]>
Commit | Line | Data |
---|---|---|
64db4cff PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
18 | * Copyright IBM Corporation, 2008 | |
19 | * | |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | |
23 | * | |
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | |
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | |
26 | * | |
27 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff PM |
29 | */ |
30 | #include <linux/types.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/smp.h> | |
35 | #include <linux/rcupdate.h> | |
36 | #include <linux/interrupt.h> | |
37 | #include <linux/sched.h> | |
c1dc0b9c | 38 | #include <linux/nmi.h> |
64db4cff PM |
39 | #include <asm/atomic.h> |
40 | #include <linux/bitops.h> | |
41 | #include <linux/module.h> | |
42 | #include <linux/completion.h> | |
43 | #include <linux/moduleparam.h> | |
44 | #include <linux/percpu.h> | |
45 | #include <linux/notifier.h> | |
46 | #include <linux/cpu.h> | |
47 | #include <linux/mutex.h> | |
48 | #include <linux/time.h> | |
6ebb237b | 49 | #include <linux/kernel_stat.h> |
64db4cff | 50 | |
9f77da9f PM |
51 | #include "rcutree.h" |
52 | ||
64db4cff PM |
53 | /* Data structures. */ |
54 | ||
b668c9cf | 55 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 56 | |
64db4cff PM |
57 | #define RCU_STATE_INITIALIZER(name) { \ |
58 | .level = { &name.node[0] }, \ | |
59 | .levelcnt = { \ | |
60 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
61 | NUM_RCU_LVL_1, \ | |
62 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
63 | NUM_RCU_LVL_3, \ |
64 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 65 | }, \ |
83f5b01f | 66 | .signaled = RCU_GP_IDLE, \ |
64db4cff PM |
67 | .gpnum = -300, \ |
68 | .completed = -300, \ | |
69 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ | |
e74f4c45 PM |
70 | .orphan_cbs_list = NULL, \ |
71 | .orphan_cbs_tail = &name.orphan_cbs_list, \ | |
72 | .orphan_qlen = 0, \ | |
64db4cff PM |
73 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ |
74 | .n_force_qs = 0, \ | |
75 | .n_force_qs_ngp = 0, \ | |
76 | } | |
77 | ||
d6714c22 PM |
78 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
79 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 80 | |
6258c4fb IM |
81 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
82 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 83 | |
6ebb237b PM |
84 | static int rcu_scheduler_active __read_mostly; |
85 | ||
f41d911f | 86 | |
fc2219d4 PM |
87 | /* |
88 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
89 | * permit this function to be invoked without holding the root rcu_node | |
90 | * structure's ->lock, but of course results can be subject to change. | |
91 | */ | |
92 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
93 | { | |
94 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
95 | } | |
96 | ||
b1f77b05 | 97 | /* |
d6714c22 | 98 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 99 | * how many quiescent states passed, just if there was at least |
d6714c22 | 100 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 101 | */ |
d6714c22 | 102 | void rcu_sched_qs(int cpu) |
b1f77b05 | 103 | { |
f41d911f PM |
104 | struct rcu_data *rdp; |
105 | ||
f41d911f | 106 | rdp = &per_cpu(rcu_sched_data, cpu); |
c64ac3ce | 107 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
108 | barrier(); |
109 | rdp->passed_quiesc = 1; | |
110 | rcu_preempt_note_context_switch(cpu); | |
b1f77b05 IM |
111 | } |
112 | ||
d6714c22 | 113 | void rcu_bh_qs(int cpu) |
b1f77b05 | 114 | { |
f41d911f PM |
115 | struct rcu_data *rdp; |
116 | ||
f41d911f | 117 | rdp = &per_cpu(rcu_bh_data, cpu); |
c64ac3ce | 118 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
119 | barrier(); |
120 | rdp->passed_quiesc = 1; | |
b1f77b05 | 121 | } |
64db4cff PM |
122 | |
123 | #ifdef CONFIG_NO_HZ | |
90a4d2c0 PM |
124 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
125 | .dynticks_nesting = 1, | |
126 | .dynticks = 1, | |
127 | }; | |
64db4cff PM |
128 | #endif /* #ifdef CONFIG_NO_HZ */ |
129 | ||
130 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
131 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
132 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
133 | ||
3d76c082 PM |
134 | module_param(blimit, int, 0); |
135 | module_param(qhimark, int, 0); | |
136 | module_param(qlowmark, int, 0); | |
137 | ||
64db4cff | 138 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 139 | static int rcu_pending(int cpu); |
64db4cff PM |
140 | |
141 | /* | |
d6714c22 | 142 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 143 | */ |
d6714c22 | 144 | long rcu_batches_completed_sched(void) |
64db4cff | 145 | { |
d6714c22 | 146 | return rcu_sched_state.completed; |
64db4cff | 147 | } |
d6714c22 | 148 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
149 | |
150 | /* | |
151 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
152 | */ | |
153 | long rcu_batches_completed_bh(void) | |
154 | { | |
155 | return rcu_bh_state.completed; | |
156 | } | |
157 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
158 | ||
bf66f18e PM |
159 | /* |
160 | * Force a quiescent state for RCU BH. | |
161 | */ | |
162 | void rcu_bh_force_quiescent_state(void) | |
163 | { | |
164 | force_quiescent_state(&rcu_bh_state, 0); | |
165 | } | |
166 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
167 | ||
168 | /* | |
169 | * Force a quiescent state for RCU-sched. | |
170 | */ | |
171 | void rcu_sched_force_quiescent_state(void) | |
172 | { | |
173 | force_quiescent_state(&rcu_sched_state, 0); | |
174 | } | |
175 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
176 | ||
64db4cff PM |
177 | /* |
178 | * Does the CPU have callbacks ready to be invoked? | |
179 | */ | |
180 | static int | |
181 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
182 | { | |
183 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
184 | } | |
185 | ||
186 | /* | |
187 | * Does the current CPU require a yet-as-unscheduled grace period? | |
188 | */ | |
189 | static int | |
190 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
191 | { | |
fc2219d4 | 192 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
193 | } |
194 | ||
195 | /* | |
196 | * Return the root node of the specified rcu_state structure. | |
197 | */ | |
198 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
199 | { | |
200 | return &rsp->node[0]; | |
201 | } | |
202 | ||
203 | #ifdef CONFIG_SMP | |
204 | ||
205 | /* | |
206 | * If the specified CPU is offline, tell the caller that it is in | |
207 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
208 | * Grace periods can end up waiting on an offline CPU when that | |
209 | * CPU is in the process of coming online -- it will be added to the | |
210 | * rcu_node bitmasks before it actually makes it online. The same thing | |
211 | * can happen while a CPU is in the process of coming online. Because this | |
212 | * race is quite rare, we check for it after detecting that the grace | |
213 | * period has been delayed rather than checking each and every CPU | |
214 | * each and every time we start a new grace period. | |
215 | */ | |
216 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
217 | { | |
218 | /* | |
219 | * If the CPU is offline, it is in a quiescent state. We can | |
220 | * trust its state not to change because interrupts are disabled. | |
221 | */ | |
222 | if (cpu_is_offline(rdp->cpu)) { | |
223 | rdp->offline_fqs++; | |
224 | return 1; | |
225 | } | |
226 | ||
f41d911f PM |
227 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
228 | if (rdp->preemptable) | |
229 | return 0; | |
230 | ||
64db4cff PM |
231 | /* The CPU is online, so send it a reschedule IPI. */ |
232 | if (rdp->cpu != smp_processor_id()) | |
233 | smp_send_reschedule(rdp->cpu); | |
234 | else | |
235 | set_need_resched(); | |
236 | rdp->resched_ipi++; | |
237 | return 0; | |
238 | } | |
239 | ||
240 | #endif /* #ifdef CONFIG_SMP */ | |
241 | ||
242 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
243 | |
244 | /** | |
245 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
246 | * | |
247 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
248 | * read-side critical sections can occur. (Though RCU read-side | |
249 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
250 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
251 | */ | |
252 | void rcu_enter_nohz(void) | |
253 | { | |
254 | unsigned long flags; | |
255 | struct rcu_dynticks *rdtp; | |
256 | ||
257 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
258 | local_irq_save(flags); | |
259 | rdtp = &__get_cpu_var(rcu_dynticks); | |
260 | rdtp->dynticks++; | |
261 | rdtp->dynticks_nesting--; | |
86848966 | 262 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
263 | local_irq_restore(flags); |
264 | } | |
265 | ||
266 | /* | |
267 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
268 | * | |
269 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
270 | * read-side critical sections normally occur. | |
271 | */ | |
272 | void rcu_exit_nohz(void) | |
273 | { | |
274 | unsigned long flags; | |
275 | struct rcu_dynticks *rdtp; | |
276 | ||
277 | local_irq_save(flags); | |
