Commit | Line | Data |
---|---|---|
f41d911f PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion (tree-based version) | |
3 | * Internal non-public definitions that provide either classic | |
6cc68793 | 4 | * or preemptible semantics. |
f41d911f PM |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
87de1cfd PM |
17 | * along with this program; if not, you can access it online at |
18 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
f41d911f PM |
19 | * |
20 | * Copyright Red Hat, 2009 | |
21 | * Copyright IBM Corporation, 2009 | |
22 | * | |
23 | * Author: Ingo Molnar <mingo@elte.hu> | |
24 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> | |
25 | */ | |
26 | ||
d9a3da06 | 27 | #include <linux/delay.h> |
3fbfbf7a | 28 | #include <linux/gfp.h> |
b626c1b6 | 29 | #include <linux/oom.h> |
b17b0153 | 30 | #include <linux/sched/debug.h> |
62ab7072 | 31 | #include <linux/smpboot.h> |
78634061 | 32 | #include <linux/sched/isolation.h> |
ae7e81c0 | 33 | #include <uapi/linux/sched/types.h> |
4102adab | 34 | #include "../time/tick-internal.h" |
f41d911f | 35 | |
5b61b0ba | 36 | #ifdef CONFIG_RCU_BOOST |
61cfd097 | 37 | |
abaa93d9 | 38 | #include "../locking/rtmutex_common.h" |
21871d7e | 39 | |
61cfd097 PM |
40 | /* |
41 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
42 | * handle all flavors of RCU. | |
43 | */ | |
44 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
45 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
46 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
47 | DEFINE_PER_CPU(char, rcu_cpu_has_work); | |
48 | ||
727b705b PM |
49 | #else /* #ifdef CONFIG_RCU_BOOST */ |
50 | ||
51 | /* | |
52 | * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST, | |
53 | * all uses are in dead code. Provide a definition to keep the compiler | |
54 | * happy, but add WARN_ON_ONCE() to complain if used in the wrong place. | |
55 | * This probably needs to be excluded from -rt builds. | |
56 | */ | |
57 | #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; }) | |
b8869781 | 58 | #define rt_mutex_futex_unlock(x) WARN_ON_ONCE(1) |
727b705b PM |
59 | |
60 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ | |
5b61b0ba | 61 | |
3fbfbf7a PM |
62 | #ifdef CONFIG_RCU_NOCB_CPU |
63 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
1b0048a4 | 64 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
65 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
66 | ||
26845c28 PM |
67 | /* |
68 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 69 | * messages about anything out of the ordinary. |
26845c28 PM |
70 | */ |
71 | static void __init rcu_bootup_announce_oddness(void) | |
72 | { | |
ab6f5bd6 | 73 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 74 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
75 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
76 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
ab6f5bd6 | 77 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", |
05c5df31 | 78 | RCU_FANOUT); |
7fa27001 | 79 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
80 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
81 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
82 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 83 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 84 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
85 | if (RCU_NUM_LVLS >= 4) |
86 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 87 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 88 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
89 | RCU_FANOUT_LEAF); |
90 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
9a5739d7 | 91 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 92 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 93 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b PM |
94 | #ifdef CONFIG_RCU_BOOST |
95 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
96 | #endif | |
97 | if (blimit != DEFAULT_RCU_BLIMIT) | |
98 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
99 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
100 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
101 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
102 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
103 | if (jiffies_till_first_fqs != ULONG_MAX) | |
104 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
105 | if (jiffies_till_next_fqs != ULONG_MAX) | |
106 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
107 | if (rcu_kick_kthreads) | |
108 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
109 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
110 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 111 | if (gp_preinit_delay) |
17c7798b | 112 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 113 | if (gp_init_delay) |
17c7798b | 114 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 115 | if (gp_cleanup_delay) |
17c7798b PM |
116 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
117 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) | |
118 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 119 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
120 | } |
121 | ||
28f6569a | 122 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 123 | |
a41bfeb2 | 124 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
b28a7c01 | 125 | static struct rcu_state *const rcu_state_p = &rcu_preempt_state; |
2927a689 | 126 | static struct rcu_data __percpu *const rcu_data_p = &rcu_preempt_data; |
f41d911f | 127 | |
d19fb8d1 PM |
128 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
129 | bool wake); | |
d9a3da06 | 130 | |
f41d911f PM |
131 | /* |
132 | * Tell them what RCU they are running. | |
133 | */ | |
0e0fc1c2 | 134 | static void __init rcu_bootup_announce(void) |
f41d911f | 135 | { |
efc151c3 | 136 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 137 | rcu_bootup_announce_oddness(); |
f41d911f PM |
138 | } |
139 | ||
8203d6d0 PM |
140 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
141 | #define RCU_GP_TASKS 0x8 | |
142 | #define RCU_EXP_TASKS 0x4 | |
143 | #define RCU_GP_BLKD 0x2 | |
144 | #define RCU_EXP_BLKD 0x1 | |
145 | ||
146 | /* | |
147 | * Queues a task preempted within an RCU-preempt read-side critical | |
148 | * section into the appropriate location within the ->blkd_tasks list, | |
149 | * depending on the states of any ongoing normal and expedited grace | |
150 | * periods. The ->gp_tasks pointer indicates which element the normal | |
151 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
152 | * indicates which element the expedited grace period is waiting on (again, | |
153 | * NULL if none). If a grace period is waiting on a given element in the | |
154 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
155 | * adding a task to the tail of the list blocks any grace period that is | |
156 | * already waiting on one of the elements. In contrast, adding a task | |
157 | * to the head of the list won't block any grace period that is already | |
158 | * waiting on one of the elements. | |
159 | * | |
160 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
161 | * period wait for a task that is not strictly speaking blocking it. | |
162 | * Given the choice, we needlessly block a normal grace period rather than | |
163 | * blocking an expedited grace period. | |
164 | * | |
165 | * Note that an endless sequence of expedited grace periods still cannot | |
166 | * indefinitely postpone a normal grace period. Eventually, all of the | |
167 | * fixed number of preempted tasks blocking the normal grace period that are | |
168 | * not also blocking the expedited grace period will resume and complete | |
169 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
170 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
171 | * the corresponding expedited grace period will also be the end of the | |
172 | * normal grace period. | |
173 | */ | |
46a5d164 PM |
174 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
175 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
176 | { |
177 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
178 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
179 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
180 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
181 | struct task_struct *t = current; | |
182 | ||
a32e01ee | 183 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 184 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 185 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
186 | /* RCU better not be waiting on newly onlined CPUs! */ |
187 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
188 | rdp->grpmask); | |
ea9b0c8a | 189 | |
8203d6d0 PM |
190 | /* |
191 | * Decide where to queue the newly blocked task. In theory, | |
192 | * this could be an if-statement. In practice, when I tried | |
193 | * that, it was quite messy. | |
194 | */ | |
195 | switch (blkd_state) { | |
196 | case 0: | |
197 | case RCU_EXP_TASKS: | |
198 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
199 | case RCU_GP_TASKS: | |
200 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
201 | ||
202 | /* | |
203 | * Blocking neither GP, or first task blocking the normal | |
204 | * GP but not blocking the already-waiting expedited GP. | |
205 | * Queue at the head of the list to avoid unnecessarily | |
206 | * blocking the already-waiting GPs. | |
207 | */ | |
208 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
209 | break; | |
210 | ||
211 | case RCU_EXP_BLKD: | |
212 | case RCU_GP_BLKD: | |
213 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
214 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
215 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
216 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
217 | ||
218 | /* | |
219 | * First task arriving that blocks either GP, or first task | |
220 | * arriving that blocks the expedited GP (with the normal | |
221 | * GP already waiting), or a task arriving that blocks | |
222 | * both GPs with both GPs already waiting. Queue at the | |
223 | * tail of the list to avoid any GP waiting on any of the | |
224 | * already queued tasks that are not blocking it. | |
225 | */ | |
226 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
227 | break; | |
228 | ||
229 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
230 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
231 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
232 | ||
233 | /* | |
234 | * Second or subsequent task blocking the expedited GP. | |
235 | * The task either does not block the normal GP, or is the | |
236 | * first task blocking the normal GP. Queue just after | |
237 | * the first task blocking the expedited GP. | |
238 | */ | |
239 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
240 | break; | |
241 | ||
242 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
243 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
244 | ||
245 | /* | |
246 | * Second or subsequent task blocking the normal GP. | |
247 | * The task does not block the expedited GP. Queue just | |
248 | * after the first task blocking the normal GP. | |
249 | */ | |
250 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
251 | break; | |
252 | ||
253 | default: | |
254 | ||
255 | /* Yet another exercise in excessive paranoia. */ | |
256 | WARN_ON_ONCE(1); | |
257 | break; | |
258 | } | |
259 | ||
260 | /* | |
261 | * We have now queued the task. If it was the first one to | |
262 | * block either grace period, update the ->gp_tasks and/or | |
263 | * ->exp_tasks pointers, respectively, to reference the newly | |
264 | * blocked tasks. | |
265 | */ | |
4bc8d555 | 266 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
8203d6d0 | 267 | rnp->gp_tasks = &t->rcu_node_entry; |
d43a5d32 | 268 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 269 | } |
8203d6d0 PM |
270 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
271 | rnp->exp_tasks = &t->rcu_node_entry; | |
2dee9404 PM |
272 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
273 | !(rnp->qsmask & rdp->grpmask)); | |
274 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
275 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 276 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
277 | |
278 | /* | |
279 | * Report the quiescent state for the expedited GP. This expedited | |
280 | * GP should not be able to end until we report, so there should be | |
281 | * no need to check for a subsequent expedited GP. (Though we are | |
282 | * still in a quiescent state in any case.) | |
283 | */ | |
284 | if (blkd_state & RCU_EXP_BLKD && | |
285 | t->rcu_read_unlock_special.b.exp_need_qs) { | |
286 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
287 | rcu_report_exp_rdp(rdp->rsp, rdp, true); | |
288 | } else { | |
289 | WARN_ON_ONCE(t->rcu_read_unlock_special.b.exp_need_qs); | |
290 | } | |
8203d6d0 PM |
291 | } |
292 | ||
f41d911f | 293 | /* |
6cc68793 | 294 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
295 | * that this just means that the task currently running on the CPU is |
296 | * not in a quiescent state. There might be any number of tasks blocked | |
297 | * while in an RCU read-side critical section. | |
25502a6c | 298 | * |
1d082fd0 PM |
299 | * As with the other rcu_*_qs() functions, callers to this function |
300 | * must disable preemption. | |
f41d911f | 301 | */ |
284a8c93 | 302 | static void rcu_preempt_qs(void) |
f41d911f | 303 | { |
ea9b0c8a | 304 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n"); |
5b74c458 | 305 | if (__this_cpu_read(rcu_data_p->cpu_no_qs.s)) { |
284a8c93 | 306 | trace_rcu_grace_period(TPS("rcu_preempt"), |
477351f7 | 307 | __this_cpu_read(rcu_data_p->gp_seq), |
284a8c93 | 308 | TPS("cpuqs")); |
5b74c458 | 309 | __this_cpu_write(rcu_data_p->cpu_no_qs.b.norm, false); |
284a8c93 PM |
