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 | ||
d9a3da06 | 14 | #include <linux/delay.h> |
3fbfbf7a | 15 | #include <linux/gfp.h> |
b626c1b6 | 16 | #include <linux/oom.h> |
b17b0153 | 17 | #include <linux/sched/debug.h> |
62ab7072 | 18 | #include <linux/smpboot.h> |
78634061 | 19 | #include <linux/sched/isolation.h> |
ae7e81c0 | 20 | #include <uapi/linux/sched/types.h> |
4102adab | 21 | #include "../time/tick-internal.h" |
f41d911f | 22 | |
5b61b0ba | 23 | #ifdef CONFIG_RCU_BOOST |
abaa93d9 | 24 | #include "../locking/rtmutex_common.h" |
727b705b PM |
25 | #else /* #ifdef CONFIG_RCU_BOOST */ |
26 | ||
27 | /* | |
28 | * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST, | |
29 | * all uses are in dead code. Provide a definition to keep the compiler | |
30 | * happy, but add WARN_ON_ONCE() to complain if used in the wrong place. | |
31 | * This probably needs to be excluded from -rt builds. | |
32 | */ | |
33 | #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; }) | |
b8869781 | 34 | #define rt_mutex_futex_unlock(x) WARN_ON_ONCE(1) |
727b705b PM |
35 | |
36 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ | |
5b61b0ba | 37 | |
3fbfbf7a PM |
38 | #ifdef CONFIG_RCU_NOCB_CPU |
39 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
1b0048a4 | 40 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
41 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
42 | ||
26845c28 PM |
43 | /* |
44 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 45 | * messages about anything out of the ordinary. |
26845c28 PM |
46 | */ |
47 | static void __init rcu_bootup_announce_oddness(void) | |
48 | { | |
ab6f5bd6 | 49 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 50 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
51 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
52 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
a7538352 JP |
53 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n", |
54 | RCU_FANOUT); | |
7fa27001 | 55 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
56 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
57 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
58 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 59 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 60 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
61 | if (RCU_NUM_LVLS >= 4) |
62 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 63 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 64 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
65 | RCU_FANOUT_LEAF); |
66 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
a7538352 JP |
67 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", |
68 | rcu_fanout_leaf); | |
cca6f393 | 69 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 70 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b | 71 | #ifdef CONFIG_RCU_BOOST |
a7538352 JP |
72 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", |
73 | kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
17c7798b PM |
74 | #endif |
75 | if (blimit != DEFAULT_RCU_BLIMIT) | |
76 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
77 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
78 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
79 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
80 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
81 | if (jiffies_till_first_fqs != ULONG_MAX) | |
82 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
83 | if (jiffies_till_next_fqs != ULONG_MAX) | |
84 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
c06aed0e PM |
85 | if (jiffies_till_sched_qs != ULONG_MAX) |
86 | pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs); | |
17c7798b PM |
87 | if (rcu_kick_kthreads) |
88 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
89 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
90 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 91 | if (gp_preinit_delay) |
17c7798b | 92 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 93 | if (gp_init_delay) |
17c7798b | 94 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 95 | if (gp_cleanup_delay) |
17c7798b PM |
96 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
97 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) | |
98 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 99 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
100 | } |
101 | ||
28f6569a | 102 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 103 | |
63d4c8c9 | 104 | static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake); |
3949fa9b | 105 | static void rcu_read_unlock_special(struct task_struct *t); |
d9a3da06 | 106 | |
f41d911f PM |
107 | /* |
108 | * Tell them what RCU they are running. | |
109 | */ | |
0e0fc1c2 | 110 | static void __init rcu_bootup_announce(void) |
f41d911f | 111 | { |
efc151c3 | 112 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 113 | rcu_bootup_announce_oddness(); |
f41d911f PM |
114 | } |
115 | ||
8203d6d0 PM |
116 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
117 | #define RCU_GP_TASKS 0x8 | |
118 | #define RCU_EXP_TASKS 0x4 | |
119 | #define RCU_GP_BLKD 0x2 | |
120 | #define RCU_EXP_BLKD 0x1 | |
121 | ||
122 | /* | |
123 | * Queues a task preempted within an RCU-preempt read-side critical | |
124 | * section into the appropriate location within the ->blkd_tasks list, | |
125 | * depending on the states of any ongoing normal and expedited grace | |
126 | * periods. The ->gp_tasks pointer indicates which element the normal | |
127 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
128 | * indicates which element the expedited grace period is waiting on (again, | |
129 | * NULL if none). If a grace period is waiting on a given element in the | |
130 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
131 | * adding a task to the tail of the list blocks any grace period that is | |
132 | * already waiting on one of the elements. In contrast, adding a task | |
133 | * to the head of the list won't block any grace period that is already | |
134 | * waiting on one of the elements. | |
135 | * | |
136 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
137 | * period wait for a task that is not strictly speaking blocking it. | |
138 | * Given the choice, we needlessly block a normal grace period rather than | |
139 | * blocking an expedited grace period. | |
140 | * | |
141 | * Note that an endless sequence of expedited grace periods still cannot | |
142 | * indefinitely postpone a normal grace period. Eventually, all of the | |
143 | * fixed number of preempted tasks blocking the normal grace period that are | |
144 | * not also blocking the expedited grace period will resume and complete | |
145 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
146 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
147 | * the corresponding expedited grace period will also be the end of the | |
148 | * normal grace period. | |
149 | */ | |
46a5d164 PM |
150 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
151 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
152 | { |
153 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
154 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
155 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
156 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
157 | struct task_struct *t = current; | |
158 | ||
a32e01ee | 159 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 160 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 161 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
162 | /* RCU better not be waiting on newly onlined CPUs! */ |
163 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
164 | rdp->grpmask); | |
ea9b0c8a | 165 | |
8203d6d0 PM |
166 | /* |
167 | * Decide where to queue the newly blocked task. In theory, | |
168 | * this could be an if-statement. In practice, when I tried | |
169 | * that, it was quite messy. | |
170 | */ | |
171 | switch (blkd_state) { | |
172 | case 0: | |
173 | case RCU_EXP_TASKS: | |
174 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
175 | case RCU_GP_TASKS: | |
176 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
177 | ||
178 | /* | |
179 | * Blocking neither GP, or first task blocking the normal | |
180 | * GP but not blocking the already-waiting expedited GP. | |
181 | * Queue at the head of the list to avoid unnecessarily | |
182 | * blocking the already-waiting GPs. | |
183 | */ | |
184 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
185 | break; | |
186 | ||
187 | case RCU_EXP_BLKD: | |
188 | case RCU_GP_BLKD: | |
189 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
190 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
191 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
192 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
193 | ||
194 | /* | |
195 | * First task arriving that blocks either GP, or first task | |
196 | * arriving that blocks the expedited GP (with the normal | |
197 | * GP already waiting), or a task arriving that blocks | |
198 | * both GPs with both GPs already waiting. Queue at the | |
199 | * tail of the list to avoid any GP waiting on any of the | |
200 | * already queued tasks that are not blocking it. | |
201 | */ | |
202 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
203 | break; | |
204 | ||
205 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
206 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
207 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
208 | ||
209 | /* | |
210 | * Second or subsequent task blocking the expedited GP. | |
211 | * The task either does not block the normal GP, or is the | |
212 | * first task blocking the normal GP. Queue just after | |
213 | * the first task blocking the expedited GP. | |
214 | */ | |
215 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
216 | break; | |
217 | ||
218 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
219 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
220 | ||
221 | /* | |
222 | * Second or subsequent task blocking the normal GP. | |
223 | * The task does not block the expedited GP. Queue just | |
224 | * after the first task blocking the normal GP. | |
225 | */ | |
226 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
227 | break; | |
228 | ||
229 | default: | |
230 | ||
231 | /* Yet another exercise in excessive paranoia. */ | |
232 | WARN_ON_ONCE(1); | |
233 | break; | |
234 | } | |
235 | ||
236 | /* | |
237 | * We have now queued the task. If it was the first one to | |
238 | * block either grace period, update the ->gp_tasks and/or | |
239 | * ->exp_tasks pointers, respectively, to reference the newly | |
240 | * blocked tasks. | |
241 | */ | |
4bc8d555 | 242 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
8203d6d0 | 243 | rnp->gp_tasks = &t->rcu_node_entry; |
d43a5d32 | 244 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 245 | } |
8203d6d0 PM |
246 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
247 | rnp->exp_tasks = &t->rcu_node_entry; | |
2dee9404 PM |
248 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
249 | !(rnp->qsmask & rdp->grpmask)); | |
250 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
251 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 252 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
253 | |
254 | /* | |
255 | * Report the quiescent state for the expedited GP. This expedited | |
256 | * GP should not be able to end until we report, so there should be | |
257 | * no need to check for a subsequent expedited GP. (Though we are | |
258 | * still in a quiescent state in any case.) | |
259 | */ | |
fcc878e4 | 260 | if (blkd_state & RCU_EXP_BLKD && rdp->deferred_qs) |
63d4c8c9 | 261 | rcu_report_exp_rdp(rdp); |
fcc878e4 PM |
262 | else |
263 | WARN_ON_ONCE(rdp->deferred_qs); | |
8203d6d0 PM |
264 | } |
265 | ||
f41d911f | 266 | /* |
c7037ff5 PM |
267 | * Record a preemptible-RCU quiescent state for the specified CPU. |
268 | * Note that this does not necessarily mean that the task currently running | |
269 | * on the CPU is in a quiescent state: Instead, it means that the current | |
270 | * grace period need not wait on any RCU read-side critical section that | |
271 | * starts later on this CPU. It also means that if the current task is | |
272 | * in an RCU read-side critical section, it has already added itself to | |
273 | * some leaf rcu_node structure's ->blkd_tasks list. In addition to the | |
274 | * current task, there might be any number of other tasks blocked while | |
275 | * in an RCU read-side critical section. | |
25502a6c | 276 | * |
c7037ff5 | 277 | * Callers to this function must disable preemption. |
f41d911f | 278 | */ |
45975c7d | 279 | static void rcu_qs(void) |
f41d911f | 280 | { |
45975c7d | 281 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n"); |
2280ee5a | 282 | if (__this_cpu_read(rcu_data.cpu_no_qs.s)) { |
284a8c93 | 283 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2280ee5a | 284 | __this_cpu_read(rcu_data.gp_seq), |
284a8c93 | 285 | TPS("cpuqs")); |
2280ee5a | 286 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); |
c98cac60 | 287 | barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */ |
284a8c93 PM |
288 | current->rcu_read_unlock_special.b.need_qs = false; |
289 | } | |
f41d911f PM |
290 | } |
291 | ||
292 | /* | |
c3422bea PM |
293 | * We have entered the scheduler, and the current task might soon be |
294 | * context-switched away from. If this task is in an RCU read-side | |
295 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
296 | * record that fact, so we enqueue the task on the blkd_tasks list. |
297 | * The task will dequeue itself when it exits the outermost enclosing | |
298 | * RCU read-side critical section. Therefore, the current grace period | |
299 | * cannot be permitted to complete until the blkd_tasks list entries | |