278 | rdtp = &__get_cpu_var(rcu_dynticks); | |
279 | rdtp->dynticks++; | |
280 | rdtp->dynticks_nesting++; | |
86848966 | 281 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
282 | local_irq_restore(flags); |
283 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
284 | } | |
285 | ||
286 | /** | |
287 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
288 | * | |
289 | * If the CPU was idle with dynamic ticks active, and there is no | |
290 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
291 | * RCU grace-period handling know that the CPU is active. | |
292 | */ | |
293 | void rcu_nmi_enter(void) | |
294 | { | |
295 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
296 | ||
297 | if (rdtp->dynticks & 0x1) | |
298 | return; | |
299 | rdtp->dynticks_nmi++; | |
86848966 | 300 | WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1)); |
64db4cff PM |
301 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
302 | } | |
303 | ||
304 | /** | |
305 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
306 | * | |
307 | * If the CPU was idle with dynamic ticks active, and there is no | |
308 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
309 | * RCU grace-period handling know that the CPU is no longer active. | |
310 | */ | |
311 | void rcu_nmi_exit(void) | |
312 | { | |
313 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
314 | ||
315 | if (rdtp->dynticks & 0x1) | |
316 | return; | |
317 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
318 | rdtp->dynticks_nmi++; | |
86848966 | 319 | WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1); |
64db4cff PM |
320 | } |
321 | ||
322 | /** | |
323 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
324 | * | |
325 | * If the CPU was idle with dynamic ticks active, this updates the | |
326 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
327 | */ | |
328 | void rcu_irq_enter(void) | |
329 | { | |
330 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
331 | ||
332 | if (rdtp->dynticks_nesting++) | |
333 | return; | |
334 | rdtp->dynticks++; | |
86848966 | 335 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
336 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
337 | } | |
338 | ||
339 | /** | |
340 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
341 | * | |
342 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
343 | * to put let the RCU handling be aware that the CPU is going back to idle | |
344 | * with no ticks. | |
345 | */ | |
346 | void rcu_irq_exit(void) | |
347 | { | |
348 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
349 | ||
350 | if (--rdtp->dynticks_nesting) | |
351 | return; | |
352 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
353 | rdtp->dynticks++; | |
86848966 | 354 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
355 | |
356 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 357 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
358 | __get_cpu_var(rcu_bh_data).nxtlist) |
359 | set_need_resched(); | |
360 | } | |
361 | ||
64db4cff PM |
362 | #ifdef CONFIG_SMP |
363 | ||
64db4cff PM |
364 | /* |
365 | * Snapshot the specified CPU's dynticks counter so that we can later | |
366 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 367 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
368 | */ |
369 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
370 | { | |
371 | int ret; | |
372 | int snap; | |
373 | int snap_nmi; | |
374 | ||
375 | snap = rdp->dynticks->dynticks; | |
376 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
377 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
378 | rdp->dynticks_snap = snap; | |
379 | rdp->dynticks_nmi_snap = snap_nmi; | |
380 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
381 | if (ret) | |
382 | rdp->dynticks_fqs++; | |
383 | return ret; | |
384 | } | |
385 | ||
386 | /* | |
387 | * Return true if the specified CPU has passed through a quiescent | |
388 | * state by virtue of being in or having passed through an dynticks | |
389 | * idle state since the last call to dyntick_save_progress_counter() | |
390 | * for this same CPU. | |
391 | */ | |
392 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
393 | { | |
394 | long curr; | |
395 | long curr_nmi; | |
396 | long snap; | |
397 | long snap_nmi; | |
398 | ||
399 | curr = rdp->dynticks->dynticks; | |
400 | snap = rdp->dynticks_snap; | |
401 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
402 | snap_nmi = rdp->dynticks_nmi_snap; | |
403 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
404 | ||
405 | /* | |
406 | * If the CPU passed through or entered a dynticks idle phase with | |
407 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
408 | * already acknowledged the request to pass through a quiescent | |
409 | * state. Either way, that CPU cannot possibly be in an RCU | |
410 | * read-side critical section that started before the beginning | |
411 | * of the current RCU grace period. | |
412 | */ | |
413 | if ((curr != snap || (curr & 0x1) == 0) && | |
414 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
415 | rdp->dynticks_fqs++; | |
416 | return 1; | |
417 | } | |
418 | ||
419 | /* Go check for the CPU being offline. */ | |
420 | return rcu_implicit_offline_qs(rdp); | |
421 | } | |
422 | ||
423 | #endif /* #ifdef CONFIG_SMP */ | |
424 | ||
425 | #else /* #ifdef CONFIG_NO_HZ */ | |
426 | ||
64db4cff PM |
427 | #ifdef CONFIG_SMP |
428 | ||
64db4cff PM |
429 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
430 | { | |
431 | return 0; | |
432 | } | |
433 | ||
434 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
435 | { | |
436 | return rcu_implicit_offline_qs(rdp); | |
437 | } | |
438 | ||
439 | #endif /* #ifdef CONFIG_SMP */ | |
440 | ||
441 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
442 | ||
443 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
444 | ||
445 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
446 | { | |
447 | rsp->gp_start = jiffies; | |
448 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
449 | } | |
450 | ||
451 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
452 | { | |
453 | int cpu; | |
454 | long delta; | |
455 | unsigned long flags; | |
456 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
457 | |
458 | /* Only let one CPU complain about others per time interval. */ | |
459 | ||
460 | spin_lock_irqsave(&rnp->lock, flags); | |
461 | delta = jiffies - rsp->jiffies_stall; | |
fc2219d4 | 462 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
64db4cff PM |
463 | spin_unlock_irqrestore(&rnp->lock, flags); |
464 | return; | |
465 | } | |
466 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
467 | |
468 | /* | |
469 | * Now rat on any tasks that got kicked up to the root rcu_node | |
470 | * due to CPU offlining. | |
471 | */ | |
472 | rcu_print_task_stall(rnp); | |
64db4cff PM |
473 | spin_unlock_irqrestore(&rnp->lock, flags); |
474 | ||
475 | /* OK, time to rat on our buddy... */ | |
476 | ||
477 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | |
a0b6c9a7 | 478 | rcu_for_each_leaf_node(rsp, rnp) { |
f41d911f | 479 | rcu_print_task_stall(rnp); |
a0b6c9a7 | 480 | if (rnp->qsmask == 0) |
64db4cff | 481 | continue; |
a0b6c9a7 PM |
482 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
483 | if (rnp->qsmask & (1UL << cpu)) | |
484 | printk(" %d", rnp->grplo + cpu); | |
64db4cff PM |
485 | } |
486 | printk(" (detected by %d, t=%ld jiffies)\n", | |
487 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | |
c1dc0b9c IM |
488 | trigger_all_cpu_backtrace(); |
489 | ||
64db4cff PM |
490 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
491 | } | |
492 | ||
493 | static void print_cpu_stall(struct rcu_state *rsp) | |
494 | { | |
495 | unsigned long flags; | |
496 | struct rcu_node *rnp = rcu_get_root(rsp); | |
497 | ||
498 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | |
499 | smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
500 | trigger_all_cpu_backtrace(); |
501 | ||
64db4cff PM |
502 | spin_lock_irqsave(&rnp->lock, flags); |
503 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) | |
504 | rsp->jiffies_stall = | |
505 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
506 | spin_unlock_irqrestore(&rnp->lock, flags); | |
c1dc0b9c | 507 | |
64db4cff PM |
508 | set_need_resched(); /* kick ourselves to get things going. */ |
509 | } | |
510 | ||
511 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
512 | { | |
513 | long delta; | |
514 | struct rcu_node *rnp; | |
515 | ||
516 | delta = jiffies - rsp->jiffies_stall; | |
517 | rnp = rdp->mynode; | |
518 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
519 | ||
520 | /* We haven't checked in, so go dump stack. */ | |
521 | print_cpu_stall(rsp); | |
522 | ||
fc2219d4 | 523 | } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) { |
64db4cff PM |
524 | |
525 | /* They had two time units to dump stack, so complain. */ | |
526 | print_other_cpu_stall(rsp); | |
527 | } | |
528 | } | |
529 | ||
530 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
531 | ||
532 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
533 | { | |
534 | } | |
535 | ||
536 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
537 | { | |
538 | } | |