310 | barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */ |
311 | current->rcu_read_unlock_special.b.need_qs = false; | |
312 | } | |
f41d911f PM |
313 | } |
314 | ||
315 | /* | |
c3422bea PM |
316 | * We have entered the scheduler, and the current task might soon be |
317 | * context-switched away from. If this task is in an RCU read-side | |
318 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
319 | * record that fact, so we enqueue the task on the blkd_tasks list. |
320 | * The task will dequeue itself when it exits the outermost enclosing | |
321 | * RCU read-side critical section. Therefore, the current grace period | |
322 | * cannot be permitted to complete until the blkd_tasks list entries | |
323 | * predating the current grace period drain, in other words, until | |
324 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 325 | * |
46a5d164 | 326 | * Caller must disable interrupts. |
f41d911f | 327 | */ |
5b72f964 | 328 | static void rcu_preempt_note_context_switch(bool preempt) |
f41d911f PM |
329 | { |
330 | struct task_struct *t = current; | |
f41d911f PM |
331 | struct rcu_data *rdp; |
332 | struct rcu_node *rnp; | |
333 | ||
b04db8e1 | 334 | lockdep_assert_irqs_disabled(); |
5b72f964 | 335 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 336 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 337 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
338 | |
339 | /* Possibly blocking in an RCU read-side critical section. */ | |
e63c887c | 340 | rdp = this_cpu_ptr(rcu_state_p->rda); |
f41d911f | 341 | rnp = rdp->mynode; |
46a5d164 | 342 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 343 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 344 | t->rcu_blocked_node = rnp; |
f41d911f PM |
345 | |
346 | /* | |
8203d6d0 PM |
347 | * Verify the CPU's sanity, trace the preemption, and |
348 | * then queue the task as required based on the states | |
349 | * of any ongoing and expedited grace periods. | |
f41d911f | 350 | */ |
0aa04b05 | 351 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 352 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
d4c08f2a PM |
353 | trace_rcu_preempt_task(rdp->rsp->name, |
354 | t->pid, | |
355 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
356 | ? rnp->gp_seq |
357 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 358 | rcu_preempt_ctxt_queue(rnp, rdp); |
10f39bb1 | 359 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 360 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
361 | |
362 | /* | |
363 | * Complete exit from RCU read-side critical section on | |
364 | * behalf of preempted instance of __rcu_read_unlock(). | |
365 | */ | |
366 | rcu_read_unlock_special(t); | |
f41d911f PM |
367 | } |
368 | ||
369 | /* | |
370 | * Either we were not in an RCU read-side critical section to | |
371 | * begin with, or we have now recorded that critical section | |
372 | * globally. Either way, we can now note a quiescent state | |
373 | * for this CPU. Again, if we were in an RCU read-side critical | |
374 | * section, and if that critical section was blocking the current | |
375 | * grace period, then the fact that the task has been enqueued | |
376 | * means that we continue to block the current grace period. | |
377 | */ | |
284a8c93 | 378 | rcu_preempt_qs(); |
f41d911f PM |
379 | } |
380 | ||
fc2219d4 PM |
381 | /* |
382 | * Check for preempted RCU readers blocking the current grace period | |
383 | * for the specified rcu_node structure. If the caller needs a reliable | |
384 | * answer, it must hold the rcu_node's ->lock. | |
385 | */ | |
27f4d280 | 386 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 387 | { |
12f5f524 | 388 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
389 | } |
390 | ||
0e5da22e PM |
391 | /* |
392 | * Preemptible RCU implementation for rcu_read_lock(). | |
393 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
394 | * if we block. | |
395 | */ | |
396 | void __rcu_read_lock(void) | |
397 | { | |
398 | current->rcu_read_lock_nesting++; | |
399 | barrier(); /* critical section after entry code. */ | |
400 | } | |
401 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
402 | ||
403 | /* | |
404 | * Preemptible RCU implementation for rcu_read_unlock(). | |
405 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
406 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
407 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
408 | * in an RCU read-side critical section and other special cases. | |
409 | */ | |
410 | void __rcu_read_unlock(void) | |
411 | { | |
412 | struct task_struct *t = current; | |
413 | ||
414 | if (t->rcu_read_lock_nesting != 1) { | |
415 | --t->rcu_read_lock_nesting; | |
416 | } else { | |
417 | barrier(); /* critical section before exit code. */ | |
418 | t->rcu_read_lock_nesting = INT_MIN; | |
419 | barrier(); /* assign before ->rcu_read_unlock_special load */ | |
420 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) | |
421 | rcu_read_unlock_special(t); | |
422 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
423 | t->rcu_read_lock_nesting = 0; | |
424 | } | |
425 | #ifdef CONFIG_PROVE_LOCKING | |
426 | { | |
427 | int rrln = READ_ONCE(t->rcu_read_lock_nesting); | |
428 | ||
429 | WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2); | |
430 | } | |
431 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | |
432 | } | |
433 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
434 | ||
12f5f524 PM |
435 | /* |
436 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
437 | * returning NULL if at the end of the list. | |
438 | */ | |
439 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
440 | struct rcu_node *rnp) | |
441 | { | |
442 | struct list_head *np; | |
443 | ||
444 | np = t->rcu_node_entry.next; | |
445 | if (np == &rnp->blkd_tasks) | |
446 | np = NULL; | |
447 | return np; | |
448 | } | |
449 | ||
8af3a5e7 PM |
450 | /* |
451 | * Return true if the specified rcu_node structure has tasks that were | |
452 | * preempted within an RCU read-side critical section. | |
453 | */ | |
454 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
455 | { | |
456 | return !list_empty(&rnp->blkd_tasks); | |
457 | } | |
458 | ||
b668c9cf PM |
459 | /* |
460 | * Handle special cases during rcu_read_unlock(), such as needing to | |
461 | * notify RCU core processing or task having blocked during the RCU | |
462 | * read-side critical section. | |
463 | */ | |
2a3fa843 | 464 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f | 465 | { |
b6a932d1 PM |
466 | bool empty_exp; |
467 | bool empty_norm; | |
468 | bool empty_exp_now; | |
f41d911f | 469 | unsigned long flags; |
12f5f524 | 470 | struct list_head *np; |
abaa93d9 | 471 | bool drop_boost_mutex = false; |
8203d6d0 | 472 | struct rcu_data *rdp; |
f41d911f | 473 | struct rcu_node *rnp; |
1d082fd0 | 474 | union rcu_special special; |
f41d911f PM |
475 | |
476 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
477 | if (in_nmi()) | |
478 | return; | |
479 | ||
480 | local_irq_save(flags); | |
481 | ||
482 | /* | |
8203d6d0 PM |
483 | * If RCU core is waiting for this CPU to exit its critical section, |
484 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 485 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
486 | */ |
487 | special = t->rcu_read_unlock_special; | |
1d082fd0 | 488 | if (special.b.need_qs) { |
284a8c93 | 489 | rcu_preempt_qs(); |
c0135d07 | 490 | t->rcu_read_unlock_special.b.need_qs = false; |
1d082fd0 | 491 | if (!t->rcu_read_unlock_special.s) { |
79a62f95 LJ |
492 | local_irq_restore(flags); |
493 | return; | |
494 | } | |
f41d911f PM |
495 | } |
496 | ||
8203d6d0 PM |
497 | /* |
498 | * Respond to a request for an expedited grace period, but only if | |
499 | * we were not preempted, meaning that we were running on the same | |
500 | * CPU throughout. If we were preempted, the exp_need_qs flag | |
501 | * would have been cleared at the time of the first preemption, | |
502 | * and the quiescent state would be reported when we were dequeued. | |
503 | */ | |
504 | if (special.b.exp_need_qs) { | |
505 | WARN_ON_ONCE(special.b.blocked); | |
506 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
507 | rdp = this_cpu_ptr(rcu_state_p->rda); | |
508 | rcu_report_exp_rdp(rcu_state_p, rdp, true); | |
509 | if (!t->rcu_read_unlock_special.s) { | |
510 | local_irq_restore(flags); | |
511 | return; | |
512 | } | |
513 | } | |
514 | ||
79a62f95 | 515 | /* Hardware IRQ handlers cannot block, complain if they get here. */ |
d24209bb PM |
516 | if (in_irq() || in_serving_softirq()) { |
517 | lockdep_rcu_suspicious(__FILE__, __LINE__, | |
518 | "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n"); | |
8203d6d0 | 519 | pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n", |
d24209bb PM |
520 | t->rcu_read_unlock_special.s, |
521 | t->rcu_read_unlock_special.b.blocked, | |
8203d6d0 | 522 | t->rcu_read_unlock_special.b.exp_need_qs, |
d24209bb | 523 | t->rcu_read_unlock_special.b.need_qs); |
f41d911f PM |
524 | local_irq_restore(flags); |
525 | return; | |
526 | } | |
527 | ||
528 | /* Clean up if blocked during RCU read-side critical section. */ | |
1d082fd0 PM |
529 | if (special.b.blocked) { |
530 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 531 | |
dd5d19ba | 532 | /* |
0a0ba1c9 | 533 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
534 | * now remains queued on the rcu_node corresponding to the |
535 | * CPU it first blocked on, so there is no longer any need | |
536 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 537 | */ |
8ba9153b PM |
538 | rnp = t->rcu_blocked_node; |
539 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
540 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 541 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 542 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 543 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 544 | (!empty_norm || rnp->qsmask)); |
8203d6d0 | 545 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 546 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 547 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 548 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 549 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 550 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 551 | rnp->gp_seq, t->pid); |
12f5f524 PM |
552 | if (&t->rcu_node_entry == rnp->gp_tasks) |
553 | rnp->gp_tasks = np; | |
554 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
555 | rnp->exp_tasks = np; | |
727b705b | 556 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
557 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
558 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
559 | if (&t->rcu_node_entry == rnp->boost_tasks) |
560 | rnp->boost_tasks = np; | |
727b705b | 561 | } |
f41d911f PM |
562 | |
563 | /* | |
564 | * If this was the last task on the current list, and if | |
565 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
566 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
567 | * so we must take a snapshot of the expedited state. | |
f41d911f | 568 | */ |
8203d6d0 | 569 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 570 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 571 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 572 | rnp->gp_seq, |
d4c08f2a PM |
573 | 0, rnp->qsmask, |
574 | rnp->level, | |
575 | rnp->grplo, | |
576 | rnp->grphi, | |
577 | !!rnp->gp_tasks); | |
e63c887c | 578 | rcu_report_unblock_qs_rnp(rcu_state_p, rnp, flags); |
c701d5d9 | 579 | } else { |
67c583a7 | 580 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 581 | } |
d9a3da06 | 582 | |
27f4d280 | 583 | /* Unboost if we were boosted. */ |
727b705b | 584 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 585 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 586 | |
d9a3da06 PM |
587 | /* |
588 | * If this was the last task on the expedited lists, | |
589 | * then we need to report up the rcu_node hierarchy. | |
590 | */ | |
389abd48 | 591 | if (!empty_exp && empty_exp_now) |
e63c887c | 592 | rcu_report_exp_rnp(rcu_state_p, rnp, true); |
b668c9cf PM |
593 | } else { |
594 | local_irq_restore(flags); | |
f41d911f | 595 | } |
f41d911f PM |
596 | } |
597 | ||
1ed509a2 PM |
598 | /* |
599 | * Dump detailed information for all tasks blocking the current RCU | |
600 | * grace period on the specified rcu_node structure. | |
601 | */ | |
602 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
603 | { | |
604 | unsigned long flags; | |
1ed509a2 PM |
605 | struct task_struct *t; |
606 | ||
6cf10081 | 607 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 608 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 609 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
610 | return; |
611 | } | |
82efed06 | 612 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 613 | struct task_struct, rcu_node_entry); |
3caa973b TH |
614 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
615 | /* | |
616 | * We could be printing a lot while holding a spinlock. | |
617 | * Avoid triggering hard lockup. | |
618 | */ | |
619 | touch_nmi_watchdog(); | |
12f5f524 | 620 | sched_show_task(t); |
3caa973b | 621 | } |
67c583a7 | 622 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
623 | } |
624 | ||
625 | /* | |
626 | * Dump detailed information for all tasks blocking the current RCU | |
627 | * grace period. | |
628 | */ | |
629 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
630 | { | |
631 | struct rcu_node *rnp = rcu_get_root(rsp); | |
632 | ||
633 | rcu_print_detail_task_stall_rnp(rnp); | |
634 | rcu_for_each_leaf_node(rsp, rnp) | |
635 | rcu_print_detail_task_stall_rnp(rnp); | |
636 | } | |
637 | ||
a858af28 PM |
638 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
639 | { | |
efc151c3 | 640 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
641 | rnp->level, rnp->grplo, rnp->grphi); |
