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> | |
49 | ||
9f77da9f PM |
50 | #include "rcutree.h" |
51 | ||
64db4cff PM |
52 | /* Data structures. */ |
53 | ||
88b91c7c PZ |
54 | static struct lock_class_key rcu_root_class; |
55 | ||
64db4cff PM |
56 | #define RCU_STATE_INITIALIZER(name) { \ |
57 | .level = { &name.node[0] }, \ | |
58 | .levelcnt = { \ | |
59 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
60 | NUM_RCU_LVL_1, \ | |
61 | NUM_RCU_LVL_2, \ | |
62 | NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \ | |
63 | }, \ | |
83f5b01f | 64 | .signaled = RCU_GP_IDLE, \ |
64db4cff PM |
65 | .gpnum = -300, \ |
66 | .completed = -300, \ | |
67 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ | |
e74f4c45 PM |
68 | .orphan_cbs_list = NULL, \ |
69 | .orphan_cbs_tail = &name.orphan_cbs_list, \ | |
70 | .orphan_qlen = 0, \ | |
64db4cff PM |
71 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ |
72 | .n_force_qs = 0, \ | |
73 | .n_force_qs_ngp = 0, \ | |
74 | } | |
75 | ||
d6714c22 PM |
76 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
77 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 78 | |
6258c4fb IM |
79 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
80 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 81 | |
f41d911f | 82 | |
fc2219d4 PM |
83 | /* |
84 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
85 | * permit this function to be invoked without holding the root rcu_node | |
86 | * structure's ->lock, but of course results can be subject to change. | |
87 | */ | |
88 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
89 | { | |
90 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
91 | } | |
92 | ||
b1f77b05 | 93 | /* |
d6714c22 | 94 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 95 | * how many quiescent states passed, just if there was at least |
d6714c22 | 96 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 97 | */ |
d6714c22 | 98 | void rcu_sched_qs(int cpu) |
b1f77b05 | 99 | { |
f41d911f PM |
100 | struct rcu_data *rdp; |
101 | ||
f41d911f | 102 | rdp = &per_cpu(rcu_sched_data, cpu); |
b1f77b05 | 103 | rdp->passed_quiesc_completed = rdp->completed; |
c3422bea PM |
104 | barrier(); |
105 | rdp->passed_quiesc = 1; | |
106 | rcu_preempt_note_context_switch(cpu); | |
b1f77b05 IM |
107 | } |
108 | ||
d6714c22 | 109 | void rcu_bh_qs(int cpu) |
b1f77b05 | 110 | { |
f41d911f PM |
111 | struct rcu_data *rdp; |
112 | ||
f41d911f | 113 | rdp = &per_cpu(rcu_bh_data, cpu); |
b1f77b05 | 114 | rdp->passed_quiesc_completed = rdp->completed; |
c3422bea PM |
115 | barrier(); |
116 | rdp->passed_quiesc = 1; | |
b1f77b05 | 117 | } |
64db4cff PM |
118 | |
119 | #ifdef CONFIG_NO_HZ | |
90a4d2c0 PM |
120 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
121 | .dynticks_nesting = 1, | |
122 | .dynticks = 1, | |
123 | }; | |
64db4cff PM |
124 | #endif /* #ifdef CONFIG_NO_HZ */ |
125 | ||
126 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
127 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
128 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
129 | ||
3d76c082 PM |
130 | module_param(blimit, int, 0); |
131 | module_param(qhimark, int, 0); | |
132 | module_param(qlowmark, int, 0); | |
133 | ||
64db4cff | 134 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 135 | static int rcu_pending(int cpu); |
64db4cff PM |
136 | |
137 | /* | |
d6714c22 | 138 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 139 | */ |
d6714c22 | 140 | long rcu_batches_completed_sched(void) |
64db4cff | 141 | { |
d6714c22 | 142 | return rcu_sched_state.completed; |
64db4cff | 143 | } |
d6714c22 | 144 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
145 | |
146 | /* | |
147 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
148 | */ | |
149 | long rcu_batches_completed_bh(void) | |
150 | { | |
151 | return rcu_bh_state.completed; | |
152 | } | |
153 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
154 | ||
155 | /* | |
156 | * Does the CPU have callbacks ready to be invoked? | |
157 | */ | |
158 | static int | |
159 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
160 | { | |
161 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
162 | } | |
163 | ||
164 | /* | |
165 | * Does the current CPU require a yet-as-unscheduled grace period? | |
166 | */ | |
167 | static int | |
168 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
169 | { | |
fc2219d4 | 170 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
171 | } |
172 | ||
173 | /* | |
174 | * Return the root node of the specified rcu_state structure. | |
175 | */ | |
176 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
177 | { | |
178 | return &rsp->node[0]; | |
179 | } | |
180 | ||
281d150c PM |
181 | /* |
182 | * Record the specified "completed" value, which is later used to validate | |
183 | * dynticks counter manipulations and CPU-offline checks. Specify | |
184 | * "rsp->completed - 1" to unconditionally invalidate any future dynticks | |
185 | * manipulations and CPU-offline checks. Such invalidation is useful at | |
186 | * the beginning of a grace period. | |
187 | */ | |
188 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
189 | { | |
190 | rsp->dynticks_completed = comp; | |
191 | } | |
192 | ||
64db4cff PM |
193 | #ifdef CONFIG_SMP |
194 | ||
281d150c PM |
195 | /* |
196 | * Recall the previously recorded value of the completion for dynticks. | |
197 | */ | |
198 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
199 | { | |
200 | return rsp->dynticks_completed; | |
201 | } | |
202 | ||
64db4cff PM |
203 | /* |
204 | * If the specified CPU is offline, tell the caller that it is in | |
205 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
206 | * Grace periods can end up waiting on an offline CPU when that | |
207 | * CPU is in the process of coming online -- it will be added to the | |
208 | * rcu_node bitmasks before it actually makes it online. The same thing | |
209 | * can happen while a CPU is in the process of coming online. Because this | |
210 | * race is quite rare, we check for it after detecting that the grace | |
211 | * period has been delayed rather than checking each and every CPU | |
212 | * each and every time we start a new grace period. | |
213 | */ | |
214 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
215 | { | |
216 | /* | |
217 | * If the CPU is offline, it is in a quiescent state. We can | |
218 | * trust its state not to change because interrupts are disabled. | |
219 | */ | |
220 | if (cpu_is_offline(rdp->cpu)) { | |
221 | rdp->offline_fqs++; | |
222 | return 1; | |
223 | } | |
224 | ||
f41d911f PM |
225 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
226 | if (rdp->preemptable) | |
227 | return 0; | |
228 | ||
64db4cff PM |
229 | /* The CPU is online, so send it a reschedule IPI. */ |
230 | if (rdp->cpu != smp_processor_id()) | |
231 | smp_send_reschedule(rdp->cpu); | |
232 | else | |
233 | set_need_resched(); | |
234 | rdp->resched_ipi++; | |
235 | return 0; | |
236 | } | |
237 | ||
238 | #endif /* #ifdef CONFIG_SMP */ | |
239 | ||
240 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
241 | |
242 | /** | |
243 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
244 | * | |
245 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
246 | * read-side critical sections can occur. (Though RCU read-side | |
247 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
248 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
249 | */ | |
250 | void rcu_enter_nohz(void) | |
251 | { | |
252 | unsigned long flags; | |
253 | struct rcu_dynticks *rdtp; | |
254 | ||
255 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
256 | local_irq_save(flags); | |
257 | rdtp = &__get_cpu_var(rcu_dynticks); | |
258 | rdtp->dynticks++; | |
259 | rdtp->dynticks_nesting--; | |
86848966 | 260 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
261 | local_irq_restore(flags); |
262 | } | |
263 | ||
264 | /* | |
265 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
266 | * | |
267 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
268 | * read-side critical sections normally occur. | |