539 | ||
540 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
541 | ||
542 | /* | |
543 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
544 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
545 | * that someone else started the grace period. The caller must hold the |
546 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
547 | * and must have irqs disabled. | |
64db4cff | 548 | */ |
9160306e PM |
549 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
550 | { | |
551 | if (rdp->gpnum != rnp->gpnum) { | |
552 | rdp->qs_pending = 1; | |
553 | rdp->passed_quiesc = 0; | |
554 | rdp->gpnum = rnp->gpnum; | |
555 | } | |
556 | } | |
557 | ||
64db4cff PM |
558 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
559 | { | |
9160306e PM |
560 | unsigned long flags; |
561 | struct rcu_node *rnp; | |
562 | ||
563 | local_irq_save(flags); | |
564 | rnp = rdp->mynode; | |
565 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
566 | !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */ | |
567 | local_irq_restore(flags); | |
568 | return; | |
569 | } | |
570 | __note_new_gpnum(rsp, rnp, rdp); | |
571 | spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
572 | } |
573 | ||
574 | /* | |
575 | * Did someone else start a new RCU grace period start since we last | |
576 | * checked? Update local state appropriately if so. Must be called | |
577 | * on the CPU corresponding to rdp. | |
578 | */ | |
579 | static int | |
580 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
581 | { | |
582 | unsigned long flags; | |
583 | int ret = 0; | |
584 | ||
585 | local_irq_save(flags); | |
586 | if (rdp->gpnum != rsp->gpnum) { | |
587 | note_new_gpnum(rsp, rdp); | |
588 | ret = 1; | |
589 | } | |
590 | local_irq_restore(flags); | |
591 | return ret; | |
592 | } | |
593 | ||
d09b62df PM |
594 | /* |
595 | * Advance this CPU's callbacks, but only if the current grace period | |
596 | * has ended. This may be called only from the CPU to whom the rdp | |
597 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
598 | * ->lock must be held by the caller, with irqs disabled. | |
599 | */ | |
600 | static void | |
601 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
602 | { | |
603 | /* Did another grace period end? */ | |
604 | if (rdp->completed != rnp->completed) { | |
605 | ||
606 | /* Advance callbacks. No harm if list empty. */ | |
607 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
608 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
609 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
610 | ||
611 | /* Remember that we saw this grace-period completion. */ | |
612 | rdp->completed = rnp->completed; | |
613 | } | |
614 | } | |
615 | ||
616 | /* | |
617 | * Advance this CPU's callbacks, but only if the current grace period | |
618 | * has ended. This may be called only from the CPU to whom the rdp | |
619 | * belongs. | |
620 | */ | |
621 | static void | |
622 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
623 | { | |
624 | unsigned long flags; | |
625 | struct rcu_node *rnp; | |
626 | ||
627 | local_irq_save(flags); | |
628 | rnp = rdp->mynode; | |
629 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
630 | !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */ | |
631 | local_irq_restore(flags); | |
632 | return; | |
633 | } | |
634 | __rcu_process_gp_end(rsp, rnp, rdp); | |
635 | spin_unlock_irqrestore(&rnp->lock, flags); | |
636 | } | |
637 | ||
638 | /* | |
639 | * Do per-CPU grace-period initialization for running CPU. The caller | |
640 | * must hold the lock of the leaf rcu_node structure corresponding to | |
641 | * this CPU. | |
642 | */ | |
643 | static void | |
644 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
645 | { | |
646 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
647 | __rcu_process_gp_end(rsp, rnp, rdp); | |
648 | ||
649 | /* | |
650 | * Because this CPU just now started the new grace period, we know | |
651 | * that all of its callbacks will be covered by this upcoming grace | |
652 | * period, even the ones that were registered arbitrarily recently. | |
653 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
654 | * | |
655 | * Other CPUs cannot be sure exactly when the grace period started. | |
656 | * Therefore, their recently registered callbacks must pass through | |
657 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
658 | * by the next RCU grace period. | |
659 | */ | |
660 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
661 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
662 | |
663 | /* Set state so that this CPU will detect the next quiescent state. */ | |
664 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
665 | } |
666 | ||
64db4cff PM |
667 | /* |
668 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
669 | * in preparation for detecting the next grace period. The caller must hold | |
670 | * the root node's ->lock, which is released before return. Hard irqs must | |
671 | * be disabled. | |
672 | */ | |
673 | static void | |
674 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
675 | __releases(rcu_get_root(rsp)->lock) | |
676 | { | |
677 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
678 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff | 679 | |
07079d53 | 680 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
681 | if (cpu_needs_another_gp(rsp, rdp)) |
682 | rsp->fqs_need_gp = 1; | |
b32e9eb6 PM |
683 | if (rnp->completed == rsp->completed) { |
684 | spin_unlock_irqrestore(&rnp->lock, flags); | |
685 | return; | |
686 | } | |
687 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
688 | ||
689 | /* | |
690 | * Propagate new ->completed value to rcu_node structures | |
691 | * so that other CPUs don't have to wait until the start | |
692 | * of the next grace period to process their callbacks. | |
693 | */ | |
694 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
695 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
696 | rnp->completed = rsp->completed; | |
697 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
698 | } | |
699 | local_irq_restore(flags); | |
64db4cff PM |
700 | return; |
701 | } | |
702 | ||
703 | /* Advance to a new grace period and initialize state. */ | |
704 | rsp->gpnum++; | |
c3422bea | 705 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
706 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
707 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 708 | record_gp_stall_check_time(rsp); |
64db4cff | 709 | |
64db4cff PM |
710 | /* Special-case the common single-level case. */ |
711 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 712 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 713 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 714 | rnp->gpnum = rsp->gpnum; |
d09b62df | 715 | rnp->completed = rsp->completed; |
c12172c0 | 716 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 717 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
64db4cff PM |
718 | spin_unlock_irqrestore(&rnp->lock, flags); |
719 | return; | |
720 | } | |
721 | ||
722 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | |
723 | ||
724 | ||
725 | /* Exclude any concurrent CPU-hotplug operations. */ | |
726 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
727 | ||
728 | /* | |
b835db1f PM |
729 | * Set the quiescent-state-needed bits in all the rcu_node |
730 | * structures for all currently online CPUs in breadth-first | |
731 | * order, starting from the root rcu_node structure. This | |
732 | * operation relies on the layout of the hierarchy within the | |
733 | * rsp->node[] array. Note that other CPUs will access only | |
734 | * the leaves of the hierarchy, which still indicate that no | |
735 | * grace period is in progress, at least until the corresponding | |
736 | * leaf node has been initialized. In addition, we have excluded | |
737 | * CPU-hotplug operations. | |
64db4cff PM |
738 | * |
739 | * Note that the grace period cannot complete until we finish | |
740 | * the initialization process, as there will be at least one | |
741 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
742 | * one corresponding to this CPU, due to the fact that we have |
743 | * irqs disabled. | |
64db4cff | 744 | */ |
a0b6c9a7 | 745 | rcu_for_each_node_breadth_first(rsp, rnp) { |
83f5b01f | 746 | spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 747 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 748 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 749 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
750 | rnp->completed = rsp->completed; |
751 | if (rnp == rdp->mynode) | |
752 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
83f5b01f | 753 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
754 | } |
755 | ||
83f5b01f PM |
756 | rnp = rcu_get_root(rsp); |
757 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
64db4cff | 758 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
83f5b01f | 759 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
760 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
761 | } | |
762 | ||
f41d911f | 763 | /* |
d3f6bad3 PM |
764 | * Report a full set of quiescent states to the specified rcu_state |