642 | } | |
643 | ||
644 | static void rcu_print_task_stall_end(void) | |
645 | { | |
efc151c3 | 646 | pr_cont("\n"); |
a858af28 PM |
647 | } |
648 | ||
f41d911f PM |
649 | /* |
650 | * Scan the current list of tasks blocked within RCU read-side critical | |
651 | * sections, printing out the tid of each. | |
652 | */ | |
9bc8b558 | 653 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 654 | { |
f41d911f | 655 | struct task_struct *t; |
9bc8b558 | 656 | int ndetected = 0; |
f41d911f | 657 | |
27f4d280 | 658 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 659 | return 0; |
a858af28 | 660 | rcu_print_task_stall_begin(rnp); |
82efed06 | 661 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 662 | struct task_struct, rcu_node_entry); |
9bc8b558 | 663 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 664 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
665 | ndetected++; |
666 | } | |
a858af28 | 667 | rcu_print_task_stall_end(); |
9bc8b558 | 668 | return ndetected; |
f41d911f PM |
669 | } |
670 | ||
74611ecb PM |
671 | /* |
672 | * Scan the current list of tasks blocked within RCU read-side critical | |
673 | * sections, printing out the tid of each that is blocking the current | |
674 | * expedited grace period. | |
675 | */ | |
676 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
677 | { | |
678 | struct task_struct *t; | |
679 | int ndetected = 0; | |
680 | ||
681 | if (!rnp->exp_tasks) | |
682 | return 0; | |
683 | t = list_entry(rnp->exp_tasks->prev, | |
684 | struct task_struct, rcu_node_entry); | |
685 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
686 | pr_cont(" P%d", t->pid); | |
687 | ndetected++; | |
688 | } | |
689 | return ndetected; | |
690 | } | |
691 | ||
b0e165c0 PM |
692 | /* |
693 | * Check that the list of blocked tasks for the newly completed grace | |
694 | * period is in fact empty. It is a serious bug to complete a grace | |
695 | * period that still has RCU readers blocked! This function must be | |
ff3bb6f4 | 696 | * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock |
b0e165c0 | 697 | * must be held by the caller. |
12f5f524 PM |
698 | * |
699 | * Also, if there are blocked tasks on the list, they automatically | |
700 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
701 | */ |
702 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
703 | { | |
c5ebe66c PM |
704 | struct task_struct *t; |
705 | ||
ea9b0c8a | 706 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
4bc8d555 PM |
707 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
708 | dump_blkd_tasks(rnp, 10); | |
0b107d24 PM |
709 | if (rcu_preempt_has_tasks(rnp) && |
710 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
12f5f524 | 711 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
712 | t = container_of(rnp->gp_tasks, struct task_struct, |
713 | rcu_node_entry); | |
714 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 715 | rnp->gp_seq, t->pid); |
c5ebe66c | 716 | } |
28ecd580 | 717 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
718 | } |
719 | ||
f41d911f PM |
720 | /* |
721 | * Check for a quiescent state from the current CPU. When a task blocks, | |
722 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
723 | * which is checked elsewhere. | |
724 | * | |
725 | * Caller must disable hard irqs. | |
726 | */ | |
86aea0e6 | 727 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
728 | { |
729 | struct task_struct *t = current; | |
730 | ||
731 | if (t->rcu_read_lock_nesting == 0) { | |
284a8c93 | 732 | rcu_preempt_qs(); |
f41d911f PM |
733 | return; |
734 | } | |
10f39bb1 | 735 | if (t->rcu_read_lock_nesting > 0 && |
97c668b8 | 736 | __this_cpu_read(rcu_data_p->core_needs_qs) && |
5b74c458 | 737 | __this_cpu_read(rcu_data_p->cpu_no_qs.b.norm)) |
1d082fd0 | 738 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
739 | } |
740 | ||
a68a2bb2 PM |
741 | /** |
742 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
743 | * @head: structure to be used for queueing the RCU updates. | |
744 | * @func: actual callback function to be invoked after the grace period | |
745 | * | |
746 | * The callback function will be invoked some time after a full grace | |
747 | * period elapses, in other words after all pre-existing RCU read-side | |
748 | * critical sections have completed. However, the callback function | |
749 | * might well execute concurrently with RCU read-side critical sections | |
750 | * that started after call_rcu() was invoked. RCU read-side critical | |
751 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
752 | * and may be nested. | |
753 | * | |
754 | * Note that all CPUs must agree that the grace period extended beyond | |
755 | * all pre-existing RCU read-side critical section. On systems with more | |
756 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
757 | * guaranteed to have executed a full memory barrier since the end of its | |
758 | * last RCU read-side critical section whose beginning preceded the call | |
759 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
760 | * critical section that continues beyond the start of "func()" must have | |
761 | * executed a memory barrier after the call_rcu() but before the beginning | |
762 | * of that RCU read-side critical section. Note that these guarantees | |
763 | * include CPUs that are offline, idle, or executing in user mode, as | |
764 | * well as CPUs that are executing in the kernel. | |
765 | * | |
766 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
767 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
768 | * guaranteed to execute a full memory barrier during the time interval | |
769 | * between the call to call_rcu() and the invocation of "func()" -- even | |
770 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
771 | * more than one CPU). | |
f41d911f | 772 | */ |
b6a4ae76 | 773 | void call_rcu(struct rcu_head *head, rcu_callback_t func) |
f41d911f | 774 | { |
e63c887c | 775 | __call_rcu(head, func, rcu_state_p, -1, 0); |
f41d911f PM |
776 | } |
777 | EXPORT_SYMBOL_GPL(call_rcu); | |
778 | ||
6ebb237b PM |
779 | /** |
780 | * synchronize_rcu - wait until a grace period has elapsed. | |
781 | * | |
782 | * Control will return to the caller some time after a full grace | |
783 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
784 | * read-side critical sections have completed. Note, however, that |
785 | * upon return from synchronize_rcu(), the caller might well be executing | |
786 | * concurrently with new RCU read-side critical sections that began while | |
787 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
788 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 | 789 | * |
e28371c8 PM |
790 | * See the description of synchronize_sched() for more detailed |
791 | * information on memory-ordering guarantees. However, please note | |
792 | * that -only- the memory-ordering guarantees apply. For example, | |
793 | * synchronize_rcu() is -not- guaranteed to wait on things like code | |
794 | * protected by preempt_disable(), instead, synchronize_rcu() is -only- | |
795 | * guaranteed to wait on RCU read-side critical sections, that is, sections | |
796 | * of code protected by rcu_read_lock(). | |
6ebb237b PM |
797 | */ |
798 | void synchronize_rcu(void) | |
799 | { | |
f78f5b90 PM |
800 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
801 | lock_is_held(&rcu_lock_map) || | |
802 | lock_is_held(&rcu_sched_lock_map), | |
803 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
52d7e48b | 804 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
6ebb237b | 805 | return; |
5afff48b | 806 | if (rcu_gp_is_expedited()) |
3705b88d AM |
807 | synchronize_rcu_expedited(); |
808 | else | |
809 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
810 | } |
811 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
812 | ||
e74f4c45 PM |
813 | /** |
814 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
815 | * |
816 | * Note that this primitive does not necessarily wait for an RCU grace period | |
817 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
818 | * in the system, then rcu_barrier() is within its rights to return | |
819 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
820 | */ |
821 | void rcu_barrier(void) | |
822 | { | |
e63c887c | 823 | _rcu_barrier(rcu_state_p); |
e74f4c45 PM |
824 | } |
825 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
826 | ||
1eba8f84 | 827 | /* |
6cc68793 | 828 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
829 | */ |
830 | static void __init __rcu_init_preempt(void) | |
831 | { | |
a87f203e | 832 | rcu_init_one(rcu_state_p); |
1eba8f84 PM |
833 | } |
834 | ||
2439b696 PM |
835 | /* |
836 | * Check for a task exiting while in a preemptible-RCU read-side | |
837 | * critical section, clean up if so. No need to issue warnings, | |
838 | * as debug_check_no_locks_held() already does this if lockdep | |
839 | * is enabled. | |
840 | */ | |
841 | void exit_rcu(void) | |
842 | { | |
843 | struct task_struct *t = current; | |
844 | ||
845 | if (likely(list_empty(¤t->rcu_node_entry))) | |
846 | return; | |
847 | t->rcu_read_lock_nesting = 1; | |
848 | barrier(); | |
1d082fd0 | 849 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 PM |
850 | __rcu_read_unlock(); |
851 | } | |
852 | ||
4bc8d555 PM |
853 | /* |
854 | * Dump the blocked-tasks state, but limit the list dump to the | |
855 | * specified number of elements. | |
856 | */ | |
857 | static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck) | |
858 | { | |
859 | int i; | |
860 | struct list_head *lhp; | |
ff3cee39 | 861 | struct rcu_node *rnp1; |
4bc8d555 | 862 | |
ce11fae8 | 863 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 864 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 865 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
ff3cee39 PM |
866 | (long)rnp->gp_seq, (long)rnp->completedqs); |
867 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
868 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
869 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf PM |
870 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
871 | __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks); | |
872 | pr_info("%s: ->blkd_tasks", __func__); | |
4bc8d555 PM |
873 | i = 0; |
874 | list_for_each(lhp, &rnp->blkd_tasks) { | |
875 | pr_cont(" %p", lhp); | |
876 | if (++i >= 10) | |
877 | break; | |
878 | } | |
879 | pr_cont("\n"); | |
880 | } | |
881 | ||
28f6569a | 882 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 883 | |
b28a7c01 | 884 | static struct rcu_state *const rcu_state_p = &rcu_sched_state; |
27f4d280 | 885 | |
f41d911f PM |
886 | /* |
887 | * Tell them what RCU they are running. | |
888 | */ | |
0e0fc1c2 | 889 | static void __init rcu_bootup_announce(void) |
f41d911f | 890 | { |
efc151c3 | 891 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 892 | rcu_bootup_announce_oddness(); |
f41d911f PM |
893 | } |
894 | ||
cba6d0d6 PM |
895 | /* |
896 | * Because preemptible RCU does not exist, we never have to check for | |
897 | * CPUs being in quiescent states. | |
898 | */ | |
5b72f964 | 899 | static void rcu_preempt_note_context_switch(bool preempt) |
cba6d0d6 PM |
900 | { |
901 | } | |
902 | ||
fc2219d4 | 903 | /* |
6cc68793 | 904 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
905 | * RCU readers. |
906 | */ | |
27f4d280 | 907 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
908 | { |
909 | return 0; | |
910 | } | |
911 | ||
8af3a5e7 PM |
912 | /* |
913 | * Because there is no preemptible RCU, there can be no readers blocked. | |
914 | */ | |
915 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 916 | { |
8af3a5e7 | 917 | return false; |
b668c9cf PM |
918 | } |
919 | ||
1ed509a2 | 920 | /* |
6cc68793 | 921 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
922 | * tasks blocked within RCU read-side critical sections. |
923 | */ | |
924 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
925 | { | |
926 | } | |
927 | ||
f41d911f | 928 | /* |
6cc68793 | 929 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
930 | * tasks blocked within RCU read-side critical sections. |
931 | */ | |
9bc8b558 | 932 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 933 | { |
9bc8b558 | 934 | return 0; |
f41d911f PM |
935 | } |
936 | ||
74611ecb PM |
937 | /* |
938 | * Because preemptible RCU does not exist, we never have to check for | |
939 | * tasks blocked within RCU read-side critical sections that are | |
940 | * blocking the current expedited grace period. | |
941 | */ | |
942 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
943 | { | |
944 | return 0; | |
945 | } | |
946 | ||
b0e165c0 | 947 | /* |
6cc68793 | 948 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
949 | * so there is no need to check for blocked tasks. So check only for |
950 | * bogus qsmask values. | |
b0e165c0 PM |
951 | */ |
952 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
953 | { | |
49e29126 | 954 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
955 | } |
956 | ||
f41d911f | 957 | /* |
6cc68793 | 958 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
959 | * to check. |
960 | */ | |
86aea0e6 | 961 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
962 | { |
963 | } | |
964 | ||
e74f4c45 | 965 | /* |
6cc68793 | 966 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
967 | * another name for rcu_barrier_sched(). |
968 | */ | |
969 | void rcu_barrier(void) | |
970 | { | |
971 | rcu_barrier_sched(); | |
972 | } | |
973 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
974 | ||
1eba8f84 | 975 | /* |
6cc68793 | 976 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
977 | */ |
978 | static void __init __rcu_init_preempt(void) | |
979 | { | |
980 | } | |
981 | ||
2439b696 PM |
982 | /* |
983 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