300 | * predating the current grace period drain, in other words, until | |
301 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 302 | * |
46a5d164 | 303 | * Caller must disable interrupts. |
f41d911f | 304 | */ |
45975c7d | 305 | void rcu_note_context_switch(bool preempt) |
f41d911f PM |
306 | { |
307 | struct task_struct *t = current; | |
da1df50d | 308 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
f41d911f PM |
309 | struct rcu_node *rnp; |
310 | ||
45975c7d PM |
311 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
312 | trace_rcu_utilization(TPS("Start context switch")); | |
b04db8e1 | 313 | lockdep_assert_irqs_disabled(); |
5b72f964 | 314 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 315 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 316 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
317 | |
318 | /* Possibly blocking in an RCU read-side critical section. */ | |
f41d911f | 319 | rnp = rdp->mynode; |
46a5d164 | 320 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 321 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 322 | t->rcu_blocked_node = rnp; |
f41d911f PM |
323 | |
324 | /* | |
8203d6d0 PM |
325 | * Verify the CPU's sanity, trace the preemption, and |
326 | * then queue the task as required based on the states | |
327 | * of any ongoing and expedited grace periods. | |
f41d911f | 328 | */ |
0aa04b05 | 329 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 330 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
88d1bead | 331 | trace_rcu_preempt_task(rcu_state.name, |
d4c08f2a PM |
332 | t->pid, |
333 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
334 | ? rnp->gp_seq |
335 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 336 | rcu_preempt_ctxt_queue(rnp, rdp); |
10f39bb1 | 337 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 338 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
339 | |
340 | /* | |
341 | * Complete exit from RCU read-side critical section on | |
342 | * behalf of preempted instance of __rcu_read_unlock(). | |
343 | */ | |
344 | rcu_read_unlock_special(t); | |
3e310098 PM |
345 | rcu_preempt_deferred_qs(t); |
346 | } else { | |
347 | rcu_preempt_deferred_qs(t); | |
f41d911f PM |
348 | } |
349 | ||
350 | /* | |
351 | * Either we were not in an RCU read-side critical section to | |
352 | * begin with, or we have now recorded that critical section | |
353 | * globally. Either way, we can now note a quiescent state | |
354 | * for this CPU. Again, if we were in an RCU read-side critical | |
355 | * section, and if that critical section was blocking the current | |
356 | * grace period, then the fact that the task has been enqueued | |
357 | * means that we continue to block the current grace period. | |
358 | */ | |
45975c7d | 359 | rcu_qs(); |
ba1c64c2 | 360 | if (rdp->deferred_qs) |
63d4c8c9 | 361 | rcu_report_exp_rdp(rdp); |
45975c7d PM |
362 | trace_rcu_utilization(TPS("End context switch")); |
363 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ | |
f41d911f | 364 | } |
45975c7d | 365 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
f41d911f | 366 | |
fc2219d4 PM |
367 | /* |
368 | * Check for preempted RCU readers blocking the current grace period | |
369 | * for the specified rcu_node structure. If the caller needs a reliable | |
370 | * answer, it must hold the rcu_node's ->lock. | |
371 | */ | |
27f4d280 | 372 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 373 | { |
12f5f524 | 374 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
375 | } |
376 | ||
5f1a6ef3 PM |
377 | /* Bias and limit values for ->rcu_read_lock_nesting. */ |
378 | #define RCU_NEST_BIAS INT_MAX | |
379 | #define RCU_NEST_NMAX (-INT_MAX / 2) | |
380 | #define RCU_NEST_PMAX (INT_MAX / 2) | |
381 | ||
0e5da22e PM |
382 | /* |
383 | * Preemptible RCU implementation for rcu_read_lock(). | |
384 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
385 | * if we block. | |
386 | */ | |
387 | void __rcu_read_lock(void) | |
388 | { | |
389 | current->rcu_read_lock_nesting++; | |
5f1a6ef3 PM |
390 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) |
391 | WARN_ON_ONCE(current->rcu_read_lock_nesting > RCU_NEST_PMAX); | |
0e5da22e PM |
392 | barrier(); /* critical section after entry code. */ |
393 | } | |
394 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
395 | ||
396 | /* | |
397 | * Preemptible RCU implementation for rcu_read_unlock(). | |
398 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
399 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
400 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
401 | * in an RCU read-side critical section and other special cases. | |
402 | */ | |
403 | void __rcu_read_unlock(void) | |
404 | { | |
405 | struct task_struct *t = current; | |
406 | ||
407 | if (t->rcu_read_lock_nesting != 1) { | |
408 | --t->rcu_read_lock_nesting; | |
409 | } else { | |
410 | barrier(); /* critical section before exit code. */ | |
5f1a6ef3 | 411 | t->rcu_read_lock_nesting = -RCU_NEST_BIAS; |
0e5da22e PM |
412 | barrier(); /* assign before ->rcu_read_unlock_special load */ |
413 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) | |
414 | rcu_read_unlock_special(t); | |
415 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
416 | t->rcu_read_lock_nesting = 0; | |
417 | } | |
5f1a6ef3 PM |
418 | if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { |
419 | int rrln = t->rcu_read_lock_nesting; | |
0e5da22e | 420 | |
5f1a6ef3 | 421 | WARN_ON_ONCE(rrln < 0 && rrln > RCU_NEST_NMAX); |
0e5da22e | 422 | } |
0e5da22e PM |
423 | } |
424 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
425 | ||
12f5f524 PM |
426 | /* |
427 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
428 | * returning NULL if at the end of the list. | |
429 | */ | |
430 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
431 | struct rcu_node *rnp) | |
432 | { | |
433 | struct list_head *np; | |
434 | ||
435 | np = t->rcu_node_entry.next; | |
436 | if (np == &rnp->blkd_tasks) | |
437 | np = NULL; | |
438 | return np; | |
439 | } | |
440 | ||
8af3a5e7 PM |
441 | /* |
442 | * Return true if the specified rcu_node structure has tasks that were | |
443 | * preempted within an RCU read-side critical section. | |
444 | */ | |
445 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
446 | { | |
447 | return !list_empty(&rnp->blkd_tasks); | |
448 | } | |
449 | ||
b668c9cf | 450 | /* |
3e310098 PM |
451 | * Report deferred quiescent states. The deferral time can |
452 | * be quite short, for example, in the case of the call from | |
453 | * rcu_read_unlock_special(). | |
b668c9cf | 454 | */ |
3e310098 PM |
455 | static void |
456 | rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags) | |
f41d911f | 457 | { |
b6a932d1 PM |
458 | bool empty_exp; |
459 | bool empty_norm; | |
460 | bool empty_exp_now; | |
12f5f524 | 461 | struct list_head *np; |
abaa93d9 | 462 | bool drop_boost_mutex = false; |
8203d6d0 | 463 | struct rcu_data *rdp; |
f41d911f | 464 | struct rcu_node *rnp; |
1d082fd0 | 465 | union rcu_special special; |
f41d911f | 466 | |
f41d911f | 467 | /* |
8203d6d0 PM |
468 | * If RCU core is waiting for this CPU to exit its critical section, |
469 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 470 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
471 | */ |
472 | special = t->rcu_read_unlock_special; | |
da1df50d | 473 | rdp = this_cpu_ptr(&rcu_data); |
3e310098 PM |
474 | if (!special.s && !rdp->deferred_qs) { |
475 | local_irq_restore(flags); | |
476 | return; | |
477 | } | |
1d082fd0 | 478 | if (special.b.need_qs) { |
45975c7d | 479 | rcu_qs(); |
c0135d07 | 480 | t->rcu_read_unlock_special.b.need_qs = false; |
3e310098 | 481 | if (!t->rcu_read_unlock_special.s && !rdp->deferred_qs) { |
79a62f95 LJ |
482 | local_irq_restore(flags); |
483 | return; | |
484 | } | |
f41d911f PM |
485 | } |
486 | ||
8203d6d0 | 487 | /* |
3e310098 PM |
488 | * Respond to a request by an expedited grace period for a |
489 | * quiescent state from this CPU. Note that requests from | |
490 | * tasks are handled when removing the task from the | |
491 | * blocked-tasks list below. | |
8203d6d0 | 492 | */ |
fcc878e4 | 493 | if (rdp->deferred_qs) { |
63d4c8c9 | 494 | rcu_report_exp_rdp(rdp); |
8203d6d0 PM |
495 | if (!t->rcu_read_unlock_special.s) { |
496 | local_irq_restore(flags); | |
497 | return; | |
498 | } | |
499 | } | |
500 | ||
f41d911f | 501 | /* Clean up if blocked during RCU read-side critical section. */ |
1d082fd0 PM |
502 | if (special.b.blocked) { |
503 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 504 | |
dd5d19ba | 505 | /* |
0a0ba1c9 | 506 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
507 | * now remains queued on the rcu_node corresponding to the |
508 | * CPU it first blocked on, so there is no longer any need | |
509 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 510 | */ |
8ba9153b PM |
511 | rnp = t->rcu_blocked_node; |
512 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
513 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 514 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 515 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 516 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 517 | (!empty_norm || rnp->qsmask)); |
8203d6d0 | 518 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 519 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 520 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 521 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 522 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 523 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 524 | rnp->gp_seq, t->pid); |
12f5f524 PM |
525 | if (&t->rcu_node_entry == rnp->gp_tasks) |
526 | rnp->gp_tasks = np; | |
527 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
528 | rnp->exp_tasks = np; | |
727b705b | 529 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
530 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
531 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
532 | if (&t->rcu_node_entry == rnp->boost_tasks) |
533 | rnp->boost_tasks = np; | |
727b705b | 534 | } |
f41d911f PM |
535 | |
536 | /* | |
537 | * If this was the last task on the current list, and if | |
538 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
539 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
540 | * so we must take a snapshot of the expedited state. | |
f41d911f | 541 | */ |
8203d6d0 | 542 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 543 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 544 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 545 | rnp->gp_seq, |
d4c08f2a PM |
546 | 0, rnp->qsmask, |
547 | rnp->level, | |
548 | rnp->grplo, | |
549 | rnp->grphi, | |
550 | !!rnp->gp_tasks); | |
139ad4da | 551 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 552 | } else { |
67c583a7 | 553 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 554 | } |
d9a3da06 | 555 | |
27f4d280 | 556 | /* Unboost if we were boosted. */ |
727b705b | 557 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 558 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 559 | |
d9a3da06 PM |
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); |
b668c9cf PM |
566 | } else { |
567 | local_irq_restore(flags); | |
f41d911f | 568 | } |
f41d911f PM |
569 | } |
570 | ||
3e310098 PM |
571 | /* |
572 | * Is a deferred quiescent-state pending, and are we also not in | |
573 | * an RCU read-side critical section? It is the caller's responsibility | |
574 | * to ensure it is otherwise safe to report any deferred quiescent | |
575 | * states. The reason for this is that it is safe to report a | |
576 | * quiescent state during context switch even though preemption | |
577 | * is disabled. This function cannot be expected to understand these | |
578 | * nuances, so the caller must handle them. | |
579 | */ | |
580 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
581 | { | |
117f683c | 582 | return (__this_cpu_read(rcu_data.deferred_qs) || |
3e310098 | 583 | READ_ONCE(t->rcu_read_unlock_special.s)) && |
27c744e3 | 584 | t->rcu_read_lock_nesting <= 0; |
3e310098 PM |
585 | } |
586 | ||
587 | /* | |
588 | * Report a deferred quiescent state if needed and safe to do so. | |
589 | * As with rcu_preempt_need_deferred_qs(), "safe" involves only | |
590 | * not being in an RCU read-side critical section. The caller must | |
591 | * evaluate safety in terms of interrupt, softirq, and preemption | |
592 | * disabling. | |
593 | */ | |
594 | static void rcu_preempt_deferred_qs(struct task_struct *t) | |
595 | { | |
596 | unsigned long flags; | |
597 | bool couldrecurse = t->rcu_read_lock_nesting >= 0; | |
598 | ||
599 | if (!rcu_preempt_need_deferred_qs(t)) | |
600 | return; | |
601 | if (couldrecurse) | |
5f1a6ef3 | 602 | t->rcu_read_lock_nesting -= RCU_NEST_BIAS; |
3e310098 PM |
603 | local_irq_save(flags); |
604 | rcu_preempt_deferred_qs_irqrestore(t, flags); | |
605 | if (couldrecurse) | |
5f1a6ef3 | 606 | t->rcu_read_lock_nesting += RCU_NEST_BIAS; |
3e310098 PM |
607 | } |
608 | ||
609 | /* | |
610 | * Handle special cases during rcu_read_unlock(), such as needing to | |
611 | * notify RCU core processing or task having blocked during the RCU | |
612 | * read-side critical section. | |
613 | */ | |
614 | static void rcu_read_unlock_special(struct task_struct *t) | |
615 | { | |
616 | unsigned long flags; | |
617 | bool preempt_bh_were_disabled = | |
618 | !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)); | |
619 | bool irqs_were_disabled; | |
620 | ||
621 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
622 | if (in_nmi()) | |
623 | return; | |
624 | ||
625 | local_irq_save(flags); | |
626 | irqs_were_disabled = irqs_disabled_flags(flags); | |
05f41571 PM |
627 | if (preempt_bh_were_disabled || irqs_were_disabled) { |
628 | WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false); | |
