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> | |
bbad9379 | 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 | |
4300aa64 PM |
57 | #define RCU_STATE_INITIALIZER(structname) { \ |
58 | .level = { &structname.node[0] }, \ | |
64db4cff PM |
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, \ | |
4300aa64 | 69 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname.onofflock), \ |
e74f4c45 | 70 | .orphan_cbs_list = NULL, \ |
4300aa64 | 71 | .orphan_cbs_tail = &structname.orphan_cbs_list, \ |
e74f4c45 | 72 | .orphan_qlen = 0, \ |
4300aa64 | 73 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname.fqslock), \ |
64db4cff PM |
74 | .n_force_qs = 0, \ |
75 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 76 | .name = #structname, \ |
64db4cff PM |
77 | } |
78 | ||
d6714c22 PM |
79 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
80 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 81 | |
6258c4fb IM |
82 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
83 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 84 | |
bbad9379 PM |
85 | int rcu_scheduler_active __read_mostly; |
86 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
87 | ||
fc2219d4 PM |
88 | /* |
89 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
90 | * permit this function to be invoked without holding the root rcu_node | |
91 | * structure's ->lock, but of course results can be subject to change. | |
92 | */ | |
93 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
94 | { | |
95 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
96 | } | |
97 | ||
b1f77b05 | 98 | /* |
d6714c22 | 99 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 100 | * how many quiescent states passed, just if there was at least |
d6714c22 | 101 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 102 | */ |
d6714c22 | 103 | void rcu_sched_qs(int cpu) |
b1f77b05 | 104 | { |
25502a6c | 105 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 106 | |
c64ac3ce | 107 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
108 | barrier(); |
109 | rdp->passed_quiesc = 1; | |
b1f77b05 IM |
110 | } |
111 | ||
d6714c22 | 112 | void rcu_bh_qs(int cpu) |
b1f77b05 | 113 | { |
25502a6c | 114 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 115 | |
c64ac3ce | 116 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
117 | barrier(); |
118 | rdp->passed_quiesc = 1; | |
b1f77b05 | 119 | } |
64db4cff | 120 | |
25502a6c PM |
121 | /* |
122 | * Note a context switch. This is a quiescent state for RCU-sched, | |
123 | * and requires special handling for preemptible RCU. | |
124 | */ | |
125 | void rcu_note_context_switch(int cpu) | |
126 | { | |
127 | rcu_sched_qs(cpu); | |
128 | rcu_preempt_note_context_switch(cpu); | |
129 | } | |
130 | ||
64db4cff | 131 | #ifdef CONFIG_NO_HZ |
90a4d2c0 PM |
132 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
133 | .dynticks_nesting = 1, | |
134 | .dynticks = 1, | |
135 | }; | |
64db4cff PM |
136 | #endif /* #ifdef CONFIG_NO_HZ */ |
137 | ||
138 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
139 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
140 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
141 | ||
3d76c082 PM |
142 | module_param(blimit, int, 0); |
143 | module_param(qhimark, int, 0); | |
144 | module_param(qlowmark, int, 0); | |
145 | ||
742734ee | 146 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
910b1b7e | 147 | int rcu_cpu_stall_suppress __read_mostly = RCU_CPU_STALL_SUPPRESS_INIT; |
f2e0dd70 | 148 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee PM |
149 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
150 | ||
64db4cff | 151 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 152 | static int rcu_pending(int cpu); |
64db4cff PM |
153 | |
154 | /* | |
d6714c22 | 155 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 156 | */ |
d6714c22 | 157 | long rcu_batches_completed_sched(void) |
64db4cff | 158 | { |
d6714c22 | 159 | return rcu_sched_state.completed; |
64db4cff | 160 | } |
d6714c22 | 161 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
162 | |
163 | /* | |
164 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
165 | */ | |
166 | long rcu_batches_completed_bh(void) | |
167 | { | |
168 | return rcu_bh_state.completed; | |
169 | } | |
170 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
171 | ||
bf66f18e PM |
172 | /* |
173 | * Force a quiescent state for RCU BH. | |
174 | */ | |
175 | void rcu_bh_force_quiescent_state(void) | |
176 | { | |
177 | force_quiescent_state(&rcu_bh_state, 0); | |
178 | } | |
179 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
180 | ||
181 | /* | |
182 | * Force a quiescent state for RCU-sched. | |
183 | */ | |
184 | void rcu_sched_force_quiescent_state(void) | |
185 | { | |
186 | force_quiescent_state(&rcu_sched_state, 0); | |
187 | } | |
188 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
189 | ||
64db4cff PM |
190 | /* |
191 | * Does the CPU have callbacks ready to be invoked? | |
192 | */ | |
193 | static int | |
194 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
195 | { | |
196 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
197 | } | |
198 | ||
199 | /* | |
200 | * Does the current CPU require a yet-as-unscheduled grace period? | |
201 | */ | |
202 | static int | |
203 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
204 | { | |
fc2219d4 | 205 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
206 | } |
207 | ||
208 | /* | |
209 | * Return the root node of the specified rcu_state structure. | |
210 | */ | |
211 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
212 | { | |
213 | return &rsp->node[0]; | |
214 | } | |
215 | ||
216 | #ifdef CONFIG_SMP | |
217 | ||
218 | /* | |
219 | * If the specified CPU is offline, tell the caller that it is in | |
220 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
221 | * Grace periods can end up waiting on an offline CPU when that | |
222 | * CPU is in the process of coming online -- it will be added to the | |
223 | * rcu_node bitmasks before it actually makes it online. The same thing | |
224 | * can happen while a CPU is in the process of coming online. Because this | |
225 | * race is quite rare, we check for it after detecting that the grace | |
226 | * period has been delayed rather than checking each and every CPU | |
227 | * each and every time we start a new grace period. | |
228 | */ | |
229 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
230 | { | |
231 | /* | |
232 | * If the CPU is offline, it is in a quiescent state. We can | |
233 | * trust its state not to change because interrupts are disabled. | |
234 | */ | |
235 | if (cpu_is_offline(rdp->cpu)) { | |
236 | rdp->offline_fqs++; | |
237 | return 1; | |
238 | } | |
239 | ||
f41d911f PM |
240 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
241 | if (rdp->preemptable) | |
242 | return 0; | |
243 | ||
64db4cff PM |
244 | /* The CPU is online, so send it a reschedule IPI. */ |
245 | if (rdp->cpu != smp_processor_id()) | |
246 | smp_send_reschedule(rdp->cpu); | |
247 | else | |
248 | set_need_resched(); | |
249 | rdp->resched_ipi++; | |
250 | return 0; | |
251 | } | |
252 | ||
253 | #endif /* #ifdef CONFIG_SMP */ | |
254 | ||
255 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
256 | |
257 | /** | |
258 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
259 | * | |
260 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
261 | * read-side critical sections can occur. (Though RCU read-side | |
262 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
263 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
264 | */ | |
265 | void rcu_enter_nohz(void) | |
266 | { | |
267 | unsigned long flags; | |
268 | struct rcu_dynticks *rdtp; | |
269 | ||
270 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
271 | local_irq_save(flags); | |
272 | rdtp = &__get_cpu_var(rcu_dynticks); | |
273 | rdtp->dynticks++; | |
274 | rdtp->dynticks_nesting--; | |
86848966 | 275 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
276 | local_irq_restore(flags); |
277 | } | |
278 | ||
279 | /* | |
280 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
281 | * | |
282 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
283 | * read-side critical sections normally occur. | |
284 | */ | |
285 | void rcu_exit_nohz(void) | |
286 | { | |
287 | unsigned long flags; | |
288 | struct rcu_dynticks *rdtp; | |
289 | ||
290 | local_irq_save(flags); | |
291 | rdtp = &__get_cpu_var(rcu_dynticks); | |
292 | rdtp->dynticks++; | |
293 | rdtp->dynticks_nesting++; | |
86848966 | 294 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
295 | local_irq_restore(flags); |
296 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
297 | } | |
298 | ||
299 | /** | |
300 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
301 | * | |
302 | * If the CPU was idle with dynamic ticks active, and there is no | |
303 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
304 | * RCU grace-period handling know that the CPU is active. | |
305 | */ | |
306 | void rcu_nmi_enter(void) | |
307 | { | |
308 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
309 | ||
310 | if (rdtp->dynticks & 0x1) | |
311 | return; | |
312 | rdtp->dynticks_nmi++; | |
86848966 | 313 | WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1)); |
64db4cff PM |