984 | * while in preemptible RCU read-side critical sections. | |
985 | */ | |
986 | void exit_rcu(void) | |
987 | { | |
988 | } | |
989 | ||
4bc8d555 PM |
990 | /* |
991 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
992 | */ | |
993 | static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck) | |
994 | { | |
995 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
996 | } | |
997 | ||
28f6569a | 998 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 999 | |
27f4d280 PM |
1000 | #ifdef CONFIG_RCU_BOOST |
1001 | ||
5d01bbd1 TG |
1002 | static void rcu_wake_cond(struct task_struct *t, int status) |
1003 | { | |
1004 | /* | |
1005 | * If the thread is yielding, only wake it when this | |
1006 | * is invoked from idle | |
1007 | */ | |
1008 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1009 | wake_up_process(t); | |
1010 | } | |
1011 | ||
27f4d280 PM |
1012 | /* |
1013 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1014 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1015 | * ->blkd_tasks list. | |
1016 | * | |
1017 | * Note that irqs must be enabled: boosting the task can block. | |
1018 | * Returns 1 if there are more tasks needing to be boosted. | |
1019 | */ | |
1020 | static int rcu_boost(struct rcu_node *rnp) | |
1021 | { | |
1022 | unsigned long flags; | |
27f4d280 PM |
1023 | struct task_struct *t; |
1024 | struct list_head *tb; | |
1025 | ||
7d0ae808 PM |
1026 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1027 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1028 | return 0; /* Nothing left to boost. */ |
1029 | ||
2a67e741 | 1030 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1031 | |
1032 | /* | |
1033 | * Recheck under the lock: all tasks in need of boosting | |
1034 | * might exit their RCU read-side critical sections on their own. | |
1035 | */ | |
1036 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1037 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1038 | return 0; |
1039 | } | |
1040 | ||
1041 | /* | |
1042 | * Preferentially boost tasks blocking expedited grace periods. | |
1043 | * This cannot starve the normal grace periods because a second | |
1044 | * expedited grace period must boost all blocked tasks, including | |
1045 | * those blocking the pre-existing normal grace period. | |
1046 | */ | |
bec06785 | 1047 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1048 | tb = rnp->exp_tasks; |
bec06785 | 1049 | else |
27f4d280 PM |
1050 | tb = rnp->boost_tasks; |
1051 | ||
1052 | /* | |
1053 | * We boost task t by manufacturing an rt_mutex that appears to | |
1054 | * be held by task t. We leave a pointer to that rt_mutex where | |
1055 | * task t can find it, and task t will release the mutex when it | |
1056 | * exits its outermost RCU read-side critical section. Then | |
1057 | * simply acquiring this artificial rt_mutex will boost task | |
1058 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1059 | * | |
1060 | * Note that task t must acquire rnp->lock to remove itself from | |
1061 | * the ->blkd_tasks list, which it will do from exit() if from | |
1062 | * nowhere else. We therefore are guaranteed that task t will | |
1063 | * stay around at least until we drop rnp->lock. Note that | |
1064 | * rnp->lock also resolves races between our priority boosting | |
1065 | * and task t's exiting its outermost RCU read-side critical | |
1066 | * section. | |
1067 | */ | |
1068 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1069 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1070 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1071 | /* Lock only for side effect: boosts task t's priority. */ |
1072 | rt_mutex_lock(&rnp->boost_mtx); | |
1073 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1074 | |
7d0ae808 PM |
1075 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1076 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1077 | } |
1078 | ||
27f4d280 | 1079 | /* |
bc17ea10 | 1080 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1081 | */ |
1082 | static int rcu_boost_kthread(void *arg) | |
1083 | { | |
1084 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1085 | int spincnt = 0; | |
1086 | int more2boost; | |
1087 | ||
f7f7bac9 | 1088 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1089 | for (;;) { |
d71df90e | 1090 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1091 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1092 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1093 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1094 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1095 | more2boost = rcu_boost(rnp); |
1096 | if (more2boost) | |
1097 | spincnt++; | |
1098 | else | |
1099 | spincnt = 0; | |
1100 | if (spincnt > 10) { | |
5d01bbd1 | 1101 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1102 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1103 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1104 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1105 | spincnt = 0; |
1106 | } | |
1107 | } | |
1217ed1b | 1108 | /* NOTREACHED */ |
f7f7bac9 | 1109 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1110 | return 0; |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Check to see if it is time to start boosting RCU readers that are | |
1115 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1116 | * kthread to start boosting them. If there is an expedited grace | |
1117 | * period in progress, it is always time to boost. | |
1118 | * | |
b065a853 PM |
1119 | * The caller must hold rnp->lock, which this function releases. |
1120 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1121 | * about it going away. | |
27f4d280 | 1122 | */ |
1217ed1b | 1123 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1124 | __releases(rnp->lock) |
27f4d280 PM |
1125 | { |
1126 | struct task_struct *t; | |
1127 | ||
a32e01ee | 1128 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1129 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1130 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1131 | return; |
0ea1f2eb | 1132 | } |
27f4d280 PM |
1133 | if (rnp->exp_tasks != NULL || |
1134 | (rnp->gp_tasks != NULL && | |
1135 | rnp->boost_tasks == NULL && | |
1136 | rnp->qsmask == 0 && | |
1137 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1138 | if (rnp->exp_tasks == NULL) | |
1139 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1140 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1141 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1142 | if (t) |
1143 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1144 | } else { |
67c583a7 | 1145 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1146 | } |
27f4d280 PM |
1147 | } |
1148 | ||
a46e0899 PM |
1149 | /* |
1150 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1151 | */ | |
1152 | static void invoke_rcu_callbacks_kthread(void) | |
1153 | { | |
1154 | unsigned long flags; | |
1155 | ||
1156 | local_irq_save(flags); | |
1157 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1158 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1159 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1160 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1161 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1162 | } | |
a46e0899 PM |
1163 | local_irq_restore(flags); |
1164 | } | |
1165 | ||
dff1672d PM |
1166 | /* |
1167 | * Is the current CPU running the RCU-callbacks kthread? | |
1168 | * Caller must have preemption disabled. | |
1169 | */ | |
1170 | static bool rcu_is_callbacks_kthread(void) | |
1171 | { | |
c9d4b0af | 1172 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1173 | } |
1174 | ||
27f4d280 PM |
1175 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1176 | ||
1177 | /* | |
1178 | * Do priority-boost accounting for the start of a new grace period. | |
1179 | */ | |
1180 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1181 | { | |
1182 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1183 | } | |
1184 | ||
27f4d280 PM |
1185 | /* |
1186 | * Create an RCU-boost kthread for the specified node if one does not | |
1187 | * already exist. We only create this kthread for preemptible RCU. | |
1188 | * Returns zero if all is well, a negated errno otherwise. | |
1189 | */ | |
49fb4c62 | 1190 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
0aa04b05 | 1191 | struct rcu_node *rnp) |
27f4d280 | 1192 | { |
5d01bbd1 | 1193 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1194 | unsigned long flags; |
1195 | struct sched_param sp; | |
1196 | struct task_struct *t; | |
1197 | ||
e63c887c | 1198 | if (rcu_state_p != rsp) |
27f4d280 | 1199 | return 0; |
5d01bbd1 | 1200 | |
0aa04b05 | 1201 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1202 | return 0; |
1203 | ||
a46e0899 | 1204 | rsp->boost = 1; |
27f4d280 PM |
1205 | if (rnp->boost_kthread_task != NULL) |
1206 | return 0; | |
1207 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1208 | "rcub/%d", rnp_index); |
27f4d280 PM |
1209 | if (IS_ERR(t)) |
1210 | return PTR_ERR(t); | |
2a67e741 | 1211 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1212 | rnp->boost_kthread_task = t; |
67c583a7 | 1213 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1214 | sp.sched_priority = kthread_prio; |
27f4d280 | 1215 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1216 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1217 | return 0; |
1218 | } | |
1219 | ||
f8b7fc6b PM |
1220 | static void rcu_kthread_do_work(void) |
1221 | { | |
c9d4b0af CL |
1222 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1223 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
be01b4ca | 1224 | rcu_do_batch(&rcu_preempt_state, this_cpu_ptr(&rcu_preempt_data)); |
f8b7fc6b PM |
1225 | } |
1226 | ||
62ab7072 | 1227 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1228 | { |
f8b7fc6b | 1229 | struct sched_param sp; |
f8b7fc6b | 1230 | |
21871d7e | 1231 | sp.sched_priority = kthread_prio; |
62ab7072 | 1232 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1233 | } |
1234 | ||
62ab7072 | 1235 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1236 | { |
62ab7072 | 1237 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1238 | } |
1239 | ||
62ab7072 | 1240 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1241 | { |
c9d4b0af | 1242 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1243 | } |
1244 | ||
1245 | /* | |
1246 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1247 | * RCU softirq used in flavors and configurations of RCU that do not |
1248 | * support RCU priority boosting. | |
f8b7fc6b | 1249 | */ |
62ab7072 | 1250 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1251 | { |
c9d4b0af CL |
1252 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1253 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1254 | int spincnt; |
f8b7fc6b | 1255 | |
62ab7072 | 1256 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1257 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1258 | local_bh_disable(); |
f8b7fc6b | 1259 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1260 | this_cpu_inc(rcu_cpu_kthread_loops); |
1261 | local_irq_disable(); | |
f8b7fc6b PM |
1262 | work = *workp; |
1263 | *workp = 0; | |
62ab7072 | 1264 | local_irq_enable(); |
f8b7fc6b PM |
1265 | if (work) |
1266 | rcu_kthread_do_work(); | |
1267 | local_bh_enable(); | |
62ab7072 | 1268 | if (*workp == 0) { |
f7f7bac9 | 1269 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1270 | *statusp = RCU_KTHREAD_WAITING; |
1271 | return; | |
f8b7fc6b PM |
1272 | } |
1273 | } | |
62ab7072 | 1274 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1275 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1276 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1277 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1278 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1279 | } |
1280 | ||
1281 | /* | |
1282 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1283 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1284 | * held, so the value of rnp->qsmaskinit will be stable. | |
1285 | * | |
1286 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1287 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1288 | * this function allows the kthread to execute on any CPU. | |
1289 | */ | |
5d01bbd1 | 1290 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1291 | { |
5d01bbd1 | 1292 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1293 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1294 | cpumask_var_t cm; |
1295 | int cpu; | |
f8b7fc6b | 1296 | |
5d01bbd1 | 1297 | if (!t) |
f8b7fc6b | 1298 | return; |
5d01bbd1 | 1299 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1300 | return; |
bc75e999 MR |
1301 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1302 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1303 | cpu != outgoingcpu) | |
f8b7fc6b | 1304 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1305 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1306 | cpumask_setall(cm); |
5d01bbd1 | 1307 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1308 | free_cpumask_var(cm); |
1309 | } | |
1310 | ||
62ab7072 PM |
1311 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1312 | .store = &rcu_cpu_kthread_task, | |
1313 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1314 | .thread_fn = rcu_cpu_kthread, | |
1315 | .thread_comm = "rcuc/%u", | |
1316 | .setup = rcu_cpu_kthread_setup, | |
1317 | .park = rcu_cpu_kthread_park, | |
1318 | }; | |
f8b7fc6b PM |
1319 | |
1320 | /* | |
9386c0b7 | 1321 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1322 | */ |
9386c0b7 | 1323 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1324 | { |
f8b7fc6b | 1325 | struct rcu_node *rnp; |
5d01bbd1 | 1326 | int cpu; |
f8b7fc6b | 1327 | |
62ab7072 | 1328 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1329 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1330 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