3e310098 | 629 | /* Need to defer quiescent state until everything is enabled. */ |
05f41571 PM |
630 | if (irqs_were_disabled) { |
631 | /* Enabling irqs does not reschedule, so... */ | |
632 | raise_softirq_irqoff(RCU_SOFTIRQ); | |
633 | } else { | |
634 | /* Enabling BH or preempt does reschedule, so... */ | |
635 | set_tsk_need_resched(current); | |
636 | set_preempt_need_resched(); | |
637 | } | |
3e310098 PM |
638 | local_irq_restore(flags); |
639 | return; | |
640 | } | |
05f41571 | 641 | WRITE_ONCE(t->rcu_read_unlock_special.b.exp_hint, false); |
3e310098 PM |
642 | rcu_preempt_deferred_qs_irqrestore(t, flags); |
643 | } | |
644 | ||
1ed509a2 PM |
645 | /* |
646 | * Dump detailed information for all tasks blocking the current RCU | |
647 | * grace period on the specified rcu_node structure. | |
648 | */ | |
649 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
650 | { | |
651 | unsigned long flags; | |
1ed509a2 PM |
652 | struct task_struct *t; |
653 | ||
6cf10081 | 654 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 655 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 656 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
657 | return; |
658 | } | |
82efed06 | 659 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 660 | struct task_struct, rcu_node_entry); |
3caa973b TH |
661 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
662 | /* | |
663 | * We could be printing a lot while holding a spinlock. | |
664 | * Avoid triggering hard lockup. | |
665 | */ | |
666 | touch_nmi_watchdog(); | |
12f5f524 | 667 | sched_show_task(t); |
3caa973b | 668 | } |
67c583a7 | 669 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
670 | } |
671 | ||
672 | /* | |
673 | * Dump detailed information for all tasks blocking the current RCU | |
674 | * grace period. | |
675 | */ | |
a2887cd8 | 676 | static void rcu_print_detail_task_stall(void) |
1ed509a2 | 677 | { |
336a4f6c | 678 | struct rcu_node *rnp = rcu_get_root(); |
1ed509a2 PM |
679 | |
680 | rcu_print_detail_task_stall_rnp(rnp); | |
aedf4ba9 | 681 | rcu_for_each_leaf_node(rnp) |
1ed509a2 PM |
682 | rcu_print_detail_task_stall_rnp(rnp); |
683 | } | |
684 | ||
a858af28 PM |
685 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
686 | { | |
efc151c3 | 687 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
688 | rnp->level, rnp->grplo, rnp->grphi); |
689 | } | |
690 | ||
691 | static void rcu_print_task_stall_end(void) | |
692 | { | |
efc151c3 | 693 | pr_cont("\n"); |
a858af28 PM |
694 | } |
695 | ||
f41d911f PM |
696 | /* |
697 | * Scan the current list of tasks blocked within RCU read-side critical | |
698 | * sections, printing out the tid of each. | |
699 | */ | |
9bc8b558 | 700 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 701 | { |
f41d911f | 702 | struct task_struct *t; |
9bc8b558 | 703 | int ndetected = 0; |
f41d911f | 704 | |
27f4d280 | 705 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 706 | return 0; |
a858af28 | 707 | rcu_print_task_stall_begin(rnp); |
82efed06 | 708 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 709 | struct task_struct, rcu_node_entry); |
9bc8b558 | 710 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 711 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
712 | ndetected++; |
713 | } | |
a858af28 | 714 | rcu_print_task_stall_end(); |
9bc8b558 | 715 | return ndetected; |
f41d911f PM |
716 | } |
717 | ||
74611ecb PM |
718 | /* |
719 | * Scan the current list of tasks blocked within RCU read-side critical | |
720 | * sections, printing out the tid of each that is blocking the current | |
721 | * expedited grace period. | |
722 | */ | |
723 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
724 | { | |
725 | struct task_struct *t; | |
726 | int ndetected = 0; | |
727 | ||
728 | if (!rnp->exp_tasks) | |
729 | return 0; | |
730 | t = list_entry(rnp->exp_tasks->prev, | |
731 | struct task_struct, rcu_node_entry); | |
732 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
733 | pr_cont(" P%d", t->pid); | |
734 | ndetected++; | |
735 | } | |
736 | return ndetected; | |
737 | } | |
738 | ||
b0e165c0 PM |
739 | /* |
740 | * Check that the list of blocked tasks for the newly completed grace | |
741 | * period is in fact empty. It is a serious bug to complete a grace | |
742 | * period that still has RCU readers blocked! This function must be | |
ff3bb6f4 | 743 | * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock |
b0e165c0 | 744 | * must be held by the caller. |
12f5f524 PM |
745 | * |
746 | * Also, if there are blocked tasks on the list, they automatically | |
747 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 748 | */ |
81ab59a3 | 749 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 750 | { |
c5ebe66c PM |
751 | struct task_struct *t; |
752 | ||
ea9b0c8a | 753 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
4bc8d555 | 754 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 755 | dump_blkd_tasks(rnp, 10); |
0b107d24 PM |
756 | if (rcu_preempt_has_tasks(rnp) && |
757 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
12f5f524 | 758 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
759 | t = container_of(rnp->gp_tasks, struct task_struct, |
760 | rcu_node_entry); | |
761 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 762 | rnp->gp_seq, t->pid); |
c5ebe66c | 763 | } |
28ecd580 | 764 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
765 | } |
766 | ||
f41d911f | 767 | /* |
c98cac60 PM |
768 | * Check for a quiescent state from the current CPU, including voluntary |
769 | * context switches for Tasks RCU. When a task blocks, the task is | |
770 | * recorded in the corresponding CPU's rcu_node structure, which is checked | |
771 | * elsewhere, hence this function need only check for quiescent states | |
772 | * related to the current CPU, not to those related to tasks. | |
f41d911f | 773 | */ |
c98cac60 | 774 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f PM |
775 | { |
776 | struct task_struct *t = current; | |
777 | ||
45975c7d PM |
778 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
779 | rcu_note_voluntary_context_switch(current); | |
780 | } | |
3e310098 PM |
781 | if (t->rcu_read_lock_nesting > 0 || |
782 | (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) { | |
783 | /* No QS, force context switch if deferred. */ | |
fced9c8c PM |
784 | if (rcu_preempt_need_deferred_qs(t)) { |
785 | set_tsk_need_resched(t); | |
786 | set_preempt_need_resched(); | |
787 | } | |
3e310098 PM |
788 | } else if (rcu_preempt_need_deferred_qs(t)) { |
789 | rcu_preempt_deferred_qs(t); /* Report deferred QS. */ | |
790 | return; | |
791 | } else if (!t->rcu_read_lock_nesting) { | |
45975c7d | 792 | rcu_qs(); /* Report immediate QS. */ |
f41d911f PM |
793 | return; |
794 | } | |
3e310098 PM |
795 | |
796 | /* If GP is oldish, ask for help from rcu_read_unlock_special(). */ | |
10f39bb1 | 797 | if (t->rcu_read_lock_nesting > 0 && |
2280ee5a PM |
798 | __this_cpu_read(rcu_data.core_needs_qs) && |
799 | __this_cpu_read(rcu_data.cpu_no_qs.b.norm) && | |
15651201 | 800 | !t->rcu_read_unlock_special.b.need_qs && |
564a9ae6 | 801 | time_after(jiffies, rcu_state.gp_start + HZ)) |
1d082fd0 | 802 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
803 | } |
804 | ||
2439b696 PM |
805 | /* |
806 | * Check for a task exiting while in a preemptible-RCU read-side | |
884157ce PM |
807 | * critical section, clean up if so. No need to issue warnings, as |
808 | * debug_check_no_locks_held() already does this if lockdep is enabled. | |
809 | * Besides, if this function does anything other than just immediately | |
810 | * return, there was a bug of some sort. Spewing warnings from this | |
811 | * function is like as not to simply obscure important prior warnings. | |
2439b696 PM |
812 | */ |
813 | void exit_rcu(void) | |
814 | { | |
815 | struct task_struct *t = current; | |
816 | ||
884157ce PM |
817 | if (unlikely(!list_empty(¤t->rcu_node_entry))) { |
818 | t->rcu_read_lock_nesting = 1; | |
819 | barrier(); | |
820 | t->rcu_read_unlock_special.b.blocked = true; | |
821 | } else if (unlikely(t->rcu_read_lock_nesting)) { | |
822 | t->rcu_read_lock_nesting = 1; | |
823 | } else { | |
2439b696 | 824 | return; |
884157ce | 825 | } |
2439b696 | 826 | __rcu_read_unlock(); |
3e310098 | 827 | rcu_preempt_deferred_qs(current); |
2439b696 PM |
828 | } |
829 | ||
4bc8d555 PM |
830 | /* |
831 | * Dump the blocked-tasks state, but limit the list dump to the | |
832 | * specified number of elements. | |
833 | */ | |
57738942 | 834 | static void |
81ab59a3 | 835 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 | 836 | { |
57738942 | 837 | int cpu; |
4bc8d555 PM |
838 | int i; |
839 | struct list_head *lhp; | |
57738942 PM |
840 | bool onl; |
841 | struct rcu_data *rdp; | |
ff3cee39 | 842 | struct rcu_node *rnp1; |
4bc8d555 | 843 | |
ce11fae8 | 844 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 845 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 846 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
ff3cee39 PM |
847 | (long)rnp->gp_seq, (long)rnp->completedqs); |
848 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
849 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
850 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf PM |
851 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
852 | __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks); | |
853 | pr_info("%s: ->blkd_tasks", __func__); | |
4bc8d555 PM |
854 | i = 0; |
855 | list_for_each(lhp, &rnp->blkd_tasks) { | |
856 | pr_cont(" %p", lhp); | |
857 | if (++i >= 10) | |
858 | break; | |
859 | } | |
860 | pr_cont("\n"); | |
57738942 | 861 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
da1df50d | 862 | rdp = per_cpu_ptr(&rcu_data, cpu); |
57738942 PM |
863 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); |
864 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
865 | cpu, ".o"[onl], | |
866 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
867 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
868 | } | |
4bc8d555 PM |
869 | } |
870 | ||
28f6569a | 871 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f PM |
872 | |
873 | /* | |
874 | * Tell them what RCU they are running. | |
875 | */ | |
0e0fc1c2 | 876 | static void __init rcu_bootup_announce(void) |
f41d911f | 877 | { |
efc151c3 | 878 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 879 | rcu_bootup_announce_oddness(); |
f41d911f PM |
880 | } |
881 | ||
45975c7d PM |
882 | /* |
883 | * Note a quiescent state for PREEMPT=n. Because we do not need to know | |
884 | * how many quiescent states passed, just if there was at least one since | |
885 | * the start of the grace period, this just sets a flag. The caller must | |
886 | * have disabled preemption. | |
887 | */ | |
888 | static void rcu_qs(void) | |
d28139c4 | 889 | { |
45975c7d PM |
890 | RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!"); |
891 | if (!__this_cpu_read(rcu_data.cpu_no_qs.s)) | |
892 | return; | |
893 | trace_rcu_grace_period(TPS("rcu_sched"), | |
894 | __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs")); | |
895 | __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false); | |
896 | if (!__this_cpu_read(rcu_data.cpu_no_qs.b.exp)) | |
897 | return; | |
898 | __this_cpu_write(rcu_data.cpu_no_qs.b.exp, false); | |
63d4c8c9 | 899 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
d28139c4 PM |
900 | } |
901 | ||
395a2f09 PM |
902 | /* |
903 | * Register an urgently needed quiescent state. If there is an | |
904 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight | |
905 | * dyntick-idle quiescent state visible to other CPUs, which will in | |
906 | * some cases serve for expedited as well as normal grace periods. | |
907 | * Either way, register a lightweight quiescent state. | |
908 | * | |
909 | * The barrier() calls are redundant in the common case when this is | |
910 | * called externally, but just in case this is called from within this | |
911 | * file. | |
912 | * | |
913 | */ | |
914 | void rcu_all_qs(void) | |
915 | { | |
916 | unsigned long flags; | |
917 | ||
2dba13f0 | 918 | if (!raw_cpu_read(rcu_data.rcu_urgent_qs)) |
395a2f09 PM |
919 | return; |
920 | preempt_disable(); | |
921 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 922 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
395a2f09 PM |
923 | preempt_enable(); |
924 | return; | |
925 | } | |
2dba13f0 | 926 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
395a2f09 | 927 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
2dba13f0 | 928 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) { |
395a2f09 PM |
929 | local_irq_save(flags); |
930 | rcu_momentary_dyntick_idle(); | |
931 | local_irq_restore(flags); | |
932 | } | |
7e28c5af | 933 | rcu_qs(); |
395a2f09 PM |
934 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
935 | preempt_enable(); | |
936 | } | |
937 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
938 | ||
cba6d0d6 | 939 | /* |
45975c7d | 940 | * Note a PREEMPT=n context switch. The caller must have disabled interrupts. |
cba6d0d6 | 941 | */ |
45975c7d | 942 | void rcu_note_context_switch(bool preempt) |
cba6d0d6 | 943 | { |
45975c7d PM |
944 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
945 | trace_rcu_utilization(TPS("Start context switch")); | |
946 | rcu_qs(); | |
947 | /* Load rcu_urgent_qs before other flags. */ | |
2dba13f0 | 948 | if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) |
45975c7d | 949 | goto out; |
2dba13f0 PM |
950 | this_cpu_write(rcu_data.rcu_urgent_qs, false); |
951 | if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) | |
45975c7d | 952 | rcu_momentary_dyntick_idle(); |
45975c7d PM |
953 | if (!preempt) |
954 | rcu_tasks_qs(current); | |
955 | out: | |
956 | trace_rcu_utilization(TPS("End context switch")); | |
957 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ | |
