user | |
960 | * mode or from the idle loop, and if this is not a | |
961 | * nested interrupt. In this case, the CPU is in | |
962 | * a quiescent state, so note it. | |
963 | * | |
964 | * No memory barrier is required here because rcu_qs() | |
965 | * references only CPU-local variables that other CPUs | |
966 | * neither access nor modify, at least not while the | |
967 | * corresponding CPU is online. | |
968 | */ | |
969 | ||
970 | rcu_qs(); | |
971 | } | |
e74f4c45 | 972 | } |
e74f4c45 | 973 | |
2439b696 PM |
974 | /* |
975 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
976 | * while in preemptible RCU read-side critical sections. | |
977 | */ | |
978 | void exit_rcu(void) | |
979 | { | |
980 | } | |
981 | ||
4bc8d555 PM |
982 | /* |
983 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
984 | */ | |
57738942 | 985 | static void |
81ab59a3 | 986 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
987 | { |
988 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
989 | } | |
990 | ||
28f6569a | 991 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 992 | |
48d07c04 SAS |
993 | /* |
994 | * If boosting, set rcuc kthreads to realtime priority. | |
995 | */ | |
996 | static void rcu_cpu_kthread_setup(unsigned int cpu) | |
997 | { | |
c9515875 | 998 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27f4d280 | 999 | #ifdef CONFIG_RCU_BOOST |
48d07c04 | 1000 | struct sched_param sp; |
27f4d280 | 1001 | |
48d07c04 SAS |
1002 | sp.sched_priority = kthread_prio; |
1003 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
1004 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
c9515875 Z |
1005 | |
1006 | WRITE_ONCE(rdp->rcuc_activity, jiffies); | |
5d01bbd1 TG |
1007 | } |
1008 | ||
48d07c04 SAS |
1009 | #ifdef CONFIG_RCU_BOOST |
1010 | ||
27f4d280 PM |
1011 | /* |
1012 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1013 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1014 | * ->blkd_tasks list. | |
1015 | * | |
1016 | * Note that irqs must be enabled: boosting the task can block. | |
1017 | * Returns 1 if there are more tasks needing to be boosted. | |
1018 | */ | |
1019 | static int rcu_boost(struct rcu_node *rnp) | |
1020 | { | |
1021 | unsigned long flags; | |
27f4d280 PM |
1022 | struct task_struct *t; |
1023 | struct list_head *tb; | |
1024 | ||
7d0ae808 PM |
1025 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1026 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1027 | return 0; /* Nothing left to boost. */ |
1028 | ||
2a67e741 | 1029 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1030 | |
1031 | /* | |
1032 | * Recheck under the lock: all tasks in need of boosting | |
1033 | * might exit their RCU read-side critical sections on their own. | |
1034 | */ | |
1035 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1036 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1037 | return 0; |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Preferentially boost tasks blocking expedited grace periods. | |
1042 | * This cannot starve the normal grace periods because a second | |
1043 | * expedited grace period must boost all blocked tasks, including | |
1044 | * those blocking the pre-existing normal grace period. | |
1045 | */ | |
bec06785 | 1046 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1047 | tb = rnp->exp_tasks; |
bec06785 | 1048 | else |
27f4d280 PM |
1049 | tb = rnp->boost_tasks; |
1050 | ||
1051 | /* | |
1052 | * We boost task t by manufacturing an rt_mutex that appears to | |
1053 | * be held by task t. We leave a pointer to that rt_mutex where | |
1054 | * task t can find it, and task t will release the mutex when it | |
1055 | * exits its outermost RCU read-side critical section. Then | |
1056 | * simply acquiring this artificial rt_mutex will boost task | |
1057 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1058 | * | |
1059 | * Note that task t must acquire rnp->lock to remove itself from | |