269 | */ | |
270 | void rcu_exit_nohz(void) | |
271 | { | |
272 | unsigned long flags; | |
273 | struct rcu_dynticks *rdtp; | |
274 | ||
275 | local_irq_save(flags); | |
276 | rdtp = &__get_cpu_var(rcu_dynticks); | |
277 | rdtp->dynticks++; | |
278 | rdtp->dynticks_nesting++; | |
86848966 | 279 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
280 | local_irq_restore(flags); |
281 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
282 | } | |
283 | ||
284 | /** | |
285 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
286 | * | |
287 | * If the CPU was idle with dynamic ticks active, and there is no | |
288 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
289 | * RCU grace-period handling know that the CPU is active. | |
290 | */ | |
291 | void rcu_nmi_enter(void) | |
292 | { | |
293 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
294 | ||
295 | if (rdtp->dynticks & 0x1) | |
296 | return; | |
297 | rdtp->dynticks_nmi++; | |
86848966 | 298 | WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1)); |
64db4cff PM |
299 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
300 | } | |
301 | ||
302 | /** | |
303 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
304 | * | |
305 | * If the CPU was idle with dynamic ticks active, and there is no | |
306 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
307 | * RCU grace-period handling know that the CPU is no longer active. | |
308 | */ | |
309 | void rcu_nmi_exit(void) | |
310 | { | |
311 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
312 | ||
313 | if (rdtp->dynticks & 0x1) | |
314 | return; | |
315 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
316 | rdtp->dynticks_nmi++; | |
86848966 | 317 | WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1); |
64db4cff PM |
318 | } |
319 | ||
320 | /** | |
321 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
322 | * | |
323 | * If the CPU was idle with dynamic ticks active, this updates the | |
324 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
325 | */ | |
326 | void rcu_irq_enter(void) | |
327 | { | |
328 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
329 | ||
330 | if (rdtp->dynticks_nesting++) | |
331 | return; | |
332 | rdtp->dynticks++; | |
86848966 | 333 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
334 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
335 | } | |
336 | ||
337 | /** | |
338 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
339 | * | |
340 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
341 | * to put let the RCU handling be aware that the CPU is going back to idle | |
342 | * with no ticks. | |
343 | */ | |
344 | void rcu_irq_exit(void) | |
345 | { | |
346 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
347 | ||
348 | if (--rdtp->dynticks_nesting) | |
349 | return; | |
350 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
351 | rdtp->dynticks++; | |
86848966 | 352 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
353 | |
354 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 355 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
356 | __get_cpu_var(rcu_bh_data).nxtlist) |
357 | set_need_resched(); | |
358 | } | |
359 | ||
64db4cff PM |
360 | #ifdef CONFIG_SMP |
361 | ||
64db4cff PM |
362 | /* |
363 | * Snapshot the specified CPU's dynticks counter so that we can later | |
364 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 365 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
366 | */ |
367 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
368 | { | |
369 | int ret; | |
370 | int snap; | |
371 | int snap_nmi; | |
372 | ||
373 | snap = rdp->dynticks->dynticks; | |
374 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
375 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
376 | rdp->dynticks_snap = snap; | |
377 | rdp->dynticks_nmi_snap = snap_nmi; | |
378 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
379 | if (ret) | |
380 | rdp->dynticks_fqs++; | |
381 | return ret; | |
382 | } | |
383 | ||
384 | /* | |
385 | * Return true if the specified CPU has passed through a quiescent | |
386 | * state by virtue of being in or having passed through an dynticks | |
387 | * idle state since the last call to dyntick_save_progress_counter() | |
388 | * for this same CPU. | |
389 | */ | |
390 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
391 | { | |
392 | long curr; | |
393 | long curr_nmi; | |
394 | long snap; | |
395 | long snap_nmi; | |
396 | ||
397 | curr = rdp->dynticks->dynticks; | |
398 | snap = rdp->dynticks_snap; | |
399 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
400 | snap_nmi = rdp->dynticks_nmi_snap; | |
401 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
402 | ||
403 | /* | |
404 | * If the CPU passed through or entered a dynticks idle phase with | |
405 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
406 | * already acknowledged the request to pass through a quiescent | |
407 | * state. Either way, that CPU cannot possibly be in an RCU | |
408 | * read-side critical section that started before the beginning | |
409 | * of the current RCU grace period. | |
410 | */ | |
411 | if ((curr != snap || (curr & 0x1) == 0) && | |
412 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
413 | rdp->dynticks_fqs++; | |
414 | return 1; | |
415 | } | |
416 | ||
417 | /* Go check for the CPU being offline. */ | |
418 | return rcu_implicit_offline_qs(rdp); | |
419 | } | |
420 | ||
421 | #endif /* #ifdef CONFIG_SMP */ | |
422 | ||
423 | #else /* #ifdef CONFIG_NO_HZ */ | |
424 | ||
64db4cff PM |
425 | #ifdef CONFIG_SMP |
426 | ||
64db4cff PM |
427 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
428 | { | |
429 | return 0; | |
430 | } | |
431 | ||
432 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
433 | { | |
434 | return rcu_implicit_offline_qs(rdp); | |
435 | } | |
436 | ||
437 | #endif /* #ifdef CONFIG_SMP */ | |
438 | ||
439 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
440 | ||
441 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
442 | ||
443 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
444 | { | |
445 | rsp->gp_start = jiffies; | |
446 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
447 | } | |
448 | ||
449 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
450 | { | |
451 | int cpu; | |
452 | long delta; | |
453 | unsigned long flags; | |
454 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
455 | |
456 | /* Only let one CPU complain about others per time interval. */ | |
457 | ||
458 | spin_lock_irqsave(&rnp->lock, flags); | |
459 | delta = jiffies - rsp->jiffies_stall; | |
fc2219d4 | 460 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
64db4cff PM |
461 | spin_unlock_irqrestore(&rnp->lock, flags); |
462 | return; | |
463 | } | |
464 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
465 | |
466 | /* | |
467 | * Now rat on any tasks that got kicked up to the root rcu_node | |
468 | * due to CPU offlining. | |
469 | */ | |
470 | rcu_print_task_stall(rnp); | |
64db4cff PM |
471 | spin_unlock_irqrestore(&rnp->lock, flags); |
472 | ||
473 | /* OK, time to rat on our buddy... */ | |
474 | ||
475 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | |
a0b6c9a7 | 476 | rcu_for_each_leaf_node(rsp, rnp) { |
f41d911f | 477 | rcu_print_task_stall(rnp); |
a0b6c9a7 | 478 | if (rnp->qsmask == 0) |
64db4cff | 479 | continue; |
a0b6c9a7 PM |
480 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
481 | if (rnp->qsmask & (1UL << cpu)) | |
482 | printk(" %d", rnp->grplo + cpu); | |
64db4cff PM |
483 | } |
484 | printk(" (detected by %d, t=%ld jiffies)\n", | |
485 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | |
c1dc0b9c IM |
486 | trigger_all_cpu_backtrace(); |
487 | ||
64db4cff PM |
488 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
489 | } | |
490 | ||
491 | static void print_cpu_stall(struct rcu_state *rsp) | |
492 | { | |
493 | unsigned long flags; | |
494 | struct rcu_node *rnp = rcu_get_root(rsp); | |
495 | ||
496 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | |
497 | smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