cba6d0d6 | 958 | } |
45975c7d | 959 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
cba6d0d6 | 960 | |
fc2219d4 | 961 | /* |
6cc68793 | 962 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
963 | * RCU readers. |
964 | */ | |
27f4d280 | 965 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
966 | { |
967 | return 0; | |
968 | } | |
969 | ||
8af3a5e7 PM |
970 | /* |
971 | * Because there is no preemptible RCU, there can be no readers blocked. | |
972 | */ | |
973 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 974 | { |
8af3a5e7 | 975 | return false; |
b668c9cf PM |
976 | } |
977 | ||
3e310098 PM |
978 | /* |
979 | * Because there is no preemptible RCU, there can be no deferred quiescent | |
980 | * states. | |
981 | */ | |
982 | static bool rcu_preempt_need_deferred_qs(struct task_struct *t) | |
983 | { | |
984 | return false; | |
985 | } | |
986 | static void rcu_preempt_deferred_qs(struct task_struct *t) { } | |
987 | ||
1ed509a2 | 988 | /* |
6cc68793 | 989 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
990 | * tasks blocked within RCU read-side critical sections. |
991 | */ | |
a2887cd8 | 992 | static void rcu_print_detail_task_stall(void) |
1ed509a2 PM |
993 | { |
994 | } | |
995 | ||
f41d911f | 996 | /* |
6cc68793 | 997 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
998 | * tasks blocked within RCU read-side critical sections. |
999 | */ | |
9bc8b558 | 1000 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 1001 | { |
9bc8b558 | 1002 | return 0; |
f41d911f PM |
1003 | } |
1004 | ||
74611ecb PM |
1005 | /* |
1006 | * Because preemptible RCU does not exist, we never have to check for | |
1007 | * tasks blocked within RCU read-side critical sections that are | |
1008 | * blocking the current expedited grace period. | |
1009 | */ | |
1010 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
1011 | { | |
1012 | return 0; | |
1013 | } | |
1014 | ||
b0e165c0 | 1015 | /* |
6cc68793 | 1016 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
1017 | * so there is no need to check for blocked tasks. So check only for |
1018 | * bogus qsmask values. | |
b0e165c0 | 1019 | */ |
81ab59a3 | 1020 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) |
b0e165c0 | 1021 | { |
49e29126 | 1022 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
1023 | } |
1024 | ||
f41d911f | 1025 | /* |
c98cac60 PM |
1026 | * Check to see if this CPU is in a non-context-switch quiescent state, |
1027 | * namely user mode and idle loop. | |
f41d911f | 1028 | */ |
c98cac60 | 1029 | static void rcu_flavor_sched_clock_irq(int user) |
f41d911f | 1030 | { |
45975c7d | 1031 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
f41d911f | 1032 | |
45975c7d PM |
1033 | /* |
1034 | * Get here if this CPU took its interrupt from user | |
1035 | * mode or from the idle loop, and if this is not a | |
1036 | * nested interrupt. In this case, the CPU is in | |
1037 | * a quiescent state, so note it. | |
1038 | * | |
1039 | * No memory barrier is required here because rcu_qs() | |
1040 | * references only CPU-local variables that other CPUs | |
1041 | * neither access nor modify, at least not while the | |
1042 | * corresponding CPU is online. | |
1043 | */ | |
1044 | ||
1045 | rcu_qs(); | |
1046 | } | |
e74f4c45 | 1047 | } |
e74f4c45 | 1048 | |
2439b696 PM |
1049 | /* |
1050 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
1051 | * while in preemptible RCU read-side critical sections. | |
1052 | */ | |
1053 | void exit_rcu(void) | |
1054 | { | |
1055 | } | |
1056 | ||
4bc8d555 PM |
1057 | /* |
1058 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
1059 | */ | |
57738942 | 1060 | static void |
81ab59a3 | 1061 | dump_blkd_tasks(struct rcu_node *rnp, int ncheck) |
4bc8d555 PM |
1062 | { |
1063 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
1064 | } | |
1065 | ||
28f6569a | 1066 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 1067 | |
27f4d280 PM |
1068 | #ifdef CONFIG_RCU_BOOST |
1069 | ||
5d01bbd1 TG |
1070 | static void rcu_wake_cond(struct task_struct *t, int status) |
1071 | { | |
1072 | /* | |
1073 | * If the thread is yielding, only wake it when this | |
1074 | * is invoked from idle | |
1075 | */ | |
1076 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1077 | wake_up_process(t); | |
1078 | } | |
1079 | ||
27f4d280 PM |
1080 | /* |
1081 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1082 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1083 | * ->blkd_tasks list. | |
1084 | * | |
1085 | * Note that irqs must be enabled: boosting the task can block. | |
1086 | * Returns 1 if there are more tasks needing to be boosted. | |
1087 | */ | |
1088 | static int rcu_boost(struct rcu_node *rnp) | |
1089 | { | |
1090 | unsigned long flags; | |
27f4d280 PM |
1091 | struct task_struct *t; |
1092 | struct list_head *tb; | |
1093 | ||
7d0ae808 PM |
1094 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1095 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1096 | return 0; /* Nothing left to boost. */ |
1097 | ||
2a67e741 | 1098 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1099 | |
1100 | /* | |
1101 | * Recheck under the lock: all tasks in need of boosting | |
1102 | * might exit their RCU read-side critical sections on their own. | |
1103 | */ | |
1104 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1105 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1106 | return 0; |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Preferentially boost tasks blocking expedited grace periods. | |
1111 | * This cannot starve the normal grace periods because a second | |
1112 | * expedited grace period must boost all blocked tasks, including | |
1113 | * those blocking the pre-existing normal grace period. | |
1114 | */ | |
bec06785 | 1115 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1116 | tb = rnp->exp_tasks; |
bec06785 | 1117 | else |
27f4d280 PM |
1118 | tb = rnp->boost_tasks; |
1119 | ||
1120 | /* | |
1121 | * We boost task t by manufacturing an rt_mutex that appears to | |
1122 | * be held by task t. We leave a pointer to that rt_mutex where | |
1123 | * task t can find it, and task t will release the mutex when it | |
1124 | * exits its outermost RCU read-side critical section. Then | |
1125 | * simply acquiring this artificial rt_mutex will boost task | |
1126 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1127 | * | |
1128 | * Note that task t must acquire rnp->lock to remove itself from | |
1129 | * the ->blkd_tasks list, which it will do from exit() if from | |
1130 | * nowhere else. We therefore are guaranteed that task t will | |
1131 | * stay around at least until we drop rnp->lock. Note that | |
1132 | * rnp->lock also resolves races between our priority boosting | |
1133 | * and task t's exiting its outermost RCU read-side critical | |
1134 | * section. | |
1135 | */ | |
1136 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1137 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1138 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1139 | /* Lock only for side effect: boosts task t's priority. */ |
1140 | rt_mutex_lock(&rnp->boost_mtx); | |
1141 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1142 | |
7d0ae808 PM |
1143 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1144 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1145 | } |
1146 | ||
27f4d280 | 1147 | /* |
bc17ea10 | 1148 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1149 | */ |
1150 | static int rcu_boost_kthread(void *arg) | |
1151 | { | |
1152 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1153 | int spincnt = 0; | |
1154 | int more2boost; | |
1155 | ||
f7f7bac9 | 1156 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1157 | for (;;) { |
d71df90e | 1158 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1159 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1160 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1161 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1162 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1163 | more2boost = rcu_boost(rnp); |
1164 | if (more2boost) | |
1165 | spincnt++; | |
1166 | else | |
1167 | spincnt = 0; | |
1168 | if (spincnt > 10) { | |
5d01bbd1 | 1169 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1170 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1171 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1172 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1173 | spincnt = 0; |
1174 | } | |
1175 | } | |
1217ed1b | 1176 | /* NOTREACHED */ |
f7f7bac9 | 1177 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1178 | return 0; |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * Check to see if it is time to start boosting RCU readers that are | |
1183 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1184 | * kthread to start boosting them. If there is an expedited grace | |
1185 | * period in progress, it is always time to boost. | |
1186 | * | |
b065a853 PM |
1187 | * The caller must hold rnp->lock, which this function releases. |
1188 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1189 | * about it going away. | |
27f4d280 | 1190 | */ |
1217ed1b | 1191 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1192 | __releases(rnp->lock) |
27f4d280 | 1193 | { |
a32e01ee | 1194 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1195 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1196 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1197 | return; |
0ea1f2eb | 1198 | } |
27f4d280 PM |
1199 | if (rnp->exp_tasks != NULL || |
1200 | (rnp->gp_tasks != NULL && | |
1201 | rnp->boost_tasks == NULL && | |
1202 | rnp->qsmask == 0 && | |
1203 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1204 | if (rnp->exp_tasks == NULL) | |
1205 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1206 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
a2badefa PM |
1207 | rcu_wake_cond(rnp->boost_kthread_task, |
1208 | rnp->boost_kthread_status); | |
1217ed1b | 1209 | } else { |
67c583a7 | 1210 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1211 | } |
27f4d280 PM |
1212 | } |
1213 | ||
a46e0899 PM |
1214 | /* |
1215 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1216 | */ | |
1217 | static void invoke_rcu_callbacks_kthread(void) | |
1218 | { | |
1219 | unsigned long flags; | |
1220 | ||
1221 | local_irq_save(flags); | |
f7e972ee | 1222 | __this_cpu_write(rcu_data.rcu_cpu_has_work, 1); |
37f62d7c PM |
1223 | if (__this_cpu_read(rcu_data.rcu_cpu_kthread_task) != NULL && |
1224 | current != __this_cpu_read(rcu_data.rcu_cpu_kthread_task)) { | |
1225 | rcu_wake_cond(__this_cpu_read(rcu_data.rcu_cpu_kthread_task), | |
6ffdde28 | 1226 | __this_cpu_read(rcu_data.rcu_cpu_kthread_status)); |
5d01bbd1 | 1227 | } |
a46e0899 PM |
1228 | local_irq_restore(flags); |
1229 | } | |
1230 | ||
dff1672d PM |
1231 | /* |
1232 | * Is the current CPU running the RCU-callbacks kthread? | |
1233 | * Caller must have preemption disabled. | |
1234 | */ | |
1235 | static bool rcu_is_callbacks_kthread(void) | |
1236 | { | |
37f62d7c | 1237 | return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current; |
dff1672d PM |
1238 | } |
1239 | ||
27f4d280 PM |
1240 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1241 | ||
1242 | /* | |
1243 | * Do priority-boost accounting for the start of a new grace period. | |
1244 | */ | |
1245 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1246 | { | |
1247 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1248 | } | |
1249 | ||
27f4d280 PM |
1250 | /* |
1251 | * Create an RCU-boost kthread for the specified node if one does not | |
1252 | * already exist. We only create this kthread for preemptible RCU. | |
1253 | * Returns zero if all is well, a negated errno otherwise. | |
1254 | */ | |
6dbfdc14 | 1255 | static int rcu_spawn_one_boost_kthread(struct rcu_node *rnp) |
27f4d280 | 1256 | { |
6dbfdc14 | 1257 | int rnp_index = rnp - rcu_get_root(); |
27f4d280 PM |
1258 | unsigned long flags; |
1259 | struct sched_param sp; | |
1260 | struct task_struct *t; | |
1261 | ||
6dbfdc14 | 1262 | if (!IS_ENABLED(CONFIG_PREEMPT_RCU)) |
27f4d280 | 1263 | return 0; |
5d01bbd1 | 1264 | |
0aa04b05 | 1265 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1266 | return 0; |
1267 | ||
6dbfdc14 | 1268 | rcu_state.boost = 1; |
27f4d280 PM |
1269 | if (rnp->boost_kthread_task != NULL) |
1270 | return 0; | |
1271 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1272 | "rcub/%d", rnp_index); |
27f4d280 PM |
1273 | if (IS_ERR(t)) |
1274 | return PTR_ERR(t); | |
2a67e741 | 1275 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1276 | rnp->boost_kthread_task = t; |
67c583a7 | 1277 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1278 | sp.sched_priority = kthread_prio; |
27f4d280 | 1279 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1280 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1281 | return 0; |
1282 | } | |
1283 | ||
62ab7072 | 1284 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1285 | { |
f8b7fc6b | 1286 | struct sched_param sp; |
f8b7fc6b | 1287 | |
21871d7e | 1288 | sp.sched_priority = kthread_prio; |
62ab7072 | 1289 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1290 | } |
1291 | ||
62ab7072 | 1292 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1293 | { |
6ffdde28 | 1294 | per_cpu(rcu_data.rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1295 | } |
1296 | ||
62ab7072 | 1297 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1298 | { |
f7e972ee | 1299 | return __this_cpu_read(rcu_data.rcu_cpu_has_work); |
f8b7fc6b PM |
1300 | } |
1301 | ||
1302 | /* | |
0ae86a27 PM |
1303 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces |
1304 | * the RCU softirq used in configurations of RCU that do not support RCU | |
1305 | * priority boosting. | |
f8b7fc6b | 1306 | */ |
62ab7072 | 1307 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1308 | { |
