]>
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 - | |
28 | * Documentation/RCU | |
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> | |
38 | #include <asm/atomic.h> | |
39 | #include <linux/bitops.h> | |
40 | #include <linux/module.h> | |
41 | #include <linux/completion.h> | |
42 | #include <linux/moduleparam.h> | |
43 | #include <linux/percpu.h> | |
44 | #include <linux/notifier.h> | |
45 | #include <linux/cpu.h> | |
46 | #include <linux/mutex.h> | |
47 | #include <linux/time.h> | |
48 | ||
9f77da9f PM |
49 | #include "rcutree.h" |
50 | ||
64db4cff PM |
51 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
52 | static struct lock_class_key rcu_lock_key; | |
53 | struct lockdep_map rcu_lock_map = | |
54 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | |
55 | EXPORT_SYMBOL_GPL(rcu_lock_map); | |
56 | #endif | |
57 | ||
58 | /* Data structures. */ | |
59 | ||
60 | #define RCU_STATE_INITIALIZER(name) { \ | |
61 | .level = { &name.node[0] }, \ | |
62 | .levelcnt = { \ | |
63 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
64 | NUM_RCU_LVL_1, \ | |
65 | NUM_RCU_LVL_2, \ | |
66 | NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \ | |
67 | }, \ | |
68 | .signaled = RCU_SIGNAL_INIT, \ | |
69 | .gpnum = -300, \ | |
70 | .completed = -300, \ | |
71 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ | |
72 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ | |
73 | .n_force_qs = 0, \ | |
74 | .n_force_qs_ngp = 0, \ | |
75 | } | |
76 | ||
d6714c22 PM |
77 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
78 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 79 | |
6258c4fb IM |
80 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
81 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 82 | |
f41d911f PM |
83 | extern long rcu_batches_completed_sched(void); |
84 | static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, | |
85 | struct rcu_node *rnp, unsigned long flags); | |
86 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags); | |
c935a331 | 87 | #ifdef CONFIG_HOTPLUG_CPU |
33f76148 | 88 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp); |
c935a331 | 89 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
f41d911f PM |
90 | static void __rcu_process_callbacks(struct rcu_state *rsp, |
91 | struct rcu_data *rdp); | |
92 | static void __call_rcu(struct rcu_head *head, | |
93 | void (*func)(struct rcu_head *rcu), | |
94 | struct rcu_state *rsp); | |
95 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp); | |
96 | static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp, | |
97 | int preemptable); | |
98 | ||
99 | #include "rcutree_plugin.h" | |
100 | ||
b1f77b05 | 101 | /* |
d6714c22 | 102 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 103 | * how many quiescent states passed, just if there was at least |
d6714c22 | 104 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 105 | */ |
d6714c22 | 106 | void rcu_sched_qs(int cpu) |
b1f77b05 | 107 | { |
f41d911f PM |
108 | unsigned long flags; |
109 | struct rcu_data *rdp; | |
110 | ||
111 | local_irq_save(flags); | |
112 | rdp = &per_cpu(rcu_sched_data, cpu); | |
b1f77b05 IM |
113 | rdp->passed_quiesc = 1; |
114 | rdp->passed_quiesc_completed = rdp->completed; | |
f41d911f PM |
115 | rcu_preempt_qs(cpu); |
116 | local_irq_restore(flags); | |
b1f77b05 IM |
117 | } |
118 | ||
d6714c22 | 119 | void rcu_bh_qs(int cpu) |
b1f77b05 | 120 | { |
f41d911f PM |
121 | unsigned long flags; |
122 | struct rcu_data *rdp; | |
123 | ||
124 | local_irq_save(flags); | |
125 | rdp = &per_cpu(rcu_bh_data, cpu); | |
b1f77b05 IM |
126 | rdp->passed_quiesc = 1; |
127 | rdp->passed_quiesc_completed = rdp->completed; | |
f41d911f | 128 | local_irq_restore(flags); |
b1f77b05 | 129 | } |
64db4cff PM |
130 | |
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 | ||
142 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); | |
a157229c | 143 | static int rcu_pending(int cpu); |
64db4cff | 144 | |
d6714c22 PM |
145 | /* |
146 | * Return the number of RCU-sched batches processed thus far for debug & stats. | |
147 | */ | |
148 | long rcu_batches_completed_sched(void) | |
149 | { | |
150 | return rcu_sched_state.completed; | |
151 | } | |
152 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); | |