498 | trigger_all_cpu_backtrace(); |
499 | ||
64db4cff PM |
500 | spin_lock_irqsave(&rnp->lock, flags); |
501 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) | |
502 | rsp->jiffies_stall = | |
503 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
504 | spin_unlock_irqrestore(&rnp->lock, flags); | |
c1dc0b9c | 505 | |
64db4cff PM |
506 | set_need_resched(); /* kick ourselves to get things going. */ |
507 | } | |
508 | ||
509 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
510 | { | |
511 | long delta; | |
512 | struct rcu_node *rnp; | |
513 | ||
514 | delta = jiffies - rsp->jiffies_stall; | |
515 | rnp = rdp->mynode; | |
516 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
517 | ||
518 | /* We haven't checked in, so go dump stack. */ | |
519 | print_cpu_stall(rsp); | |
520 | ||
fc2219d4 | 521 | } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) { |
64db4cff PM |
522 | |
523 | /* They had two time units to dump stack, so complain. */ | |
524 | print_other_cpu_stall(rsp); | |
525 | } | |
526 | } | |
527 | ||
528 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
529 | ||
530 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
531 | { | |
532 | } | |
533 | ||
534 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
535 | { | |
536 | } | |
537 | ||
538 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
539 | ||
540 | /* | |
541 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
542 | * This is used both when we started the grace period and when we notice | |
543 | * that someone else started the grace period. | |
544 | */ | |
545 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) | |
546 | { | |
547 | rdp->qs_pending = 1; | |
548 | rdp->passed_quiesc = 0; | |
549 | rdp->gpnum = rsp->gpnum; | |
64db4cff PM |
550 | } |
551 | ||
552 | /* | |
553 | * Did someone else start a new RCU grace period start since we last | |
554 | * checked? Update local state appropriately if so. Must be called | |
555 | * on the CPU corresponding to rdp. | |
556 | */ | |
557 | static int | |
558 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
559 | { | |
560 | unsigned long flags; | |
561 | int ret = 0; | |
562 | ||
563 | local_irq_save(flags); | |
564 | if (rdp->gpnum != rsp->gpnum) { | |
565 | note_new_gpnum(rsp, rdp); | |
566 | ret = 1; | |
567 | } | |
568 | local_irq_restore(flags); | |
569 | return ret; | |
570 | } | |
571 | ||
572 | /* | |
573 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
574 | * in preparation for detecting the next grace period. The caller must hold | |
575 | * the root node's ->lock, which is released before return. Hard irqs must | |
576 | * be disabled. | |
577 | */ | |
578 | static void | |
579 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
580 | __releases(rcu_get_root(rsp)->lock) | |
581 | { | |
582 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
583 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
584 | |
585 | if (!cpu_needs_another_gp(rsp, rdp)) { | |
586 | spin_unlock_irqrestore(&rnp->lock, flags); | |
587 | return; | |
588 | } | |
589 | ||
590 | /* Advance to a new grace period and initialize state. */ | |
591 | rsp->gpnum++; | |
c3422bea | 592 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
593 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
594 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
595 | record_gp_stall_check_time(rsp); |
596 | dyntick_record_completed(rsp, rsp->completed - 1); | |
597 | note_new_gpnum(rsp, rdp); | |
598 | ||
599 | /* | |
1eba8f84 PM |
600 | * Because this CPU just now started the new grace period, we know |
601 | * that all of its callbacks will be covered by this upcoming grace | |
602 | * period, even the ones that were registered arbitrarily recently. | |
603 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
604 | * | |
605 | * Other CPUs cannot be sure exactly when the grace period started. | |
606 | * Therefore, their recently registered callbacks must pass through | |
607 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
608 | * by the next RCU grace period. | |
64db4cff PM |
609 | */ |
610 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
611 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
612 | ||
613 | /* Special-case the common single-level case. */ | |
614 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 615 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 616 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 617 | rnp->gpnum = rsp->gpnum; |
c12172c0 | 618 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
64db4cff PM |
619 | spin_unlock_irqrestore(&rnp->lock, flags); |
620 | return; | |
621 | } | |
622 | ||
623 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | |
624 | ||
625 | ||
626 | /* Exclude any concurrent CPU-hotplug operations. */ | |
627 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
628 | ||
629 | /* | |
b835db1f PM |
630 | * Set the quiescent-state-needed bits in all the rcu_node |
631 | * structures for all currently online CPUs in breadth-first | |
632 | * order, starting from the root rcu_node structure. This | |
633 | * operation relies on the layout of the hierarchy within the | |
634 | * rsp->node[] array. Note that other CPUs will access only | |
635 | * the leaves of the hierarchy, which still indicate that no | |
636 | * grace period is in progress, at least until the corresponding | |
637 | * leaf node has been initialized. In addition, we have excluded | |
638 | * CPU-hotplug operations. | |
64db4cff PM |
639 | * |
640 | * Note that the grace period cannot complete until we finish | |
641 | * the initialization process, as there will be at least one | |
642 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
643 | * one corresponding to this CPU, due to the fact that we have |
644 | * irqs disabled. | |
64db4cff | 645 | */ |
a0b6c9a7 | 646 | rcu_for_each_node_breadth_first(rsp, rnp) { |
83f5b01f | 647 | spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 648 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 649 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 650 | rnp->gpnum = rsp->gpnum; |
83f5b01f | 651 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
652 | } |
653 | ||
83f5b01f PM |
654 | rnp = rcu_get_root(rsp); |
655 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
64db4cff | 656 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
83f5b01f | 657 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
658 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
659 | } | |
660 | ||
661 | /* | |
662 | * Advance this CPU's callbacks, but only if the current grace period | |
663 | * has ended. This may be called only from the CPU to whom the rdp | |
664 | * belongs. | |
665 | */ | |
666 | static void | |
667 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
668 | { | |
669 | long completed_snap; | |
670 | unsigned long flags; | |
671 | ||
672 | local_irq_save(flags); | |
673 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ | |
674 | ||
675 | /* Did another grace period end? */ | |
676 | if (rdp->completed != completed_snap) { | |
677 | ||
678 | /* Advance callbacks. No harm if list empty. */ | |
679 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
680 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
681 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
682 | ||
683 | /* Remember that we saw this grace-period completion. */ | |
684 | rdp->completed = completed_snap; | |
685 | } | |
686 | local_irq_restore(flags); | |
687 | } | |
688 | ||
f41d911f PM |
689 | /* |
690 | * Clean up after the prior grace period and let rcu_start_gp() start up | |
691 | * the next grace period if one is needed. Note that the caller must | |
692 | * hold rnp->lock, as required by rcu_start_gp(), which will release it. | |
693 | */ | |
694 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) | |
fc2219d4 | 695 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 696 | { |
fc2219d4 | 697 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
f41d911f | 698 | rsp->completed = rsp->gpnum; |