314 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
315 | } | |
316 | ||
317 | /** | |
318 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
319 | * | |
320 | * If the CPU was idle with dynamic ticks active, and there is no | |
321 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
322 | * RCU grace-period handling know that the CPU is no longer active. | |
323 | */ | |
324 | void rcu_nmi_exit(void) | |
325 | { | |
326 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
327 | ||
328 | if (rdtp->dynticks & 0x1) | |
329 | return; | |
330 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
331 | rdtp->dynticks_nmi++; | |
86848966 | 332 | WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1); |
64db4cff PM |
333 | } |
334 | ||
335 | /** | |
336 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
337 | * | |
338 | * If the CPU was idle with dynamic ticks active, this updates the | |
339 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
340 | */ | |
341 | void rcu_irq_enter(void) | |
342 | { | |
343 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
344 | ||
345 | if (rdtp->dynticks_nesting++) | |
346 | return; | |
347 | rdtp->dynticks++; | |
86848966 | 348 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
349 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
350 | } | |
351 | ||
352 | /** | |
353 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
354 | * | |
355 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
356 | * to put let the RCU handling be aware that the CPU is going back to idle | |
357 | * with no ticks. | |
358 | */ | |
359 | void rcu_irq_exit(void) | |
360 | { | |
361 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
362 | ||
363 | if (--rdtp->dynticks_nesting) | |
364 | return; | |
365 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
366 | rdtp->dynticks++; | |
86848966 | 367 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
368 | |
369 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 370 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
371 | __get_cpu_var(rcu_bh_data).nxtlist) |
372 | set_need_resched(); | |
373 | } | |
374 | ||
64db4cff PM |
375 | #ifdef CONFIG_SMP |
376 | ||
64db4cff PM |
377 | /* |
378 | * Snapshot the specified CPU's dynticks counter so that we can later | |
379 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 380 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
381 | */ |
382 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
383 | { | |
384 | int ret; | |
385 | int snap; | |
386 | int snap_nmi; | |
387 | ||
388 | snap = rdp->dynticks->dynticks; | |
389 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
390 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
391 | rdp->dynticks_snap = snap; | |
392 | rdp->dynticks_nmi_snap = snap_nmi; | |
393 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
394 | if (ret) | |
395 | rdp->dynticks_fqs++; | |
396 | return ret; | |
397 | } | |
398 | ||
399 | /* | |
400 | * Return true if the specified CPU has passed through a quiescent | |
401 | * state by virtue of being in or having passed through an dynticks | |
402 | * idle state since the last call to dyntick_save_progress_counter() | |
403 | * for this same CPU. | |
404 | */ | |
405 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
406 | { | |
407 | long curr; | |
408 | long curr_nmi; | |
409 | long snap; | |
410 | long snap_nmi; | |
411 | ||
412 | curr = rdp->dynticks->dynticks; | |
413 | snap = rdp->dynticks_snap; | |
414 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
415 | snap_nmi = rdp->dynticks_nmi_snap; | |
416 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
417 | ||
418 | /* | |
419 | * If the CPU passed through or entered a dynticks idle phase with | |
420 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
421 | * already acknowledged the request to pass through a quiescent | |
422 | * state. Either way, that CPU cannot possibly be in an RCU | |
423 | * read-side critical section that started before the beginning | |
424 | * of the current RCU grace period. | |
425 | */ | |
426 | if ((curr != snap || (curr & 0x1) == 0) && | |
427 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
428 | rdp->dynticks_fqs++; | |
429 | return 1; | |
430 | } | |
431 | ||
432 | /* Go check for the CPU being offline. */ | |
433 | return rcu_implicit_offline_qs(rdp); | |
434 | } | |
435 | ||
436 | #endif /* #ifdef CONFIG_SMP */ | |
437 | ||
438 | #else /* #ifdef CONFIG_NO_HZ */ | |
439 | ||
64db4cff PM |
440 | #ifdef CONFIG_SMP |
441 | ||
64db4cff PM |
442 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
443 | { | |
444 | return 0; | |
445 | } | |
446 | ||
447 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
448 | { | |
449 | return rcu_implicit_offline_qs(rdp); | |
450 | } | |
451 | ||
452 | #endif /* #ifdef CONFIG_SMP */ | |
453 | ||
454 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
455 | ||
456 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
457 | ||
742734ee | 458 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 459 | |
64db4cff PM |
460 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
461 | { | |
462 | rsp->gp_start = jiffies; | |
463 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
464 | } | |
465 | ||
466 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
467 | { | |
468 | int cpu; | |
469 | long delta; | |
470 | unsigned long flags; | |
471 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
472 | |
473 | /* Only let one CPU complain about others per time interval. */ | |
474 | ||
1304afb2 | 475 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 476 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 477 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 478 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
479 | return; |
480 | } | |
481 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
482 | |
483 | /* | |
484 | * Now rat on any tasks that got kicked up to the root rcu_node | |
485 | * due to CPU offlining. | |
486 | */ | |
487 | rcu_print_task_stall(rnp); | |
1304afb2 | 488 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 489 | |
8cdd32a9 PM |
490 | /* |
491 | * OK, time to rat on our buddy... | |
492 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
493 | * RCU CPU stall warnings. | |
494 | */ | |
4300aa64 PM |
495 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
496 | rsp->name); | |
a0b6c9a7 | 497 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 498 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 499 | rcu_print_task_stall(rnp); |
3acd9eb3 | 500 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 501 | if (rnp->qsmask == 0) |
64db4cff | 502 | continue; |
a0b6c9a7 PM |
503 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
504 | if (rnp->qsmask & (1UL << cpu)) | |
505 | printk(" %d", rnp->grplo + cpu); | |
64db4cff | 506 | } |
4300aa64 | 507 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 508 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
c1dc0b9c IM |
509 | trigger_all_cpu_backtrace(); |
510 | ||
1ed509a2 PM |
511 | /* If so configured, complain about tasks blocking the grace period. */ |
512 | ||
513 | rcu_print_detail_task_stall(rsp); | |
514 | ||
64db4cff PM |
515 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
516 | } | |
517 | ||
518 | static void print_cpu_stall(struct rcu_state *rsp) | |
519 | { | |
520 | unsigned long flags; | |
521 | struct rcu_node *rnp = rcu_get_root(rsp); | |
522 | ||
8cdd32a9 PM |
523 | /* |
524 | * OK, time to rat on ourselves... | |
525 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
526 | * RCU CPU stall warnings. | |
527 | */ | |
4300aa64 PM |
528 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
529 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
530 | trigger_all_cpu_backtrace(); |
531 | ||
1304afb2 | 532 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 533 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
534 | rsp->jiffies_stall = |
535 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 536 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 537 | |
64db4cff PM |
538 | set_need_resched(); /* kick ourselves to get things going. */ |
539 | } | |
540 | ||
541 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
542 | { | |
543 | long delta; | |
544 | struct rcu_node *rnp; | |
545 | ||
742734ee | 546 | if (rcu_cpu_stall_suppress) |
c68de209 | 547 | return; |
64db4cff PM |
548 | delta = jiffies - rsp->jiffies_stall; |
549 | rnp = rdp->mynode; | |
550 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
551 | ||
552 | /* We haven't checked in, so go dump stack. */ | |
553 | print_cpu_stall(rsp); | |
554 | ||
fc2219d4 | 555 | } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) { |
64db4cff PM |
556 | |
557 | /* They had two time units to dump stack, so complain. */ | |
558 | print_other_cpu_stall(rsp); | |
559 | } | |
560 | } | |
561 | ||
c68de209 PM |
562 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
563 | { | |
742734ee | 564 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
565 | return NOTIFY_DONE; |
566 | } | |
567 | ||
568 | static struct notifier_block rcu_panic_block = { | |
569 | .notifier_call = rcu_panic, | |
570 | }; | |
571 | ||
572 | static void __init check_cpu_stall_init(void) | |
573 | { | |
574 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
575 | } | |