153 | ||
64db4cff PM |
154 | /* |
155 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
156 | */ | |
157 | long rcu_batches_completed_bh(void) | |
158 | { | |
159 | return rcu_bh_state.completed; | |
160 | } | |
161 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
162 | ||
163 | /* | |
164 | * Does the CPU have callbacks ready to be invoked? | |
165 | */ | |
166 | static int | |
167 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
168 | { | |
169 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
170 | } | |
171 | ||
172 | /* | |
173 | * Does the current CPU require a yet-as-unscheduled grace period? | |
174 | */ | |
175 | static int | |
176 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
177 | { | |
178 | /* ACCESS_ONCE() because we are accessing outside of lock. */ | |
179 | return *rdp->nxttail[RCU_DONE_TAIL] && | |
180 | ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum); | |
181 | } | |
182 | ||
183 | /* | |
184 | * Return the root node of the specified rcu_state structure. | |
185 | */ | |
186 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
187 | { | |
188 | return &rsp->node[0]; | |
189 | } | |
190 | ||
191 | #ifdef CONFIG_SMP | |
192 | ||
193 | /* | |
194 | * If the specified CPU is offline, tell the caller that it is in | |
195 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
196 | * Grace periods can end up waiting on an offline CPU when that | |
197 | * CPU is in the process of coming online -- it will be added to the | |
198 | * rcu_node bitmasks before it actually makes it online. The same thing | |
199 | * can happen while a CPU is in the process of coming online. Because this | |
200 | * race is quite rare, we check for it after detecting that the grace | |
201 | * period has been delayed rather than checking each and every CPU | |
202 | * each and every time we start a new grace period. | |
203 | */ | |
204 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
205 | { | |
206 | /* | |
207 | * If the CPU is offline, it is in a quiescent state. We can | |
208 | * trust its state not to change because interrupts are disabled. | |
209 | */ | |
210 | if (cpu_is_offline(rdp->cpu)) { | |
211 | rdp->offline_fqs++; | |
212 | return 1; | |
213 | } | |
214 | ||
f41d911f PM |
215 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
216 | if (rdp->preemptable) | |
217 | return 0; | |
218 | ||
64db4cff PM |
219 | /* The CPU is online, so send it a reschedule IPI. */ |
220 | if (rdp->cpu != smp_processor_id()) | |
221 | smp_send_reschedule(rdp->cpu); | |
222 | else | |
223 | set_need_resched(); | |
224 | rdp->resched_ipi++; | |
225 | return 0; | |
226 | } | |
227 | ||
228 | #endif /* #ifdef CONFIG_SMP */ | |
229 | ||
230 | #ifdef CONFIG_NO_HZ | |
231 | static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5); | |
232 | ||
233 | /** | |
234 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
235 | * | |
236 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
237 | * read-side critical sections can occur. (Though RCU read-side | |
238 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
239 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
240 | */ | |
241 | void rcu_enter_nohz(void) | |
242 | { | |
243 | unsigned long flags; | |
244 | struct rcu_dynticks *rdtp; | |
245 | ||
246 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
247 | local_irq_save(flags); | |
248 | rdtp = &__get_cpu_var(rcu_dynticks); | |
249 | rdtp->dynticks++; | |
250 | rdtp->dynticks_nesting--; | |
251 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); | |
252 | local_irq_restore(flags); | |
253 | } | |
254 | ||
255 | /* | |
256 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
257 | * | |
258 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
259 | * read-side critical sections normally occur. | |
260 | */ | |
261 | void rcu_exit_nohz(void) | |
262 | { | |
263 | unsigned long flags; | |
264 | struct rcu_dynticks *rdtp; | |
265 | ||
266 | local_irq_save(flags); | |
267 | rdtp = &__get_cpu_var(rcu_dynticks); | |
268 | rdtp->dynticks++; | |
269 | rdtp->dynticks_nesting++; | |
270 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); | |
271 | local_irq_restore(flags); | |