83f5b01f | 699 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
700 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); |
701 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ | |
702 | } | |
703 | ||
64db4cff PM |
704 | /* |
705 | * Similar to cpu_quiet(), for which it is a helper function. Allows | |
706 | * a group of CPUs to be quieted at one go, though all the CPUs in the | |
707 | * group must be represented by the same leaf rcu_node structure. | |
708 | * That structure's lock must be held upon entry, and it is released | |
709 | * before return. | |
710 | */ | |
711 | static void | |
712 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, | |
713 | unsigned long flags) | |
714 | __releases(rnp->lock) | |
715 | { | |
28ecd580 PM |
716 | struct rcu_node *rnp_c; |
717 | ||
64db4cff PM |
718 | /* Walk up the rcu_node hierarchy. */ |
719 | for (;;) { | |
720 | if (!(rnp->qsmask & mask)) { | |
721 | ||
722 | /* Our bit has already been cleared, so done. */ | |
723 | spin_unlock_irqrestore(&rnp->lock, flags); | |
724 | return; | |
725 | } | |
726 | rnp->qsmask &= ~mask; | |
f41d911f | 727 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
728 | |
729 | /* Other bits still set at this level, so done. */ | |
730 | spin_unlock_irqrestore(&rnp->lock, flags); | |
731 | return; | |
732 | } | |
733 | mask = rnp->grpmask; | |
734 | if (rnp->parent == NULL) { | |
735 | ||
736 | /* No more levels. Exit loop holding root lock. */ | |
737 | ||
738 | break; | |
739 | } | |
740 | spin_unlock_irqrestore(&rnp->lock, flags); | |
28ecd580 | 741 | rnp_c = rnp; |
64db4cff PM |
742 | rnp = rnp->parent; |
743 | spin_lock_irqsave(&rnp->lock, flags); | |
28ecd580 | 744 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
745 | } |
746 | ||
747 | /* | |
748 | * Get here if we are the last CPU to pass through a quiescent | |
f41d911f PM |
749 | * state for this grace period. Invoke cpu_quiet_msk_finish() |
750 | * to clean up and start the next grace period if one is needed. | |
64db4cff | 751 | */ |
f41d911f | 752 | cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
753 | } |
754 | ||
755 | /* | |
756 | * Record a quiescent state for the specified CPU, which must either be | |
e7d8842e PM |
757 | * the current CPU. The lastcomp argument is used to make sure we are |
758 | * still in the grace period of interest. We don't want to end the current | |
759 | * grace period based on quiescent states detected in an earlier grace | |
760 | * period! | |
64db4cff PM |
761 | */ |
762 | static void | |
763 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) | |
764 | { | |
765 | unsigned long flags; | |
766 | unsigned long mask; | |
767 | struct rcu_node *rnp; | |
768 | ||
769 | rnp = rdp->mynode; | |
770 | spin_lock_irqsave(&rnp->lock, flags); | |
771 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { | |
772 | ||
773 | /* | |
774 | * Someone beat us to it for this grace period, so leave. | |
775 | * The race with GP start is resolved by the fact that we | |
776 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
777 | * cannot yet be initialized -- so we would simply find our | |
778 | * CPU's bit already cleared in cpu_quiet_msk() if this race | |
779 | * occurred. | |
780 | */ | |
781 | rdp->passed_quiesc = 0; /* try again later! */ | |
782 | spin_unlock_irqrestore(&rnp->lock, flags); | |
783 | return; | |
784 | } | |
785 | mask = rdp->grpmask; | |
786 | if ((rnp->qsmask & mask) == 0) { | |
787 | spin_unlock_irqrestore(&rnp->lock, flags); | |
788 | } else { | |
789 | rdp->qs_pending = 0; | |
790 | ||
791 | /* | |
792 | * This GP can't end until cpu checks in, so all of our | |
793 | * callbacks can be processed during the next GP. | |
794 | */ | |
64db4cff PM |
795 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
796 | ||
797 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ | |
798 | } | |
799 | } | |
800 | ||
801 | /* | |
802 | * Check to see if there is a new grace period of which this CPU | |
803 | * is not yet aware, and if so, set up local rcu_data state for it. | |
804 | * Otherwise, see if this CPU has just passed through its first | |
805 | * quiescent state for this grace period, and record that fact if so. | |
806 | */ | |
807 | static void | |
808 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
809 | { | |
810 | /* If there is now a new grace period, record and return. */ | |
811 | if (check_for_new_grace_period(rsp, rdp)) | |
812 | return; | |
813 | ||
814 | /* | |
815 | * Does this CPU still need to do its part for current grace period? | |
816 | * If no, return and let the other CPUs do their part as well. | |
817 | */ | |
818 | if (!rdp->qs_pending) | |
819 | return; | |
820 | ||
821 | /* | |
822 | * Was there a quiescent state since the beginning of the grace | |
823 | * period? If no, then exit and wait for the next call. | |
824 | */ | |
825 | if (!rdp->passed_quiesc) | |
826 | return; | |
827 | ||
828 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ | |
829 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
830 | } | |
831 | ||
832 | #ifdef CONFIG_HOTPLUG_CPU | |
833 | ||
e74f4c45 PM |
834 | /* |
835 | * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the | |
836 | * specified flavor of RCU. The callbacks will be adopted by the next | |
837 | * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever | |
838 | * comes first. Because this is invoked from the CPU_DYING notifier, | |
839 | * irqs are already disabled. | |
840 | */ | |
841 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) | |
842 | { | |
843 | int i; | |
844 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
845 | ||
846 | if (rdp->nxtlist == NULL) | |
847 | return; /* irqs disabled, so comparison is stable. */ | |
848 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
849 | *rsp->orphan_cbs_tail = rdp->nxtlist; | |
850 | rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL]; | |
851 | rdp->nxtlist = NULL; | |
852 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
853 | rdp->nxttail[i] = &rdp->nxtlist; | |
854 | rsp->orphan_qlen += rdp->qlen; | |
855 | rdp->qlen = 0; | |
856 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
857 | } | |
858 | ||
859 | /* | |
860 | * Adopt previously orphaned RCU callbacks. | |
861 | */ | |
862 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
863 | { | |
864 | unsigned long flags; | |
865 | struct rcu_data *rdp; | |
866 | ||
867 | spin_lock_irqsave(&rsp->onofflock, flags); | |
868 | rdp = rsp->rda[smp_processor_id()]; | |
869 | if (rsp->orphan_cbs_list == NULL) { | |
870 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
871 | return; | |
872 | } | |
873 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list; | |
874 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail; | |
875 | rdp->qlen += rsp->orphan_qlen; | |
876 | rsp->orphan_cbs_list = NULL; | |
877 | rsp->orphan_cbs_tail = &rsp->orphan_cbs_list; | |
878 | rsp->orphan_qlen = 0; | |
879 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
880 | } | |
881 | ||
64db4cff PM |
882 | /* |
883 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
884 | * and move all callbacks from the outgoing CPU to the current one. | |
885 | */ | |
886 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
887 | { | |
64db4cff PM |
888 | unsigned long flags; |
889 | long lastcomp; | |
890 | unsigned long mask; | |
891 | struct rcu_data *rdp = rsp->rda[cpu]; | |
64db4cff PM |
892 | struct rcu_node *rnp; |
893 | ||
894 | /* Exclude any attempts to start a new grace period. */ | |
895 | spin_lock_irqsave(&rsp->onofflock, flags); | |
896 | ||
897 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 898 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
899 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
900 | do { | |
901 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
902 | rnp->qsmaskinit &= ~mask; | |
903 | if (rnp->qsmaskinit != 0) { | |
f41d911f | 904 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
905 | break; |
906 | } | |
237c80c5 PM |
907 | |
908 | /* | |
909 | * If there was a task blocking the current grace period, | |