576 | ||
64db4cff PM |
577 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
578 | ||
579 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
580 | { | |
581 | } | |
582 | ||
583 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
584 | { | |
585 | } | |
586 | ||
c68de209 PM |
587 | static void __init check_cpu_stall_init(void) |
588 | { | |
589 | } | |
590 | ||
64db4cff PM |
591 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
592 | ||
593 | /* | |
594 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
595 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
596 | * that someone else started the grace period. The caller must hold the |
597 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
598 | * and must have irqs disabled. | |
64db4cff | 599 | */ |
9160306e PM |
600 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
601 | { | |
602 | if (rdp->gpnum != rnp->gpnum) { | |
603 | rdp->qs_pending = 1; | |
604 | rdp->passed_quiesc = 0; | |
605 | rdp->gpnum = rnp->gpnum; | |
606 | } | |
607 | } | |
608 | ||
64db4cff PM |
609 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
610 | { | |
9160306e PM |
611 | unsigned long flags; |
612 | struct rcu_node *rnp; | |
613 | ||
614 | local_irq_save(flags); | |
615 | rnp = rdp->mynode; | |
616 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 617 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
618 | local_irq_restore(flags); |
619 | return; | |
620 | } | |
621 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 622 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
623 | } |
624 | ||
625 | /* | |
626 | * Did someone else start a new RCU grace period start since we last | |
627 | * checked? Update local state appropriately if so. Must be called | |
628 | * on the CPU corresponding to rdp. | |
629 | */ | |
630 | static int | |
631 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
632 | { | |
633 | unsigned long flags; | |
634 | int ret = 0; | |
635 | ||
636 | local_irq_save(flags); | |
637 | if (rdp->gpnum != rsp->gpnum) { | |
638 | note_new_gpnum(rsp, rdp); | |
639 | ret = 1; | |
640 | } | |
641 | local_irq_restore(flags); | |
642 | return ret; | |
643 | } | |
644 | ||
d09b62df PM |
645 | /* |
646 | * Advance this CPU's callbacks, but only if the current grace period | |
647 | * has ended. This may be called only from the CPU to whom the rdp | |
648 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
649 | * ->lock must be held by the caller, with irqs disabled. | |
650 | */ | |
651 | static void | |
652 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
653 | { | |
654 | /* Did another grace period end? */ | |
655 | if (rdp->completed != rnp->completed) { | |
656 | ||
657 | /* Advance callbacks. No harm if list empty. */ | |
658 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
659 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
660 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
661 | ||
662 | /* Remember that we saw this grace-period completion. */ | |
663 | rdp->completed = rnp->completed; | |
664 | } | |
665 | } | |
666 | ||
667 | /* | |
668 | * Advance this CPU's callbacks, but only if the current grace period | |
669 | * has ended. This may be called only from the CPU to whom the rdp | |
670 | * belongs. | |
671 | */ | |
672 | static void | |
673 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
674 | { | |
675 | unsigned long flags; | |
676 | struct rcu_node *rnp; | |
677 | ||
678 | local_irq_save(flags); | |
679 | rnp = rdp->mynode; | |
680 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 681 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
682 | local_irq_restore(flags); |
683 | return; | |
684 | } | |
685 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 686 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
687 | } |
688 | ||
689 | /* | |
690 | * Do per-CPU grace-period initialization for running CPU. The caller | |
691 | * must hold the lock of the leaf rcu_node structure corresponding to | |
692 | * this CPU. | |
693 | */ | |
694 | static void | |
695 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
696 | { | |
697 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
698 | __rcu_process_gp_end(rsp, rnp, rdp); | |
699 | ||
700 | /* | |
701 | * Because this CPU just now started the new grace period, we know | |
702 | * that all of its callbacks will be covered by this upcoming grace | |
703 | * period, even the ones that were registered arbitrarily recently. | |
704 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
705 | * | |
706 | * Other CPUs cannot be sure exactly when the grace period started. | |
707 | * Therefore, their recently registered callbacks must pass through | |
708 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
709 | * by the next RCU grace period. | |
710 | */ | |
711 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
712 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
713 | |
714 | /* Set state so that this CPU will detect the next quiescent state. */ | |
715 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
716 | } |
717 | ||
64db4cff PM |
718 | /* |
719 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
720 | * in preparation for detecting the next grace period. The caller must hold | |
721 | * the root node's ->lock, which is released before return. Hard irqs must | |
722 | * be disabled. | |
723 | */ | |
724 | static void | |
725 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
726 | __releases(rcu_get_root(rsp)->lock) | |
727 | { | |
394f99a9 | 728 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 729 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 730 | |
07079d53 | 731 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
732 | if (cpu_needs_another_gp(rsp, rdp)) |
733 | rsp->fqs_need_gp = 1; | |
b32e9eb6 | 734 | if (rnp->completed == rsp->completed) { |
1304afb2 | 735 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b32e9eb6 PM |
736 | return; |
737 | } | |
1304afb2 | 738 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
739 | |
740 | /* | |
741 | * Propagate new ->completed value to rcu_node structures | |
742 | * so that other CPUs don't have to wait until the start | |
743 | * of the next grace period to process their callbacks. | |
744 | */ | |
745 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1304afb2 | 746 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b32e9eb6 | 747 | rnp->completed = rsp->completed; |
1304afb2 | 748 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
749 | } |
750 | local_irq_restore(flags); | |
64db4cff PM |
751 | return; |
752 | } | |
753 | ||
754 | /* Advance to a new grace period and initialize state. */ | |
755 | rsp->gpnum++; | |
c3422bea | 756 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
757 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
758 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 759 | record_gp_stall_check_time(rsp); |
64db4cff | 760 | |
64db4cff PM |
761 | /* Special-case the common single-level case. */ |
762 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 763 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 764 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 765 | rnp->gpnum = rsp->gpnum; |
d09b62df | 766 | rnp->completed = rsp->completed; |
c12172c0 | 767 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 768 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
1304afb2 | 769 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
770 | return; |
771 | } | |
772 | ||
1304afb2 | 773 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
774 | |
775 | ||
776 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 777 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
778 | |
779 | /* | |
b835db1f PM |
780 | * Set the quiescent-state-needed bits in all the rcu_node |
781 | * structures for all currently online CPUs in breadth-first | |
782 | * order, starting from the root rcu_node structure. This | |
783 | * operation relies on the layout of the hierarchy within the | |
784 | * rsp->node[] array. Note that other CPUs will access only | |
785 | * the leaves of the hierarchy, which still indicate that no | |
786 | * grace period is in progress, at least until the corresponding | |
787 | * leaf node has been initialized. In addition, we have excluded | |
788 | * CPU-hotplug operations. | |
64db4cff PM |
789 | * |
790 | * Note that the grace period cannot complete until we finish | |
791 | * the initialization process, as there will be at least one | |
792 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
793 | * one corresponding to this CPU, due to the fact that we have |
794 | * irqs disabled. | |
64db4cff | 795 | */ |
a0b6c9a7 | 796 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 797 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 798 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 799 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 800 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
801 | rnp->completed = rsp->completed; |
802 | if (rnp == rdp->mynode) | |
803 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
1304afb2 | 804 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
805 | } |
806 | ||
83f5b01f | 807 | rnp = rcu_get_root(rsp); |
1304afb2 | 808 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff | 809 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