272 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
273 | } | |
274 | ||
275 | /** | |
276 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
277 | * | |
278 | * If the CPU was idle with dynamic ticks active, and there is no | |
279 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
280 | * RCU grace-period handling know that the CPU is active. | |
281 | */ | |
282 | void rcu_nmi_enter(void) | |
283 | { | |
284 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
285 | ||
286 | if (rdtp->dynticks & 0x1) | |
287 | return; | |
288 | rdtp->dynticks_nmi++; | |
289 | WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs); | |
290 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
291 | } | |
292 | ||
293 | /** | |
294 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
295 | * | |
296 | * If the CPU was idle with dynamic ticks active, and there is no | |
297 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
298 | * RCU grace-period handling know that the CPU is no longer active. | |
299 | */ | |
300 | void rcu_nmi_exit(void) | |
301 | { | |
302 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
303 | ||
304 | if (rdtp->dynticks & 0x1) | |
305 | return; | |
306 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
307 | rdtp->dynticks_nmi++; | |
308 | WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs); | |
309 | } | |
310 | ||
311 | /** | |
312 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
313 | * | |
314 | * If the CPU was idle with dynamic ticks active, this updates the | |
315 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
316 | */ | |
317 | void rcu_irq_enter(void) | |
318 | { | |
319 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
320 | ||
321 | if (rdtp->dynticks_nesting++) | |
322 | return; | |
323 | rdtp->dynticks++; | |
324 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); | |
325 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
326 | } | |
327 | ||
328 | /** | |
329 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
330 | * | |
331 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
332 | * to put let the RCU handling be aware that the CPU is going back to idle | |
333 | * with no ticks. | |
334 | */ | |
335 | void rcu_irq_exit(void) | |
336 | { | |
337 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
338 | ||
339 | if (--rdtp->dynticks_nesting) | |
340 | return; | |
341 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
342 | rdtp->dynticks++; | |
343 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); | |
344 | ||
345 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 346 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
347 | __get_cpu_var(rcu_bh_data).nxtlist) |
348 | set_need_resched(); | |
349 | } | |
350 | ||
351 | /* | |
352 | * Record the specified "completed" value, which is later used to validate | |
353 | * dynticks counter manipulations. Specify "rsp->completed - 1" to | |
354 | * unconditionally invalidate any future dynticks manipulations (which is | |
355 | * useful at the beginning of a grace period). | |
356 | */ | |
357 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
358 | { | |
359 | rsp->dynticks_completed = comp; | |
360 | } | |
361 | ||
362 | #ifdef CONFIG_SMP | |
363 | ||
364 | /* | |
365 | * Recall the previously recorded value of the completion for dynticks. | |
366 | */ | |
367 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
368 | { | |
369 | return rsp->dynticks_completed; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Snapshot the specified CPU's dynticks counter so that we can later | |
374 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
375 | * is already in a quiescent state courtesy of dynticks idle mode. | |
376 | */ | |
377 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
378 | { | |
379 | int ret; | |
380 | int snap; | |
381 | int snap_nmi; | |
382 | ||
383 | snap = rdp->dynticks->dynticks; | |
384 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
385 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
386 | rdp->dynticks_snap = snap; | |
387 | rdp->dynticks_nmi_snap = snap_nmi; | |
388 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
389 | if (ret) | |
390 | rdp->dynticks_fqs++; | |
391 | return ret; | |
392 | } | |
393 | ||