910 | * and if all CPUs have checked in, we need to propagate | |
911 | * the quiescent state up the rcu_node hierarchy. But that | |
912 | * is inconvenient at the moment due to deadlock issues if | |
913 | * this should end the current grace period. So set the | |
914 | * offlined CPU's bit in ->qsmask in order to force the | |
915 | * next force_quiescent_state() invocation to clean up this | |
916 | * mess in a deadlock-free manner. | |
917 | */ | |
918 | if (rcu_preempt_offline_tasks(rsp, rnp, rdp) && !rnp->qsmask) | |
919 | rnp->qsmask |= mask; | |
920 | ||
64db4cff | 921 | mask = rnp->grpmask; |
f41d911f | 922 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
923 | rnp = rnp->parent; |
924 | } while (rnp != NULL); | |
925 | lastcomp = rsp->completed; | |
926 | ||
e74f4c45 | 927 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff | 928 | |
e74f4c45 | 929 | rcu_adopt_orphan_cbs(rsp); |
64db4cff PM |
930 | } |
931 | ||
932 | /* | |
933 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
934 | * callbacks that it might have to the current CPU. This code assumes | |
935 | * that at least one CPU in the system will remain running at all times. | |
936 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
937 | */ | |
938 | static void rcu_offline_cpu(int cpu) | |
939 | { | |
d6714c22 | 940 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 941 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 942 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
943 | } |
944 | ||
945 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
946 | ||
e74f4c45 PM |
947 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) |
948 | { | |
949 | } | |
950 | ||
951 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
952 | { | |
953 | } | |
954 | ||
64db4cff PM |
955 | static void rcu_offline_cpu(int cpu) |
956 | { | |
957 | } | |
958 | ||
959 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
960 | ||
961 | /* | |
962 | * Invoke any RCU callbacks that have made it to the end of their grace | |
963 | * period. Thottle as specified by rdp->blimit. | |
964 | */ | |
37c72e56 | 965 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
966 | { |
967 | unsigned long flags; | |
968 | struct rcu_head *next, *list, **tail; | |
969 | int count; | |
970 | ||
971 | /* If no callbacks are ready, just return.*/ | |
972 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
973 | return; | |
974 | ||
975 | /* | |
976 | * Extract the list of ready callbacks, disabling to prevent | |
977 | * races with call_rcu() from interrupt handlers. | |
978 | */ | |
979 | local_irq_save(flags); | |
980 | list = rdp->nxtlist; | |
981 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
982 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
983 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
984 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
985 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
986 | rdp->nxttail[count] = &rdp->nxtlist; | |
987 | local_irq_restore(flags); | |
988 | ||
989 | /* Invoke callbacks. */ | |
990 | count = 0; | |
991 | while (list) { | |
992 | next = list->next; | |
993 | prefetch(next); | |
994 | list->func(list); | |
995 | list = next; | |
996 | if (++count >= rdp->blimit) | |
997 | break; | |
998 | } | |
999 | ||
1000 | local_irq_save(flags); | |
1001 | ||
1002 | /* Update count, and requeue any remaining callbacks. */ | |
1003 | rdp->qlen -= count; | |
1004 | if (list != NULL) { | |
1005 | *tail = rdp->nxtlist; | |
1006 | rdp->nxtlist = list; | |
1007 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1008 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1009 | rdp->nxttail[count] = tail; | |
1010 | else | |
1011 | break; | |
1012 | } | |
1013 | ||
1014 | /* Reinstate batch limit if we have worked down the excess. */ | |
1015 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1016 | rdp->blimit = blimit; | |
1017 | ||
37c72e56 PM |
1018 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1019 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1020 | rdp->qlen_last_fqs_check = 0; | |
1021 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1022 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1023 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1024 | ||
64db4cff PM |
1025 | local_irq_restore(flags); |
1026 | ||
1027 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1028 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
1029 | raise_softirq(RCU_SOFTIRQ); | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1034 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1035 | * Also schedule the RCU softirq handler. | |
1036 | * | |
1037 | * This function must be called with hardirqs disabled. It is normally | |
1038 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1039 | * false, there is no point in invoking rcu_check_callbacks(). | |
1040 | */ | |
1041 | void rcu_check_callbacks(int cpu, int user) | |
1042 | { | |
a157229c PM |
1043 | if (!rcu_pending(cpu)) |
1044 | return; /* if nothing for RCU to do. */ | |
64db4cff | 1045 | if (user || |
a6826048 PM |
1046 | (idle_cpu(cpu) && rcu_scheduler_active && |
1047 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1048 | |
1049 | /* | |
1050 | * Get here if this CPU took its interrupt from user | |
1051 | * mode or from the idle loop, and if this is not a | |
1052 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1053 | * a quiescent state, so note it. |
64db4cff PM |
1054 | * |
1055 | * No memory barrier is required here because both | |
d6714c22 PM |
1056 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1057 | * variables that other CPUs neither access nor modify, | |
1058 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1059 | */ |
1060 | ||
d6714c22 PM |
1061 | rcu_sched_qs(cpu); |
1062 | rcu_bh_qs(cpu); | |
64db4cff PM |
1063 | |
1064 | } else if (!in_softirq()) { | |
1065 | ||
1066 | /* | |
1067 | * Get here if this CPU did not take its interrupt from | |
1068 | * softirq, in other words, if it is not interrupting | |
1069 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1070 | * critical section, so note it. |
64db4cff PM |
1071 | */ |
1072 | ||
d6714c22 | 1073 | rcu_bh_qs(cpu); |
64db4cff | 1074 | } |
f41d911f | 1075 | rcu_preempt_check_callbacks(cpu); |
64db4cff PM |
1076 | raise_softirq(RCU_SOFTIRQ); |
1077 | } | |
1078 | ||
1079 | #ifdef CONFIG_SMP | |
1080 | ||
1081 | /* | |
1082 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1083 | * have not yet encountered a quiescent state, using the function specified. | |
1084 | * Returns 1 if the current grace period ends while scanning (possibly | |
1085 | * because we made it end). | |
1086 | */ | |
1087 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, | |
1088 | int (*f)(struct rcu_data *)) | |
1089 | { | |
1090 | unsigned long bit; | |
1091 | int cpu; | |
1092 | unsigned long flags; | |
1093 | unsigned long mask; | |
a0b6c9a7 | 1094 | struct rcu_node *rnp; |
64db4cff | 1095 | |
a0b6c9a7 | 1096 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1097 | mask = 0; |
a0b6c9a7 | 1098 | spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 1099 | if (rsp->completed != lastcomp) { |
a0b6c9a7 | 1100 | spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1101 | return 1; |
1102 | } | |
a0b6c9a7 PM |
1103 | if (rnp->qsmask == 0) { |
1104 | spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1105 | continue; |
1106 | } | |
a0b6c9a7 | 1107 | cpu = rnp->grplo; |
64db4cff | 1108 | bit = 1; |
a0b6c9a7 PM |
1109 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
1110 | if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu])) | |
64db4cff PM |
1111 | mask |= bit; |
1112 | } | |
1113 | if (mask != 0 && rsp->completed == lastcomp) { | |
1114 | ||
a0b6c9a7 PM |
1115 | /* cpu_quiet_msk() releases rnp->lock. */ |
1116 | cpu_quiet_msk(mask, rsp, rnp, flags); | |
64db4cff PM |
1117 | continue; |
1118 | } | |
a0b6c9a7 | 1119 | spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1120 | } |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Force quiescent states on reluctant CPUs, and also detect which | |