810 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
811 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
812 | } |
813 | ||
f41d911f | 814 | /* |
d3f6bad3 PM |
815 | * Report a full set of quiescent states to the specified rcu_state |
816 | * data structure. This involves cleaning up after the prior grace | |
817 | * period and letting rcu_start_gp() start up the next grace period | |
818 | * if one is needed. Note that the caller must hold rnp->lock, as | |
819 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 820 | */ |
d3f6bad3 | 821 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 822 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 823 | { |
fc2219d4 | 824 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
f41d911f | 825 | rsp->completed = rsp->gpnum; |
83f5b01f | 826 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
827 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
828 | } | |
829 | ||
64db4cff | 830 | /* |
d3f6bad3 PM |
831 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
832 | * Allows quiescent states for a group of CPUs to be reported at one go | |
833 | * to the specified rcu_node structure, though all the CPUs in the group | |
834 | * must be represented by the same rcu_node structure (which need not be | |
835 | * a leaf rcu_node structure, though it often will be). That structure's | |
836 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
837 | */ |
838 | static void | |
d3f6bad3 PM |
839 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
840 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
841 | __releases(rnp->lock) |
842 | { | |
28ecd580 PM |
843 | struct rcu_node *rnp_c; |
844 | ||
64db4cff PM |
845 | /* Walk up the rcu_node hierarchy. */ |
846 | for (;;) { | |
847 | if (!(rnp->qsmask & mask)) { | |
848 | ||
849 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 850 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
851 | return; |
852 | } | |
853 | rnp->qsmask &= ~mask; | |
f41d911f | 854 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
855 | |
856 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 857 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
858 | return; |
859 | } | |
860 | mask = rnp->grpmask; | |
861 | if (rnp->parent == NULL) { | |
862 | ||
863 | /* No more levels. Exit loop holding root lock. */ | |
864 | ||
865 | break; | |
866 | } | |
1304afb2 | 867 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 868 | rnp_c = rnp; |
64db4cff | 869 | rnp = rnp->parent; |
1304afb2 | 870 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 871 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
872 | } |
873 | ||
874 | /* | |
875 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 876 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 877 | * to clean up and start the next grace period if one is needed. |
64db4cff | 878 | */ |
d3f6bad3 | 879 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
880 | } |
881 | ||
882 | /* | |
d3f6bad3 PM |
883 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
884 | * structure. This must be either called from the specified CPU, or | |
885 | * called when the specified CPU is known to be offline (and when it is | |
886 | * also known that no other CPU is concurrently trying to help the offline | |
887 | * CPU). The lastcomp argument is used to make sure we are still in the | |
888 | * grace period of interest. We don't want to end the current grace period | |
889 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
890 | */ |
891 | static void | |
d3f6bad3 | 892 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
64db4cff PM |
893 | { |
894 | unsigned long flags; | |
895 | unsigned long mask; | |
896 | struct rcu_node *rnp; | |
897 | ||
898 | rnp = rdp->mynode; | |
1304afb2 | 899 | raw_spin_lock_irqsave(&rnp->lock, flags); |
560d4bc0 | 900 | if (lastcomp != rnp->completed) { |
64db4cff PM |
901 | |
902 | /* | |
903 | * Someone beat us to it for this grace period, so leave. | |
904 | * The race with GP start is resolved by the fact that we | |
905 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
906 | * cannot yet be initialized -- so we would simply find our | |
d3f6bad3 PM |
907 | * CPU's bit already cleared in rcu_report_qs_rnp() if this |
908 | * race occurred. | |
64db4cff PM |
909 | */ |
910 | rdp->passed_quiesc = 0; /* try again later! */ | |
1304afb2 | 911 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
912 | return; |
913 | } | |
914 | mask = rdp->grpmask; | |
915 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 916 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
917 | } else { |
918 | rdp->qs_pending = 0; | |
919 | ||
920 | /* | |
921 | * This GP can't end until cpu checks in, so all of our | |
922 | * callbacks can be processed during the next GP. | |
923 | */ | |
64db4cff PM |
924 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
925 | ||
d3f6bad3 | 926 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
927 | } |
928 | } | |
929 | ||
930 | /* | |
931 | * Check to see if there is a new grace period of which this CPU | |
932 | * is not yet aware, and if so, set up local rcu_data state for it. | |
933 | * Otherwise, see if this CPU has just passed through its first | |
934 | * quiescent state for this grace period, and record that fact if so. | |
935 | */ | |
936 | static void | |
937 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
938 | { | |
939 | /* If there is now a new grace period, record and return. */ | |
940 | if (check_for_new_grace_period(rsp, rdp)) | |
941 | return; | |
942 | ||
943 | /* | |
944 | * Does this CPU still need to do its part for current grace period? | |
945 | * If no, return and let the other CPUs do their part as well. | |
946 | */ | |
947 | if (!rdp->qs_pending) | |
948 | return; | |
949 | ||
950 | /* | |
951 | * Was there a quiescent state since the beginning of the grace | |
952 | * period? If no, then exit and wait for the next call. | |
953 | */ | |
954 | if (!rdp->passed_quiesc) | |
955 | return; | |
956 | ||
d3f6bad3 PM |
957 | /* |
958 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
959 | * judge of that). | |
960 | */ | |
961 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
64db4cff PM |
962 | } |
963 | ||
964 | #ifdef CONFIG_HOTPLUG_CPU | |
965 | ||
e74f4c45 PM |
966 | /* |
967 | * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the | |
968 | * specified flavor of RCU. The callbacks will be adopted by the next | |
969 | * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever | |
970 | * comes first. Because this is invoked from the CPU_DYING notifier, | |
971 | * irqs are already disabled. | |
972 | */ | |
973 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) | |
974 | { | |
975 | int i; | |
394f99a9 | 976 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
e74f4c45 PM |
977 | |
978 | if (rdp->nxtlist == NULL) | |
979 | return; /* irqs disabled, so comparison is stable. */ | |
1304afb2 | 980 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
e74f4c45 PM |
981 | *rsp->orphan_cbs_tail = rdp->nxtlist; |
982 | rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL]; | |
983 | rdp->nxtlist = NULL; | |
984 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
985 | rdp->nxttail[i] = &rdp->nxtlist; | |
986 | rsp->orphan_qlen += rdp->qlen; | |
987 | rdp->qlen = 0; | |
1304afb2 | 988 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
e74f4c45 PM |
989 | } |
990 | ||
991 | /* | |
992 | * Adopt previously orphaned RCU callbacks. | |
993 | */ | |
994 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
995 | { | |
996 | unsigned long flags; | |
997 | struct rcu_data *rdp; | |
998 | ||
1304afb2 | 999 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
394f99a9 | 1000 | rdp = this_cpu_ptr(rsp->rda); |
e74f4c45 | 1001 | if (rsp->orphan_cbs_list == NULL) { |
1304afb2 | 1002 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
e74f4c45 PM |
1003 | return; |
1004 | } | |
1005 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list; | |
1006 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail; | |
1007 | rdp->qlen += rsp->orphan_qlen; | |
1008 | rsp->orphan_cbs_list = NULL; | |
1009 | rsp->orphan_cbs_tail = &rsp->orphan_cbs_list; | |
1010 | rsp->orphan_qlen = 0; | |
1304afb2 | 1011 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
e74f4c45 PM |
1012 | } |
1013 | ||
64db4cff PM |
1014 | /* |
1015 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1016 | * and move all callbacks from the outgoing CPU to the current one. | |
1017 | */ | |
1018 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1019 | { | |
64db4cff | 1020 | unsigned long flags; |
64db4cff | 1021 | unsigned long mask; |
d9a3da06 | 1022 | int need_report = 0; |
394f99a9 | 1023 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
1024 | struct rcu_node *rnp; |
1025 | ||
1026 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1027 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1028 | |
1029 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1030 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1031 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1032 | do { | |
1304afb2 | 1033 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1034 | rnp->qsmaskinit &= ~mask; |