3e9f5c70 PM |
1331 | rcu_for_each_leaf_node(rcu_state_p, rnp) |
1332 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); | |
f8b7fc6b | 1333 | } |
f8b7fc6b | 1334 | |
49fb4c62 | 1335 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1336 | { |
e534165b | 1337 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
f8b7fc6b PM |
1338 | struct rcu_node *rnp = rdp->mynode; |
1339 | ||
1340 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1341 | if (rcu_scheduler_fully_active) |
e534165b | 1342 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); |
f8b7fc6b PM |
1343 | } |
1344 | ||
27f4d280 PM |
1345 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1346 | ||
1217ed1b | 1347 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1348 | __releases(rnp->lock) |
27f4d280 | 1349 | { |
67c583a7 | 1350 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1351 | } |
1352 | ||
a46e0899 | 1353 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1354 | { |
a46e0899 | 1355 | WARN_ON_ONCE(1); |
27f4d280 PM |
1356 | } |
1357 | ||
dff1672d PM |
1358 | static bool rcu_is_callbacks_kthread(void) |
1359 | { | |
1360 | return false; | |
1361 | } | |
1362 | ||
27f4d280 PM |
1363 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1364 | { | |
1365 | } | |
1366 | ||
5d01bbd1 | 1367 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1368 | { |
1369 | } | |
1370 | ||
9386c0b7 | 1371 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1372 | { |
b0d30417 | 1373 | } |
b0d30417 | 1374 | |
49fb4c62 | 1375 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1376 | { |
1377 | } | |
1378 | ||
27f4d280 PM |
1379 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1380 | ||
8bd93a2c PM |
1381 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1382 | ||
1383 | /* | |
1384 | * Check to see if any future RCU-related work will need to be done | |
1385 | * by the current CPU, even if none need be done immediately, returning | |
1386 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1387 | * an exported member of the RCU API. | |
1388 | * | |
7cb92499 PM |
1389 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1390 | * any flavor of RCU. | |
8bd93a2c | 1391 | */ |
c1ad348b | 1392 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1393 | { |
c1ad348b | 1394 | *nextevt = KTIME_MAX; |
44c65ff2 | 1395 | return rcu_cpu_has_callbacks(NULL); |
7cb92499 PM |
1396 | } |
1397 | ||
1398 | /* | |
1399 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1400 | * after it. | |
1401 | */ | |
8fa7845d | 1402 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1403 | { |
1404 | } | |
1405 | ||
aea1b35e | 1406 | /* |
a858af28 | 1407 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1408 | * is nothing. |
1409 | */ | |
198bbf81 | 1410 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1411 | { |
1412 | } | |
1413 | ||
c57afe80 PM |
1414 | /* |
1415 | * Don't bother keeping a running count of the number of RCU callbacks | |
1416 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1417 | */ | |
1418 | static void rcu_idle_count_callbacks_posted(void) | |
1419 | { | |
1420 | } | |
1421 | ||
8bd93a2c PM |
1422 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1423 | ||
f23f7fa1 PM |
1424 | /* |
1425 | * This code is invoked when a CPU goes idle, at which point we want | |
1426 | * to have the CPU do everything required for RCU so that it can enter | |
1427 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1428 | * state machine implemented by rcu_prepare_for_idle() below. | |
1429 | * | |
1430 | * The following three proprocessor symbols control this state machine: | |
1431 | * | |
f23f7fa1 PM |
1432 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1433 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1434 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1435 | * benchmarkers who might otherwise be tempted to set this to a large | |
1436 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1437 | * system. And if you are -that- concerned about energy efficiency, | |
1438 | * just power the system down and be done with it! | |
778d250a PM |
1439 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1440 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1441 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1442 | * |
1443 | * The values below work well in practice. If future workloads require | |
1444 | * adjustment, they can be converted into kernel config parameters, though | |
1445 | * making the state machine smarter might be a better option. | |
1446 | */ | |
e84c48ae | 1447 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1448 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1449 | |
5e44ce35 PM |
1450 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1451 | module_param(rcu_idle_gp_delay, int, 0644); | |
1452 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1453 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1454 | |
486e2593 | 1455 | /* |
c229828c PM |
1456 | * Try to advance callbacks for all flavors of RCU on the current CPU, but |
1457 | * only if it has been awhile since the last time we did so. Afterwards, | |
1458 | * if there are any callbacks ready for immediate invocation, return true. | |
486e2593 | 1459 | */ |
f1f399d1 | 1460 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1461 | { |
c0f4dfd4 PM |
1462 | bool cbs_ready = false; |
1463 | struct rcu_data *rdp; | |
c229828c | 1464 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1465 | struct rcu_node *rnp; |
1466 | struct rcu_state *rsp; | |
486e2593 | 1467 | |
c229828c PM |
1468 | /* Exit early if we advanced recently. */ |
1469 | if (jiffies == rdtp->last_advance_all) | |
d0bc90fd | 1470 | return false; |
c229828c PM |
1471 | rdtp->last_advance_all = jiffies; |
1472 | ||
c0f4dfd4 PM |
1473 | for_each_rcu_flavor(rsp) { |
1474 | rdp = this_cpu_ptr(rsp->rda); | |
1475 | rnp = rdp->mynode; | |
486e2593 | 1476 | |
c0f4dfd4 PM |
1477 | /* |
1478 | * Don't bother checking unless a grace period has | |
1479 | * completed since we last checked and there are | |
1480 | * callbacks not yet ready to invoke. | |
1481 | */ | |
03c8cb76 PM |
1482 | if ((rcu_seq_completed_gp(rdp->gp_seq, |
1483 | rcu_seq_current(&rnp->gp_seq)) || | |
7d0ae808 | 1484 | unlikely(READ_ONCE(rdp->gpwrap))) && |
15fecf89 | 1485 | rcu_segcblist_pend_cbs(&rdp->cblist)) |
470716fc | 1486 | note_gp_changes(rsp, rdp); |
486e2593 | 1487 | |
15fecf89 | 1488 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1489 | cbs_ready = true; |
1490 | } | |
1491 | return cbs_ready; | |
486e2593 PM |
1492 | } |
1493 | ||
aa9b1630 | 1494 | /* |
c0f4dfd4 PM |
1495 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1496 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1497 | * caller to set the timeout based on whether or not there are non-lazy | |
1498 | * callbacks. | |
aa9b1630 | 1499 | * |
c0f4dfd4 | 1500 | * The caller must have disabled interrupts. |
aa9b1630 | 1501 | */ |
c1ad348b | 1502 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1503 | { |
aa6da514 | 1504 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c1ad348b | 1505 | unsigned long dj; |
aa9b1630 | 1506 | |
b04db8e1 | 1507 | lockdep_assert_irqs_disabled(); |
3382adbc | 1508 | |
c0f4dfd4 PM |
1509 | /* Snapshot to detect later posting of non-lazy callback. */ |
1510 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1511 | ||
aa9b1630 | 1512 | /* If no callbacks, RCU doesn't need the CPU. */ |
aa6da514 | 1513 | if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) { |
c1ad348b | 1514 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1515 | return 0; |
1516 | } | |
c0f4dfd4 PM |
1517 | |
1518 | /* Attempt to advance callbacks. */ | |
1519 | if (rcu_try_advance_all_cbs()) { | |
1520 | /* Some ready to invoke, so initiate later invocation. */ | |
1521 | invoke_rcu_core(); | |
aa9b1630 PM |
1522 | return 1; |
1523 | } | |
c0f4dfd4 PM |
1524 | rdtp->last_accelerate = jiffies; |
1525 | ||
1526 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1527 | if (!rdtp->all_lazy) { |
c1ad348b | 1528 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1529 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1530 | } else { |
c1ad348b | 1531 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1532 | } |
c1ad348b | 1533 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1534 | return 0; |
1535 | } | |
1536 | ||
21e52e15 | 1537 | /* |
c0f4dfd4 PM |
1538 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1539 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1540 | * The second major task is to check to see if a non-lazy callback has | |
1541 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1542 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1543 | * any recently arrived callbacks. | |
aea1b35e PM |
1544 | * |
1545 | * The caller must have disabled interrupts. | |
8bd93a2c | 1546 | */ |
198bbf81 | 1547 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1548 | { |
48a7639c | 1549 | bool needwake; |
c0f4dfd4 | 1550 | struct rcu_data *rdp; |
198bbf81 | 1551 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1552 | struct rcu_node *rnp; |
1553 | struct rcu_state *rsp; | |
9d2ad243 PM |
1554 | int tne; |
1555 | ||
b04db8e1 | 1556 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1557 | if (rcu_is_nocb_cpu(smp_processor_id())) |
3382adbc PM |
1558 | return; |
1559 | ||
9d2ad243 | 1560 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1561 | tne = READ_ONCE(tick_nohz_active); |
9d2ad243 | 1562 | if (tne != rdtp->tick_nohz_enabled_snap) { |
aa6da514 | 1563 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 PM |
1564 | invoke_rcu_core(); /* force nohz to see update. */ |
1565 | rdtp->tick_nohz_enabled_snap = tne; | |
1566 | return; | |
1567 | } | |
1568 | if (!tne) | |
1569 | return; | |
f511fc62 | 1570 | |
c57afe80 | 1571 | /* |
c0f4dfd4 PM |
1572 | * If a non-lazy callback arrived at a CPU having only lazy |
1573 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1574 | * idle duration on re-entry to idle. | |
c57afe80 | 1575 | */ |
c0f4dfd4 PM |
1576 | if (rdtp->all_lazy && |
1577 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
c337f8f5 PM |
1578 | rdtp->all_lazy = false; |
1579 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
c0f4dfd4 | 1580 | invoke_rcu_core(); |
c57afe80 PM |
1581 | return; |
1582 | } | |
c57afe80 | 1583 | |
3084f2f8 | 1584 | /* |
c0f4dfd4 PM |
1585 | * If we have not yet accelerated this jiffy, accelerate all |
1586 | * callbacks on this CPU. | |
3084f2f8 | 1587 | */ |
c0f4dfd4 | 1588 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1589 | return; |
c0f4dfd4 PM |
1590 | rdtp->last_accelerate = jiffies; |
1591 | for_each_rcu_flavor(rsp) { | |
198bbf81 | 1592 | rdp = this_cpu_ptr(rsp->rda); |
135bd1a2 | 1593 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1594 | continue; |
1595 | rnp = rdp->mynode; | |
2a67e741 | 1596 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
48a7639c | 1597 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
67c583a7 | 1598 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c PM |
1599 | if (needwake) |
1600 | rcu_gp_kthread_wake(rsp); | |
77e38ed3 | 1601 | } |
c0f4dfd4 | 1602 | } |
3084f2f8 | 1603 | |
c0f4dfd4 PM |
1604 | /* |
1605 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1606 | * any grace periods that elapsed while the CPU was idle, and if any | |
1607 | * callbacks are now ready to invoke, initiate invocation. | |
1608 | */ | |
8fa7845d | 1609 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1610 | { |
b04db8e1 | 1611 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1612 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1613 | return; |
7a497c96 PM |
1614 | if (rcu_try_advance_all_cbs()) |
1615 | invoke_rcu_core(); | |
8bd93a2c PM |
1616 | } |
1617 | ||
c57afe80 | 1618 | /* |
98248a0e PM |
1619 | * Keep a running count of the number of non-lazy callbacks posted |
1620 | * on this CPU. This running counter (which is never decremented) allows | |
1621 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1622 | * posts a callback, even if an equal number of callbacks are invoked. | |
1623 | * Of course, callbacks should only be posted from within a trace event | |
1624 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1625 | */ |
1626 | static void rcu_idle_count_callbacks_posted(void) | |
1627 | { | |
5955f7ee | 1628 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1629 | } |
1630 | ||
b626c1b6 PM |
1631 | /* |
1632 | * Data for flushing lazy RCU callbacks at OOM time. | |
1633 | */ | |
1634 | static atomic_t oom_callback_count; | |
1635 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1636 | ||
1637 | /* | |
1638 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1639 | * wake-up if we are the last one. | |
1640 | */ | |
1641 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1642 | { | |
1643 | if (atomic_dec_and_test(&oom_callback_count)) | |
1644 | wake_up(&oom_callback_wq); | |
1645 | } | |
1646 | ||
1647 | /* | |
1648 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1649 | * least one lazy callback. This will unnecessarily post callbacks | |
1650 | * to CPUs that already have a non-lazy callback at the end of their | |
1651 | * callback list, but this is an infrequent operation, so accept some | |
1652 | * extra overhead to keep things simple. | |
1653 | */ | |
1654 | static void rcu_oom_notify_cpu(void *unused) | |
1655 | { | |
1656 | struct rcu_state *rsp; | |
1657 | struct rcu_data *rdp; | |
1658 | ||
1659 | for_each_rcu_flavor(rsp) { | |
fa07a58f | 1660 | rdp = raw_cpu_ptr(rsp->rda); |
15fecf89 | 1661 | if (rcu_segcblist_n_lazy_cbs(&rdp->cblist)) { |
b626c1b6 PM |
1662 | atomic_inc(&oom_callback_count); |
1663 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1664 | } | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | /* | |