1060 | * the ->blkd_tasks list, which it will do from exit() if from | |
1061 | * nowhere else. We therefore are guaranteed that task t will | |
1062 | * stay around at least until we drop rnp->lock. Note that | |
1063 | * rnp->lock also resolves races between our priority boosting | |
1064 | * and task t's exiting its outermost RCU read-side critical | |
1065 | * section. | |
1066 | */ | |
1067 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
830e6acc | 1068 | rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t); |
67c583a7 | 1069 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1070 | /* Lock only for side effect: boosts task t's priority. */ |
1071 | rt_mutex_lock(&rnp->boost_mtx); | |
1072 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
396eba65 | 1073 | rnp->n_boosts++; |
27f4d280 | 1074 | |
7d0ae808 PM |
1075 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1076 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1077 | } |
1078 | ||
27f4d280 | 1079 | /* |
bc17ea10 | 1080 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1081 | */ |
1082 | static int rcu_boost_kthread(void *arg) | |
1083 | { | |
1084 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1085 | int spincnt = 0; | |
1086 | int more2boost; | |
1087 | ||
f7f7bac9 | 1088 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1089 | for (;;) { |
3ca3b0e2 | 1090 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING); |
f7f7bac9 | 1091 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
065a6db1 PM |
1092 | rcu_wait(READ_ONCE(rnp->boost_tasks) || |
1093 | READ_ONCE(rnp->exp_tasks)); | |
f7f7bac9 | 1094 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
3ca3b0e2 | 1095 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING); |
27f4d280 PM |
1096 | more2boost = rcu_boost(rnp); |
1097 | if (more2boost) | |
1098 | spincnt++; | |
1099 | else | |
1100 | spincnt = 0; | |
1101 | if (spincnt > 10) { | |
3ca3b0e2 | 1102 | WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING); |
f7f7bac9 | 1103 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
a9352f72 | 1104 | schedule_timeout_idle(2); |
f7f7bac9 | 1105 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1106 | spincnt = 0; |
1107 | } | |
1108 | } | |
1217ed1b | 1109 | /* NOTREACHED */ |
f7f7bac9 | 1110 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1111 | return 0; |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Check to see if it is time to start boosting RCU readers that are | |
1116 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1117 | * kthread to start boosting them. If there is an expedited grace | |
1118 | * period in progress, it is always time to boost. | |
1119 | * | |
b065a853 PM |
1120 | * The caller must hold rnp->lock, which this function releases. |
1121 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1122 | * about it going away. | |
27f4d280 | 1123 | */ |
1217ed1b | 1124 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1125 | __releases(rnp->lock) |
27f4d280 | 1126 | { |
a32e01ee | 1127 | raw_lockdep_assert_held_rcu_node(rnp); |
88ca472f Z |
1128 | if (!rnp->boost_kthread_task || |
1129 | (!rcu_preempt_blocked_readers_cgp(rnp) && !rnp->exp_tasks)) { | |
67c583a7 | 1130 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1131 | return; |
0ea1f2eb | 1132 | } |
27f4d280 PM |
1133 | if (rnp->exp_tasks != NULL || |
1134 | (rnp->gp_tasks != NULL && | |
1135 | rnp->boost_tasks == NULL && | |
1136 | rnp->qsmask == 0 && | |
7b241311 | 1137 | (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld))) { |
27f4d280 | 1138 | if (rnp->exp_tasks == NULL) |
5822b812 | 1139 | WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks); |
67c583a7 | 1140 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
a2badefa | 1141 | rcu_wake_cond(rnp->boost_kthread_task, |
3ca3b0e2 | 1142 | READ_ONCE(rnp->boost_kthread_status)); |
1217ed1b | 1143 | } else { |
67c583a7 | 1144 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1145 | } |