765 | * data structure. This involves cleaning up after the prior grace | |
766 | * period and letting rcu_start_gp() start up the next grace period | |
767 | * if one is needed. Note that the caller must hold rnp->lock, as | |
768 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 769 | */ |
d3f6bad3 | 770 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 771 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 772 | { |
fc2219d4 | 773 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
f41d911f | 774 | rsp->completed = rsp->gpnum; |
83f5b01f | 775 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
776 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
777 | } | |
778 | ||
64db4cff | 779 | /* |
d3f6bad3 PM |
780 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
781 | * Allows quiescent states for a group of CPUs to be reported at one go | |
782 | * to the specified rcu_node structure, though all the CPUs in the group | |
783 | * must be represented by the same rcu_node structure (which need not be | |
784 | * a leaf rcu_node structure, though it often will be). That structure's | |
785 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
786 | */ |
787 | static void | |
d3f6bad3 PM |
788 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
789 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
790 | __releases(rnp->lock) |
791 | { | |
28ecd580 PM |
792 | struct rcu_node *rnp_c; |
793 | ||
64db4cff PM |
794 | /* Walk up the rcu_node hierarchy. */ |
795 | for (;;) { | |
796 | if (!(rnp->qsmask & mask)) { | |
797 | ||
798 | /* Our bit has already been cleared, so done. */ | |
799 | spin_unlock_irqrestore(&rnp->lock, flags); | |
800 | return; | |
801 | } | |
802 | rnp->qsmask &= ~mask; | |
f41d911f | 803 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
804 | |
805 | /* Other bits still set at this level, so done. */ | |
806 | spin_unlock_irqrestore(&rnp->lock, flags); | |
807 | return; | |
808 | } | |
809 | mask = rnp->grpmask; | |
810 | if (rnp->parent == NULL) { | |
811 | ||
812 | /* No more levels. Exit loop holding root lock. */ | |
813 | ||
814 | break; | |
815 | } | |
816 | spin_unlock_irqrestore(&rnp->lock, flags); | |
28ecd580 | 817 | rnp_c = rnp; |
64db4cff PM |
818 | rnp = rnp->parent; |
819 | spin_lock_irqsave(&rnp->lock, flags); | |
28ecd580 | 820 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
821 | } |
822 | ||
823 | /* | |
824 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 825 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 826 | * to clean up and start the next grace period if one is needed. |
64db4cff | 827 | */ |
d3f6bad3 | 828 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
829 | } |
830 | ||
831 | /* | |
d3f6bad3 PM |
832 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
833 | * structure. This must be either called from the specified CPU, or | |
834 | * called when the specified CPU is known to be offline (and when it is | |
835 | * also known that no other CPU is concurrently trying to help the offline | |
836 | * CPU). The lastcomp argument is used to make sure we are still in the | |
837 | * grace period of interest. We don't want to end the current grace period | |
838 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
839 | */ |
840 | static void | |
d3f6bad3 | 841 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
64db4cff PM |
842 | { |
843 | unsigned long flags; | |
844 | unsigned long mask; | |
845 | struct rcu_node *rnp; | |
846 | ||
847 | rnp = rdp->mynode; | |
848 | spin_lock_irqsave(&rnp->lock, flags); | |
560d4bc0 | 849 | if (lastcomp != rnp->completed) { |
64db4cff PM |
850 | |
851 | /* | |
852 | * Someone beat us to it for this grace period, so leave. | |
853 | * The race with GP start is resolved by the fact that we | |
854 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
855 | * cannot yet be initialized -- so we would simply find our | |
d3f6bad3 PM |
856 | * CPU's bit already cleared in rcu_report_qs_rnp() if this |
857 | * race occurred. | |
64db4cff PM |
858 | */ |
859 | rdp->passed_quiesc = 0; /* try again later! */ | |
860 | spin_unlock_irqrestore(&rnp->lock, flags); | |
861 | return; | |
862 | } | |
863 | mask = rdp->grpmask; | |
864 | if ((rnp->qsmask & mask) == 0) { | |
865 | spin_unlock_irqrestore(&rnp->lock, flags); | |
866 | } else { | |
867 | rdp->qs_pending = 0; | |
868 | ||
869 | /* | |
870 | * This GP can't end until cpu checks in, so all of our | |
871 | * callbacks can be processed during the next GP. | |
872 | */ | |
64db4cff PM |
873 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
874 | ||
d3f6bad3 | 875 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
876 | } |
877 | } | |
878 | ||
879 | /* | |
880 | * Check to see if there is a new grace period of which this CPU | |
881 | * is not yet aware, and if so, set up local rcu_data state for it. | |
882 | * Otherwise, see if this CPU has just passed through its first | |
883 | * quiescent state for this grace period, and record that fact if so. | |
884 | */ | |
885 | static void | |
886 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
887 | { | |
888 | /* If there is now a new grace period, record and return. */ | |
889 | if (check_for_new_grace_period(rsp, rdp)) | |
890 | return; | |
891 | ||
892 | /* | |
893 | * Does this CPU still need to do its part for current grace period? | |
894 | * If no, return and let the other CPUs do their part as well. | |
895 | */ | |
896 | if (!rdp->qs_pending) | |
897 | return; | |
898 | ||
899 | /* | |
900 | * Was there a quiescent state since the beginning of the grace | |
901 | * period? If no, then exit and wait for the next call. | |
902 | */ | |
903 | if (!rdp->passed_quiesc) | |
904 | return; | |
905 | ||
d3f6bad3 PM |
906 | /* |
907 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
908 | * judge of that). | |
909 | */ | |
910 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
64db4cff PM |
911 | } |
912 | ||
913 | #ifdef CONFIG_HOTPLUG_CPU | |
914 | ||
e74f4c45 PM |
915 | /* |
916 | * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the | |
917 | * specified flavor of RCU. The callbacks will be adopted by the next | |
918 | * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever | |
919 | * comes first. Because this is invoked from the CPU_DYING notifier, | |
920 | * irqs are already disabled. | |
921 | */ | |
922 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) | |
923 | { | |
924 | int i; | |
925 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
926 | ||
927 | if (rdp->nxtlist == NULL) | |
928 | return; /* irqs disabled, so comparison is stable. */ | |
929 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
930 | *rsp->orphan_cbs_tail = rdp->nxtlist; | |
931 | rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL]; | |
932 | rdp->nxtlist = NULL; | |
933 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
934 | rdp->nxttail[i] = &rdp->nxtlist; | |
935 | rsp->orphan_qlen += rdp->qlen; | |
936 | rdp->qlen = 0; | |
937 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
938 | } | |
939 | ||
940 | /* | |
941 | * Adopt previously orphaned RCU callbacks. | |
942 | */ | |
943 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
944 | { | |
945 | unsigned long flags; | |
946 | struct rcu_data *rdp; | |
947 | ||
948 | spin_lock_irqsave(&rsp->onofflock, flags); | |
949 | rdp = rsp->rda[smp_processor_id()]; | |
950 | if (rsp->orphan_cbs_list == NULL) { | |
951 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
952 | return; | |
953 | } | |
954 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list; | |
955 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail; | |
956 | rdp->qlen += rsp->orphan_qlen; | |
957 | rsp->orphan_cbs_list = NULL; | |
958 | rsp->orphan_cbs_tail = &rsp->orphan_cbs_list; | |
959 | rsp->orphan_qlen = 0; | |
960 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
961 | } | |
962 | ||
64db4cff PM |
963 | /* |
964 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
965 | * and move all callbacks from the outgoing CPU to the current one. | |
966 | */ | |
967 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
968 | { | |
64db4cff | 969 | unsigned long flags; |
64db4cff | 970 | unsigned long mask; |
d9a3da06 | 971 | int need_report = 0; |
64db4cff | 972 | struct rcu_data *rdp = rsp->rda[cpu]; |
64db4cff PM |
973 | struct rcu_node *rnp; |
974 | ||
975 | /* Exclude any attempts to start a new grace period. */ | |
976 | spin_lock_irqsave(&rsp->onofflock, flags); | |
977 | ||
978 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 979 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
980 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
981 | do { | |
982 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
983 | rnp->qsmaskinit &= ~mask; | |
984 | if (rnp->qsmaskinit != 0) { | |
b668c9cf PM |
985 | if (rnp != rdp->mynode) |
986 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
64db4cff PM |
987 | break; |
988 | } | |
b668c9cf | 989 | if (rnp == rdp->mynode) |
d9a3da06 | 990 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
b668c9cf PM |
991 | else |
992 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