1126 | * CPUs are in dyntick-idle mode. | |
1127 | */ | |
1128 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1129 | { | |
1130 | unsigned long flags; | |
1131 | long lastcomp; | |
64db4cff PM |
1132 | struct rcu_node *rnp = rcu_get_root(rsp); |
1133 | u8 signaled; | |
281d150c | 1134 | u8 forcenow; |
64db4cff | 1135 | |
fc2219d4 | 1136 | if (!rcu_gp_in_progress(rsp)) |
64db4cff PM |
1137 | return; /* No grace period in progress, nothing to force. */ |
1138 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | |
1139 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | |
1140 | return; /* Someone else is already on the job. */ | |
1141 | } | |
1142 | if (relaxed && | |
ef631b0c | 1143 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
64db4cff PM |
1144 | goto unlock_ret; /* no emergency and done recently. */ |
1145 | rsp->n_force_qs++; | |
1146 | spin_lock(&rnp->lock); | |
1147 | lastcomp = rsp->completed; | |
1148 | signaled = rsp->signaled; | |
1149 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
1150 | if (lastcomp == rsp->gpnum) { |
1151 | rsp->n_force_qs_ngp++; | |
1152 | spin_unlock(&rnp->lock); | |
1153 | goto unlock_ret; /* no GP in progress, time updated. */ | |
1154 | } | |
1155 | spin_unlock(&rnp->lock); | |
1156 | switch (signaled) { | |
83f5b01f | 1157 | case RCU_GP_IDLE: |
64db4cff PM |
1158 | case RCU_GP_INIT: |
1159 | ||
83f5b01f | 1160 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1161 | |
1162 | case RCU_SAVE_DYNTICK: | |
1163 | ||
1164 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) | |
1165 | break; /* So gcc recognizes the dead code. */ | |
1166 | ||
1167 | /* Record dyntick-idle state. */ | |
1168 | if (rcu_process_dyntick(rsp, lastcomp, | |
1169 | dyntick_save_progress_counter)) | |
1170 | goto unlock_ret; | |
281d150c PM |
1171 | /* fall into next case. */ |
1172 | ||
1173 | case RCU_SAVE_COMPLETED: | |
64db4cff PM |
1174 | |
1175 | /* Update state, record completion counter. */ | |
281d150c | 1176 | forcenow = 0; |
64db4cff | 1177 | spin_lock(&rnp->lock); |
83f5b01f | 1178 | if (lastcomp == rsp->completed && |
281d150c | 1179 | rsp->signaled == signaled) { |
64db4cff PM |
1180 | rsp->signaled = RCU_FORCE_QS; |
1181 | dyntick_record_completed(rsp, lastcomp); | |
281d150c | 1182 | forcenow = signaled == RCU_SAVE_COMPLETED; |
64db4cff PM |
1183 | } |
1184 | spin_unlock(&rnp->lock); | |
281d150c PM |
1185 | if (!forcenow) |
1186 | break; | |
1187 | /* fall into next case. */ | |
64db4cff PM |
1188 | |
1189 | case RCU_FORCE_QS: | |
1190 | ||
1191 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1192 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), | |
1193 | rcu_implicit_dynticks_qs)) | |
1194 | goto unlock_ret; | |
1195 | ||
1196 | /* Leave state in case more forcing is required. */ | |
1197 | ||
1198 | break; | |
1199 | } | |
1200 | unlock_ret: | |
1201 | spin_unlock_irqrestore(&rsp->fqslock, flags); | |
1202 | } | |
1203 | ||
1204 | #else /* #ifdef CONFIG_SMP */ | |
1205 | ||
1206 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1207 | { | |
1208 | set_need_resched(); | |
1209 | } | |
1210 | ||
1211 | #endif /* #else #ifdef CONFIG_SMP */ | |
1212 | ||
1213 | /* | |
1214 | * This does the RCU processing work from softirq context for the | |
1215 | * specified rcu_state and rcu_data structures. This may be called | |
1216 | * only from the CPU to whom the rdp belongs. | |
1217 | */ | |
1218 | static void | |
1219 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1220 | { | |
1221 | unsigned long flags; | |
1222 | ||
2e597558 PM |
1223 | WARN_ON_ONCE(rdp->beenonline == 0); |
1224 | ||
64db4cff PM |
1225 | /* |
1226 | * If an RCU GP has gone long enough, go check for dyntick | |
1227 | * idle CPUs and, if needed, send resched IPIs. | |
1228 | */ | |
ef631b0c | 1229 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1230 | force_quiescent_state(rsp, 1); |
1231 | ||
1232 | /* | |
1233 | * Advance callbacks in response to end of earlier grace | |
1234 | * period that some other CPU ended. | |
1235 | */ | |
1236 | rcu_process_gp_end(rsp, rdp); | |
1237 | ||
1238 | /* Update RCU state based on any recent quiescent states. */ | |
1239 | rcu_check_quiescent_state(rsp, rdp); | |
1240 | ||
1241 | /* Does this CPU require a not-yet-started grace period? */ | |
1242 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1243 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | |
1244 | rcu_start_gp(rsp, flags); /* releases above lock */ | |
1245 | } | |
1246 | ||
1247 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1248 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * Do softirq processing for the current CPU. | |
1253 | */ | |
1254 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1255 | { | |
1256 | /* | |
1257 | * Memory references from any prior RCU read-side critical sections | |
1258 | * executed by the interrupted code must be seen before any RCU | |
1259 | * grace-period manipulations below. | |
1260 | */ | |
1261 | smp_mb(); /* See above block comment. */ | |
1262 | ||
d6714c22 PM |
1263 | __rcu_process_callbacks(&rcu_sched_state, |
1264 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1265 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1266 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1267 | |
1268 | /* | |
1269 | * Memory references from any later RCU read-side critical sections | |
1270 | * executed by the interrupted code must be seen after any RCU | |
1271 | * grace-period manipulations above. | |
1272 | */ | |
1273 | smp_mb(); /* See above block comment. */ | |
1274 | } | |
1275 | ||
1276 | static void | |
1277 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1278 | struct rcu_state *rsp) | |
1279 | { | |
1280 | unsigned long flags; | |
1281 | struct rcu_data *rdp; | |
1282 | ||
1283 | head->func = func; | |
1284 | head->next = NULL; | |
1285 | ||
1286 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1287 | ||
1288 | /* | |
1289 | * Opportunistically note grace-period endings and beginnings. | |
1290 | * Note that we might see a beginning right after we see an | |
1291 | * end, but never vice versa, since this CPU has to pass through | |
1292 | * a quiescent state betweentimes. | |
1293 | */ | |
1294 | local_irq_save(flags); | |
1295 | rdp = rsp->rda[smp_processor_id()]; | |
1296 | rcu_process_gp_end(rsp, rdp); | |
1297 | check_for_new_grace_period(rsp, rdp); | |
1298 | ||
1299 | /* Add the callback to our list. */ | |
1300 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1301 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1302 | ||
1303 | /* Start a new grace period if one not already started. */ | |
fc2219d4 | 1304 | if (!rcu_gp_in_progress(rsp)) { |
64db4cff PM |
1305 | unsigned long nestflag; |
1306 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1307 | ||
1308 | spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1309 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | |
1310 | } | |
1311 | ||
37c72e56 PM |
1312 | /* |
1313 | * Force the grace period if too many callbacks or too long waiting. | |
1314 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1315 | * if some other CPU has recently done so. Also, don't bother | |
1316 | * invoking force_quiescent_state() if the newly enqueued callback | |
1317 | * is the only one waiting for a grace period to complete. | |
1318 | */ | |
1319 | if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { | |
64db4cff | 1320 | rdp->blimit = LONG_MAX; |
37c72e56 PM |
1321 | if (rsp->n_force_qs == rdp->n_force_qs_snap && |
1322 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1323 | force_quiescent_state(rsp, 0); | |
1324 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1325 | rdp->qlen_last_fqs_check = rdp->qlen; | |
ef631b0c | 1326 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1327 | force_quiescent_state(rsp, 1); |
1328 | local_irq_restore(flags); | |
1329 | } | |
1330 | ||
1331 | /* | |
d6714c22 | 1332 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1333 | */ |
d6714c22 | 1334 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1335 | { |