27f4d280 PM |
1146 | } |
1147 | ||
dff1672d PM |
1148 | /* |
1149 | * Is the current CPU running the RCU-callbacks kthread? | |
1150 | * Caller must have preemption disabled. | |
1151 | */ | |
1152 | static bool rcu_is_callbacks_kthread(void) | |
1153 | { | |
37f62d7c | 1154 | return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current; |
dff1672d PM |
1155 | } |
1156 | ||
27f4d280 PM |
1157 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1158 | ||
1159 | /* | |
1160 | * Do priority-boost accounting for the start of a new grace period. | |
1161 | */ | |
1162 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1163 | { | |
1164 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1165 | } | |
1166 | ||
27f4d280 PM |
1167 | /* |
1168 | * Create an RCU-boost kthread for the specified node if one does not | |
1169 | * already exist. We only create this kthread for preemptible RCU. | |
27f4d280 | 1170 | */ |
3545832f | 1171 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 PM |
1172 | { |
1173 | unsigned long flags; | |
3ef5a1c3 | 1174 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1175 | struct sched_param sp; |
1176 | struct task_struct *t; | |
1177 | ||
218b957a | 1178 | mutex_lock(&rnp->boost_kthread_mutex); |
3ef5a1c3 | 1179 | if (rnp->boost_kthread_task || !rcu_scheduler_fully_active) |
218b957a | 1180 | goto out; |
3545832f | 1181 | |
27f4d280 | 1182 | t = kthread_create(rcu_boost_kthread, (void *)rnp, |
5b61b0ba | 1183 | "rcub/%d", rnp_index); |
3545832f | 1184 | if (WARN_ON_ONCE(IS_ERR(t))) |
218b957a | 1185 | goto out; |
3545832f | 1186 | |
2a67e741 | 1187 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1188 | rnp->boost_kthread_task = t; |
67c583a7 | 1189 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1190 | sp.sched_priority = kthread_prio; |
27f4d280 | 1191 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1192 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
218b957a DW |
1193 | |
1194 | out: | |
1195 | mutex_unlock(&rnp->boost_kthread_mutex); | |
27f4d280 PM |
1196 | } |
1197 | ||
f8b7fc6b PM |
1198 | /* |
1199 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1200 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1201 | * held, so the value of rnp->qsmaskinit will be stable. | |
1202 | * | |
1203 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1204 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1205 | * this function allows the kthread to execute on any CPU. | |
1206 | */ | |
5d01bbd1 | 1207 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1208 | { |
5d01bbd1 | 1209 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1210 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1211 | cpumask_var_t cm; |
1212 | int cpu; | |
f8b7fc6b | 1213 | |
5d01bbd1 | 1214 | if (!t) |
f8b7fc6b | 1215 | return; |
5d01bbd1 | 1216 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1217 | return; |
218b957a | 1218 | mutex_lock(&rnp->boost_kthread_mutex); |
bc75e999 MR |
1219 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1220 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1221 | cpu != outgoingcpu) | |
f8b7fc6b | 1222 | cpumask_set_cpu(cpu, cm); |
04d4e665 | 1223 | cpumask_and(cm, cm, housekeeping_cpumask(HK_TYPE_RCU)); |
6a2c1d45 | 1224 | if (cpumask_empty(cm)) |
04d4e665 | 1225 | cpumask_copy(cm, housekeeping_cpumask(HK_TYPE_RCU)); |
5d01bbd1 | 1226 | set_cpus_allowed_ptr(t, cm); |
218b957a | 1227 | mutex_unlock(&rnp->boost_kthread_mutex); |
f8b7fc6b PM |
1228 | free_cpumask_var(cm); |
1229 | } | |
1230 | ||
f8b7fc6b | 1231 | /* |
9386c0b7 | 1232 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1233 | */ |
9386c0b7 | 1234 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1235 | { |
f8b7fc6b PM |
1236 | struct rcu_node *rnp; |
1237 | ||
aedf4ba9 | 1238 | rcu_for_each_leaf_node(rnp) |
3ef5a1c3 PM |
1239 | if (rcu_rnp_online_cpus(rnp)) |
1240 | rcu_spawn_one_boost_kthread(rnp); | |