64db4cff | 993 | mask = rnp->grpmask; |
64db4cff PM |
994 | rnp = rnp->parent; |
995 | } while (rnp != NULL); | |
64db4cff | 996 | |
b668c9cf PM |
997 | /* |
998 | * We still hold the leaf rcu_node structure lock here, and | |
999 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1000 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1001 | * held leads to deadlock. | |
b668c9cf PM |
1002 | */ |
1003 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
1004 | rnp = rdp->mynode; | |
d9a3da06 | 1005 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1006 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf PM |
1007 | else |
1008 | spin_unlock_irqrestore(&rnp->lock, flags); | |
d9a3da06 PM |
1009 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1010 | rcu_report_exp_rnp(rsp, rnp); | |
64db4cff | 1011 | |
e74f4c45 | 1012 | rcu_adopt_orphan_cbs(rsp); |
64db4cff PM |
1013 | } |
1014 | ||
1015 | /* | |
1016 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1017 | * callbacks that it might have to the current CPU. This code assumes | |
1018 | * that at least one CPU in the system will remain running at all times. | |
1019 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1020 | */ | |
1021 | static void rcu_offline_cpu(int cpu) | |
1022 | { | |
d6714c22 | 1023 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1024 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1025 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1026 | } |
1027 | ||
1028 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1029 | ||
e74f4c45 PM |
1030 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) |
1031 | { | |
1032 | } | |
1033 | ||
1034 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
1035 | { | |
1036 | } | |
1037 | ||
64db4cff PM |
1038 | static void rcu_offline_cpu(int cpu) |
1039 | { | |
1040 | } | |
1041 | ||
1042 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1043 | ||
1044 | /* | |
1045 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1046 | * period. Thottle as specified by rdp->blimit. | |
1047 | */ | |
37c72e56 | 1048 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1049 | { |
1050 | unsigned long flags; | |
1051 | struct rcu_head *next, *list, **tail; | |
1052 | int count; | |
1053 | ||
1054 | /* If no callbacks are ready, just return.*/ | |
1055 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
1056 | return; | |
1057 | ||
1058 | /* | |
1059 | * Extract the list of ready callbacks, disabling to prevent | |
1060 | * races with call_rcu() from interrupt handlers. | |
1061 | */ | |
1062 | local_irq_save(flags); | |
1063 | list = rdp->nxtlist; | |
1064 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1065 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1066 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1067 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1068 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1069 | rdp->nxttail[count] = &rdp->nxtlist; | |
1070 | local_irq_restore(flags); | |
1071 | ||
1072 | /* Invoke callbacks. */ | |
1073 | count = 0; | |
1074 | while (list) { | |
1075 | next = list->next; | |
1076 | prefetch(next); | |
1077 | list->func(list); | |
1078 | list = next; | |
1079 | if (++count >= rdp->blimit) | |
1080 | break; | |
1081 | } | |
1082 | ||
1083 | local_irq_save(flags); | |
1084 | ||
1085 | /* Update count, and requeue any remaining callbacks. */ | |
1086 | rdp->qlen -= count; | |
1087 | if (list != NULL) { | |
1088 | *tail = rdp->nxtlist; | |
1089 | rdp->nxtlist = list; | |
1090 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1091 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1092 | rdp->nxttail[count] = tail; | |
1093 | else | |
1094 | break; | |
1095 | } | |
1096 | ||
1097 | /* Reinstate batch limit if we have worked down the excess. */ | |
1098 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1099 | rdp->blimit = blimit; | |
1100 | ||
37c72e56 PM |
1101 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1102 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1103 | rdp->qlen_last_fqs_check = 0; | |
1104 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1105 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1106 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1107 | ||
64db4cff PM |
1108 | local_irq_restore(flags); |
1109 | ||
1110 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1111 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
1112 | raise_softirq(RCU_SOFTIRQ); | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1117 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1118 | * Also schedule the RCU softirq handler. | |
1119 | * | |
1120 | * This function must be called with hardirqs disabled. It is normally | |
1121 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1122 | * false, there is no point in invoking rcu_check_callbacks(). | |
1123 | */ | |
1124 | void rcu_check_callbacks(int cpu, int user) | |
1125 | { | |
a157229c PM |
1126 | if (!rcu_pending(cpu)) |
1127 | return; /* if nothing for RCU to do. */ | |
64db4cff | 1128 | if (user || |
a6826048 PM |
1129 | (idle_cpu(cpu) && rcu_scheduler_active && |
1130 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1131 | |
1132 | /* | |
1133 | * Get here if this CPU took its interrupt from user | |
1134 | * mode or from the idle loop, and if this is not a | |
1135 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1136 | * a quiescent state, so note it. |
64db4cff PM |
1137 | * |
1138 | * No memory barrier is required here because both | |
d6714c22 PM |
1139 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1140 | * variables that other CPUs neither access nor modify, | |
1141 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1142 | */ |
1143 | ||
d6714c22 PM |
1144 | rcu_sched_qs(cpu); |
1145 | rcu_bh_qs(cpu); | |
64db4cff PM |
1146 | |
1147 | } else if (!in_softirq()) { | |
1148 | ||
1149 | /* | |
1150 | * Get here if this CPU did not take its interrupt from | |
1151 | * softirq, in other words, if it is not interrupting | |
1152 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1153 | * critical section, so note it. |
64db4cff PM |
1154 | */ |
1155 | ||
d6714c22 | 1156 | rcu_bh_qs(cpu); |
64db4cff | 1157 | } |
f41d911f | 1158 | rcu_preempt_check_callbacks(cpu); |
64db4cff PM |
1159 | raise_softirq(RCU_SOFTIRQ); |
1160 | } | |
1161 | ||
1162 | #ifdef CONFIG_SMP | |
1163 | ||
1164 | /* | |
1165 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1166 | * have not yet encountered a quiescent state, using the function specified. | |
ee47eb9f | 1167 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1168 | */ |
45f014c5 | 1169 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1170 | { |
1171 | unsigned long bit; | |
1172 | int cpu; | |
1173 | unsigned long flags; | |
1174 | unsigned long mask; | |
a0b6c9a7 | 1175 | struct rcu_node *rnp; |
64db4cff | 1176 | |
a0b6c9a7 | 1177 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1178 | mask = 0; |
a0b6c9a7 | 1179 | spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1180 | if (!rcu_gp_in_progress(rsp)) { |
a0b6c9a7 | 1181 | spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1182 | return; |
64db4cff | 1183 | } |
a0b6c9a7 PM |
1184 | if (rnp->qsmask == 0) { |
1185 | spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1186 | continue; |
1187 | } | |
a0b6c9a7 | 1188 | cpu = rnp->grplo; |
64db4cff | 1189 | bit = 1; |
a0b6c9a7 PM |
1190 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
1191 | if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu])) | |
64db4cff PM |
1192 | mask |= bit; |
1193 | } | |
45f014c5 | 1194 | if (mask != 0) { |
64db4cff | 1195 | |
d3f6bad3 PM |
1196 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1197 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1198 | continue; |
1199 | } | |
a0b6c9a7 | 1200 | spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1201 | } |
64db4cff PM |
1202 | } |
1203 | ||
1204 | /* | |
1205 | * Force quiescent states on reluctant CPUs, and also detect which | |
1206 | * CPUs are in dyntick-idle mode. | |
1207 | */ | |
1208 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1209 | { | |
1210 | unsigned long flags; | |
64db4cff | 1211 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1212 | |
fc2219d4 | 1213 | if (!rcu_gp_in_progress(rsp)) |
64db4cff PM |
1214 | return; /* No grace period in progress, nothing to force. */ |
1215 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | |
1216 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | |
1217 | return; /* Someone else is already on the job. */ | |
1218 | } | |
1219 | if (relaxed && | |
ef631b0c | 1220 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
f96e9232 | 1221 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1222 | rsp->n_force_qs++; |
07079d53 | 1223 | spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1224 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1225 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1226 | rsp->n_force_qs_ngp++; |
07079d53 | 1227 | spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1228 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1229 | } |
07079d53 | 1230 | rsp->fqs_active = 1; |