6ffdde28 | 1309 | unsigned int *statusp = this_cpu_ptr(&rcu_data.rcu_cpu_kthread_status); |
f7e972ee | 1310 | char work, *workp = this_cpu_ptr(&rcu_data.rcu_cpu_has_work); |
62ab7072 | 1311 | int spincnt; |
f8b7fc6b | 1312 | |
62ab7072 | 1313 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1314 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1315 | local_bh_disable(); |
f8b7fc6b | 1316 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 | 1317 | local_irq_disable(); |
f8b7fc6b PM |
1318 | work = *workp; |
1319 | *workp = 0; | |
62ab7072 | 1320 | local_irq_enable(); |
f8b7fc6b | 1321 | if (work) |
c46f497a | 1322 | rcu_do_batch(this_cpu_ptr(&rcu_data)); |
f8b7fc6b | 1323 | local_bh_enable(); |
62ab7072 | 1324 | if (*workp == 0) { |
f7f7bac9 | 1325 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1326 | *statusp = RCU_KTHREAD_WAITING; |
1327 | return; | |
f8b7fc6b PM |
1328 | } |
1329 | } | |
62ab7072 | 1330 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1331 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1332 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1333 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1334 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1339 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1340 | * held, so the value of rnp->qsmaskinit will be stable. | |
1341 | * | |
1342 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1343 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1344 | * this function allows the kthread to execute on any CPU. | |
1345 | */ | |
5d01bbd1 | 1346 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1347 | { |
5d01bbd1 | 1348 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1349 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1350 | cpumask_var_t cm; |
1351 | int cpu; | |
f8b7fc6b | 1352 | |
5d01bbd1 | 1353 | if (!t) |
f8b7fc6b | 1354 | return; |
5d01bbd1 | 1355 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1356 | return; |
bc75e999 MR |
1357 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1358 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1359 | cpu != outgoingcpu) | |
f8b7fc6b | 1360 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1361 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1362 | cpumask_setall(cm); |
5d01bbd1 | 1363 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1364 | free_cpumask_var(cm); |
1365 | } | |
1366 | ||
62ab7072 | 1367 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
37f62d7c | 1368 | .store = &rcu_data.rcu_cpu_kthread_task, |
62ab7072 PM |
1369 | .thread_should_run = rcu_cpu_kthread_should_run, |
1370 | .thread_fn = rcu_cpu_kthread, | |
1371 | .thread_comm = "rcuc/%u", | |
1372 | .setup = rcu_cpu_kthread_setup, | |
1373 | .park = rcu_cpu_kthread_park, | |
1374 | }; | |
f8b7fc6b PM |
1375 | |
1376 | /* | |
9386c0b7 | 1377 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1378 | */ |
9386c0b7 | 1379 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1380 | { |
f8b7fc6b | 1381 | struct rcu_node *rnp; |
5d01bbd1 | 1382 | int cpu; |
f8b7fc6b | 1383 | |
62ab7072 | 1384 | for_each_possible_cpu(cpu) |
f7e972ee | 1385 | per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0; |
9213784b PM |
1386 | if (WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), "%s: Could not start rcub kthread, OOM is now expected behavior\n", __func__)) |
1387 | return; | |
aedf4ba9 | 1388 | rcu_for_each_leaf_node(rnp) |
6dbfdc14 | 1389 | (void)rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b | 1390 | } |
f8b7fc6b | 1391 | |
49fb4c62 | 1392 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1393 | { |
da1df50d | 1394 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
f8b7fc6b PM |
1395 | struct rcu_node *rnp = rdp->mynode; |
1396 | ||
1397 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1398 | if (rcu_scheduler_fully_active) |
6dbfdc14 | 1399 | (void)rcu_spawn_one_boost_kthread(rnp); |
f8b7fc6b PM |
1400 | } |
1401 | ||
27f4d280 PM |
1402 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1403 | ||
1217ed1b | 1404 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1405 | __releases(rnp->lock) |
27f4d280 | 1406 | { |
67c583a7 | 1407 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1408 | } |
1409 | ||
a46e0899 | 1410 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1411 | { |
a46e0899 | 1412 | WARN_ON_ONCE(1); |
27f4d280 PM |
1413 | } |
1414 | ||
dff1672d PM |
1415 | static bool rcu_is_callbacks_kthread(void) |
1416 | { | |
1417 | return false; | |
1418 | } | |
1419 | ||
27f4d280 PM |
1420 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1421 | { | |
1422 | } | |
1423 | ||
5d01bbd1 | 1424 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1425 | { |
1426 | } | |
1427 | ||
9386c0b7 | 1428 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1429 | { |
b0d30417 | 1430 | } |
b0d30417 | 1431 | |
49fb4c62 | 1432 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1433 | { |
1434 | } | |
1435 | ||
27f4d280 PM |
1436 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1437 | ||
8bd93a2c PM |
1438 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1439 | ||
1440 | /* | |
1441 | * Check to see if any future RCU-related work will need to be done | |
1442 | * by the current CPU, even if none need be done immediately, returning | |
1443 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1444 | * an exported member of the RCU API. | |
1445 | * | |
0ae86a27 PM |
1446 | * Because we not have RCU_FAST_NO_HZ, just check whether or not this |
1447 | * CPU has RCU callbacks queued. | |
8bd93a2c | 1448 | */ |
c1ad348b | 1449 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1450 | { |
c1ad348b | 1451 | *nextevt = KTIME_MAX; |
260e1e4f | 1452 | return !rcu_segcblist_empty(&this_cpu_ptr(&rcu_data)->cblist); |
7cb92499 PM |
1453 | } |
1454 | ||
1455 | /* | |
1456 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1457 | * after it. | |
1458 | */ | |
8fa7845d | 1459 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1460 | { |
1461 | } | |
1462 | ||
aea1b35e | 1463 | /* |
a858af28 | 1464 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1465 | * is nothing. |
1466 | */ | |
198bbf81 | 1467 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1468 | { |
1469 | } | |
1470 | ||
8bd93a2c PM |
1471 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1472 | ||
f23f7fa1 PM |
1473 | /* |
1474 | * This code is invoked when a CPU goes idle, at which point we want | |
1475 | * to have the CPU do everything required for RCU so that it can enter | |
1476 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1477 | * state machine implemented by rcu_prepare_for_idle() below. | |
1478 | * | |
1479 | * The following three proprocessor symbols control this state machine: | |
1480 | * | |
f23f7fa1 PM |
1481 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1482 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1483 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1484 | * benchmarkers who might otherwise be tempted to set this to a large | |
1485 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1486 | * system. And if you are -that- concerned about energy efficiency, | |
1487 | * just power the system down and be done with it! | |
778d250a PM |
1488 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1489 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1490 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1491 | * |
1492 | * The values below work well in practice. If future workloads require | |
1493 | * adjustment, they can be converted into kernel config parameters, though | |
1494 | * making the state machine smarter might be a better option. | |
1495 | */ | |
e84c48ae | 1496 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1497 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1498 | |
5e44ce35 PM |
1499 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1500 | module_param(rcu_idle_gp_delay, int, 0644); | |
1501 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1502 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1503 | |
486e2593 | 1504 | /* |
0ae86a27 PM |
1505 | * Try to advance callbacks on the current CPU, but only if it has been |
1506 | * awhile since the last time we did so. Afterwards, if there are any | |
1507 | * callbacks ready for immediate invocation, return true. | |
486e2593 | 1508 | */ |
f1f399d1 | 1509 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1510 | { |
c0f4dfd4 | 1511 | bool cbs_ready = false; |
5998a75a | 1512 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1513 | struct rcu_node *rnp; |
486e2593 | 1514 | |
c229828c | 1515 | /* Exit early if we advanced recently. */ |
5998a75a | 1516 | if (jiffies == rdp->last_advance_all) |
d0bc90fd | 1517 | return false; |
5998a75a | 1518 | rdp->last_advance_all = jiffies; |
c229828c | 1519 | |
b97d23c5 | 1520 | rnp = rdp->mynode; |
486e2593 | 1521 | |
b97d23c5 PM |
1522 | /* |
1523 | * Don't bother checking unless a grace period has | |
1524 | * completed since we last checked and there are | |
1525 | * callbacks not yet ready to invoke. | |
1526 | */ | |
1527 | if ((rcu_seq_completed_gp(rdp->gp_seq, | |
1528 | rcu_seq_current(&rnp->gp_seq)) || | |
1529 | unlikely(READ_ONCE(rdp->gpwrap))) && | |
1530 | rcu_segcblist_pend_cbs(&rdp->cblist)) | |
1531 | note_gp_changes(rdp); | |
1532 | ||
1533 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
1534 | cbs_ready = true; | |
c0f4dfd4 | 1535 | return cbs_ready; |
486e2593 PM |
1536 | } |
1537 | ||
aa9b1630 | 1538 | /* |
c0f4dfd4 PM |
1539 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1540 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1541 | * caller to set the timeout based on whether or not there are non-lazy | |
1542 | * callbacks. | |
aa9b1630 | 1543 | * |
c0f4dfd4 | 1544 | * The caller must have disabled interrupts. |
aa9b1630 | 1545 | */ |
c1ad348b | 1546 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1547 | { |
5998a75a | 1548 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c1ad348b | 1549 | unsigned long dj; |
aa9b1630 | 1550 | |
b04db8e1 | 1551 | lockdep_assert_irqs_disabled(); |
3382adbc | 1552 | |
aa9b1630 | 1553 | /* If no callbacks, RCU doesn't need the CPU. */ |
260e1e4f | 1554 | if (rcu_segcblist_empty(&rdp->cblist)) { |
c1ad348b | 1555 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1556 | return 0; |
1557 | } | |
c0f4dfd4 PM |
1558 | |
1559 | /* Attempt to advance callbacks. */ | |
1560 | if (rcu_try_advance_all_cbs()) { | |
1561 | /* Some ready to invoke, so initiate later invocation. */ | |
1562 | invoke_rcu_core(); | |
aa9b1630 PM |
1563 | return 1; |
1564 | } | |
5998a75a | 1565 | rdp->last_accelerate = jiffies; |
c0f4dfd4 PM |
1566 | |
1567 | /* Request timer delay depending on laziness, and round. */ | |
260e1e4f PM |
1568 | rdp->all_lazy = !rcu_segcblist_n_nonlazy_cbs(&rdp->cblist); |
1569 | if (rdp->all_lazy) { | |
1570 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; | |
1571 | } else { | |
c1ad348b | 1572 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1573 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1574 | } |
c1ad348b | 1575 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1576 | return 0; |
1577 | } | |
1578 | ||
21e52e15 | 1579 | /* |
c0f4dfd4 PM |
1580 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1581 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1582 | * The second major task is to check to see if a non-lazy callback has | |
1583 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1584 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1585 | * any recently arrived callbacks. | |
aea1b35e PM |
1586 | * |
1587 | * The caller must have disabled interrupts. | |
8bd93a2c | 1588 | */ |
198bbf81 | 1589 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1590 | { |
48a7639c | 1591 | bool needwake; |
0fd79e75 | 1592 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
c0f4dfd4 | 1593 | struct rcu_node *rnp; |
9d2ad243 PM |
1594 | int tne; |
1595 | ||
b04db8e1 | 1596 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1597 | if (rcu_is_nocb_cpu(smp_processor_id())) |
3382adbc PM |
1598 | return; |
1599 | ||
9d2ad243 | 1600 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1601 | tne = READ_ONCE(tick_nohz_active); |
0fd79e75 | 1602 | if (tne != rdp->tick_nohz_enabled_snap) { |
260e1e4f | 1603 | if (!rcu_segcblist_empty(&rdp->cblist)) |
9d2ad243 | 1604 | invoke_rcu_core(); /* force nohz to see update. */ |
0fd79e75 | 1605 | rdp->tick_nohz_enabled_snap = tne; |
9d2ad243 PM |
1606 | return; |
1607 | } | |
1608 | if (!tne) | |
1609 | return; | |
f511fc62 | 1610 | |
c57afe80 | 1611 | /* |
c0f4dfd4 PM |
1612 | * If a non-lazy callback arrived at a CPU having only lazy |
1613 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1614 | * idle duration on re-entry to idle. | |
c57afe80 | 1615 | */ |
260e1e4f | 1616 | if (rdp->all_lazy && rcu_segcblist_n_nonlazy_cbs(&rdp->cblist)) { |
c458a89e | 1617 | rdp->all_lazy = false; |
c0f4dfd4 | 1618 | invoke_rcu_core(); |
c57afe80 PM |
1619 | return; |
1620 | } | |
c57afe80 | 1621 | |
3084f2f8 | 1622 | /* |
c0f4dfd4 PM |
1623 | * If we have not yet accelerated this jiffy, accelerate all |
1624 | * callbacks on this CPU. | |
3084f2f8 | 1625 | */ |
5998a75a | 1626 | if (rdp->last_accelerate == jiffies) |
aea1b35e | 1627 | return; |
5998a75a | 1628 | rdp->last_accelerate = jiffies; |
b97d23c5 | 1629 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) { |
c0f4dfd4 | 1630 | rnp = rdp->mynode; |
2a67e741 | 1631 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
02f50142 | 1632 | needwake = rcu_accelerate_cbs(rnp, rdp); |
67c583a7 | 1633 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c | 1634 | if (needwake) |
532c00c9 | 1635 | rcu_gp_kthread_wake(); |
77e38ed3 | 1636 | } |
c0f4dfd4 | 1637 | } |
3084f2f8 | 1638 | |
c0f4dfd4 PM |
1639 | /* |