f8b7fc6b PM |
1241 | } |
1242 | ||
27f4d280 PM |
1243 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1244 | ||
1217ed1b | 1245 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1246 | __releases(rnp->lock) |
27f4d280 | 1247 | { |
67c583a7 | 1248 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1249 | } |
1250 | ||
dff1672d PM |
1251 | static bool rcu_is_callbacks_kthread(void) |
1252 | { | |
1253 | return false; | |
1254 | } | |
1255 | ||
27f4d280 PM |
1256 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1257 | { | |
1258 | } | |
1259 | ||
3ef5a1c3 | 1260 | static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
f8b7fc6b PM |
1261 | { |
1262 | } | |
1263 | ||
3ef5a1c3 | 1264 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
b0d30417 | 1265 | { |
b0d30417 | 1266 | } |
b0d30417 | 1267 | |
3ef5a1c3 | 1268 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b PM |
1269 | { |
1270 | } | |
1271 | ||
27f4d280 PM |
1272 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1273 | ||
a096932f PM |
1274 | /* |
1275 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
1276 | * grace-period kthread will do force_quiescent_state() processing? | |
1277 | * The idea is to avoid waking up RCU core processing on such a | |
1278 | * CPU unless the grace period has extended for too long. | |
1279 | * | |
1280 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
17ea3718 | 1281 | * RCU_NOCB_CPU CPUs. |
a096932f | 1282 | */ |
4580b054 | 1283 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
1284 | { |
1285 | #ifdef CONFIG_NO_HZ_FULL | |
1286 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 1287 | (!rcu_gp_in_progress() || |
e2f3ccfa | 1288 | time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 1289 | return true; |
a096932f | 1290 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 1291 | return false; |
a096932f | 1292 | } |
5057f55e PM |
1293 | |
1294 | /* | |
265f5f28 | 1295 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
1296 | */ |
1297 | static void rcu_bind_gp_kthread(void) | |
1298 | { | |
c0f489d2 | 1299 | if (!tick_nohz_full_enabled()) |
5057f55e | 1300 | return; |
04d4e665 | 1301 | housekeeping_affine(current, HK_TYPE_RCU); |
5057f55e | 1302 | } |
176f8f7a PM |
1303 | |
1304 | /* Record the current task on dyntick-idle entry. */ | |
7663ad9a | 1305 | static __always_inline void rcu_dynticks_task_enter(void) |
176f8f7a PM |
1306 | { |
1307 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 1308 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
1309 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
1310 | } | |
1311 | ||
1312 | /* Record no current task on dyntick-idle exit. */ | |
7663ad9a | 1313 | static __always_inline void rcu_dynticks_task_exit(void) |
176f8f7a PM |
1314 | { |
1315 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 1316 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
1317 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
1318 | } | |
7d0c9c50 PM |
1319 | |
1320 | /* Turn on heavyweight RCU tasks trace readers on idle/user entry. */ | |
7663ad9a | 1321 | static __always_inline void rcu_dynticks_task_trace_enter(void) |
7d0c9c50 | 1322 | { |
fed31a4d | 1323 | #ifdef CONFIG_TASKS_TRACE_RCU |
7d0c9c50 PM |
1324 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
1325 | current->trc_reader_special.b.need_mb = true; | |
fed31a4d | 1326 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
7d0c9c50 PM |
1327 | } |
1328 | ||
1329 | /* Turn off heavyweight RCU tasks trace readers on idle/user exit. */ | |
7663ad9a | 1330 | static __always_inline void rcu_dynticks_task_trace_exit(void) |
7d0c9c50 | 1331 | { |
fed31a4d | 1332 | #ifdef CONFIG_TASKS_TRACE_RCU |
7d0c9c50 PM |
1333 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
1334 | current->trc_reader_special.b.need_mb = false; | |
fed31a4d | 1335 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
7d0c9c50 | 1336 | } |