1035 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1036 | if (rnp != rdp->mynode) |
1304afb2 | 1037 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1038 | break; |
1039 | } | |
b668c9cf | 1040 | if (rnp == rdp->mynode) |
d9a3da06 | 1041 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
b668c9cf | 1042 | else |
1304afb2 | 1043 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1044 | mask = rnp->grpmask; |
64db4cff PM |
1045 | rnp = rnp->parent; |
1046 | } while (rnp != NULL); | |
64db4cff | 1047 | |
b668c9cf PM |
1048 | /* |
1049 | * We still hold the leaf rcu_node structure lock here, and | |
1050 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1051 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1052 | * held leads to deadlock. | |
b668c9cf | 1053 | */ |
1304afb2 | 1054 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1055 | rnp = rdp->mynode; |
d9a3da06 | 1056 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1057 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1058 | else |
1304afb2 | 1059 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
1060 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1061 | rcu_report_exp_rnp(rsp, rnp); | |
64db4cff | 1062 | |
e74f4c45 | 1063 | rcu_adopt_orphan_cbs(rsp); |
64db4cff PM |
1064 | } |
1065 | ||
1066 | /* | |
1067 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1068 | * callbacks that it might have to the current CPU. This code assumes | |
1069 | * that at least one CPU in the system will remain running at all times. | |
1070 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1071 | */ | |
1072 | static void rcu_offline_cpu(int cpu) | |
1073 | { | |
d6714c22 | 1074 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1075 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1076 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1077 | } |
1078 | ||
1079 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1080 | ||
e74f4c45 PM |
1081 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) |
1082 | { | |
1083 | } | |
1084 | ||
1085 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
1086 | { | |
1087 | } | |
1088 | ||
64db4cff PM |
1089 | static void rcu_offline_cpu(int cpu) |
1090 | { | |
1091 | } | |
1092 | ||
1093 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1094 | ||
1095 | /* | |
1096 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1097 | * period. Thottle as specified by rdp->blimit. | |
1098 | */ | |
37c72e56 | 1099 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1100 | { |
1101 | unsigned long flags; | |
1102 | struct rcu_head *next, *list, **tail; | |
1103 | int count; | |
1104 | ||
1105 | /* If no callbacks are ready, just return.*/ | |
1106 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
1107 | return; | |
1108 | ||
1109 | /* | |
1110 | * Extract the list of ready callbacks, disabling to prevent | |
1111 | * races with call_rcu() from interrupt handlers. | |
1112 | */ | |
1113 | local_irq_save(flags); | |
1114 | list = rdp->nxtlist; | |
1115 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1116 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1117 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1118 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1119 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1120 | rdp->nxttail[count] = &rdp->nxtlist; | |
1121 | local_irq_restore(flags); | |
1122 | ||
1123 | /* Invoke callbacks. */ | |
1124 | count = 0; | |
1125 | while (list) { | |
1126 | next = list->next; | |
1127 | prefetch(next); | |
551d55a9 | 1128 | debug_rcu_head_unqueue(list); |
64db4cff PM |
1129 | list->func(list); |
1130 | list = next; | |
1131 | if (++count >= rdp->blimit) | |
1132 | break; | |
1133 | } | |
1134 | ||
1135 | local_irq_save(flags); | |
1136 | ||
1137 | /* Update count, and requeue any remaining callbacks. */ | |
1138 | rdp->qlen -= count; | |
1139 | if (list != NULL) { | |
1140 | *tail = rdp->nxtlist; | |
1141 | rdp->nxtlist = list; | |
1142 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1143 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1144 | rdp->nxttail[count] = tail; | |
1145 | else | |
1146 | break; | |
1147 | } | |
1148 | ||
1149 | /* Reinstate batch limit if we have worked down the excess. */ | |
1150 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1151 | rdp->blimit = blimit; | |
1152 | ||
37c72e56 PM |
1153 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1154 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1155 | rdp->qlen_last_fqs_check = 0; | |
1156 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1157 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1158 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1159 | ||
64db4cff PM |
1160 | local_irq_restore(flags); |
1161 | ||
1162 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1163 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
1164 | raise_softirq(RCU_SOFTIRQ); | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1169 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1170 | * Also schedule the RCU softirq handler. | |
1171 | * | |
1172 | * This function must be called with hardirqs disabled. It is normally | |
1173 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1174 | * false, there is no point in invoking rcu_check_callbacks(). | |
1175 | */ | |
1176 | void rcu_check_callbacks(int cpu, int user) | |
1177 | { | |
1178 | if (user || | |
a6826048 PM |
1179 | (idle_cpu(cpu) && rcu_scheduler_active && |
1180 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1181 | |
1182 | /* | |
1183 | * Get here if this CPU took its interrupt from user | |
1184 | * mode or from the idle loop, and if this is not a | |
1185 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1186 | * a quiescent state, so note it. |
64db4cff PM |
1187 | * |
1188 | * No memory barrier is required here because both | |
d6714c22 PM |
1189 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1190 | * variables that other CPUs neither access nor modify, | |
1191 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1192 | */ |
1193 | ||
d6714c22 PM |
1194 | rcu_sched_qs(cpu); |
1195 | rcu_bh_qs(cpu); | |
64db4cff PM |
1196 | |
1197 | } else if (!in_softirq()) { | |
1198 | ||
1199 | /* | |
1200 | * Get here if this CPU did not take its interrupt from | |
1201 | * softirq, in other words, if it is not interrupting | |
1202 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1203 | * critical section, so note it. |
64db4cff PM |
1204 | */ |
1205 | ||
d6714c22 | 1206 | rcu_bh_qs(cpu); |
64db4cff | 1207 | } |
f41d911f | 1208 | rcu_preempt_check_callbacks(cpu); |
d21670ac PM |
1209 | if (rcu_pending(cpu)) |
1210 | raise_softirq(RCU_SOFTIRQ); | |
64db4cff PM |
1211 | } |
1212 | ||
1213 | #ifdef CONFIG_SMP | |
1214 | ||
1215 | /* | |
1216 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1217 | * have not yet encountered a quiescent state, using the function specified. | |
ee47eb9f | 1218 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1219 | */ |
45f014c5 | 1220 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1221 | { |
1222 | unsigned long bit; | |
1223 | int cpu; | |
1224 | unsigned long flags; | |
1225 | unsigned long mask; | |
a0b6c9a7 | 1226 | struct rcu_node *rnp; |
64db4cff | 1227 | |
a0b6c9a7 | 1228 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1229 | mask = 0; |
1304afb2 | 1230 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1231 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1232 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1233 | return; |
64db4cff | 1234 | } |
a0b6c9a7 | 1235 | if (rnp->qsmask == 0) { |
1304afb2 | 1236 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1237 | continue; |
1238 | } | |
a0b6c9a7 | 1239 | cpu = rnp->grplo; |
64db4cff | 1240 | bit = 1; |
a0b6c9a7 | 1241 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1242 | if ((rnp->qsmask & bit) != 0 && |
1243 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1244 | mask |= bit; |
1245 | } | |
45f014c5 | 1246 | if (mask != 0) { |
64db4cff | 1247 | |
d3f6bad3 PM |
1248 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1249 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1250 | continue; |
1251 | } | |
1304afb2 | 1252 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1253 | } |
64db4cff PM |
1254 | } |
1255 | ||
1256 | /* | |
1257 | * Force quiescent states on reluctant CPUs, and also detect which | |
1258 | * CPUs are in dyntick-idle mode. | |
1259 | */ | |
1260 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1261 | { | |
1262 | unsigned long flags; | |
64db4cff | 1263 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1264 | |
fc2219d4 | 1265 | if (!rcu_gp_in_progress(rsp)) |
64db4cff | 1266 | return; /* No grace period in progress, nothing to force. */ |
1304afb2 | 1267 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff PM |
1268 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1269 | return; /* Someone else is already on the job. */ | |
1270 | } | |
20133cfc | 1271 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1272 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1273 | rsp->n_force_qs++; |
1304afb2 | 1274 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1275 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1276 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1277 | rsp->n_force_qs_ngp++; |