d6714c22 | 1336 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1337 | } |
d6714c22 | 1338 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1339 | |
1340 | /* | |
1341 | * Queue an RCU for invocation after a quicker grace period. | |
1342 | */ | |
1343 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1344 | { | |
1345 | __call_rcu(head, func, &rcu_bh_state); | |
1346 | } | |
1347 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1348 | ||
1349 | /* | |
1350 | * Check to see if there is any immediate RCU-related work to be done | |
1351 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1352 | * The checks are in order of increasing expense: checks that can be | |
1353 | * carried out against CPU-local state are performed first. However, | |
1354 | * we must check for CPU stalls first, else we might not get a chance. | |
1355 | */ | |
1356 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1357 | { | |
1358 | rdp->n_rcu_pending++; | |
1359 | ||
1360 | /* Check for CPU stalls, if enabled. */ | |
1361 | check_cpu_stall(rsp, rdp); | |
1362 | ||
1363 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
7ba5c840 PM |
1364 | if (rdp->qs_pending) { |
1365 | rdp->n_rp_qs_pending++; | |
64db4cff | 1366 | return 1; |
7ba5c840 | 1367 | } |
64db4cff PM |
1368 | |
1369 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1370 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1371 | rdp->n_rp_cb_ready++; | |
64db4cff | 1372 | return 1; |
7ba5c840 | 1373 | } |
64db4cff PM |
1374 | |
1375 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1376 | if (cpu_needs_another_gp(rsp, rdp)) { |
1377 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1378 | return 1; |
7ba5c840 | 1379 | } |
64db4cff PM |
1380 | |
1381 | /* Has another RCU grace period completed? */ | |
7ba5c840 PM |
1382 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ |
1383 | rdp->n_rp_gp_completed++; | |
64db4cff | 1384 | return 1; |
7ba5c840 | 1385 | } |
64db4cff PM |
1386 | |
1387 | /* Has a new RCU grace period started? */ | |
7ba5c840 PM |
1388 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ |
1389 | rdp->n_rp_gp_started++; | |
64db4cff | 1390 | return 1; |
7ba5c840 | 1391 | } |
64db4cff PM |
1392 | |
1393 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1394 | if (rcu_gp_in_progress(rsp) && |
7ba5c840 PM |
1395 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
1396 | rdp->n_rp_need_fqs++; | |
64db4cff | 1397 | return 1; |
7ba5c840 | 1398 | } |
64db4cff PM |
1399 | |
1400 | /* nothing to do */ | |
7ba5c840 | 1401 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1402 | return 0; |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * Check to see if there is any immediate RCU-related work to be done | |
1407 | * by the current CPU, returning 1 if so. This function is part of the | |
1408 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1409 | */ | |
a157229c | 1410 | static int rcu_pending(int cpu) |
64db4cff | 1411 | { |
d6714c22 | 1412 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1413 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1414 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1415 | } |
1416 | ||
1417 | /* | |
1418 | * Check to see if any future RCU-related work will need to be done | |
1419 | * by the current CPU, even if none need be done immediately, returning | |
1420 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1421 | * an exported member of the RCU API. | |
1422 | */ | |
1423 | int rcu_needs_cpu(int cpu) | |
1424 | { | |
1425 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1426 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1427 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1428 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1429 | } |
1430 | ||
d0ec774c PM |
1431 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1432 | static atomic_t rcu_barrier_cpu_count; | |
1433 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1434 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1435 | |
1436 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1437 | { | |
1438 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1439 | complete(&rcu_barrier_completion); | |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1444 | */ | |
1445 | static void rcu_barrier_func(void *type) | |
1446 | { | |
1447 | int cpu = smp_processor_id(); | |
1448 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1449 | void (*call_rcu_func)(struct rcu_head *head, | |
1450 | void (*func)(struct rcu_head *head)); | |
1451 | ||
1452 | atomic_inc(&rcu_barrier_cpu_count); | |
1453 | call_rcu_func = type; | |
1454 | call_rcu_func(head, rcu_barrier_callback); | |
1455 | } | |
1456 | ||
d0ec774c PM |
1457 | /* |
1458 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1459 | * RCU callbacks of the specified type to complete. | |
1460 | */ | |
e74f4c45 PM |
1461 | static void _rcu_barrier(struct rcu_state *rsp, |
1462 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1463 | void (*func)(struct rcu_head *head))) |
1464 | { | |
1465 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1466 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1467 | mutex_lock(&rcu_barrier_mutex); |
1468 | init_completion(&rcu_barrier_completion); | |
1469 | /* | |
1470 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1471 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1472 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1473 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1474 | * might complete its grace period before all of the other CPUs | |
1475 | * did their increment, causing this function to return too | |
1476 | * early. | |
1477 | */ | |
1478 | atomic_set(&rcu_barrier_cpu_count, 1); | |
e74f4c45 PM |
1479 | preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */ |
1480 | rcu_adopt_orphan_cbs(rsp); | |
d0ec774c | 1481 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); |
e74f4c45 | 1482 | preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */ |
d0ec774c PM |
1483 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1484 | complete(&rcu_barrier_completion); | |
1485 | wait_for_completion(&rcu_barrier_completion); | |
1486 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1487 | } |
d0ec774c PM |
1488 | |
1489 | /** | |
1490 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1491 | */ | |
1492 | void rcu_barrier_bh(void) | |
1493 | { | |
e74f4c45 | 1494 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1495 | } |
1496 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
1497 | ||
1498 | /** | |
1499 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
1500 | */ | |
1501 | void rcu_barrier_sched(void) | |
1502 | { | |
e74f4c45 | 1503 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
1504 | } |
1505 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
1506 | ||
64db4cff | 1507 | /* |
27569620 | 1508 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1509 | */ |
27569620 PM |
1510 | static void __init |
1511 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1512 | { |
1513 | unsigned long flags; | |
1514 | int i; | |
27569620 PM |
1515 | struct rcu_data *rdp = rsp->rda[cpu]; |
1516 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1517 | ||
1518 | /* Set up local state, ensuring consistent view of global state. */ | |
1519 | spin_lock_irqsave(&rnp->lock, flags); | |
1520 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | |
1521 | rdp->nxtlist = NULL; | |
1522 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1523 | rdp->nxttail[i] = &rdp->nxtlist; | |
1524 | rdp->qlen = 0; | |
1525 | #ifdef CONFIG_NO_HZ | |
1526 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1527 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1528 | rdp->cpu = cpu; | |
1529 | spin_unlock_irqrestore(&rnp->lock, flags); | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1534 | * offline event can be happening at a given time. Note also that we | |
1535 | * can accept some slop in the rsp->completed access due to the fact | |
1536 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1537 | */ |
e4fa4c97 | 1538 | static void __cpuinit |