1669 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1670 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1671 | * ensuring that they free up the corresponding memory in a timely manner. | |
1672 | * Because an uncertain amount of memory will be freed in some uncertain | |
1673 | * timeframe, we do not claim to have freed anything. | |
1674 | */ | |
1675 | static int rcu_oom_notify(struct notifier_block *self, | |
1676 | unsigned long notused, void *nfreed) | |
1677 | { | |
1678 | int cpu; | |
1679 | ||
1680 | /* Wait for callbacks from earlier instance to complete. */ | |
1681 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
78e4bc34 | 1682 | smp_mb(); /* Ensure callback reuse happens after callback invocation. */ |
b626c1b6 PM |
1683 | |
1684 | /* | |
1685 | * Prevent premature wakeup: ensure that all increments happen | |
1686 | * before there is a chance of the counter reaching zero. | |
1687 | */ | |
1688 | atomic_set(&oom_callback_count, 1); | |
1689 | ||
b626c1b6 PM |
1690 | for_each_online_cpu(cpu) { |
1691 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
cee43939 | 1692 | cond_resched_tasks_rcu_qs(); |
b626c1b6 | 1693 | } |
b626c1b6 PM |
1694 | |
1695 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1696 | atomic_dec(&oom_callback_count); | |
1697 | ||
1698 | return NOTIFY_OK; | |
1699 | } | |
1700 | ||
1701 | static struct notifier_block rcu_oom_nb = { | |
1702 | .notifier_call = rcu_oom_notify | |
1703 | }; | |
1704 | ||
1705 | static int __init rcu_register_oom_notifier(void) | |
1706 | { | |
1707 | register_oom_notifier(&rcu_oom_nb); | |
1708 | return 0; | |
1709 | } | |
1710 | early_initcall(rcu_register_oom_notifier); | |
1711 | ||
8bd93a2c | 1712 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 | 1713 | |
a858af28 PM |
1714 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1715 | ||
1716 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1717 | { | |
5955f7ee | 1718 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1719 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1720 | |
c0f4dfd4 PM |
1721 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1722 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1723 | ulong2long(nlpd), | |
1724 | rdtp->all_lazy ? 'L' : '.', | |
1725 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1726 | } |
1727 | ||
1728 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1729 | ||
1730 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1731 | { | |
1c17e4d4 | 1732 | *cp = '\0'; |
a858af28 PM |
1733 | } |
1734 | ||
1735 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1736 | ||
1737 | /* Initiate the stall-info list. */ | |
1738 | static void print_cpu_stall_info_begin(void) | |
1739 | { | |
efc151c3 | 1740 | pr_cont("\n"); |
a858af28 PM |
1741 | } |
1742 | ||
1743 | /* | |
1744 | * Print out diagnostic information for the specified stalled CPU. | |
1745 | * | |
1746 | * If the specified CPU is aware of the current RCU grace period | |
1747 | * (flavor specified by rsp), then print the number of scheduling | |
1748 | * clock interrupts the CPU has taken during the time that it has | |
1749 | * been aware. Otherwise, print the number of RCU grace periods | |
1750 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1751 | * aware of the previous grace period. | |
1752 | * | |
1753 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1754 | */ | |
1755 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1756 | { | |
9b9500da | 1757 | unsigned long delta; |
a858af28 PM |
1758 | char fast_no_hz[72]; |
1759 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1760 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1761 | char *ticks_title; | |
1762 | unsigned long ticks_value; | |
1763 | ||
3caa973b TH |
1764 | /* |
1765 | * We could be printing a lot while holding a spinlock. Avoid | |
1766 | * triggering hard lockup. | |
1767 | */ | |
1768 | touch_nmi_watchdog(); | |
1769 | ||
471f87c3 PM |
1770 | ticks_value = rcu_seq_ctr(rsp->gp_seq - rdp->gp_seq); |
1771 | if (ticks_value) { | |
1772 | ticks_title = "GPs behind"; | |
1773 | } else { | |
a858af28 PM |
1774 | ticks_title = "ticks this GP"; |
1775 | ticks_value = rdp->ticks_this_gp; | |
a858af28 PM |
1776 | } |
1777 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
8aa670cd | 1778 | delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq); |
84585aa8 | 1779 | pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%ld softirq=%u/%u fqs=%ld %s\n", |
7f21aeef PM |
1780 | cpu, |
1781 | "O."[!!cpu_online(cpu)], | |
1782 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1783 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
9b9500da PM |
1784 | !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' : |
1785 | rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : | |
1786 | "!."[!delta], | |
7f21aeef | 1787 | ticks_value, ticks_title, |
02a5c550 | 1788 | rcu_dynticks_snap(rdtp) & 0xfff, |
a858af28 | 1789 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, |
6231069b | 1790 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
7d0ae808 | 1791 | READ_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart, |
a858af28 PM |
1792 | fast_no_hz); |
1793 | } | |
1794 | ||
1795 | /* Terminate the stall-info list. */ | |
1796 | static void print_cpu_stall_info_end(void) | |
1797 | { | |
efc151c3 | 1798 | pr_err("\t"); |
a858af28 PM |
1799 | } |
1800 | ||
1801 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1802 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1803 | { | |
1804 | rdp->ticks_this_gp = 0; | |
6231069b | 1805 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1806 | } |
1807 | ||
1808 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1809 | static void increment_cpu_stall_ticks(void) | |
1810 | { | |
115f7a7c PM |
1811 | struct rcu_state *rsp; |
1812 | ||
1813 | for_each_rcu_flavor(rsp) | |
fa07a58f | 1814 | raw_cpu_inc(rsp->rda->ticks_this_gp); |
a858af28 PM |
1815 | } |
1816 | ||
3fbfbf7a PM |
1817 | #ifdef CONFIG_RCU_NOCB_CPU |
1818 | ||
1819 | /* | |
1820 | * Offload callback processing from the boot-time-specified set of CPUs | |
1821 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1822 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1823 | * waits for a grace period to elapse, and invokes the callbacks. | |
1824 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1825 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1826 | * has been specified, in which case each kthread actively polls its | |
1827 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1828 | * reduce RCU's overhead on that CPU.) | |
1829 | * | |
1830 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1831 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1832 | * running CPU-bound user-mode computations. | |
1833 | * | |
1834 | * Offloading of callback processing could also in theory be used as | |
1835 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1836 | * queued are more aggressive about entering dyntick-idle mode. | |
1837 | */ | |
1838 | ||
1839 | ||
1840 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1841 | static int __init rcu_nocb_setup(char *str) | |
1842 | { | |
1843 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
3fbfbf7a PM |
1844 | cpulist_parse(str, rcu_nocb_mask); |
1845 | return 1; | |
1846 | } | |
1847 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1848 | ||
1b0048a4 PG |
1849 | static int __init parse_rcu_nocb_poll(char *arg) |
1850 | { | |
5455a7f6 | 1851 | rcu_nocb_poll = true; |
1b0048a4 PG |
1852 | return 0; |
1853 | } | |
1854 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1855 | ||
dae6e64d | 1856 | /* |
0446be48 PM |
1857 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1858 | * grace period. | |
dae6e64d | 1859 | */ |
abedf8e2 | 1860 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1861 | { |
abedf8e2 | 1862 | swake_up_all(sq); |
dae6e64d PM |
1863 | } |
1864 | ||
abedf8e2 | 1865 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1866 | { |
e0da2374 | 1867 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1868 | } |
1869 | ||
dae6e64d | 1870 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1871 | { |
abedf8e2 PG |
1872 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1873 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1874 | } |
1875 | ||
24342c96 | 1876 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1877 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1878 | { |
84b12b75 | 1879 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1880 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1881 | return false; | |
1882 | } | |
1883 | ||
fbce7497 | 1884 | /* |
8be6e1b1 PM |
1885 | * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock |
1886 | * and this function releases it. | |
fbce7497 | 1887 | */ |
8be6e1b1 PM |
1888 | static void __wake_nocb_leader(struct rcu_data *rdp, bool force, |
1889 | unsigned long flags) | |
1890 | __releases(rdp->nocb_lock) | |
fbce7497 PM |
1891 | { |
1892 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1893 | ||
8be6e1b1 PM |
1894 | lockdep_assert_held(&rdp->nocb_lock); |
1895 | if (!READ_ONCE(rdp_leader->nocb_kthread)) { | |
1896 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 1897 | return; |
8be6e1b1 PM |
1898 | } |
1899 | if (rdp_leader->nocb_leader_sleep || force) { | |
39953dfd | 1900 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1901 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
8be6e1b1 PM |
1902 | del_timer(&rdp->nocb_timer); |
1903 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
6b5fc3a1 | 1904 | smp_mb(); /* ->nocb_leader_sleep before swake_up(). */ |
abedf8e2 | 1905 | swake_up(&rdp_leader->nocb_wq); |
8be6e1b1 PM |
1906 | } else { |
1907 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 PM |
1908 | } |
1909 | } | |
1910 | ||
8be6e1b1 PM |
1911 | /* |
1912 | * Kick the leader kthread for this NOCB group, but caller has not | |
1913 | * acquired locks. | |
1914 | */ | |
1915 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1916 | { | |
1917 | unsigned long flags; | |
1918 | ||
1919 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1920 | __wake_nocb_leader(rdp, force, flags); | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * Arrange to wake the leader kthread for this NOCB group at some | |
1925 | * future time when it is safe to do so. | |
1926 | */ | |
1927 | static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype, | |
1928 | const char *reason) | |
1929 | { | |
1930 | unsigned long flags; | |
1931 | ||
1932 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1933 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) | |
1934 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
1935 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
1936 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, reason); | |
1937 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
1938 | } | |
1939 | ||
d7e29933 PM |
1940 | /* |
1941 | * Does the specified CPU need an RCU callback for the specified flavor | |
1942 | * of rcu_barrier()? | |
1943 | */ | |
1944 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) | |
1945 | { | |
1946 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
41050a00 PM |
1947 | unsigned long ret; |
1948 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1949 | struct rcu_head *rhp; |
41050a00 | 1950 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1951 | |
41050a00 PM |
1952 | /* |
1953 | * Check count of all no-CBs callbacks awaiting invocation. | |
1954 | * There needs to be a barrier before this function is called, | |
1955 | * but associated with a prior determination that no more | |
1956 | * callbacks would be posted. In the worst case, the first | |
1957 | * barrier in _rcu_barrier() suffices (but the caller cannot | |
1958 | * necessarily rely on this, not a substitute for the caller | |
1959 | * getting the concurrency design right!). There must also be | |
1960 | * a barrier between the following load an posting of a callback | |
1961 | * (if a callback is in fact needed). This is associated with an | |
1962 | * atomic_inc() in the caller. | |
1963 | */ | |
1964 | ret = atomic_long_read(&rdp->nocb_q_count); | |
d7e29933 | 1965 | |
41050a00 | 1966 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 1967 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 1968 | if (!rhp) |
7d0ae808 | 1969 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 1970 | if (!rhp) |
7d0ae808 | 1971 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
1972 | |
1973 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 1974 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 1975 | rcu_scheduler_fully_active) { |
d7e29933 PM |
1976 | /* RCU callback enqueued before CPU first came online??? */ |
1977 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1978 | cpu, rhp->func); | |
1979 | WARN_ON_ONCE(1); | |
1980 | } | |
41050a00 | 1981 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1982 | |
41050a00 | 1983 | return !!ret; |
d7e29933 PM |
1984 | } |
1985 | ||
3fbfbf7a PM |
1986 | /* |
1987 | * Enqueue the specified string of rcu_head structures onto the specified | |
1988 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
1989 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
1990 | * counts are supplied by rhcount and rhcount_lazy. | |
1991 | * | |
1992 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1993 | */ | |
1994 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
1995 | struct rcu_head *rhp, | |
1996 | struct rcu_head **rhtp, | |
96d3fd0d PM |
1997 | int rhcount, int rhcount_lazy, |
1998 | unsigned long flags) | |
3fbfbf7a PM |
1999 | { |
2000 | int len; | |
2001 | struct rcu_head **old_rhpp; | |
2002 | struct task_struct *t; | |
2003 | ||