f3a8b5c6 | 1231 | switch (rsp->signaled) { |
83f5b01f | 1232 | case RCU_GP_IDLE: |
64db4cff PM |
1233 | case RCU_GP_INIT: |
1234 | ||
83f5b01f | 1235 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1236 | |
1237 | case RCU_SAVE_DYNTICK: | |
1238 | ||
07079d53 | 1239 | spin_unlock(&rnp->lock); /* irqs remain disabled */ |
64db4cff PM |
1240 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1241 | break; /* So gcc recognizes the dead code. */ | |
1242 | ||
1243 | /* Record dyntick-idle state. */ | |
45f014c5 | 1244 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
07079d53 | 1245 | spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1246 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1247 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1248 | break; |
64db4cff PM |
1249 | |
1250 | case RCU_FORCE_QS: | |
1251 | ||
1252 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
07079d53 | 1253 | spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1254 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1255 | |
1256 | /* Leave state in case more forcing is required. */ | |
1257 | ||
07079d53 | 1258 | spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1259 | break; |
64db4cff | 1260 | } |
07079d53 | 1261 | rsp->fqs_active = 0; |
46a1e34e PM |
1262 | if (rsp->fqs_need_gp) { |
1263 | spin_unlock(&rsp->fqslock); /* irqs remain disabled */ | |
1264 | rsp->fqs_need_gp = 0; | |
1265 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
1266 | return; | |
1267 | } | |
07079d53 | 1268 | spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1269 | unlock_fqs_ret: |
64db4cff PM |
1270 | spin_unlock_irqrestore(&rsp->fqslock, flags); |
1271 | } | |
1272 | ||
1273 | #else /* #ifdef CONFIG_SMP */ | |
1274 | ||
1275 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1276 | { | |
1277 | set_need_resched(); | |
1278 | } | |
1279 | ||
1280 | #endif /* #else #ifdef CONFIG_SMP */ | |
1281 | ||
1282 | /* | |
1283 | * This does the RCU processing work from softirq context for the | |
1284 | * specified rcu_state and rcu_data structures. This may be called | |
1285 | * only from the CPU to whom the rdp belongs. | |
1286 | */ | |
1287 | static void | |
1288 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1289 | { | |
1290 | unsigned long flags; | |
1291 | ||
2e597558 PM |
1292 | WARN_ON_ONCE(rdp->beenonline == 0); |
1293 | ||
64db4cff PM |
1294 | /* |
1295 | * If an RCU GP has gone long enough, go check for dyntick | |
1296 | * idle CPUs and, if needed, send resched IPIs. | |
1297 | */ | |
ef631b0c | 1298 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1299 | force_quiescent_state(rsp, 1); |
1300 | ||
1301 | /* | |
1302 | * Advance callbacks in response to end of earlier grace | |
1303 | * period that some other CPU ended. | |
1304 | */ | |
1305 | rcu_process_gp_end(rsp, rdp); | |
1306 | ||
1307 | /* Update RCU state based on any recent quiescent states. */ | |
1308 | rcu_check_quiescent_state(rsp, rdp); | |
1309 | ||
1310 | /* Does this CPU require a not-yet-started grace period? */ | |
1311 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1312 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | |
1313 | rcu_start_gp(rsp, flags); /* releases above lock */ | |
1314 | } | |
1315 | ||
1316 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1317 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1318 | } |
1319 | ||
1320 | /* | |
1321 | * Do softirq processing for the current CPU. | |
1322 | */ | |
1323 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1324 | { | |
1325 | /* | |
1326 | * Memory references from any prior RCU read-side critical sections | |
1327 | * executed by the interrupted code must be seen before any RCU | |
1328 | * grace-period manipulations below. | |
1329 | */ | |
1330 | smp_mb(); /* See above block comment. */ | |
1331 | ||
d6714c22 PM |
1332 | __rcu_process_callbacks(&rcu_sched_state, |
1333 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1334 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1335 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1336 | |
1337 | /* | |
1338 | * Memory references from any later RCU read-side critical sections | |
1339 | * executed by the interrupted code must be seen after any RCU | |
1340 | * grace-period manipulations above. | |
1341 | */ | |
1342 | smp_mb(); /* See above block comment. */ | |
1343 | } | |
1344 | ||
1345 | static void | |
1346 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1347 | struct rcu_state *rsp) | |
1348 | { | |
1349 | unsigned long flags; | |
1350 | struct rcu_data *rdp; | |
1351 | ||
1352 | head->func = func; | |
1353 | head->next = NULL; | |
1354 | ||
1355 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1356 | ||
1357 | /* | |
1358 | * Opportunistically note grace-period endings and beginnings. | |
1359 | * Note that we might see a beginning right after we see an | |
1360 | * end, but never vice versa, since this CPU has to pass through | |
1361 | * a quiescent state betweentimes. | |
1362 | */ | |
1363 | local_irq_save(flags); | |
1364 | rdp = rsp->rda[smp_processor_id()]; | |
1365 | rcu_process_gp_end(rsp, rdp); | |
1366 | check_for_new_grace_period(rsp, rdp); | |
1367 | ||
1368 | /* Add the callback to our list. */ | |
1369 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1370 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1371 | ||
1372 | /* Start a new grace period if one not already started. */ | |
fc2219d4 | 1373 | if (!rcu_gp_in_progress(rsp)) { |
64db4cff PM |
1374 | unsigned long nestflag; |
1375 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1376 | ||
1377 | spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1378 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | |
1379 | } | |
1380 | ||
37c72e56 PM |
1381 | /* |
1382 | * Force the grace period if too many callbacks or too long waiting. | |
1383 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1384 | * if some other CPU has recently done so. Also, don't bother | |
1385 | * invoking force_quiescent_state() if the newly enqueued callback | |
1386 | * is the only one waiting for a grace period to complete. | |
1387 | */ | |
1388 | if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { | |
64db4cff | 1389 | rdp->blimit = LONG_MAX; |
37c72e56 PM |
1390 | if (rsp->n_force_qs == rdp->n_force_qs_snap && |
1391 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1392 | force_quiescent_state(rsp, 0); | |
1393 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1394 | rdp->qlen_last_fqs_check = rdp->qlen; | |
ef631b0c | 1395 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1396 | force_quiescent_state(rsp, 1); |
1397 | local_irq_restore(flags); | |
1398 | } | |
1399 | ||
1400 | /* | |
d6714c22 | 1401 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1402 | */ |
d6714c22 | 1403 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1404 | { |
d6714c22 | 1405 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1406 | } |
d6714c22 | 1407 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1408 | |
1409 | /* | |
1410 | * Queue an RCU for invocation after a quicker grace period. | |
1411 | */ | |
1412 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1413 | { | |
1414 | __call_rcu(head, func, &rcu_bh_state); | |
1415 | } | |
1416 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1417 | ||
6ebb237b PM |
1418 | /** |
1419 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1420 | * | |
1421 | * Control will return to the caller some time after a full rcu-sched | |
1422 | * grace period has elapsed, in other words after all currently executing | |
1423 | * rcu-sched read-side critical sections have completed. These read-side | |
1424 | * critical sections are delimited by rcu_read_lock_sched() and | |
1425 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1426 | * local_irq_disable(), and so on may be used in place of | |
1427 | * rcu_read_lock_sched(). | |
1428 | * | |
1429 | * This means that all preempt_disable code sequences, including NMI and | |
1430 | * hardware-interrupt handlers, in progress on entry will have completed | |
1431 | * before this primitive returns. However, this does not guarantee that | |
1432 | * softirq handlers will have completed, since in some kernels, these | |
1433 | * handlers can run in process context, and can block. | |
1434 | * | |
1435 | * This primitive provides the guarantees made by the (now removed) | |
1436 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1437 | * guarantees that rcu_read_lock() sections will have completed. | |
1438 | * In "classic RCU", these two guarantees happen to be one and | |
1439 | * the same, but can differ in realtime RCU implementations. | |
1440 | */ | |
1441 | void synchronize_sched(void) | |
1442 | { | |
1443 | struct rcu_synchronize rcu; | |
1444 | ||
1445 | if (rcu_blocking_is_gp()) | |
1446 | return; | |
1447 | ||
1448 | init_completion(&rcu.completion); | |
1449 | /* Will wake me after RCU finished. */ | |
1450 | call_rcu_sched(&rcu.head, wakeme_after_rcu); | |
1451 | /* Wait for it. */ | |
1452 | wait_for_completion(&rcu.completion); | |
1453 | } | |
1454 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1455 | ||