394 | /* | |
395 | * Return true if the specified CPU has passed through a quiescent | |
396 | * state by virtue of being in or having passed through an dynticks | |
397 | * idle state since the last call to dyntick_save_progress_counter() | |
398 | * for this same CPU. | |
399 | */ | |
400 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
401 | { | |
402 | long curr; | |
403 | long curr_nmi; | |
404 | long snap; | |
405 | long snap_nmi; | |
406 | ||
407 | curr = rdp->dynticks->dynticks; | |
408 | snap = rdp->dynticks_snap; | |
409 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
410 | snap_nmi = rdp->dynticks_nmi_snap; | |
411 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
412 | ||
413 | /* | |
414 | * If the CPU passed through or entered a dynticks idle phase with | |
415 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
416 | * already acknowledged the request to pass through a quiescent | |
417 | * state. Either way, that CPU cannot possibly be in an RCU | |
418 | * read-side critical section that started before the beginning | |
419 | * of the current RCU grace period. | |
420 | */ | |
421 | if ((curr != snap || (curr & 0x1) == 0) && | |
422 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
423 | rdp->dynticks_fqs++; | |
424 | return 1; | |
425 | } | |
426 | ||
427 | /* Go check for the CPU being offline. */ | |
428 | return rcu_implicit_offline_qs(rdp); | |
429 | } | |
430 | ||
431 | #endif /* #ifdef CONFIG_SMP */ | |
432 | ||
433 | #else /* #ifdef CONFIG_NO_HZ */ | |
434 | ||
435 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
436 | { | |
437 | } | |
438 | ||
439 | #ifdef CONFIG_SMP | |
440 | ||
441 | /* | |
442 | * If there are no dynticks, then the only way that a CPU can passively | |
443 | * be in a quiescent state is to be offline. Unlike dynticks idle, which | |
444 | * is a point in time during the prior (already finished) grace period, | |
445 | * an offline CPU is always in a quiescent state, and thus can be | |
446 | * unconditionally applied. So just return the current value of completed. | |
447 | */ | |
448 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
449 | { | |
450 | return rsp->completed; | |
451 | } | |
452 | ||
453 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
454 | { | |
455 | return 0; | |
456 | } | |
457 | ||
458 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
459 | { | |
460 | return rcu_implicit_offline_qs(rdp); | |
461 | } | |
462 | ||
463 | #endif /* #ifdef CONFIG_SMP */ | |
464 | ||
465 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
466 | ||
467 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
468 | ||
469 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
470 | { | |
471 | rsp->gp_start = jiffies; | |
472 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
473 | } | |
474 | ||
475 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
476 | { | |
477 | int cpu; | |
478 | long delta; | |
479 | unsigned long flags; | |
480 | struct rcu_node *rnp = rcu_get_root(rsp); | |
481 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
482 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
483 | ||
484 | /* Only let one CPU complain about others per time interval. */ | |
485 | ||
486 | spin_lock_irqsave(&rnp->lock, flags); | |
487 | delta = jiffies - rsp->jiffies_stall; | |
488 | if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { | |
489 | spin_unlock_irqrestore(&rnp->lock, flags); | |
490 | return; | |
491 | } | |
492 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
493 | spin_unlock_irqrestore(&rnp->lock, flags); | |
494 | ||
495 | /* OK, time to rat on our buddy... */ | |
496 | ||
497 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | |
498 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
f41d911f | 499 | rcu_print_task_stall(rnp); |
64db4cff PM |
500 | if (rnp_cur->qsmask == 0) |
501 | continue; | |
502 | for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) | |
503 | if (rnp_cur->qsmask & (1UL << cpu)) | |
504 | printk(" %d", rnp_cur->grplo + cpu); | |
505 | } | |
506 | printk(" (detected by %d, t=%ld jiffies)\n", | |
507 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | |
508 | force_quiescent_state(rsp, 0); /* Kick them all. */ | |