f41d911f | 1539 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1540 | { |
1541 | unsigned long flags; | |
64db4cff PM |
1542 | long lastcomp; |
1543 | unsigned long mask; | |
1544 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1545 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1546 | ||
1547 | /* Set up local state, ensuring consistent view of global state. */ | |
1548 | spin_lock_irqsave(&rnp->lock, flags); | |
1549 | lastcomp = rsp->completed; | |
1550 | rdp->completed = lastcomp; | |
1551 | rdp->gpnum = lastcomp; | |
1552 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ | |
1553 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1554 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1555 | rdp->preemptable = preemptable; |
64db4cff | 1556 | rdp->passed_quiesc_completed = lastcomp - 1; |
37c72e56 PM |
1557 | rdp->qlen_last_fqs_check = 0; |
1558 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 1559 | rdp->blimit = blimit; |
64db4cff PM |
1560 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1561 | ||
1562 | /* | |
1563 | * A new grace period might start here. If so, we won't be part | |
1564 | * of it, but that is OK, as we are currently in a quiescent state. | |
1565 | */ | |
1566 | ||
1567 | /* Exclude any attempts to start a new GP on large systems. */ | |
1568 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
1569 | ||
1570 | /* Add CPU to rcu_node bitmasks. */ | |
1571 | rnp = rdp->mynode; | |
1572 | mask = rdp->grpmask; | |
1573 | do { | |
1574 | /* Exclude any attempts to start a new GP on small systems. */ | |
1575 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1576 | rnp->qsmaskinit |= mask; | |
1577 | mask = rnp->grpmask; | |
1578 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | |
1579 | rnp = rnp->parent; | |
1580 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1581 | ||
e7d8842e | 1582 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
1583 | } |
1584 | ||
1585 | static void __cpuinit rcu_online_cpu(int cpu) | |
1586 | { | |
f41d911f PM |
1587 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1588 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1589 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1590 | } |
1591 | ||
1592 | /* | |
f41d911f | 1593 | * Handle CPU online/offline notification events. |
64db4cff | 1594 | */ |
2e597558 PM |
1595 | int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1596 | unsigned long action, void *hcpu) | |
64db4cff PM |
1597 | { |
1598 | long cpu = (long)hcpu; | |
1599 | ||
1600 | switch (action) { | |
1601 | case CPU_UP_PREPARE: | |
1602 | case CPU_UP_PREPARE_FROZEN: | |
1603 | rcu_online_cpu(cpu); | |
1604 | break; | |
d0ec774c PM |
1605 | case CPU_DYING: |
1606 | case CPU_DYING_FROZEN: | |
1607 | /* | |
e74f4c45 | 1608 | * preempt_disable() in _rcu_barrier() prevents stop_machine(), |
d0ec774c | 1609 | * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);" |
e74f4c45 PM |
1610 | * returns, all online cpus have queued rcu_barrier_func(). |
1611 | * The dying CPU clears its cpu_online_mask bit and | |
1612 | * moves all of its RCU callbacks to ->orphan_cbs_list | |
1613 | * in the context of stop_machine(), so subsequent calls | |
1614 | * to _rcu_barrier() will adopt these callbacks and only | |
1615 | * then queue rcu_barrier_func() on all remaining CPUs. | |
d0ec774c | 1616 | */ |
e74f4c45 PM |
1617 | rcu_send_cbs_to_orphanage(&rcu_bh_state); |
1618 | rcu_send_cbs_to_orphanage(&rcu_sched_state); | |
1619 | rcu_preempt_send_cbs_to_orphanage(); | |
d0ec774c | 1620 | break; |
64db4cff PM |
1621 | case CPU_DEAD: |
1622 | case CPU_DEAD_FROZEN: | |
1623 | case CPU_UP_CANCELED: | |
1624 | case CPU_UP_CANCELED_FROZEN: | |
1625 | rcu_offline_cpu(cpu); | |
1626 | break; | |
1627 | default: | |
1628 | break; | |
1629 | } | |
1630 | return NOTIFY_OK; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * Compute the per-level fanout, either using the exact fanout specified | |
1635 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1636 | */ | |
1637 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1638 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1639 | { | |
1640 | int i; | |
1641 | ||
1642 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1643 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1644 | } | |
1645 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1646 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1647 | { | |
1648 | int ccur; | |
1649 | int cprv; | |
1650 | int i; | |
1651 | ||
1652 | cprv = NR_CPUS; | |
1653 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1654 | ccur = rsp->levelcnt[i]; | |
1655 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1656 | cprv = ccur; | |
1657 | } | |
1658 | } | |
1659 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1660 | ||
1661 | /* | |
1662 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1663 | */ | |
1664 | static void __init rcu_init_one(struct rcu_state *rsp) | |
1665 | { | |
1666 | int cpustride = 1; | |
1667 | int i; | |
1668 | int j; | |
1669 | struct rcu_node *rnp; | |
1670 | ||
1671 | /* Initialize the level-tracking arrays. */ | |
1672 | ||
1673 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1674 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1675 | rcu_init_levelspread(rsp); | |
1676 | ||
1677 | /* Initialize the elements themselves, starting from the leaves. */ | |
1678 | ||
1679 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1680 | cpustride *= rsp->levelspread[i]; | |
1681 | rnp = rsp->level[i]; | |
1682 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
88b91c7c | 1683 | spin_lock_init(&rnp->lock); |
f41d911f | 1684 | rnp->gpnum = 0; |
64db4cff PM |
1685 | rnp->qsmask = 0; |
1686 | rnp->qsmaskinit = 0; | |
1687 | rnp->grplo = j * cpustride; | |
1688 | rnp->grphi = (j + 1) * cpustride - 1; | |
1689 | if (rnp->grphi >= NR_CPUS) | |
1690 | rnp->grphi = NR_CPUS - 1; | |
1691 | if (i == 0) { | |
1692 | rnp->grpnum = 0; | |
1693 | rnp->grpmask = 0; | |
1694 | rnp->parent = NULL; | |
1695 | } else { | |
1696 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1697 | rnp->grpmask = 1UL << rnp->grpnum; | |
1698 | rnp->parent = rsp->level[i - 1] + | |
1699 | j / rsp->levelspread[i - 1]; | |
1700 | } | |
1701 | rnp->level = i; | |
f41d911f PM |
1702 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1703 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
64db4cff PM |
1704 | } |
1705 | } | |
88b91c7c | 1706 | lockdep_set_class(&rcu_get_root(rsp)->lock, &rcu_root_class); |
64db4cff PM |
1707 | } |
1708 | ||
1709 | /* | |
f41d911f PM |
1710 | * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used |
1711 | * nowhere else! Assigns leaf node pointers into each CPU's rcu_data | |
1712 | * structure. | |
64db4cff | 1713 | */ |
65cf8f86 | 1714 | #define RCU_INIT_FLAVOR(rsp, rcu_data) \ |
64db4cff | 1715 | do { \ |
a0b6c9a7 PM |
1716 | int i; \ |
1717 | int j; \ | |
1718 | struct rcu_node *rnp; \ | |
1719 | \ | |
65cf8f86 | 1720 | rcu_init_one(rsp); \ |
64db4cff PM |
1721 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
1722 | j = 0; \ | |
1723 | for_each_possible_cpu(i) { \ | |
1724 | if (i > rnp[j].grphi) \ | |
1725 | j++; \ | |
1726 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | |
1727 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | |
65cf8f86 | 1728 | rcu_boot_init_percpu_data(i, rsp); \ |
64db4cff PM |
1729 | } \ |
1730 | } while (0) | |
1731 | ||
64db4cff PM |
1732 | void __init __rcu_init(void) |
1733 | { | |
f41d911f | 1734 | rcu_bootup_announce(); |
64db4cff PM |
1735 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
1736 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | |
1737 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
65cf8f86 PM |
1738 | RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); |
1739 | RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); | |
f41d911f | 1740 | __rcu_init_preempt(); |
2e597558 | 1741 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
64db4cff PM |
1742 | } |
1743 | ||
1eba8f84 | 1744 | #include "rcutree_plugin.h" |