1456 | /** | |
1457 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1458 | * | |
1459 | * Control will return to the caller some time after a full rcu_bh grace | |
1460 | * period has elapsed, in other words after all currently executing rcu_bh | |
1461 | * read-side critical sections have completed. RCU read-side critical | |
1462 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1463 | * and may be nested. | |
1464 | */ | |
1465 | void synchronize_rcu_bh(void) | |
1466 | { | |
1467 | struct rcu_synchronize rcu; | |
1468 | ||
1469 | if (rcu_blocking_is_gp()) | |
1470 | return; | |
1471 | ||
1472 | init_completion(&rcu.completion); | |
1473 | /* Will wake me after RCU finished. */ | |
1474 | call_rcu_bh(&rcu.head, wakeme_after_rcu); | |
1475 | /* Wait for it. */ | |
1476 | wait_for_completion(&rcu.completion); | |
1477 | } | |
1478 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1479 | ||
64db4cff PM |
1480 | /* |
1481 | * Check to see if there is any immediate RCU-related work to be done | |
1482 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1483 | * The checks are in order of increasing expense: checks that can be | |
1484 | * carried out against CPU-local state are performed first. However, | |
1485 | * we must check for CPU stalls first, else we might not get a chance. | |
1486 | */ | |
1487 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1488 | { | |
2f51f988 PM |
1489 | struct rcu_node *rnp = rdp->mynode; |
1490 | ||
64db4cff PM |
1491 | rdp->n_rcu_pending++; |
1492 | ||
1493 | /* Check for CPU stalls, if enabled. */ | |
1494 | check_cpu_stall(rsp, rdp); | |
1495 | ||
1496 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
7ba5c840 PM |
1497 | if (rdp->qs_pending) { |
1498 | rdp->n_rp_qs_pending++; | |
64db4cff | 1499 | return 1; |
7ba5c840 | 1500 | } |
64db4cff PM |
1501 | |
1502 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1503 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1504 | rdp->n_rp_cb_ready++; | |
64db4cff | 1505 | return 1; |
7ba5c840 | 1506 | } |
64db4cff PM |
1507 | |
1508 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1509 | if (cpu_needs_another_gp(rsp, rdp)) { |
1510 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1511 | return 1; |
7ba5c840 | 1512 | } |
64db4cff PM |
1513 | |
1514 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1515 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1516 | rdp->n_rp_gp_completed++; |
64db4cff | 1517 | return 1; |
7ba5c840 | 1518 | } |
64db4cff PM |
1519 | |
1520 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1521 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1522 | rdp->n_rp_gp_started++; |
64db4cff | 1523 | return 1; |
7ba5c840 | 1524 | } |
64db4cff PM |
1525 | |
1526 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1527 | if (rcu_gp_in_progress(rsp) && |
7ba5c840 PM |
1528 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
1529 | rdp->n_rp_need_fqs++; | |
64db4cff | 1530 | return 1; |
7ba5c840 | 1531 | } |
64db4cff PM |
1532 | |
1533 | /* nothing to do */ | |
7ba5c840 | 1534 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1535 | return 0; |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Check to see if there is any immediate RCU-related work to be done | |
1540 | * by the current CPU, returning 1 if so. This function is part of the | |
1541 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1542 | */ | |
a157229c | 1543 | static int rcu_pending(int cpu) |
64db4cff | 1544 | { |
d6714c22 | 1545 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1546 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1547 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1548 | } |
1549 | ||
1550 | /* | |
1551 | * Check to see if any future RCU-related work will need to be done | |
1552 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1553 | * 1 if so. |
64db4cff | 1554 | */ |
8bd93a2c | 1555 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
1556 | { |
1557 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1558 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1559 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1560 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1561 | } |
1562 | ||
6ebb237b PM |
1563 | /* |
1564 | * This function is invoked towards the end of the scheduler's initialization | |
1565 | * process. Before this is called, the idle task might contain | |
1566 | * RCU read-side critical sections (during which time, this idle | |
1567 | * task is booting the system). After this function is called, the | |
1568 | * idle tasks are prohibited from containing RCU read-side critical | |
1569 | * sections. | |
1570 | */ | |
1571 | void rcu_scheduler_starting(void) | |
1572 | { | |
1573 | WARN_ON(num_online_cpus() != 1); | |
1574 | WARN_ON(nr_context_switches() > 0); | |
1575 | rcu_scheduler_active = 1; | |
1576 | } | |
1577 | ||
d0ec774c PM |
1578 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1579 | static atomic_t rcu_barrier_cpu_count; | |
1580 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1581 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1582 | |
1583 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1584 | { | |
1585 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1586 | complete(&rcu_barrier_completion); | |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1591 | */ | |
1592 | static void rcu_barrier_func(void *type) | |
1593 | { | |
1594 | int cpu = smp_processor_id(); | |
1595 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1596 | void (*call_rcu_func)(struct rcu_head *head, | |
1597 | void (*func)(struct rcu_head *head)); | |
1598 | ||
1599 | atomic_inc(&rcu_barrier_cpu_count); | |
1600 | call_rcu_func = type; | |
1601 | call_rcu_func(head, rcu_barrier_callback); | |
1602 | } | |
1603 | ||
d0ec774c PM |
1604 | /* |
1605 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1606 | * RCU callbacks of the specified type to complete. | |
1607 | */ | |
e74f4c45 PM |
1608 | static void _rcu_barrier(struct rcu_state *rsp, |
1609 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1610 | void (*func)(struct rcu_head *head))) |
1611 | { | |
1612 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1613 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1614 | mutex_lock(&rcu_barrier_mutex); |
1615 | init_completion(&rcu_barrier_completion); | |
1616 | /* | |
1617 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1618 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1619 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1620 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1621 | * might complete its grace period before all of the other CPUs | |
1622 | * did their increment, causing this function to return too | |
1623 | * early. | |
1624 | */ | |
1625 | atomic_set(&rcu_barrier_cpu_count, 1); | |
e74f4c45 PM |
1626 | preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */ |
1627 | rcu_adopt_orphan_cbs(rsp); | |
d0ec774c | 1628 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); |
e74f4c45 | 1629 | preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */ |
d0ec774c PM |
1630 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1631 | complete(&rcu_barrier_completion); | |
1632 | wait_for_completion(&rcu_barrier_completion); | |
1633 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1634 | } |
d0ec774c PM |
1635 | |
1636 | /** | |
1637 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1638 | */ | |
1639 | void rcu_barrier_bh(void) | |
1640 | { | |
e74f4c45 | 1641 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1642 | } |
1643 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
1644 | ||
1645 | /** | |
1646 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
1647 | */ | |
1648 | void rcu_barrier_sched(void) | |
1649 | { | |
e74f4c45 | 1650 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
1651 | } |
1652 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
1653 | ||
64db4cff | 1654 | /* |
27569620 | 1655 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1656 | */ |
27569620 PM |
1657 | static void __init |
1658 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1659 | { |
1660 | unsigned long flags; | |
1661 | int i; | |
27569620 PM |
1662 | struct rcu_data *rdp = rsp->rda[cpu]; |
1663 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1664 | ||
1665 | /* Set up local state, ensuring consistent view of global state. */ | |
1666 | spin_lock_irqsave(&rnp->lock, flags); | |
1667 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | |
1668 | rdp->nxtlist = NULL; | |
1669 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1670 | rdp->nxttail[i] = &rdp->nxtlist; | |
1671 | rdp->qlen = 0; | |
1672 | #ifdef CONFIG_NO_HZ | |
1673 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1674 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1675 | rdp->cpu = cpu; | |
1676 | spin_unlock_irqrestore(&rnp->lock, flags); | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1681 | * offline event can be happening at a given time. Note also that we | |
1682 | * can accept some slop in the rsp->completed access due to the fact | |
1683 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1684 | */ |