509 | } | |
510 | ||
511 | static void print_cpu_stall(struct rcu_state *rsp) | |
512 | { | |
513 | unsigned long flags; | |
514 | struct rcu_node *rnp = rcu_get_root(rsp); | |
515 | ||
516 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | |
517 | smp_processor_id(), jiffies - rsp->gp_start); | |
518 | dump_stack(); | |
519 | spin_lock_irqsave(&rnp->lock, flags); | |
520 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) | |
521 | rsp->jiffies_stall = | |
522 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
523 | spin_unlock_irqrestore(&rnp->lock, flags); | |
524 | set_need_resched(); /* kick ourselves to get things going. */ | |
525 | } | |
526 | ||
527 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
528 | { | |
529 | long delta; | |
530 | struct rcu_node *rnp; | |
531 | ||
532 | delta = jiffies - rsp->jiffies_stall; | |
533 | rnp = rdp->mynode; | |
534 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
535 | ||
536 | /* We haven't checked in, so go dump stack. */ | |
537 | print_cpu_stall(rsp); | |
538 | ||
539 | } else if (rsp->gpnum != rsp->completed && | |
540 | delta >= RCU_STALL_RAT_DELAY) { | |
541 | ||
542 | /* They had two time units to dump stack, so complain. */ | |
543 | print_other_cpu_stall(rsp); | |
544 | } | |
545 | } | |
546 | ||
547 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
548 | ||
549 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
550 | { | |
551 | } | |
552 | ||
553 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
554 | { | |
555 | } | |
556 | ||
557 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
558 | ||
559 | /* | |
560 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
561 | * This is used both when we started the grace period and when we notice | |
562 | * that someone else started the grace period. | |
563 | */ | |
564 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) | |
565 | { | |
566 | rdp->qs_pending = 1; | |
567 | rdp->passed_quiesc = 0; | |
568 | rdp->gpnum = rsp->gpnum; | |
64db4cff PM |
569 | } |
570 | ||
571 | /* | |
572 | * Did someone else start a new RCU grace period start since we last | |
573 | * checked? Update local state appropriately if so. Must be called | |
574 | * on the CPU corresponding to rdp. | |
575 | */ | |
576 | static int | |
577 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
578 | { | |
579 | unsigned long flags; | |
580 | int ret = 0; | |
581 | ||
582 | local_irq_save(flags); | |
583 | if (rdp->gpnum != rsp->gpnum) { | |
584 | note_new_gpnum(rsp, rdp); | |
585 | ret = 1; | |
586 | } | |
587 | local_irq_restore(flags); | |
588 | return ret; | |
589 | } | |
590 | ||
591 | /* | |
592 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
593 | * in preparation for detecting the next grace period. The caller must hold | |
594 | * the root node's ->lock, which is released before return. Hard irqs must | |
595 | * be disabled. | |
596 | */ | |
597 | static void | |
598 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
599 | __releases(rcu_get_root(rsp)->lock) | |
600 | { | |
601 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
602 | struct rcu_node *rnp = rcu_get_root(rsp); | |
603 | struct rcu_node *rnp_cur; | |
604 | struct rcu_node *rnp_end; | |
605 | ||
606 | if (!cpu_needs_another_gp(rsp, rdp)) { | |
607 | spin_unlock_irqrestore(&rnp->lock, flags); | |
608 | return; | |
609 | } | |
610 | ||
611 | /* Advance to a new grace period and initialize state. */ | |
612 | rsp->gpnum++; | |
613 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
614 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
615 | record_gp_stall_check_time(rsp); |
616 | dyntick_record_completed(rsp, rsp->completed - 1); | |
617 | note_new_gpnum(rsp, rdp); | |
618 | ||
619 | /* | |
620 | * Because we are first, we know that all our callbacks will | |
621 | * be covered by this upcoming grace period, even the ones | |
622 | * that were registered arbitrarily recently. | |
623 | */ | |
624 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
625 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
626 | ||
627 | /* Special-case the common single-level case. */ | |
628 | if (NUM_RCU_NODES == 1) { | |
629 | rnp->qsmask = rnp->qsmaskinit; | |