2004 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
2005 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
2006 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 2007 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 2008 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 2009 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 2010 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
2011 | |
2012 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 2013 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 2014 | if (rcu_nocb_poll || !t) { |
9261dd0d PM |
2015 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2016 | TPS("WakeNotPoll")); | |
3fbfbf7a | 2017 | return; |
9261dd0d | 2018 | } |
3fbfbf7a PM |
2019 | len = atomic_long_read(&rdp->nocb_q_count); |
2020 | if (old_rhpp == &rdp->nocb_head) { | |
96d3fd0d | 2021 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
2022 | /* ... if queue was empty ... */ |
2023 | wake_nocb_leader(rdp, false); | |
96d3fd0d PM |
2024 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2025 | TPS("WakeEmpty")); | |
2026 | } else { | |
8be6e1b1 PM |
2027 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
2028 | TPS("WakeEmptyIsDeferred")); | |
96d3fd0d | 2029 | } |
3fbfbf7a PM |
2030 | rdp->qlen_last_fqs_check = 0; |
2031 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 2032 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
2033 | if (!irqs_disabled_flags(flags)) { |
2034 | wake_nocb_leader(rdp, true); | |
2035 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2036 | TPS("WakeOvf")); | |
2037 | } else { | |
efcd2d54 | 2038 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE_FORCE, |
8be6e1b1 | 2039 | TPS("WakeOvfIsDeferred")); |
9fdd3bc9 | 2040 | } |
3fbfbf7a | 2041 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d PM |
2042 | } else { |
2043 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot")); | |
3fbfbf7a PM |
2044 | } |
2045 | return; | |
2046 | } | |
2047 | ||
2048 | /* | |
2049 | * This is a helper for __call_rcu(), which invokes this when the normal | |
2050 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
2051 | * function returns failure back to __call_rcu(), which can complain | |
2052 | * appropriately. | |
2053 | * | |
2054 | * Otherwise, this function queues the callback where the corresponding | |
2055 | * "rcuo" kthread can find it. | |
2056 | */ | |
2057 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 2058 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
2059 | { |
2060 | ||
d1e43fa5 | 2061 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 2062 | return false; |
96d3fd0d | 2063 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 PM |
2064 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
2065 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
2066 | (unsigned long)rhp->func, | |
756cbf6b PM |
2067 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2068 | -atomic_long_read(&rdp->nocb_q_count)); | |
21e7a608 PM |
2069 | else |
2070 | trace_rcu_callback(rdp->rsp->name, rhp, | |
756cbf6b PM |
2071 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2072 | -atomic_long_read(&rdp->nocb_q_count)); | |
1772947b PM |
2073 | |
2074 | /* | |
2075 | * If called from an extended quiescent state with interrupts | |
2076 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2077 | * deferred-wakeup check to function. | |
2078 | */ | |
2079 | if (irqs_disabled_flags(flags) && | |
2080 | !rcu_is_watching() && | |
2081 | cpu_online(smp_processor_id())) | |
2082 | invoke_rcu_core(); | |
2083 | ||
c271d3a9 | 2084 | return true; |
3fbfbf7a PM |
2085 | } |
2086 | ||
2087 | /* | |
2088 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2089 | * not a no-CBs CPU. | |
2090 | */ | |
b1a2d79f | 2091 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2092 | struct rcu_data *rdp, |
2093 | unsigned long flags) | |
3fbfbf7a | 2094 | { |
b04db8e1 | 2095 | lockdep_assert_irqs_disabled(); |
d1e43fa5 | 2096 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
b1a2d79f PM |
2097 | return false; /* Not NOCBs CPU, caller must migrate CBs. */ |
2098 | __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist), | |
2099 | rcu_segcblist_tail(&rdp->cblist), | |
2100 | rcu_segcblist_n_cbs(&rdp->cblist), | |
2101 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags); | |
2102 | rcu_segcblist_init(&rdp->cblist); | |
2103 | rcu_segcblist_disable(&rdp->cblist); | |
0a9e1e11 | 2104 | return true; |
3fbfbf7a PM |
2105 | } |
2106 | ||
2107 | /* | |
34ed6246 PM |
2108 | * If necessary, kick off a new grace period, and either way wait |
2109 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2110 | */ |
34ed6246 | 2111 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2112 | { |
34ed6246 | 2113 | unsigned long c; |
dae6e64d | 2114 | bool d; |
34ed6246 | 2115 | unsigned long flags; |
48a7639c | 2116 | bool needwake; |
34ed6246 PM |
2117 | struct rcu_node *rnp = rdp->mynode; |
2118 | ||
ab5e869c | 2119 | local_irq_save(flags); |
29365e56 | 2120 | c = rcu_seq_snap(&rdp->rsp->gp_seq); |
ab5e869c PM |
2121 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
2122 | local_irq_restore(flags); | |
2123 | } else { | |
2124 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
2125 | needwake = rcu_start_this_gp(rnp, rdp, c); | |
2126 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2127 | if (needwake) | |
2128 | rcu_gp_kthread_wake(rdp->rsp); | |
2129 | } | |
3fbfbf7a PM |
2130 | |
2131 | /* | |
34ed6246 PM |
2132 | * Wait for the grace period. Do so interruptibly to avoid messing |
2133 | * up the load average. | |
3fbfbf7a | 2134 | */ |
41e80595 | 2135 | trace_rcu_this_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2136 | for (;;) { |
abedf8e2 | 2137 | swait_event_interruptible( |
29365e56 PM |
2138 | rnp->nocb_gp_wq[rcu_seq_ctr(c) & 0x1], |
2139 | (d = rcu_seq_done(&rnp->gp_seq, c))); | |
dae6e64d | 2140 | if (likely(d)) |
34ed6246 | 2141 | break; |
73a860cd | 2142 | WARN_ON(signal_pending(current)); |
41e80595 | 2143 | trace_rcu_this_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2144 | } |
41e80595 | 2145 | trace_rcu_this_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2146 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2147 | } |
2148 | ||
fbce7497 PM |
2149 | /* |
2150 | * Leaders come here to wait for additional callbacks to show up. | |
2151 | * This function does not return until callbacks appear. | |
2152 | */ | |
2153 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2154 | { | |
2155 | bool firsttime = true; | |
8be6e1b1 | 2156 | unsigned long flags; |
fbce7497 PM |
2157 | bool gotcbs; |
2158 | struct rcu_data *rdp; | |
2159 | struct rcu_head **tail; | |
2160 | ||
2161 | wait_again: | |
2162 | ||
2163 | /* Wait for callbacks to appear. */ | |
2164 | if (!rcu_nocb_poll) { | |
bedbb648 | 2165 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Sleep")); |
abedf8e2 | 2166 | swait_event_interruptible(my_rdp->nocb_wq, |
7d0ae808 | 2167 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
8be6e1b1 PM |
2168 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); |
2169 | my_rdp->nocb_leader_sleep = true; | |
2170 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); | |
2171 | del_timer(&my_rdp->nocb_timer); | |
2172 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); | |
fbce7497 PM |
2173 | } else if (firsttime) { |
2174 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
bedbb648 | 2175 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Poll")); |
fbce7497 PM |
2176 | } |
2177 | ||
2178 | /* | |
2179 | * Each pass through the following loop checks a follower for CBs. | |
2180 | * We are our own first follower. Any CBs found are moved to | |
2181 | * nocb_gp_head, where they await a grace period. | |
2182 | */ | |
2183 | gotcbs = false; | |
8be6e1b1 | 2184 | smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */ |
fbce7497 | 2185 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2186 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2187 | if (!rdp->nocb_gp_head) |
2188 | continue; /* No CBs here, try next follower. */ | |
2189 | ||
2190 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2191 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2192 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2193 | gotcbs = true; |
2194 | } | |
2195 | ||
8be6e1b1 | 2196 | /* No callbacks? Sleep a bit if polling, and go retry. */ |
fbce7497 | 2197 | if (unlikely(!gotcbs)) { |
73a860cd | 2198 | WARN_ON(signal_pending(current)); |
8be6e1b1 PM |
2199 | if (rcu_nocb_poll) { |
2200 | schedule_timeout_interruptible(1); | |
2201 | } else { | |
fbce7497 | 2202 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, |
bedbb648 | 2203 | TPS("WokeEmpty")); |
8be6e1b1 | 2204 | } |
fbce7497 PM |
2205 | goto wait_again; |
2206 | } | |
2207 | ||
2208 | /* Wait for one grace period. */ | |
2209 | rcu_nocb_wait_gp(my_rdp); | |
2210 | ||
fbce7497 PM |
2211 | /* Each pass through the following loop wakes a follower, if needed. */ |
2212 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
8be6e1b1 PM |
2213 | if (!rcu_nocb_poll && |
2214 | READ_ONCE(rdp->nocb_head) && | |
2215 | READ_ONCE(my_rdp->nocb_leader_sleep)) { | |
2216 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); | |
11ed7f93 | 2217 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
8be6e1b1 PM |
2218 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); |
2219 | } | |
fbce7497 PM |
2220 | if (!rdp->nocb_gp_head) |
2221 | continue; /* No CBs, so no need to wake follower. */ | |
2222 | ||
2223 | /* Append callbacks to follower's "done" list. */ | |
8be6e1b1 PM |
2224 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2225 | tail = rdp->nocb_follower_tail; | |
2226 | rdp->nocb_follower_tail = rdp->nocb_gp_tail; | |
fbce7497 | 2227 | *tail = rdp->nocb_gp_head; |
8be6e1b1 | 2228 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
fbce7497 | 2229 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
8be6e1b1 | 2230 | /* List was empty, so wake up the follower. */ |
abedf8e2 | 2231 | swake_up(&rdp->nocb_wq); |
fbce7497 PM |
2232 | } |
2233 | } | |
2234 | ||
2235 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2236 | if (!my_rdp->nocb_follower_head) | |
2237 | goto wait_again; | |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * Followers come here to wait for additional callbacks to show up. | |
2242 | * This function does not return until callbacks appear. | |
2243 | */ | |
2244 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2245 | { | |
fbce7497 | 2246 | for (;;) { |
bedbb648 | 2247 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("FollowerSleep")); |
8be6e1b1 PM |
2248 | swait_event_interruptible(rdp->nocb_wq, |
2249 | READ_ONCE(rdp->nocb_follower_head)); | |
fbce7497 PM |
2250 | if (smp_load_acquire(&rdp->nocb_follower_head)) { |
2251 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2252 | return; | |
2253 | } | |
73a860cd | 2254 | WARN_ON(signal_pending(current)); |
bedbb648 | 2255 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeEmpty")); |
fbce7497 PM |
2256 | } |
2257 | } | |
2258 | ||
3fbfbf7a PM |
2259 | /* |
2260 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2261 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2262 | * an optional leader-follower relationship so that the grace-period | |
2263 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2264 | */ |
2265 | static int rcu_nocb_kthread(void *arg) | |
2266 | { | |
2267 | int c, cl; | |
8be6e1b1 | 2268 | unsigned long flags; |
3fbfbf7a PM |
2269 | struct rcu_head *list; |
2270 | struct rcu_head *next; | |
2271 | struct rcu_head **tail; | |
2272 | struct rcu_data *rdp = arg; | |
2273 | ||
2274 | /* Each pass through this loop invokes one batch of callbacks */ | |
2275 | for (;;) { | |
fbce7497 PM |
2276 | /* Wait for callbacks. */ |
2277 | if (rdp->nocb_leader == rdp) | |
2278 | nocb_leader_wait(rdp); | |
2279 | else | |
2280 | nocb_follower_wait(rdp); | |
2281 | ||
2282 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
8be6e1b1 PM |
2283 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2284 | list = rdp->nocb_follower_head; | |
2285 | rdp->nocb_follower_head = NULL; | |
2286 | tail = rdp->nocb_follower_tail; | |
2287 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; | |
2288 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 2289 | BUG_ON(!list); |
bedbb648 | 2290 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeNonEmpty")); |
3fbfbf7a PM |
2291 | |
2292 | /* Each pass through the following loop invokes a callback. */ | |
41050a00 PM |
2293 | trace_rcu_batch_start(rdp->rsp->name, |
2294 | atomic_long_read(&rdp->nocb_q_count_lazy), | |
2295 | atomic_long_read(&rdp->nocb_q_count), -1); | |
3fbfbf7a PM |
2296 | c = cl = 0; |
2297 | while (list) { | |
2298 | next = list->next; | |
2299 | /* Wait for enqueuing to complete, if needed. */ | |
2300 | while (next == NULL && &list->next != tail) { | |
69a79bb1 PM |
2301 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2302 | TPS("WaitQueue")); | |
3fbfbf7a | 2303 | schedule_timeout_interruptible(1); |
69a79bb1 PM |
2304 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2305 | TPS("WokeQueue")); | |
3fbfbf7a PM |
2306 | next = list->next; |
2307 | } | |
2308 | debug_rcu_head_unqueue(list); | |
2309 | local_bh_disable(); | |
2310 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2311 | cl++; | |
2312 | c++; | |
2313 | local_bh_enable(); | |
cee43939 | 2314 | cond_resched_tasks_rcu_qs(); |
3fbfbf7a PM |
2315 | list = next; |
2316 | } | |
2317 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
41050a00 PM |
2318 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2319 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2320 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
3fbfbf7a PM |
2321 | } |
2322 | return 0; | |
2323 | } | |