1640 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1641 | * any grace periods that elapsed while the CPU was idle, and if any | |
1642 | * callbacks are now ready to invoke, initiate invocation. | |
1643 | */ | |
8fa7845d | 1644 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1645 | { |
b04db8e1 | 1646 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1647 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1648 | return; |
7a497c96 PM |
1649 | if (rcu_try_advance_all_cbs()) |
1650 | invoke_rcu_core(); | |
8bd93a2c PM |
1651 | } |
1652 | ||
1653 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ | |
a858af28 | 1654 | |
a858af28 PM |
1655 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1656 | ||
1657 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1658 | { | |
0fd79e75 | 1659 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
a858af28 | 1660 | |
260e1e4f | 1661 | sprintf(cp, "last_accelerate: %04lx/%04lx, Nonlazy posted: %c%c%c", |
5998a75a | 1662 | rdp->last_accelerate & 0xffff, jiffies & 0xffff, |
260e1e4f PM |
1663 | ".l"[rdp->all_lazy], |
1664 | ".L"[!rcu_segcblist_n_nonlazy_cbs(&rdp->cblist)], | |
1665 | ".D"[!rdp->tick_nohz_enabled_snap]); | |
a858af28 PM |
1666 | } |
1667 | ||
1668 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1669 | ||
1670 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1671 | { | |
1c17e4d4 | 1672 | *cp = '\0'; |
a858af28 PM |
1673 | } |
1674 | ||
1675 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1676 | ||
1677 | /* Initiate the stall-info list. */ | |
1678 | static void print_cpu_stall_info_begin(void) | |
1679 | { | |
efc151c3 | 1680 | pr_cont("\n"); |
a858af28 PM |
1681 | } |
1682 | ||
1683 | /* | |
1684 | * Print out diagnostic information for the specified stalled CPU. | |
1685 | * | |
564a9ae6 PM |
1686 | * If the specified CPU is aware of the current RCU grace period, then |
1687 | * print the number of scheduling clock interrupts the CPU has taken | |
1688 | * during the time that it has been aware. Otherwise, print the number | |
1689 | * of RCU grace periods that this CPU is ignorant of, for example, "1" | |
1690 | * if the CPU was aware of the previous grace period. | |
a858af28 PM |
1691 | * |
1692 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1693 | */ | |
b21ebed9 | 1694 | static void print_cpu_stall_info(int cpu) |
a858af28 | 1695 | { |
9b9500da | 1696 | unsigned long delta; |
a858af28 | 1697 | char fast_no_hz[72]; |
da1df50d | 1698 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
a858af28 PM |
1699 | char *ticks_title; |
1700 | unsigned long ticks_value; | |
1701 | ||
3caa973b TH |
1702 | /* |
1703 | * We could be printing a lot while holding a spinlock. Avoid | |
1704 | * triggering hard lockup. | |
1705 | */ | |
1706 | touch_nmi_watchdog(); | |
1707 | ||
b21ebed9 | 1708 | ticks_value = rcu_seq_ctr(rcu_state.gp_seq - rdp->gp_seq); |
471f87c3 PM |
1709 | if (ticks_value) { |
1710 | ticks_title = "GPs behind"; | |
1711 | } else { | |
a858af28 PM |
1712 | ticks_title = "ticks this GP"; |
1713 | ticks_value = rdp->ticks_this_gp; | |
a858af28 PM |
1714 | } |
1715 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
8aa670cd | 1716 | delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq); |
89b4cd4b | 1717 | pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%#lx softirq=%u/%u fqs=%ld %s\n", |
7f21aeef PM |
1718 | cpu, |
1719 | "O."[!!cpu_online(cpu)], | |
1720 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1721 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
9b9500da PM |
1722 | !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' : |
1723 | rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : | |
1724 | "!."[!delta], | |
7f21aeef | 1725 | ticks_value, ticks_title, |
dc5a4f29 | 1726 | rcu_dynticks_snap(rdp) & 0xfff, |
4c5273bf | 1727 | rdp->dynticks_nesting, rdp->dynticks_nmi_nesting, |
6231069b | 1728 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
b21ebed9 | 1729 | READ_ONCE(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart, |
a858af28 PM |
1730 | fast_no_hz); |
1731 | } | |
1732 | ||
1733 | /* Terminate the stall-info list. */ | |
1734 | static void print_cpu_stall_info_end(void) | |
1735 | { | |
efc151c3 | 1736 | pr_err("\t"); |
a858af28 PM |
1737 | } |
1738 | ||
0ae86a27 | 1739 | /* Zero ->ticks_this_gp and snapshot the number of RCU softirq handlers. */ |
a858af28 PM |
1740 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) |
1741 | { | |
1742 | rdp->ticks_this_gp = 0; | |
6231069b | 1743 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
d3052109 | 1744 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
a858af28 PM |
1745 | } |
1746 | ||
3fbfbf7a PM |
1747 | #ifdef CONFIG_RCU_NOCB_CPU |
1748 | ||
1749 | /* | |
1750 | * Offload callback processing from the boot-time-specified set of CPUs | |
a9fefdb2 PM |
1751 | * specified by rcu_nocb_mask. For the CPUs in the set, there are kthreads |
1752 | * created that pull the callbacks from the corresponding CPU, wait for | |
1753 | * a grace period to elapse, and invoke the callbacks. These kthreads | |
1754 | * are organized into leaders, which manage incoming callbacks, wait for | |
1755 | * grace periods, and awaken followers, and the followers, which only | |
1756 | * invoke callbacks. Each leader is its own follower. The no-CBs CPUs | |
1757 | * do a wake_up() on their kthread when they insert a callback into any | |
1758 | * empty list, unless the rcu_nocb_poll boot parameter has been specified, | |
1759 | * in which case each kthread actively polls its CPU. (Which isn't so great | |
1760 | * for energy efficiency, but which does reduce RCU's overhead on that CPU.) | |
3fbfbf7a PM |
1761 | * |
1762 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1763 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1764 | * running CPU-bound user-mode computations. | |
1765 | * | |
a9fefdb2 PM |
1766 | * Offloading of callbacks can also be used as an energy-efficiency |
1767 | * measure because CPUs with no RCU callbacks queued are more aggressive | |
1768 | * about entering dyntick-idle mode. | |
3fbfbf7a PM |
1769 | */ |
1770 | ||
1771 | ||
497e4260 PM |
1772 | /* |
1773 | * Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. | |
1774 | * The string after the "rcu_nocbs=" is either "all" for all CPUs, or a | |
1775 | * comma-separated list of CPUs and/or CPU ranges. If an invalid list is | |
1776 | * given, a warning is emitted and all CPUs are offloaded. | |
1777 | */ | |
3fbfbf7a PM |
1778 | static int __init rcu_nocb_setup(char *str) |
1779 | { | |
1780 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
da8739f2 PM |
1781 | if (!strcasecmp(str, "all")) |
1782 | cpumask_setall(rcu_nocb_mask); | |
1783 | else | |
497e4260 PM |
1784 | if (cpulist_parse(str, rcu_nocb_mask)) { |
1785 | pr_warn("rcu_nocbs= bad CPU range, all CPUs set\n"); | |
1786 | cpumask_setall(rcu_nocb_mask); | |
1787 | } | |
3fbfbf7a PM |
1788 | return 1; |
1789 | } | |
1790 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1791 | ||
1b0048a4 PG |
1792 | static int __init parse_rcu_nocb_poll(char *arg) |
1793 | { | |
5455a7f6 | 1794 | rcu_nocb_poll = true; |
1b0048a4 PG |
1795 | return 0; |
1796 | } | |
1797 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1798 | ||
dae6e64d | 1799 | /* |
0446be48 PM |
1800 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1801 | * grace period. | |
dae6e64d | 1802 | */ |
abedf8e2 | 1803 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1804 | { |
abedf8e2 | 1805 | swake_up_all(sq); |
dae6e64d PM |
1806 | } |
1807 | ||
abedf8e2 | 1808 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1809 | { |
e0da2374 | 1810 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1811 | } |
1812 | ||
dae6e64d | 1813 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1814 | { |
abedf8e2 PG |
1815 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1816 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1817 | } |
1818 | ||
24342c96 | 1819 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1820 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1821 | { |
84b12b75 | 1822 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1823 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1824 | return false; | |
1825 | } | |
1826 | ||
fbce7497 | 1827 | /* |
8be6e1b1 PM |
1828 | * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock |
1829 | * and this function releases it. | |
fbce7497 | 1830 | */ |
8be6e1b1 PM |
1831 | static void __wake_nocb_leader(struct rcu_data *rdp, bool force, |
1832 | unsigned long flags) | |
1833 | __releases(rdp->nocb_lock) | |
fbce7497 PM |
1834 | { |
1835 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1836 | ||
8be6e1b1 PM |
1837 | lockdep_assert_held(&rdp->nocb_lock); |
1838 | if (!READ_ONCE(rdp_leader->nocb_kthread)) { | |
1839 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 1840 | return; |
8be6e1b1 PM |
1841 | } |
1842 | if (rdp_leader->nocb_leader_sleep || force) { | |
39953dfd | 1843 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1844 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
8be6e1b1 PM |
1845 | del_timer(&rdp->nocb_timer); |
1846 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
b3dae109 PZ |
1847 | smp_mb(); /* ->nocb_leader_sleep before swake_up_one(). */ |
1848 | swake_up_one(&rdp_leader->nocb_wq); | |
8be6e1b1 PM |
1849 | } else { |
1850 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 PM |
1851 | } |
1852 | } | |
1853 | ||
8be6e1b1 PM |
1854 | /* |
1855 | * Kick the leader kthread for this NOCB group, but caller has not | |
1856 | * acquired locks. | |
1857 | */ | |
1858 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1859 | { | |
1860 | unsigned long flags; | |
1861 | ||
1862 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1863 | __wake_nocb_leader(rdp, force, flags); | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * Arrange to wake the leader kthread for this NOCB group at some | |
1868 | * future time when it is safe to do so. | |
1869 | */ | |
1870 | static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype, | |
1871 | const char *reason) | |
1872 | { | |
1873 | unsigned long flags; | |
1874 | ||
1875 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1876 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) | |
1877 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
1878 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
88d1bead | 1879 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, reason); |
8be6e1b1 PM |
1880 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
1881 | } | |
1882 | ||
a9fefdb2 | 1883 | /* Does rcu_barrier need to queue an RCU callback on the specified CPU? */ |
4580b054 | 1884 | static bool rcu_nocb_cpu_needs_barrier(int cpu) |
d7e29933 | 1885 | { |
da1df50d | 1886 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
41050a00 PM |
1887 | unsigned long ret; |
1888 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1889 | struct rcu_head *rhp; |
41050a00 | 1890 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1891 | |
41050a00 PM |
1892 | /* |
1893 | * Check count of all no-CBs callbacks awaiting invocation. | |
1894 | * There needs to be a barrier before this function is called, | |
1895 | * but associated with a prior determination that no more | |
1896 | * callbacks would be posted. In the worst case, the first | |
dd46a788 | 1897 | * barrier in rcu_barrier() suffices (but the caller cannot |
41050a00 | 1898 | * necessarily rely on this, not a substitute for the caller |
a9fefdb2 PM |
1899 | * getting the concurrency design right!). There must also be a |
1900 | * barrier between the following load and posting of a callback | |
41050a00 PM |
1901 | * (if a callback is in fact needed). This is associated with an |
1902 | * atomic_inc() in the caller. | |
1903 | */ | |
903ee83d | 1904 | ret = rcu_get_n_cbs_nocb_cpu(rdp); |
d7e29933 | 1905 | |
41050a00 | 1906 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 1907 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 1908 | if (!rhp) |
7d0ae808 | 1909 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 1910 | if (!rhp) |
7d0ae808 | 1911 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
1912 | |
1913 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 1914 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 1915 | rcu_scheduler_fully_active) { |
d7e29933 PM |
1916 | /* RCU callback enqueued before CPU first came online??? */ |
1917 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1918 | cpu, rhp->func); | |
1919 | WARN_ON_ONCE(1); | |
1920 | } | |
41050a00 | 1921 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1922 | |
41050a00 | 1923 | return !!ret; |
d7e29933 PM |
1924 | } |
1925 | ||
3fbfbf7a PM |
1926 | /* |
1927 | * Enqueue the specified string of rcu_head structures onto the specified | |
1928 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
1929 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
1930 | * counts are supplied by rhcount and rhcount_lazy. | |
1931 | * | |
1932 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1933 | */ | |
1934 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
1935 | struct rcu_head *rhp, | |
1936 | struct rcu_head **rhtp, | |
96d3fd0d PM |
1937 | int rhcount, int rhcount_lazy, |
1938 | unsigned long flags) | |
3fbfbf7a PM |
1939 | { |
1940 | int len; | |
1941 | struct rcu_head **old_rhpp; | |
1942 | struct task_struct *t; | |
1943 | ||
1944 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
1945 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
1946 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 1947 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 1948 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 1949 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 1950 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
1951 | |