c12172c0 | 630 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
64db4cff PM |
631 | spin_unlock_irqrestore(&rnp->lock, flags); |
632 | return; | |
633 | } | |
634 | ||
635 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | |
636 | ||
637 | ||
638 | /* Exclude any concurrent CPU-hotplug operations. */ | |
639 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
640 | ||
641 | /* | |
642 | * Set the quiescent-state-needed bits in all the non-leaf RCU | |
643 | * nodes for all currently online CPUs. This operation relies | |
644 | * on the layout of the hierarchy within the rsp->node[] array. | |
645 | * Note that other CPUs will access only the leaves of the | |
646 | * hierarchy, which still indicate that no grace period is in | |
647 | * progress. In addition, we have excluded CPU-hotplug operations. | |
648 | * | |
649 | * We therefore do not need to hold any locks. Any required | |
650 | * memory barriers will be supplied by the locks guarding the | |
651 | * leaf rcu_nodes in the hierarchy. | |
652 | */ | |
653 | ||
654 | rnp_end = rsp->level[NUM_RCU_LVLS - 1]; | |
655 | for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) | |
656 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | |
657 | ||
658 | /* | |
659 | * Now set up the leaf nodes. Here we must be careful. First, | |
660 | * we need to hold the lock in order to exclude other CPUs, which | |
661 | * might be contending for the leaf nodes' locks. Second, as | |
662 | * soon as we initialize a given leaf node, its CPUs might run | |
663 | * up the rest of the hierarchy. We must therefore acquire locks | |
664 | * for each node that we touch during this stage. (But we still | |
665 | * are excluding CPU-hotplug operations.) | |
666 | * | |
667 | * Note that the grace period cannot complete until we finish | |
668 | * the initialization process, as there will be at least one | |
669 | * qsmask bit set in the root node until that time, namely the | |
670 | * one corresponding to this CPU. | |
671 | */ | |
672 | rnp_end = &rsp->node[NUM_RCU_NODES]; | |
673 | rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
674 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
675 | spin_lock(&rnp_cur->lock); /* irqs already disabled. */ | |
676 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | |
677 | spin_unlock(&rnp_cur->lock); /* irqs already disabled. */ | |
678 | } | |
679 | ||
680 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ | |
681 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
682 | } | |
683 | ||
684 | /* | |
685 | * Advance this CPU's callbacks, but only if the current grace period | |
686 | * has ended. This may be called only from the CPU to whom the rdp | |
687 | * belongs. | |
688 | */ | |
689 | static void | |
690 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
691 | { | |
692 | long completed_snap; | |
693 | unsigned long flags; | |
694 | ||
695 | local_irq_save(flags); | |
696 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ | |
697 | ||
698 | /* Did another grace period end? */ | |
699 | if (rdp->completed != completed_snap) { | |
700 | ||
701 | /* Advance callbacks. No harm if list empty. */ | |
702 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
703 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
704 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
705 | ||
706 | /* Remember that we saw this grace-period completion. */ | |
707 | rdp->completed = completed_snap; | |
708 | } | |
709 | local_irq_restore(flags); | |
710 | } | |
711 | ||
f41d911f PM |
712 | /* |
713 | * Clean up after the prior grace period and let rcu_start_gp() start up | |
714 | * the next grace period if one is needed. Note that the caller must | |
715 | * hold rnp->lock, as required by rcu_start_gp(), which will release it. | |
716 | */ | |
717 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) | |
718 | __releases(rnp->lock) | |
719 | { | |
720 | rsp->completed = rsp->gpnum; | |
721 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); | |
722 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ | |
723 | } | |
724 | ||
64db4cff PM |
725 | /* |
726 | * Similar to cpu_quiet(), for which it is a helper function. Allows | |
727 | * a group of CPUs to be quieted at one go, though all the CPUs in the | |