2324 | ||
96d3fd0d | 2325 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2326 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2327 | { |
7d0ae808 | 2328 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2329 | } |
2330 | ||
2331 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2332 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2333 | { |
8be6e1b1 | 2334 | unsigned long flags; |
9fdd3bc9 PM |
2335 | int ndw; |
2336 | ||
8be6e1b1 PM |
2337 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2338 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { | |
2339 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
96d3fd0d | 2340 | return; |
8be6e1b1 | 2341 | } |
7d0ae808 | 2342 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2343 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
8be6e1b1 | 2344 | __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
9fdd3bc9 | 2345 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2346 | } |
2347 | ||
8be6e1b1 | 2348 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2349 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2350 | { |
fd30b717 KC |
2351 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2352 | ||
2353 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2354 | } |
2355 | ||
2356 | /* | |
2357 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2358 | * This means we do an inexact common-case check. Note that if | |
2359 | * we miss, ->nocb_timer will eventually clean things up. | |
2360 | */ | |
2361 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2362 | { | |
2363 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2364 | do_nocb_deferred_wakeup_common(rdp); | |
2365 | } | |
2366 | ||
f4579fc5 PM |
2367 | void __init rcu_init_nohz(void) |
2368 | { | |
2369 | int cpu; | |
ef126206 | 2370 | bool need_rcu_nocb_mask = false; |
f4579fc5 PM |
2371 | struct rcu_state *rsp; |
2372 | ||
f4579fc5 PM |
2373 | #if defined(CONFIG_NO_HZ_FULL) |
2374 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2375 | need_rcu_nocb_mask = true; | |
2376 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2377 | ||
84b12b75 | 2378 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
2379 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2380 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2381 | return; | |
2382 | } | |
f4579fc5 | 2383 | } |
84b12b75 | 2384 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
2385 | return; |
2386 | ||
f4579fc5 PM |
2387 | #if defined(CONFIG_NO_HZ_FULL) |
2388 | if (tick_nohz_full_running) | |
2389 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2390 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2391 | ||
2392 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 2393 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
2394 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
2395 | rcu_nocb_mask); | |
2396 | } | |
3016611e PM |
2397 | if (cpumask_empty(rcu_nocb_mask)) |
2398 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
2399 | else | |
2400 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
2401 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2402 | if (rcu_nocb_poll) |
2403 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2404 | ||
2405 | for_each_rcu_flavor(rsp) { | |
34404ca8 PM |
2406 | for_each_cpu(cpu, rcu_nocb_mask) |
2407 | init_nocb_callback_list(per_cpu_ptr(rsp->rda, cpu)); | |
35ce7f29 | 2408 | rcu_organize_nocb_kthreads(rsp); |
f4579fc5 | 2409 | } |
96d3fd0d PM |
2410 | } |
2411 | ||
3fbfbf7a PM |
2412 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2413 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2414 | { | |
2415 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2416 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2417 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
8be6e1b1 | 2418 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 2419 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
2420 | } |
2421 | ||
35ce7f29 PM |
2422 | /* |
2423 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2424 | * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are | |
2425 | * brought online out of order, this can require re-organizing the | |
2426 | * leader-follower relationships. | |
2427 | */ | |
2428 | static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu) | |
2429 | { | |
2430 | struct rcu_data *rdp; | |
2431 | struct rcu_data *rdp_last; | |
2432 | struct rcu_data *rdp_old_leader; | |
2433 | struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu); | |
2434 | struct task_struct *t; | |
2435 | ||
2436 | /* | |
2437 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2438 | * then nothing to do. | |
2439 | */ | |
2440 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2441 | return; | |
2442 | ||
2443 | /* If we didn't spawn the leader first, reorganize! */ | |
2444 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2445 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2446 | rdp_last = NULL; | |
2447 | rdp = rdp_old_leader; | |
2448 | do { | |
2449 | rdp->nocb_leader = rdp_spawn; | |
2450 | if (rdp_last && rdp != rdp_spawn) | |
2451 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2452 | if (rdp == rdp_spawn) { |
2453 | rdp = rdp->nocb_next_follower; | |
2454 | } else { | |
2455 | rdp_last = rdp; | |
2456 | rdp = rdp->nocb_next_follower; | |
2457 | rdp_last->nocb_next_follower = NULL; | |
2458 | } | |
35ce7f29 PM |
2459 | } while (rdp); |
2460 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2461 | } | |
2462 | ||
2463 | /* Spawn the kthread for this CPU and RCU flavor. */ | |
2464 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, | |
2465 | "rcuo%c/%d", rsp->abbr, cpu); | |
2466 | BUG_ON(IS_ERR(t)); | |
7d0ae808 | 2467 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2468 | } |
2469 | ||
2470 | /* | |
2471 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2472 | * rcuo kthreads, spawn them. | |
2473 | */ | |
2474 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2475 | { | |
2476 | struct rcu_state *rsp; | |
2477 | ||
2478 | if (rcu_scheduler_fully_active) | |
2479 | for_each_rcu_flavor(rsp) | |
2480 | rcu_spawn_one_nocb_kthread(rsp, cpu); | |
2481 | } | |
2482 | ||
2483 | /* | |
2484 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2485 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2486 | * non-boot CPUs come online -- if this changes, we will need to add | |
2487 | * some mutual exclusion. | |
2488 | */ | |
2489 | static void __init rcu_spawn_nocb_kthreads(void) | |
2490 | { | |
2491 | int cpu; | |
2492 | ||
2493 | for_each_online_cpu(cpu) | |
2494 | rcu_spawn_all_nocb_kthreads(cpu); | |
2495 | } | |
2496 | ||
fbce7497 PM |
2497 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2498 | static int rcu_nocb_leader_stride = -1; | |
2499 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2500 | ||
2501 | /* | |
35ce7f29 | 2502 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2503 | */ |
35ce7f29 | 2504 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp) |
3fbfbf7a PM |
2505 | { |
2506 | int cpu; | |
fbce7497 PM |
2507 | int ls = rcu_nocb_leader_stride; |
2508 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2509 | struct rcu_data *rdp; |
fbce7497 PM |
2510 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2511 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2512 | |
84b12b75 | 2513 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2514 | return; |
fbce7497 PM |
2515 | if (ls == -1) { |
2516 | ls = int_sqrt(nr_cpu_ids); | |
2517 | rcu_nocb_leader_stride = ls; | |
2518 | } | |
2519 | ||
2520 | /* | |
9831ce3b PM |
2521 | * Each pass through this loop sets up one rcu_data structure. |
2522 | * Should the corresponding CPU come online in the future, then | |
2523 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2524 | */ |
3fbfbf7a PM |
2525 | for_each_cpu(cpu, rcu_nocb_mask) { |
2526 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
fbce7497 PM |
2527 | if (rdp->cpu >= nl) { |
2528 | /* New leader, set up for followers & next leader. */ | |
2529 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2530 | rdp->nocb_leader = rdp; | |
2531 | rdp_leader = rdp; | |
2532 | } else { | |
2533 | /* Another follower, link to previous leader. */ | |
2534 | rdp->nocb_leader = rdp_leader; | |
2535 | rdp_prev->nocb_next_follower = rdp; | |
2536 | } | |
2537 | rdp_prev = rdp; | |
3fbfbf7a PM |
2538 | } |
2539 | } | |
2540 | ||
2541 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2542 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2543 | { |
22c2f669 | 2544 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2545 | return false; |
22c2f669 | 2546 | |
34404ca8 | 2547 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2548 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2549 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2550 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2551 | atomic_long_set(&rdp->nocb_q_count, | |
2552 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2553 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2554 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2555 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2556 | } |
15fecf89 | 2557 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2558 | return true; |
3fbfbf7a PM |
2559 | } |
2560 | ||
34ed6246 PM |
2561 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2562 | ||
d7e29933 PM |
2563 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) |
2564 | { | |
2565 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2566 | return false; | |
2567 | } | |
2568 | ||
abedf8e2 | 2569 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2570 | { |
3fbfbf7a PM |
2571 | } |
2572 | ||
abedf8e2 | 2573 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2574 | { |
2575 | return NULL; | |
2576 | } | |
2577 | ||
dae6e64d PM |
2578 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2579 | { | |
2580 | } | |
3fbfbf7a | 2581 | |
3fbfbf7a | 2582 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2583 | bool lazy, unsigned long flags) |
3fbfbf7a | 2584 | { |
4afc7e26 | 2585 | return false; |
3fbfbf7a PM |
2586 | } |
2587 | ||
b1a2d79f | 2588 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2589 | struct rcu_data *rdp, |
2590 | unsigned long flags) | |
3fbfbf7a | 2591 | { |
f4aa84ba | 2592 | return false; |
3fbfbf7a PM |
2593 | } |
2594 | ||
3fbfbf7a PM |
2595 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2596 | { | |
2597 | } | |
2598 | ||
9fdd3bc9 | 2599 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2600 | { |
2601 | return false; | |
2602 | } | |
2603 | ||
2604 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2605 | { | |
2606 | } | |
2607 | ||
35ce7f29 PM |
2608 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2609 | { | |
2610 | } | |
2611 | ||
2612 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2613 | { |
2614 | } | |
2615 | ||
34ed6246 | 2616 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2617 | { |
34ed6246 | 2618 | return false; |
3fbfbf7a PM |
2619 | } |
2620 | ||
2621 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2622 | |
2623 | /* | |
2624 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2625 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2626 | * off. RCU will be paying attention to this CPU because it is in the | |
2627 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2628 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2629 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2630 | * period and has not be idle from an RCU perspective, kick it. | |
2631 | */ | |
4a81e832 | 2632 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu) |
65d798f0 PM |
2633 | { |
2634 | #ifdef CONFIG_NO_HZ_FULL | |
2635 | if (tick_nohz_full_cpu(cpu)) | |
2636 | smp_send_reschedule(cpu); | |
2637 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2638 | } | |
2333210b | 2639 | |
a096932f PM |
2640 | /* |
2641 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2642 | * grace-period kthread will do force_quiescent_state() processing? | |
2643 | * The idea is to avoid waking up RCU core processing on such a | |
2644 | * CPU unless the grace period has extended for too long. | |
2645 | * | |
2646 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2647 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f PM |
2648 | */ |
2649 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp) | |
2650 | { | |
2651 | #ifdef CONFIG_NO_HZ_FULL | |
2652 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
2653 | (!rcu_gp_in_progress(rsp) || | |
7d0ae808 | 2654 | ULONG_CMP_LT(jiffies, READ_ONCE(rsp->gp_start) + HZ))) |
5ce035fb | 2655 | return true; |
a096932f | 2656 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2657 | return false; |
a096932f | 2658 | } |
5057f55e PM |
2659 | |
2660 | /* | |
265f5f28 | 2661 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2662 | */ |
2663 | static void rcu_bind_gp_kthread(void) | |
2664 | { | |
c0f489d2 | 2665 | int __maybe_unused cpu; |
5057f55e | 2666 | |
c0f489d2 | 2667 | if (!tick_nohz_full_enabled()) |
5057f55e | 2668 | return; |
de201559 | 2669 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2670 | } |
176f8f7a PM |
2671 | |
2672 | /* Record the current task on dyntick-idle entry. */ | |
2673 | static void rcu_dynticks_task_enter(void) | |
2674 | { | |
2675 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2676 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2677 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2678 | } | |
2679 | ||
2680 | /* Record no current task on dyntick-idle exit. */ | |
2681 | static void rcu_dynticks_task_exit(void) | |
2682 | { | |
2683 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2684 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2685 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2686 | } |