1952 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 1953 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 1954 | if (rcu_nocb_poll || !t) { |
88d1bead | 1955 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9261dd0d | 1956 | TPS("WakeNotPoll")); |
3fbfbf7a | 1957 | return; |
9261dd0d | 1958 | } |
903ee83d | 1959 | len = rcu_get_n_cbs_nocb_cpu(rdp); |
3fbfbf7a | 1960 | if (old_rhpp == &rdp->nocb_head) { |
96d3fd0d | 1961 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
1962 | /* ... if queue was empty ... */ |
1963 | wake_nocb_leader(rdp, false); | |
88d1bead | 1964 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
96d3fd0d PM |
1965 | TPS("WakeEmpty")); |
1966 | } else { | |
8be6e1b1 PM |
1967 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
1968 | TPS("WakeEmptyIsDeferred")); | |
96d3fd0d | 1969 | } |
3fbfbf7a PM |
1970 | rdp->qlen_last_fqs_check = 0; |
1971 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 1972 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
1973 | if (!irqs_disabled_flags(flags)) { |
1974 | wake_nocb_leader(rdp, true); | |
88d1bead | 1975 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
9fdd3bc9 PM |
1976 | TPS("WakeOvf")); |
1977 | } else { | |
efcd2d54 | 1978 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE_FORCE, |
8be6e1b1 | 1979 | TPS("WakeOvfIsDeferred")); |
9fdd3bc9 | 1980 | } |
3fbfbf7a | 1981 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d | 1982 | } else { |
88d1bead | 1983 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WakeNot")); |
3fbfbf7a PM |
1984 | } |
1985 | return; | |
1986 | } | |
1987 | ||
1988 | /* | |
1989 | * This is a helper for __call_rcu(), which invokes this when the normal | |
1990 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
1991 | * function returns failure back to __call_rcu(), which can complain | |
1992 | * appropriately. | |
1993 | * | |
1994 | * Otherwise, this function queues the callback where the corresponding | |
1995 | * "rcuo" kthread can find it. | |
1996 | */ | |
1997 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 1998 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
1999 | { |
2000 | ||
d1e43fa5 | 2001 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 2002 | return false; |
96d3fd0d | 2003 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 | 2004 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
88d1bead | 2005 | trace_rcu_kfree_callback(rcu_state.name, rhp, |
21e7a608 | 2006 | (unsigned long)rhp->func, |
756cbf6b | 2007 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2008 | -rcu_get_n_cbs_nocb_cpu(rdp)); |
21e7a608 | 2009 | else |
88d1bead | 2010 | trace_rcu_callback(rcu_state.name, rhp, |
756cbf6b | 2011 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2012 | -rcu_get_n_cbs_nocb_cpu(rdp)); |
1772947b PM |
2013 | |
2014 | /* | |
2015 | * If called from an extended quiescent state with interrupts | |
2016 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2017 | * deferred-wakeup check to function. | |
2018 | */ | |
2019 | if (irqs_disabled_flags(flags) && | |
2020 | !rcu_is_watching() && | |
2021 | cpu_online(smp_processor_id())) | |
2022 | invoke_rcu_core(); | |
2023 | ||
c271d3a9 | 2024 | return true; |
3fbfbf7a PM |
2025 | } |
2026 | ||
2027 | /* | |
2028 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2029 | * not a no-CBs CPU. | |
2030 | */ | |
b1a2d79f | 2031 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2032 | struct rcu_data *rdp, |
2033 | unsigned long flags) | |
3fbfbf7a | 2034 | { |
b04db8e1 | 2035 | lockdep_assert_irqs_disabled(); |
d1e43fa5 | 2036 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
b1a2d79f PM |
2037 | return false; /* Not NOCBs CPU, caller must migrate CBs. */ |
2038 | __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist), | |
2039 | rcu_segcblist_tail(&rdp->cblist), | |
2040 | rcu_segcblist_n_cbs(&rdp->cblist), | |
2041 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags); | |
2042 | rcu_segcblist_init(&rdp->cblist); | |
2043 | rcu_segcblist_disable(&rdp->cblist); | |
0a9e1e11 | 2044 | return true; |
3fbfbf7a PM |
2045 | } |
2046 | ||
2047 | /* | |
34ed6246 PM |
2048 | * If necessary, kick off a new grace period, and either way wait |
2049 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2050 | */ |
34ed6246 | 2051 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2052 | { |
34ed6246 | 2053 | unsigned long c; |
dae6e64d | 2054 | bool d; |
34ed6246 | 2055 | unsigned long flags; |
48a7639c | 2056 | bool needwake; |
34ed6246 PM |
2057 | struct rcu_node *rnp = rdp->mynode; |
2058 | ||
ab5e869c | 2059 | local_irq_save(flags); |
88d1bead | 2060 | c = rcu_seq_snap(&rcu_state.gp_seq); |
ab5e869c PM |
2061 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
2062 | local_irq_restore(flags); | |
2063 | } else { | |
2064 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
2065 | needwake = rcu_start_this_gp(rnp, rdp, c); | |
2066 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2067 | if (needwake) | |
532c00c9 | 2068 | rcu_gp_kthread_wake(); |
ab5e869c | 2069 | } |
3fbfbf7a PM |
2070 | |
2071 | /* | |
34ed6246 PM |
2072 | * Wait for the grace period. Do so interruptibly to avoid messing |
2073 | * up the load average. | |
3fbfbf7a | 2074 | */ |
41e80595 | 2075 | trace_rcu_this_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2076 | for (;;) { |
b3dae109 | 2077 | swait_event_interruptible_exclusive( |
29365e56 PM |
2078 | rnp->nocb_gp_wq[rcu_seq_ctr(c) & 0x1], |
2079 | (d = rcu_seq_done(&rnp->gp_seq, c))); | |
dae6e64d | 2080 | if (likely(d)) |
34ed6246 | 2081 | break; |
73a860cd | 2082 | WARN_ON(signal_pending(current)); |
41e80595 | 2083 | trace_rcu_this_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2084 | } |
41e80595 | 2085 | trace_rcu_this_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2086 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2087 | } |
2088 | ||
fbce7497 PM |
2089 | /* |
2090 | * Leaders come here to wait for additional callbacks to show up. | |
2091 | * This function does not return until callbacks appear. | |
2092 | */ | |
2093 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2094 | { | |
2095 | bool firsttime = true; | |
8be6e1b1 | 2096 | unsigned long flags; |
fbce7497 PM |
2097 | bool gotcbs; |
2098 | struct rcu_data *rdp; | |
2099 | struct rcu_head **tail; | |
2100 | ||
2101 | wait_again: | |
2102 | ||
2103 | /* Wait for callbacks to appear. */ | |
2104 | if (!rcu_nocb_poll) { | |
88d1bead | 2105 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Sleep")); |
b3dae109 | 2106 | swait_event_interruptible_exclusive(my_rdp->nocb_wq, |
7d0ae808 | 2107 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
8be6e1b1 PM |
2108 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); |
2109 | my_rdp->nocb_leader_sleep = true; | |
2110 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); | |
2111 | del_timer(&my_rdp->nocb_timer); | |
2112 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); | |
fbce7497 PM |
2113 | } else if (firsttime) { |
2114 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
88d1bead | 2115 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, TPS("Poll")); |
fbce7497 PM |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * Each pass through the following loop checks a follower for CBs. | |
2120 | * We are our own first follower. Any CBs found are moved to | |
2121 | * nocb_gp_head, where they await a grace period. | |
2122 | */ | |
2123 | gotcbs = false; | |
8be6e1b1 | 2124 | smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */ |
fbce7497 | 2125 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2126 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2127 | if (!rdp->nocb_gp_head) |
2128 | continue; /* No CBs here, try next follower. */ | |
2129 | ||
2130 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2131 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2132 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2133 | gotcbs = true; |
2134 | } | |
2135 | ||
8be6e1b1 | 2136 | /* No callbacks? Sleep a bit if polling, and go retry. */ |
fbce7497 | 2137 | if (unlikely(!gotcbs)) { |
73a860cd | 2138 | WARN_ON(signal_pending(current)); |
8be6e1b1 PM |
2139 | if (rcu_nocb_poll) { |
2140 | schedule_timeout_interruptible(1); | |
2141 | } else { | |
88d1bead | 2142 | trace_rcu_nocb_wake(rcu_state.name, my_rdp->cpu, |
bedbb648 | 2143 | TPS("WokeEmpty")); |
8be6e1b1 | 2144 | } |
fbce7497 PM |
2145 | goto wait_again; |
2146 | } | |
2147 | ||
2148 | /* Wait for one grace period. */ | |
2149 | rcu_nocb_wait_gp(my_rdp); | |
2150 | ||
fbce7497 PM |
2151 | /* Each pass through the following loop wakes a follower, if needed. */ |
2152 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
8be6e1b1 PM |
2153 | if (!rcu_nocb_poll && |
2154 | READ_ONCE(rdp->nocb_head) && | |
2155 | READ_ONCE(my_rdp->nocb_leader_sleep)) { | |
2156 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); | |
11ed7f93 | 2157 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
8be6e1b1 PM |
2158 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); |
2159 | } | |
fbce7497 PM |
2160 | if (!rdp->nocb_gp_head) |
2161 | continue; /* No CBs, so no need to wake follower. */ | |
2162 | ||
2163 | /* Append callbacks to follower's "done" list. */ | |
8be6e1b1 PM |
2164 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2165 | tail = rdp->nocb_follower_tail; | |
2166 | rdp->nocb_follower_tail = rdp->nocb_gp_tail; | |
fbce7497 | 2167 | *tail = rdp->nocb_gp_head; |
8be6e1b1 | 2168 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
fbce7497 | 2169 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
8be6e1b1 | 2170 | /* List was empty, so wake up the follower. */ |
b3dae109 | 2171 | swake_up_one(&rdp->nocb_wq); |
fbce7497 PM |
2172 | } |
2173 | } | |
2174 | ||
2175 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2176 | if (!my_rdp->nocb_follower_head) | |
2177 | goto wait_again; | |
2178 | } | |
2179 | ||
2180 | /* | |
2181 | * Followers come here to wait for additional callbacks to show up. | |
2182 | * This function does not return until callbacks appear. | |
2183 | */ | |
2184 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2185 | { | |
fbce7497 | 2186 | for (;;) { |
88d1bead | 2187 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("FollowerSleep")); |
b3dae109 | 2188 | swait_event_interruptible_exclusive(rdp->nocb_wq, |
8be6e1b1 | 2189 | READ_ONCE(rdp->nocb_follower_head)); |
fbce7497 PM |
2190 | if (smp_load_acquire(&rdp->nocb_follower_head)) { |
2191 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2192 | return; | |
2193 | } | |
73a860cd | 2194 | WARN_ON(signal_pending(current)); |
88d1bead | 2195 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty")); |
fbce7497 PM |
2196 | } |
2197 | } | |
2198 | ||
3fbfbf7a PM |
2199 | /* |
2200 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2201 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2202 | * an optional leader-follower relationship so that the grace-period | |
2203 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2204 | */ |
2205 | static int rcu_nocb_kthread(void *arg) | |
2206 | { | |
2207 | int c, cl; | |
8be6e1b1 | 2208 | unsigned long flags; |
3fbfbf7a PM |
2209 | struct rcu_head *list; |
2210 | struct rcu_head *next; | |
2211 | struct rcu_head **tail; | |
2212 | struct rcu_data *rdp = arg; | |
2213 | ||
2214 | /* Each pass through this loop invokes one batch of callbacks */ | |
2215 | for (;;) { | |
fbce7497 PM |
2216 | /* Wait for callbacks. */ |
2217 | if (rdp->nocb_leader == rdp) | |
2218 | nocb_leader_wait(rdp); | |
2219 | else | |
2220 | nocb_follower_wait(rdp); | |
2221 | ||
2222 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
8be6e1b1 PM |
2223 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2224 | list = rdp->nocb_follower_head; | |
2225 | rdp->nocb_follower_head = NULL; | |
2226 | tail = rdp->nocb_follower_tail; | |
2227 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; | |
2228 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
9213784b PM |
2229 | if (WARN_ON_ONCE(!list)) |
2230 | continue; | |
88d1bead | 2231 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeNonEmpty")); |
3fbfbf7a PM |
2232 | |
2233 | /* Each pass through the following loop invokes a callback. */ | |
88d1bead | 2234 | trace_rcu_batch_start(rcu_state.name, |
41050a00 | 2235 | atomic_long_read(&rdp->nocb_q_count_lazy), |
903ee83d | 2236 | rcu_get_n_cbs_nocb_cpu(rdp), -1); |
3fbfbf7a PM |
2237 | c = cl = 0; |
2238 | while (list) { | |
2239 | next = list->next; | |
2240 | /* Wait for enqueuing to complete, if needed. */ | |
2241 | while (next == NULL && &list->next != tail) { | |
88d1bead | 2242 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
69a79bb1 | 2243 | TPS("WaitQueue")); |
3fbfbf7a | 2244 | schedule_timeout_interruptible(1); |
88d1bead | 2245 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, |
69a79bb1 | 2246 | TPS("WokeQueue")); |
3fbfbf7a PM |
2247 | next = list->next; |
2248 | } | |
2249 | debug_rcu_head_unqueue(list); | |
2250 | local_bh_disable(); | |
88d1bead | 2251 | if (__rcu_reclaim(rcu_state.name, list)) |
3fbfbf7a PM |
2252 | cl++; |
2253 | c++; | |
2254 | local_bh_enable(); | |
cee43939 | 2255 | cond_resched_tasks_rcu_qs(); |
3fbfbf7a PM |
2256 | list = next; |
2257 | } | |
88d1bead | 2258 | trace_rcu_batch_end(rcu_state.name, c, !!list, 0, 0, 1); |
41050a00 PM |
2259 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2260 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2261 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
3fbfbf7a PM |
2262 | } |
2263 | return 0; | |
2264 | } | |
2265 | ||
96d3fd0d | 2266 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2267 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2268 | { |
7d0ae808 | 2269 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2270 | } |
2271 | ||