728 | * group must be represented by the same leaf rcu_node structure. | |
729 | * That structure's lock must be held upon entry, and it is released | |
730 | * before return. | |
731 | */ | |
732 | static void | |
733 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, | |
734 | unsigned long flags) | |
735 | __releases(rnp->lock) | |
736 | { | |
737 | /* Walk up the rcu_node hierarchy. */ | |
738 | for (;;) { | |
739 | if (!(rnp->qsmask & mask)) { | |
740 | ||
741 | /* Our bit has already been cleared, so done. */ | |
742 | spin_unlock_irqrestore(&rnp->lock, flags); | |
743 | return; | |
744 | } | |
745 | rnp->qsmask &= ~mask; | |
f41d911f | 746 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
747 | |
748 | /* Other bits still set at this level, so done. */ | |
749 | spin_unlock_irqrestore(&rnp->lock, flags); | |
750 | return; | |
751 | } | |
752 | mask = rnp->grpmask; | |
753 | if (rnp->parent == NULL) { | |
754 | ||
755 | /* No more levels. Exit loop holding root lock. */ | |
756 | ||
757 | break; | |
758 | } | |
759 | spin_unlock_irqrestore(&rnp->lock, flags); | |
760 | rnp = rnp->parent; | |
761 | spin_lock_irqsave(&rnp->lock, flags); | |
762 | } | |
763 | ||
764 | /* | |
765 | * Get here if we are the last CPU to pass through a quiescent | |
f41d911f PM |
766 | * state for this grace period. Invoke cpu_quiet_msk_finish() |
767 | * to clean up and start the next grace period if one is needed. | |
64db4cff | 768 | */ |
f41d911f | 769 | cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
770 | } |
771 | ||
772 | /* | |
773 | * Record a quiescent state for the specified CPU, which must either be | |
774 | * the current CPU or an offline CPU. The lastcomp argument is used to | |
775 | * make sure we are still in the grace period of interest. We don't want | |
776 | * to end the current grace period based on quiescent states detected in | |
777 | * an earlier grace period! | |
778 | */ | |
779 | static void | |
780 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) | |
781 | { | |
782 | unsigned long flags; | |
783 | unsigned long mask; | |
784 | struct rcu_node *rnp; | |
785 | ||
786 | rnp = rdp->mynode; | |
787 | spin_lock_irqsave(&rnp->lock, flags); | |
788 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { | |
789 | ||
790 | /* | |
791 | * Someone beat us to it for this grace period, so leave. | |
792 | * The race with GP start is resolved by the fact that we | |
793 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
794 | * cannot yet be initialized -- so we would simply find our | |
795 | * CPU's bit already cleared in cpu_quiet_msk() if this race | |
796 | * occurred. | |
797 | */ | |
798 | rdp->passed_quiesc = 0; /* try again later! */ | |
799 | spin_unlock_irqrestore(&rnp->lock, flags); | |
800 | return; | |
801 | } | |
802 | mask = rdp->grpmask; | |
803 | if ((rnp->qsmask & mask) == 0) { | |
804 | spin_unlock_irqrestore(&rnp->lock, flags); | |
805 | } else { | |
806 | rdp->qs_pending = 0; | |
807 | ||
808 | /* | |
809 | * This GP can't end until cpu checks in, so all of our | |
810 | * callbacks can be processed during the next GP. | |
811 | */ | |
812 | rdp = rsp->rda[smp_processor_id()]; | |
813 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
814 | ||
815 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ | |
816 | } | |
817 | } | |
818 | ||
819 | /* | |
820 | * Check to see if there is a new grace period of which this CPU | |
821 | * is not yet aware, and if so, set up local rcu_data state for it. | |
822 | * Otherwise, see if this CPU has just passed through its first | |
823 | * quiescent state for this grace period, and record that fact if so. | |
824 | */ | |
825 | static void | |
826 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
827 | { | |
828 | /* If there is now a new grace period, record and return. */ | |
829 | if (check_for_new_grace_period(rsp, rdp)) | |
830 | return; | |
831 | ||
832 | /* | |
833 | * Does this CPU still need to do its part for current grace period? | |
834 | * If no, return and let the other CPUs do their part as well. | |
835 | */ | |
836 | if (!rdp->qs_pending) | |
837 | return; | |
838 | ||
839 | /* | |
840 | * Was there a quiescent state since the beginning of the grace | |