1304afb2 | 1278 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1279 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1280 | } |
07079d53 | 1281 | rsp->fqs_active = 1; |
f3a8b5c6 | 1282 | switch (rsp->signaled) { |
83f5b01f | 1283 | case RCU_GP_IDLE: |
64db4cff PM |
1284 | case RCU_GP_INIT: |
1285 | ||
83f5b01f | 1286 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1287 | |
1288 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1289 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1290 | break; /* So gcc recognizes the dead code. */ | |
1291 | ||
f261414f LJ |
1292 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1293 | ||
64db4cff | 1294 | /* Record dyntick-idle state. */ |
45f014c5 | 1295 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1296 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1297 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1298 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1299 | break; |
64db4cff PM |
1300 | |
1301 | case RCU_FORCE_QS: | |
1302 | ||
1303 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1304 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1305 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1306 | |
1307 | /* Leave state in case more forcing is required. */ | |
1308 | ||
1304afb2 | 1309 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1310 | break; |
64db4cff | 1311 | } |
07079d53 | 1312 | rsp->fqs_active = 0; |
46a1e34e | 1313 | if (rsp->fqs_need_gp) { |
1304afb2 | 1314 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1315 | rsp->fqs_need_gp = 0; |
1316 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
1317 | return; | |
1318 | } | |
1304afb2 | 1319 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1320 | unlock_fqs_ret: |
1304afb2 | 1321 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
64db4cff PM |
1322 | } |
1323 | ||
1324 | #else /* #ifdef CONFIG_SMP */ | |
1325 | ||
1326 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1327 | { | |
1328 | set_need_resched(); | |
1329 | } | |
1330 | ||
1331 | #endif /* #else #ifdef CONFIG_SMP */ | |
1332 | ||
1333 | /* | |
1334 | * This does the RCU processing work from softirq context for the | |
1335 | * specified rcu_state and rcu_data structures. This may be called | |
1336 | * only from the CPU to whom the rdp belongs. | |
1337 | */ | |
1338 | static void | |
1339 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1340 | { | |
1341 | unsigned long flags; | |
1342 | ||
2e597558 PM |
1343 | WARN_ON_ONCE(rdp->beenonline == 0); |
1344 | ||
64db4cff PM |
1345 | /* |
1346 | * If an RCU GP has gone long enough, go check for dyntick | |
1347 | * idle CPUs and, if needed, send resched IPIs. | |
1348 | */ | |
20133cfc | 1349 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1350 | force_quiescent_state(rsp, 1); |
1351 | ||
1352 | /* | |
1353 | * Advance callbacks in response to end of earlier grace | |
1354 | * period that some other CPU ended. | |
1355 | */ | |
1356 | rcu_process_gp_end(rsp, rdp); | |
1357 | ||
1358 | /* Update RCU state based on any recent quiescent states. */ | |
1359 | rcu_check_quiescent_state(rsp, rdp); | |
1360 | ||
1361 | /* Does this CPU require a not-yet-started grace period? */ | |
1362 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1363 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1364 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1365 | } | |
1366 | ||
1367 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1368 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1369 | } |
1370 | ||
1371 | /* | |
1372 | * Do softirq processing for the current CPU. | |
1373 | */ | |
1374 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1375 | { | |
1376 | /* | |
1377 | * Memory references from any prior RCU read-side critical sections | |
1378 | * executed by the interrupted code must be seen before any RCU | |
1379 | * grace-period manipulations below. | |
1380 | */ | |
1381 | smp_mb(); /* See above block comment. */ | |
1382 | ||
d6714c22 PM |
1383 | __rcu_process_callbacks(&rcu_sched_state, |
1384 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1385 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1386 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1387 | |
1388 | /* | |
1389 | * Memory references from any later RCU read-side critical sections | |
1390 | * executed by the interrupted code must be seen after any RCU | |
1391 | * grace-period manipulations above. | |
1392 | */ | |
1393 | smp_mb(); /* See above block comment. */ | |
a47cd880 PM |
1394 | |
1395 | /* If we are last CPU on way to dyntick-idle mode, accelerate it. */ | |
1396 | rcu_needs_cpu_flush(); | |
64db4cff PM |
1397 | } |
1398 | ||
1399 | static void | |
1400 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1401 | struct rcu_state *rsp) | |
1402 | { | |
1403 | unsigned long flags; | |
1404 | struct rcu_data *rdp; | |
1405 | ||
551d55a9 | 1406 | debug_rcu_head_queue(head); |
64db4cff PM |
1407 | head->func = func; |
1408 | head->next = NULL; | |
1409 | ||
1410 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1411 | ||
1412 | /* | |
1413 | * Opportunistically note grace-period endings and beginnings. | |
1414 | * Note that we might see a beginning right after we see an | |
1415 | * end, but never vice versa, since this CPU has to pass through | |
1416 | * a quiescent state betweentimes. | |
1417 | */ | |
1418 | local_irq_save(flags); | |
394f99a9 | 1419 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1420 | rcu_process_gp_end(rsp, rdp); |
1421 | check_for_new_grace_period(rsp, rdp); | |
1422 | ||
1423 | /* Add the callback to our list. */ | |
1424 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1425 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1426 | ||
1427 | /* Start a new grace period if one not already started. */ | |
fc2219d4 | 1428 | if (!rcu_gp_in_progress(rsp)) { |
64db4cff PM |
1429 | unsigned long nestflag; |
1430 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1431 | ||
1304afb2 | 1432 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); |
64db4cff PM |
1433 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ |
1434 | } | |
1435 | ||
37c72e56 PM |
1436 | /* |
1437 | * Force the grace period if too many callbacks or too long waiting. | |
1438 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1439 | * if some other CPU has recently done so. Also, don't bother | |
1440 | * invoking force_quiescent_state() if the newly enqueued callback | |
1441 | * is the only one waiting for a grace period to complete. | |
1442 | */ | |
1443 | if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { | |
64db4cff | 1444 | rdp->blimit = LONG_MAX; |
37c72e56 PM |
1445 | if (rsp->n_force_qs == rdp->n_force_qs_snap && |
1446 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1447 | force_quiescent_state(rsp, 0); | |
1448 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1449 | rdp->qlen_last_fqs_check = rdp->qlen; | |
20133cfc | 1450 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1451 | force_quiescent_state(rsp, 1); |
1452 | local_irq_restore(flags); | |
1453 | } | |
1454 | ||
1455 | /* | |
d6714c22 | 1456 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1457 | */ |
d6714c22 | 1458 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1459 | { |
d6714c22 | 1460 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1461 | } |
d6714c22 | 1462 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1463 | |
1464 | /* | |
1465 | * Queue an RCU for invocation after a quicker grace period. | |
1466 | */ | |
1467 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1468 | { | |
1469 | __call_rcu(head, func, &rcu_bh_state); | |
1470 | } | |
1471 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1472 | ||
6ebb237b PM |
1473 | /** |
1474 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1475 | * | |
1476 | * Control will return to the caller some time after a full rcu-sched | |
1477 | * grace period has elapsed, in other words after all currently executing | |
1478 | * rcu-sched read-side critical sections have completed. These read-side | |
1479 | * critical sections are delimited by rcu_read_lock_sched() and | |
1480 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1481 | * local_irq_disable(), and so on may be used in place of | |
1482 | * rcu_read_lock_sched(). | |
1483 | * | |
1484 | * This means that all preempt_disable code sequences, including NMI and | |
1485 | * hardware-interrupt handlers, in progress on entry will have completed | |
1486 | * before this primitive returns. However, this does not guarantee that | |
1487 | * softirq handlers will have completed, since in some kernels, these | |
1488 | * handlers can run in process context, and can block. | |
1489 | * | |
1490 | * This primitive provides the guarantees made by the (now removed) | |
1491 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1492 | * guarantees that rcu_read_lock() sections will have completed. | |
1493 | * In "classic RCU", these two guarantees happen to be one and | |
1494 | * the same, but can differ in realtime RCU implementations. | |
1495 | */ | |
1496 | void synchronize_sched(void) | |
1497 | { | |
1498 | struct rcu_synchronize rcu; | |
1499 | ||
1500 | if (rcu_blocking_is_gp()) | |
1501 | return; | |
1502 | ||