2272 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2273 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2274 | { |
8be6e1b1 | 2275 | unsigned long flags; |
9fdd3bc9 PM |
2276 | int ndw; |
2277 | ||
8be6e1b1 PM |
2278 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2279 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { | |
2280 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
96d3fd0d | 2281 | return; |
8be6e1b1 | 2282 | } |
7d0ae808 | 2283 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2284 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
8be6e1b1 | 2285 | __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
88d1bead | 2286 | trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2287 | } |
2288 | ||
8be6e1b1 | 2289 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2290 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2291 | { |
fd30b717 KC |
2292 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2293 | ||
2294 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2295 | } |
2296 | ||
2297 | /* | |
2298 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2299 | * This means we do an inexact common-case check. Note that if | |
2300 | * we miss, ->nocb_timer will eventually clean things up. | |
2301 | */ | |
2302 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2303 | { | |
2304 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2305 | do_nocb_deferred_wakeup_common(rdp); | |
2306 | } | |
2307 | ||
f4579fc5 PM |
2308 | void __init rcu_init_nohz(void) |
2309 | { | |
2310 | int cpu; | |
ef126206 | 2311 | bool need_rcu_nocb_mask = false; |
f4579fc5 | 2312 | |
f4579fc5 PM |
2313 | #if defined(CONFIG_NO_HZ_FULL) |
2314 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2315 | need_rcu_nocb_mask = true; | |
2316 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2317 | ||
84b12b75 | 2318 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
2319 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2320 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2321 | return; | |
2322 | } | |
f4579fc5 | 2323 | } |
84b12b75 | 2324 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
2325 | return; |
2326 | ||
f4579fc5 PM |
2327 | #if defined(CONFIG_NO_HZ_FULL) |
2328 | if (tick_nohz_full_running) | |
2329 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2330 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2331 | ||
2332 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 2333 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
2334 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
2335 | rcu_nocb_mask); | |
2336 | } | |
3016611e PM |
2337 | if (cpumask_empty(rcu_nocb_mask)) |
2338 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
2339 | else | |
2340 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
2341 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2342 | if (rcu_nocb_poll) |
2343 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2344 | ||
b97d23c5 PM |
2345 | for_each_cpu(cpu, rcu_nocb_mask) |
2346 | init_nocb_callback_list(per_cpu_ptr(&rcu_data, cpu)); | |
2347 | rcu_organize_nocb_kthreads(); | |
96d3fd0d PM |
2348 | } |
2349 | ||
3fbfbf7a PM |
2350 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2351 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2352 | { | |
2353 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2354 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2355 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
8be6e1b1 | 2356 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 2357 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
2358 | } |
2359 | ||
35ce7f29 PM |
2360 | /* |
2361 | * If the specified CPU is a no-CBs CPU that does not already have its | |
0ae86a27 PM |
2362 | * rcuo kthread, spawn it. If the CPUs are brought online out of order, |
2363 | * this can require re-organizing the leader-follower relationships. | |
35ce7f29 | 2364 | */ |
4580b054 | 2365 | static void rcu_spawn_one_nocb_kthread(int cpu) |
35ce7f29 PM |
2366 | { |
2367 | struct rcu_data *rdp; | |
2368 | struct rcu_data *rdp_last; | |
2369 | struct rcu_data *rdp_old_leader; | |
da1df50d | 2370 | struct rcu_data *rdp_spawn = per_cpu_ptr(&rcu_data, cpu); |
35ce7f29 PM |
2371 | struct task_struct *t; |
2372 | ||
2373 | /* | |
2374 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2375 | * then nothing to do. | |
2376 | */ | |
2377 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2378 | return; | |
2379 | ||
2380 | /* If we didn't spawn the leader first, reorganize! */ | |
2381 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2382 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2383 | rdp_last = NULL; | |
2384 | rdp = rdp_old_leader; | |
2385 | do { | |
2386 | rdp->nocb_leader = rdp_spawn; | |
2387 | if (rdp_last && rdp != rdp_spawn) | |
2388 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2389 | if (rdp == rdp_spawn) { |
2390 | rdp = rdp->nocb_next_follower; | |
2391 | } else { | |
2392 | rdp_last = rdp; | |
2393 | rdp = rdp->nocb_next_follower; | |
2394 | rdp_last->nocb_next_follower = NULL; | |
2395 | } | |
35ce7f29 PM |
2396 | } while (rdp); |
2397 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2398 | } | |
2399 | ||
0ae86a27 | 2400 | /* Spawn the kthread for this CPU. */ |
35ce7f29 | 2401 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, |
4580b054 | 2402 | "rcuo%c/%d", rcu_state.abbr, cpu); |
9213784b PM |
2403 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start rcuo kthread, OOM is now expected behavior\n", __func__)) |
2404 | return; | |
7d0ae808 | 2405 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2406 | } |
2407 | ||
2408 | /* | |
2409 | * If the specified CPU is a no-CBs CPU that does not already have its | |
ad368d15 | 2410 | * rcuo kthread, spawn it. |
35ce7f29 | 2411 | */ |
ad368d15 | 2412 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 | 2413 | { |
35ce7f29 | 2414 | if (rcu_scheduler_fully_active) |
b97d23c5 | 2415 | rcu_spawn_one_nocb_kthread(cpu); |
35ce7f29 PM |
2416 | } |
2417 | ||
2418 | /* | |
2419 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2420 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2421 | * non-boot CPUs come online -- if this changes, we will need to add | |
2422 | * some mutual exclusion. | |
2423 | */ | |
2424 | static void __init rcu_spawn_nocb_kthreads(void) | |
2425 | { | |
2426 | int cpu; | |
2427 | ||
2428 | for_each_online_cpu(cpu) | |
ad368d15 | 2429 | rcu_spawn_cpu_nocb_kthread(cpu); |
35ce7f29 PM |
2430 | } |
2431 | ||
fbce7497 PM |
2432 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2433 | static int rcu_nocb_leader_stride = -1; | |
2434 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2435 | ||
2436 | /* | |
35ce7f29 | 2437 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2438 | */ |
4580b054 | 2439 | static void __init rcu_organize_nocb_kthreads(void) |
3fbfbf7a PM |
2440 | { |
2441 | int cpu; | |
fbce7497 PM |
2442 | int ls = rcu_nocb_leader_stride; |
2443 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2444 | struct rcu_data *rdp; |
fbce7497 PM |
2445 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2446 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2447 | |
84b12b75 | 2448 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2449 | return; |
fbce7497 PM |
2450 | if (ls == -1) { |
2451 | ls = int_sqrt(nr_cpu_ids); | |
2452 | rcu_nocb_leader_stride = ls; | |
2453 | } | |
2454 | ||
2455 | /* | |
9831ce3b PM |
2456 | * Each pass through this loop sets up one rcu_data structure. |
2457 | * Should the corresponding CPU come online in the future, then | |
2458 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2459 | */ |
3fbfbf7a | 2460 | for_each_cpu(cpu, rcu_nocb_mask) { |
da1df50d | 2461 | rdp = per_cpu_ptr(&rcu_data, cpu); |
fbce7497 PM |
2462 | if (rdp->cpu >= nl) { |
2463 | /* New leader, set up for followers & next leader. */ | |
2464 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2465 | rdp->nocb_leader = rdp; | |
2466 | rdp_leader = rdp; | |
2467 | } else { | |
2468 | /* Another follower, link to previous leader. */ | |
2469 | rdp->nocb_leader = rdp_leader; | |
2470 | rdp_prev->nocb_next_follower = rdp; | |
2471 | } | |
2472 | rdp_prev = rdp; | |
3fbfbf7a PM |
2473 | } |
2474 | } | |
2475 | ||
2476 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2477 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2478 | { |
22c2f669 | 2479 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2480 | return false; |
22c2f669 | 2481 | |
34404ca8 | 2482 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2483 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2484 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2485 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2486 | atomic_long_set(&rdp->nocb_q_count, | |
2487 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2488 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2489 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2490 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2491 | } |
15fecf89 | 2492 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2493 | return true; |
3fbfbf7a PM |
2494 | } |
2495 | ||
5ab7ab83 PM |
2496 | /* |
2497 | * Bind the current task to the offloaded CPUs. If there are no offloaded | |
2498 | * CPUs, leave the task unbound. Splat if the bind attempt fails. | |
2499 | */ | |
2500 | void rcu_bind_current_to_nocb(void) | |
2501 | { | |
2502 | if (cpumask_available(rcu_nocb_mask) && cpumask_weight(rcu_nocb_mask)) | |
2503 | WARN_ON(sched_setaffinity(current->pid, rcu_nocb_mask)); | |
2504 | } | |
2505 | EXPORT_SYMBOL_GPL(rcu_bind_current_to_nocb); | |
2506 | ||
903ee83d PM |
2507 | /* |
2508 | * Return the number of RCU callbacks still queued from the specified | |
2509 | * CPU, which must be a nocbs CPU. | |
2510 | */ | |
2511 | static unsigned long rcu_get_n_cbs_nocb_cpu(struct rcu_data *rdp) | |
2512 | { | |
2513 | return atomic_long_read(&rdp->nocb_q_count); | |
2514 | } | |
2515 | ||
34ed6246 PM |
2516 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2517 | ||
4580b054 | 2518 | static bool rcu_nocb_cpu_needs_barrier(int cpu) |
d7e29933 PM |
2519 | { |
2520 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2521 | return false; | |
2522 | } | |
2523 | ||
abedf8e2 | 2524 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2525 | { |
3fbfbf7a PM |
2526 | } |
2527 | ||
abedf8e2 | 2528 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2529 | { |
2530 | return NULL; | |
2531 | } | |
2532 | ||
dae6e64d PM |
2533 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2534 | { | |
2535 | } | |
3fbfbf7a | 2536 | |
3fbfbf7a | 2537 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2538 | bool lazy, unsigned long flags) |
3fbfbf7a | 2539 | { |
4afc7e26 | 2540 | return false; |
3fbfbf7a PM |
2541 | } |
2542 | ||
b1a2d79f | 2543 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2544 | struct rcu_data *rdp, |
2545 | unsigned long flags) | |
3fbfbf7a | 2546 | { |
f4aa84ba | 2547 | return false; |
3fbfbf7a PM |
2548 | } |
2549 | ||
3fbfbf7a PM |
2550 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2551 | { | |
2552 | } | |
2553 | ||
9fdd3bc9 | 2554 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2555 | { |
2556 | return false; | |
2557 | } | |
2558 | ||
2559 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2560 | { | |
2561 | } | |
2562 | ||
ad368d15 | 2563 | static void rcu_spawn_cpu_nocb_kthread(int cpu) |
35ce7f29 PM |
2564 | { |
2565 | } | |
2566 | ||
2567 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2568 | { |
2569 | } | |
2570 | ||
34ed6246 | 2571 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2572 | { |
34ed6246 | 2573 | return false; |
3fbfbf7a PM |
2574 | } |
2575 | ||
903ee83d PM |
2576 | static unsigned long rcu_get_n_cbs_nocb_cpu(struct rcu_data *rdp) |
2577 | { | |
2578 | return 0; | |
2579 | } | |
2580 | ||
3fbfbf7a | 2581 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ |
65d798f0 | 2582 | |
a096932f PM |
2583 | /* |
2584 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2585 | * grace-period kthread will do force_quiescent_state() processing? | |
2586 | * The idea is to avoid waking up RCU core processing on such a | |
2587 | * CPU unless the grace period has extended for too long. | |
2588 | * | |
2589 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2590 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f | 2591 | */ |
4580b054 | 2592 | static bool rcu_nohz_full_cpu(void) |
a096932f PM |
2593 | { |
2594 | #ifdef CONFIG_NO_HZ_FULL | |
2595 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
de8e8730 | 2596 | (!rcu_gp_in_progress() || |
4580b054 | 2597 | ULONG_CMP_LT(jiffies, READ_ONCE(rcu_state.gp_start) + HZ))) |
5ce035fb | 2598 | return true; |
a096932f | 2599 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2600 | return false; |
a096932f | 2601 | } |
5057f55e PM |
2602 | |
2603 | /* | |
265f5f28 | 2604 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2605 | */ |
2606 | static void rcu_bind_gp_kthread(void) | |
2607 | { | |
c0f489d2 | 2608 | if (!tick_nohz_full_enabled()) |
5057f55e | 2609 | return; |
de201559 | 2610 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2611 | } |
176f8f7a PM |
2612 | |
2613 | /* Record the current task on dyntick-idle entry. */ | |
2614 | static void rcu_dynticks_task_enter(void) | |
2615 | { | |
2616 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2617 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2618 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2619 | } | |
2620 | ||
2621 | /* Record no current task on dyntick-idle exit. */ | |
2622 | static void rcu_dynticks_task_exit(void) | |
2623 | { | |
2624 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2625 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2626 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2627 | } |