e4fa4c97 | 1685 | static void __cpuinit |
f41d911f | 1686 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1687 | { |
1688 | unsigned long flags; | |
64db4cff PM |
1689 | unsigned long mask; |
1690 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1691 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1692 | ||
1693 | /* Set up local state, ensuring consistent view of global state. */ | |
1694 | spin_lock_irqsave(&rnp->lock, flags); | |
64db4cff PM |
1695 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
1696 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1697 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1698 | rdp->preemptable = preemptable; |
37c72e56 PM |
1699 | rdp->qlen_last_fqs_check = 0; |
1700 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 1701 | rdp->blimit = blimit; |
64db4cff PM |
1702 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1703 | ||
1704 | /* | |
1705 | * A new grace period might start here. If so, we won't be part | |
1706 | * of it, but that is OK, as we are currently in a quiescent state. | |
1707 | */ | |
1708 | ||
1709 | /* Exclude any attempts to start a new GP on large systems. */ | |
1710 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
1711 | ||
1712 | /* Add CPU to rcu_node bitmasks. */ | |
1713 | rnp = rdp->mynode; | |
1714 | mask = rdp->grpmask; | |
1715 | do { | |
1716 | /* Exclude any attempts to start a new GP on small systems. */ | |
1717 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1718 | rnp->qsmaskinit |= mask; | |
1719 | mask = rnp->grpmask; | |
d09b62df PM |
1720 | if (rnp == rdp->mynode) { |
1721 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
1722 | rdp->completed = rnp->completed; | |
1723 | rdp->passed_quiesc_completed = rnp->completed - 1; | |
1724 | } | |
64db4cff PM |
1725 | spin_unlock(&rnp->lock); /* irqs already disabled. */ |
1726 | rnp = rnp->parent; | |
1727 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1728 | ||
e7d8842e | 1729 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
1730 | } |
1731 | ||
1732 | static void __cpuinit rcu_online_cpu(int cpu) | |
1733 | { | |
f41d911f PM |
1734 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1735 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1736 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1737 | } |
1738 | ||
1739 | /* | |
f41d911f | 1740 | * Handle CPU online/offline notification events. |
64db4cff | 1741 | */ |
9f680ab4 PM |
1742 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1743 | unsigned long action, void *hcpu) | |
64db4cff PM |
1744 | { |
1745 | long cpu = (long)hcpu; | |
1746 | ||
1747 | switch (action) { | |
1748 | case CPU_UP_PREPARE: | |
1749 | case CPU_UP_PREPARE_FROZEN: | |
1750 | rcu_online_cpu(cpu); | |
1751 | break; | |
d0ec774c PM |
1752 | case CPU_DYING: |
1753 | case CPU_DYING_FROZEN: | |
1754 | /* | |
e74f4c45 | 1755 | * preempt_disable() in _rcu_barrier() prevents stop_machine(), |
d0ec774c | 1756 | * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);" |
e74f4c45 PM |
1757 | * returns, all online cpus have queued rcu_barrier_func(). |
1758 | * The dying CPU clears its cpu_online_mask bit and | |
1759 | * moves all of its RCU callbacks to ->orphan_cbs_list | |
1760 | * in the context of stop_machine(), so subsequent calls | |
1761 | * to _rcu_barrier() will adopt these callbacks and only | |
1762 | * then queue rcu_barrier_func() on all remaining CPUs. | |
d0ec774c | 1763 | */ |
e74f4c45 PM |
1764 | rcu_send_cbs_to_orphanage(&rcu_bh_state); |
1765 | rcu_send_cbs_to_orphanage(&rcu_sched_state); | |
1766 | rcu_preempt_send_cbs_to_orphanage(); | |
d0ec774c | 1767 | break; |
64db4cff PM |
1768 | case CPU_DEAD: |
1769 | case CPU_DEAD_FROZEN: | |
1770 | case CPU_UP_CANCELED: | |
1771 | case CPU_UP_CANCELED_FROZEN: | |
1772 | rcu_offline_cpu(cpu); | |
1773 | break; | |
1774 | default: | |
1775 | break; | |
1776 | } | |
1777 | return NOTIFY_OK; | |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * Compute the per-level fanout, either using the exact fanout specified | |
1782 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1783 | */ | |
1784 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1785 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1786 | { | |
1787 | int i; | |
1788 | ||
1789 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1790 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1791 | } | |
1792 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1793 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1794 | { | |
1795 | int ccur; | |
1796 | int cprv; | |
1797 | int i; | |
1798 | ||
1799 | cprv = NR_CPUS; | |
1800 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1801 | ccur = rsp->levelcnt[i]; | |
1802 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1803 | cprv = ccur; | |
1804 | } | |
1805 | } | |
1806 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1807 | ||
1808 | /* | |
1809 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1810 | */ | |
1811 | static void __init rcu_init_one(struct rcu_state *rsp) | |
1812 | { | |
b6407e86 PM |
1813 | static char *buf[] = { "rcu_node_level_0", |
1814 | "rcu_node_level_1", | |
1815 | "rcu_node_level_2", | |
1816 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
1817 | int cpustride = 1; |
1818 | int i; | |
1819 | int j; | |
1820 | struct rcu_node *rnp; | |
1821 | ||
b6407e86 PM |
1822 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
1823 | ||
64db4cff PM |
1824 | /* Initialize the level-tracking arrays. */ |
1825 | ||
1826 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1827 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1828 | rcu_init_levelspread(rsp); | |
1829 | ||
1830 | /* Initialize the elements themselves, starting from the leaves. */ | |
1831 | ||
1832 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1833 | cpustride *= rsp->levelspread[i]; | |
1834 | rnp = rsp->level[i]; | |
1835 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
88b91c7c | 1836 | spin_lock_init(&rnp->lock); |
b6407e86 PM |
1837 | lockdep_set_class_and_name(&rnp->lock, |
1838 | &rcu_node_class[i], buf[i]); | |
f41d911f | 1839 | rnp->gpnum = 0; |
64db4cff PM |
1840 | rnp->qsmask = 0; |
1841 | rnp->qsmaskinit = 0; | |
1842 | rnp->grplo = j * cpustride; | |
1843 | rnp->grphi = (j + 1) * cpustride - 1; | |
1844 | if (rnp->grphi >= NR_CPUS) | |
1845 | rnp->grphi = NR_CPUS - 1; | |
1846 | if (i == 0) { | |
1847 | rnp->grpnum = 0; | |
1848 | rnp->grpmask = 0; | |
1849 | rnp->parent = NULL; | |
1850 | } else { | |
1851 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1852 | rnp->grpmask = 1UL << rnp->grpnum; | |
1853 | rnp->parent = rsp->level[i - 1] + | |
1854 | j / rsp->levelspread[i - 1]; | |
1855 | } | |
1856 | rnp->level = i; | |
f41d911f PM |
1857 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1858 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
d9a3da06 PM |
1859 | INIT_LIST_HEAD(&rnp->blocked_tasks[2]); |
1860 | INIT_LIST_HEAD(&rnp->blocked_tasks[3]); | |
64db4cff PM |
1861 | } |
1862 | } | |
1863 | } | |
1864 | ||
1865 | /* | |
f41d911f PM |
1866 | * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used |
1867 | * nowhere else! Assigns leaf node pointers into each CPU's rcu_data | |
1868 | * structure. | |
64db4cff | 1869 | */ |
65cf8f86 | 1870 | #define RCU_INIT_FLAVOR(rsp, rcu_data) \ |
64db4cff | 1871 | do { \ |
a0b6c9a7 PM |
1872 | int i; \ |
1873 | int j; \ | |
1874 | struct rcu_node *rnp; \ | |
1875 | \ | |
65cf8f86 | 1876 | rcu_init_one(rsp); \ |
64db4cff PM |
1877 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
1878 | j = 0; \ | |
1879 | for_each_possible_cpu(i) { \ | |
1880 | if (i > rnp[j].grphi) \ | |
1881 | j++; \ | |
1882 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | |
1883 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | |
65cf8f86 | 1884 | rcu_boot_init_percpu_data(i, rsp); \ |
64db4cff PM |
1885 | } \ |
1886 | } while (0) | |
1887 | ||
9f680ab4 | 1888 | void __init rcu_init(void) |
64db4cff | 1889 | { |
017c4261 | 1890 | int cpu; |
9f680ab4 | 1891 | |
f41d911f | 1892 | rcu_bootup_announce(); |
64db4cff PM |
1893 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
1894 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | |
1895 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
cf244dc0 PM |
1896 | #if NUM_RCU_LVL_4 != 0 |
1897 | printk(KERN_INFO "Experimental four-level hierarchy is enabled.\n"); | |
1898 | #endif /* #if NUM_RCU_LVL_4 != 0 */ | |
65cf8f86 PM |
1899 | RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); |
1900 | RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); | |
f41d911f | 1901 | __rcu_init_preempt(); |
2e597558 | 1902 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
1903 | |
1904 | /* | |
1905 | * We don't need protection against CPU-hotplug here because | |
1906 | * this is called early in boot, before either interrupts | |
1907 | * or the scheduler are operational. | |
1908 | */ | |
1909 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
1910 | for_each_online_cpu(cpu) |
1911 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
64db4cff PM |
1912 | } |
1913 | ||
1eba8f84 | 1914 | #include "rcutree_plugin.h" |