72d5a9f7 | 1503 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1504 | init_completion(&rcu.completion); |
1505 | /* Will wake me after RCU finished. */ | |
1506 | call_rcu_sched(&rcu.head, wakeme_after_rcu); | |
1507 | /* Wait for it. */ | |
1508 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1509 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1510 | } |
1511 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1512 | ||
1513 | /** | |
1514 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1515 | * | |
1516 | * Control will return to the caller some time after a full rcu_bh grace | |
1517 | * period has elapsed, in other words after all currently executing rcu_bh | |
1518 | * read-side critical sections have completed. RCU read-side critical | |
1519 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1520 | * and may be nested. | |
1521 | */ | |
1522 | void synchronize_rcu_bh(void) | |
1523 | { | |
1524 | struct rcu_synchronize rcu; | |
1525 | ||
1526 | if (rcu_blocking_is_gp()) | |
1527 | return; | |
1528 | ||
72d5a9f7 | 1529 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1530 | init_completion(&rcu.completion); |
1531 | /* Will wake me after RCU finished. */ | |
1532 | call_rcu_bh(&rcu.head, wakeme_after_rcu); | |
1533 | /* Wait for it. */ | |
1534 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1535 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1536 | } |
1537 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1538 | ||
64db4cff PM |
1539 | /* |
1540 | * Check to see if there is any immediate RCU-related work to be done | |
1541 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1542 | * The checks are in order of increasing expense: checks that can be | |
1543 | * carried out against CPU-local state are performed first. However, | |
1544 | * we must check for CPU stalls first, else we might not get a chance. | |
1545 | */ | |
1546 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1547 | { | |
2f51f988 PM |
1548 | struct rcu_node *rnp = rdp->mynode; |
1549 | ||
64db4cff PM |
1550 | rdp->n_rcu_pending++; |
1551 | ||
1552 | /* Check for CPU stalls, if enabled. */ | |
1553 | check_cpu_stall(rsp, rdp); | |
1554 | ||
1555 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
d21670ac | 1556 | if (rdp->qs_pending && !rdp->passed_quiesc) { |
d25eb944 PM |
1557 | |
1558 | /* | |
1559 | * If force_quiescent_state() coming soon and this CPU | |
1560 | * needs a quiescent state, and this is either RCU-sched | |
1561 | * or RCU-bh, force a local reschedule. | |
1562 | */ | |
d21670ac | 1563 | rdp->n_rp_qs_pending++; |
d25eb944 PM |
1564 | if (!rdp->preemptable && |
1565 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, | |
1566 | jiffies)) | |
1567 | set_need_resched(); | |
d21670ac PM |
1568 | } else if (rdp->qs_pending && rdp->passed_quiesc) { |
1569 | rdp->n_rp_report_qs++; | |
64db4cff | 1570 | return 1; |
7ba5c840 | 1571 | } |
64db4cff PM |
1572 | |
1573 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1574 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1575 | rdp->n_rp_cb_ready++; | |
64db4cff | 1576 | return 1; |
7ba5c840 | 1577 | } |
64db4cff PM |
1578 | |
1579 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1580 | if (cpu_needs_another_gp(rsp, rdp)) { |
1581 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1582 | return 1; |
7ba5c840 | 1583 | } |
64db4cff PM |
1584 | |
1585 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1586 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1587 | rdp->n_rp_gp_completed++; |
64db4cff | 1588 | return 1; |
7ba5c840 | 1589 | } |
64db4cff PM |
1590 | |
1591 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1592 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1593 | rdp->n_rp_gp_started++; |
64db4cff | 1594 | return 1; |
7ba5c840 | 1595 | } |
64db4cff PM |
1596 | |
1597 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1598 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 1599 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 1600 | rdp->n_rp_need_fqs++; |
64db4cff | 1601 | return 1; |
7ba5c840 | 1602 | } |
64db4cff PM |
1603 | |
1604 | /* nothing to do */ | |
7ba5c840 | 1605 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1606 | return 0; |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * Check to see if there is any immediate RCU-related work to be done | |
1611 | * by the current CPU, returning 1 if so. This function is part of the | |
1612 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1613 | */ | |
a157229c | 1614 | static int rcu_pending(int cpu) |
64db4cff | 1615 | { |
d6714c22 | 1616 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1617 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1618 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1619 | } |
1620 | ||
1621 | /* | |
1622 | * Check to see if any future RCU-related work will need to be done | |
1623 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1624 | * 1 if so. |
64db4cff | 1625 | */ |
8bd93a2c | 1626 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
1627 | { |
1628 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1629 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1630 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1631 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1632 | } |
1633 | ||
d0ec774c PM |
1634 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1635 | static atomic_t rcu_barrier_cpu_count; | |
1636 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1637 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1638 | |
1639 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1640 | { | |
1641 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1642 | complete(&rcu_barrier_completion); | |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1647 | */ | |
1648 | static void rcu_barrier_func(void *type) | |
1649 | { | |
1650 | int cpu = smp_processor_id(); | |
1651 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1652 | void (*call_rcu_func)(struct rcu_head *head, | |
1653 | void (*func)(struct rcu_head *head)); | |
1654 | ||
1655 | atomic_inc(&rcu_barrier_cpu_count); | |
1656 | call_rcu_func = type; | |
1657 | call_rcu_func(head, rcu_barrier_callback); | |
1658 | } | |
1659 | ||
d0ec774c PM |
1660 | /* |
1661 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1662 | * RCU callbacks of the specified type to complete. | |
1663 | */ | |
e74f4c45 PM |
1664 | static void _rcu_barrier(struct rcu_state *rsp, |
1665 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1666 | void (*func)(struct rcu_head *head))) |
1667 | { | |
1668 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1669 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1670 | mutex_lock(&rcu_barrier_mutex); |
1671 | init_completion(&rcu_barrier_completion); | |
1672 | /* | |
1673 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1674 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1675 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1676 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1677 | * might complete its grace period before all of the other CPUs | |
1678 | * did their increment, causing this function to return too | |
1679 | * early. | |
1680 | */ | |
1681 | atomic_set(&rcu_barrier_cpu_count, 1); | |
e74f4c45 PM |
1682 | preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */ |
1683 | rcu_adopt_orphan_cbs(rsp); | |
d0ec774c | 1684 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); |
e74f4c45 | 1685 | preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */ |
d0ec774c PM |
1686 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1687 | complete(&rcu_barrier_completion); | |
1688 | wait_for_completion(&rcu_barrier_completion); | |
1689 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1690 | } |
d0ec774c PM |
1691 | |
1692 | /** | |
1693 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1694 | */ | |
1695 | void rcu_barrier_bh(void) | |
1696 | { | |
e74f4c45 | 1697 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1698 | } |
1699 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
1700 | ||
1701 | /** | |
1702 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
1703 | */ | |
1704 | void rcu_barrier_sched(void) | |
1705 | { | |
e74f4c45 | 1706 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
1707 | } |
1708 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
1709 | ||
64db4cff | 1710 | /* |
27569620 | 1711 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1712 | */ |
27569620 PM |
1713 | static void __init |
1714 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1715 | { |
1716 | unsigned long flags; | |
1717 | int i; | |
394f99a9 | 1718 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
1719 | struct rcu_node *rnp = rcu_get_root(rsp); |
1720 | ||
1721 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 1722 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
1723 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
1724 | rdp->nxtlist = NULL; | |
1725 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1726 | rdp->nxttail[i] = &rdp->nxtlist; | |
1727 | rdp->qlen = 0; | |
1728 | #ifdef CONFIG_NO_HZ | |
1729 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1730 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1731 | rdp->cpu = cpu; | |
1304afb2 | 1732 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
1733 | } |
1734 | ||
1735 | /* | |
1736 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1737 | * offline event can be happening at a given time. Note also that we | |
1738 | * can accept some slop in the rsp->completed access due to the fact | |
1739 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1740 | */ |
e4fa4c97 | 1741 | static void __cpuinit |
f41d911f | 1742 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1743 | { |
1744 | unsigned long flags; | |
64db4cff | 1745 | unsigned long mask; |
394f99a9 | 1746 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
1747 | struct rcu_node *rnp = rcu_get_root(rsp); |
1748 | ||
1749 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 1750 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff PM |
1751 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
1752 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1753 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1754 | rdp->preemptable = preemptable; |
37c72e56 PM |
1755 | rdp->qlen_last_fqs_check = 0; |
1756 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 1757 | rdp->blimit = blimit; |
1304afb2 | 1758 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1759 | |
1760 | /* | |
1761 | * A new grace period might start here. If so, we won't be part | |
1762 | * of it, but that is OK, as we are currently in a quiescent state. | |
1763 | */ | |
1764 | ||
1765 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 1766 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1767 | |
1768 | /* Add CPU to rcu_node bitmasks. */ | |
1769 | rnp = rdp->mynode; | |
1770 | mask = rdp->grpmask; | |
1771 | do { | |
1772 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 1773 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1774 | rnp->qsmaskinit |= mask; |
1775 | mask = rnp->grpmask; | |
d09b62df PM |
1776 | if (rnp == rdp->mynode) { |
1777 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
1778 | rdp->completed = rnp->completed; | |
1779 | rdp->passed_quiesc_completed = rnp->completed - 1; | |
1780 | } | |
1304afb2 | 1781 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1782 | rnp = rnp->parent; |
1783 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1784 | ||
1304afb2 | 1785 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
1786 | } |
1787 | ||
1788 | static void __cpuinit rcu_online_cpu(int cpu) | |
1789 | { | |
f41d911f PM |
1790 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1791 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1792 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1793 | } |
1794 | ||
1795 | /* | |
f41d911f | 1796 | * Handle CPU online/offline notification events. |
64db4cff | 1797 | */ |
9f680ab4 PM |
1798 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1799 | unsigned long action, void *hcpu) | |
64db4cff PM |
1800 | { |
1801 | long cpu = (long)hcpu; | |
1802 | ||
1803 | switch (action) { | |
1804 | case CPU_UP_PREPARE: | |
1805 | case CPU_UP_PREPARE_FROZEN: | |
1806 | rcu_online_cpu(cpu); | |
1807 | break; | |
d0ec774c PM |
1808 | case CPU_DYING: |
1809 | case CPU_DYING_FROZEN: | |
1810 | /* | |
e74f4c45 | 1811 | * preempt_disable() in _rcu_barrier() prevents stop_machine(), |
d0ec774c | 1812 | * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);" |
e74f4c45 PM |
1813 | * returns, all online cpus have queued rcu_barrier_func(). |
1814 | * The dying CPU clears its cpu_online_mask bit and | |
1815 | * moves all of its RCU callbacks to ->orphan_cbs_list | |
1816 | * in the context of stop_machine(), so subsequent calls | |
1817 | * to _rcu_barrier() will adopt these callbacks and only | |
1818 | * then queue rcu_barrier_func() on all remaining CPUs. | |
d0ec774c | 1819 | */ |
e74f4c45 PM |
1820 | rcu_send_cbs_to_orphanage(&rcu_bh_state); |
1821 | rcu_send_cbs_to_orphanage(&rcu_sched_state); | |
1822 | rcu_preempt_send_cbs_to_orphanage(); | |
d0ec774c | 1823 | break; |
64db4cff PM |
1824 | case CPU_DEAD: |
1825 | case CPU_DEAD_FROZEN: | |
1826 | case CPU_UP_CANCELED: | |
1827 | case CPU_UP_CANCELED_FROZEN: | |
1828 | rcu_offline_cpu(cpu); | |
1829 | break; | |
1830 | default: | |
1831 | break; | |
1832 | } | |
1833 | return NOTIFY_OK; | |
1834 | } | |
1835 | ||
bbad9379 PM |
1836 | /* |
1837 | * This function is invoked towards the end of the scheduler's initialization | |
1838 | * process. Before this is called, the idle task might contain | |
1839 | * RCU read-side critical sections (during which time, this idle | |
1840 | * task is booting the system). After this function is called, the | |
1841 | * idle tasks are prohibited from containing RCU read-side critical | |
1842 | * sections. This function also enables RCU lockdep checking. | |
1843 | */ | |
1844 | void rcu_scheduler_starting(void) | |
1845 | { | |
1846 | WARN_ON(num_online_cpus() != 1); | |
1847 | WARN_ON(nr_context_switches() > 0); | |
1848 | rcu_scheduler_active = 1; | |
1849 | } | |
1850 | ||
64db4cff PM |
1851 | /* |
1852 | * Compute the per-level fanout, either using the exact fanout specified | |
1853 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1854 | */ | |
1855 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1856 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1857 | { | |
1858 | int i; | |
1859 | ||
1860 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1861 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1862 | } | |
1863 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1864 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1865 | { | |
1866 | int ccur; | |
1867 | int cprv; | |
1868 | int i; | |
1869 | ||
1870 | cprv = NR_CPUS; | |
1871 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1872 | ccur = rsp->levelcnt[i]; | |
1873 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1874 | cprv = ccur; | |
1875 | } | |
1876 | } | |
1877 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1878 | ||
1879 | /* | |
1880 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1881 | */ | |
394f99a9 LJ |
1882 | static void __init rcu_init_one(struct rcu_state *rsp, |
1883 | struct rcu_data __percpu *rda) | |
64db4cff | 1884 | { |
b6407e86 PM |
1885 | static char *buf[] = { "rcu_node_level_0", |
1886 | "rcu_node_level_1", | |
1887 | "rcu_node_level_2", | |
1888 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
1889 | int cpustride = 1; |
1890 | int i; | |
1891 | int j; | |
1892 | struct rcu_node *rnp; | |
1893 | ||
b6407e86 PM |
1894 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
1895 | ||
64db4cff PM |
1896 | /* Initialize the level-tracking arrays. */ |
1897 | ||
1898 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1899 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1900 | rcu_init_levelspread(rsp); | |
1901 | ||
1902 | /* Initialize the elements themselves, starting from the leaves. */ | |
1903 | ||
1904 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1905 | cpustride *= rsp->levelspread[i]; | |
1906 | rnp = rsp->level[i]; | |
1907 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 1908 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
1909 | lockdep_set_class_and_name(&rnp->lock, |
1910 | &rcu_node_class[i], buf[i]); | |
f41d911f | 1911 | rnp->gpnum = 0; |
64db4cff PM |
1912 | rnp->qsmask = 0; |
1913 | rnp->qsmaskinit = 0; | |
1914 | rnp->grplo = j * cpustride; | |
1915 | rnp->grphi = (j + 1) * cpustride - 1; | |
1916 | if (rnp->grphi >= NR_CPUS) | |
1917 | rnp->grphi = NR_CPUS - 1; | |
1918 | if (i == 0) { | |
1919 | rnp->grpnum = 0; | |
1920 | rnp->grpmask = 0; | |
1921 | rnp->parent = NULL; | |
1922 | } else { | |
1923 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1924 | rnp->grpmask = 1UL << rnp->grpnum; | |
1925 | rnp->parent = rsp->level[i - 1] + | |
1926 | j / rsp->levelspread[i - 1]; | |
1927 | } | |
1928 | rnp->level = i; | |
f41d911f PM |
1929 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1930 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
d9a3da06 PM |
1931 | INIT_LIST_HEAD(&rnp->blocked_tasks[2]); |
1932 | INIT_LIST_HEAD(&rnp->blocked_tasks[3]); | |
64db4cff PM |
1933 | } |
1934 | } | |
0c34029a | 1935 | |
394f99a9 | 1936 | rsp->rda = rda; |
0c34029a LJ |
1937 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
1938 | for_each_possible_cpu(i) { | |
4a90a068 | 1939 | while (i > rnp->grphi) |
0c34029a | 1940 | rnp++; |
394f99a9 | 1941 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
1942 | rcu_boot_init_percpu_data(i, rsp); |
1943 | } | |
64db4cff PM |
1944 | } |
1945 | ||
9f680ab4 | 1946 | void __init rcu_init(void) |
64db4cff | 1947 | { |
017c4261 | 1948 | int cpu; |
9f680ab4 | 1949 | |
f41d911f | 1950 | rcu_bootup_announce(); |
394f99a9 LJ |
1951 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
1952 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 1953 | __rcu_init_preempt(); |
2e597558 | 1954 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
1955 | |
1956 | /* | |
1957 | * We don't need protection against CPU-hotplug here because | |
1958 | * this is called early in boot, before either interrupts | |
1959 | * or the scheduler are operational. | |
1960 | */ | |
1961 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
1962 | for_each_online_cpu(cpu) |
1963 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 1964 | check_cpu_stall_init(); |
64db4cff PM |
1965 | } |
1966 | ||
1eba8f84 | 1967 | #include "rcutree_plugin.h" |