841 | * period? If no, then exit and wait for the next call. | |
842 | */ | |
843 | if (!rdp->passed_quiesc) | |
844 | return; | |
845 | ||
846 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ | |
847 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
848 | } | |
849 | ||
850 | #ifdef CONFIG_HOTPLUG_CPU | |
851 | ||
852 | /* | |
853 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
854 | * and move all callbacks from the outgoing CPU to the current one. | |
855 | */ | |
856 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
857 | { | |
858 | int i; | |
859 | unsigned long flags; | |
860 | long lastcomp; | |
861 | unsigned long mask; | |
862 | struct rcu_data *rdp = rsp->rda[cpu]; | |
863 | struct rcu_data *rdp_me; | |
864 | struct rcu_node *rnp; | |
865 | ||
866 | /* Exclude any attempts to start a new grace period. */ | |
867 | spin_lock_irqsave(&rsp->onofflock, flags); | |
868 | ||
869 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
870 | rnp = rdp->mynode; | |
871 | mask = rdp->grpmask; /* rnp->grplo is constant. */ | |
872 | do { | |
873 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
874 | rnp->qsmaskinit &= ~mask; | |
875 | if (rnp->qsmaskinit != 0) { | |
f41d911f | 876 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
877 | break; |
878 | } | |
879 | mask = rnp->grpmask; | |
f41d911f | 880 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
881 | rnp = rnp->parent; |
882 | } while (rnp != NULL); | |
883 | lastcomp = rsp->completed; | |
884 | ||
885 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
886 | ||
887 | /* Being offline is a quiescent state, so go record it. */ | |
888 | cpu_quiet(cpu, rsp, rdp, lastcomp); | |
889 | ||
890 | /* | |
891 | * Move callbacks from the outgoing CPU to the running CPU. | |
892 | * Note that the outgoing CPU is now quiscent, so it is now | |
d6714c22 | 893 | * (uncharacteristically) safe to access its rcu_data structure. |
64db4cff PM |
894 | * Note also that we must carefully retain the order of the |
895 | * outgoing CPU's callbacks in order for rcu_barrier() to work | |
896 | * correctly. Finally, note that we start all the callbacks | |
897 | * afresh, even those that have passed through a grace period | |
898 | * and are therefore ready to invoke. The theory is that hotplug | |
899 | * events are rare, and that if they are frequent enough to | |
900 | * indefinitely delay callbacks, you have far worse things to | |
901 | * be worrying about. | |
902 | */ | |
903 | rdp_me = rsp->rda[smp_processor_id()]; | |
904 | if (rdp->nxtlist != NULL) { | |
905 | *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
906 | rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
907 | rdp->nxtlist = NULL; | |
908 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
909 | rdp->nxttail[i] = &rdp->nxtlist; | |
910 | rdp_me->qlen += rdp->qlen; | |
911 | rdp->qlen = 0; | |
912 | } | |
913 | local_irq_restore(flags); | |
914 | } | |
915 | ||
916 | /* | |
917 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
918 | * callbacks that it might have to the current CPU. This code assumes | |
919 | * that at least one CPU in the system will remain running at all times. | |
920 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
921 | */ | |
922 | static void rcu_offline_cpu(int cpu) | |
923 | { | |
d6714c22 | 924 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 925 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 926 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
927 | } |
928 | ||
929 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
930 | ||
931 | static void rcu_offline_cpu(int cpu) | |
932 | { | |
933 | } | |
934 | ||
935 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
936 | ||
937 | /* | |
938 | * Invoke any RCU callbacks that have made it to the end of their grace | |
939 | * period. Thottle as specified by rdp->blimit. | |
940 | */ | |
941 | static void rcu_do_batch(struct rcu_data *rdp) | |
942 | { | |
943 | unsigned long flags; | |
944 | struct rcu_head *next, *list, **tail; | |
945 | int count; | |
946 | ||
947 | /* If no callbacks are ready, just return.*/ | |
948 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
949 | return; | |
950 | ||
951 | /* | |
952 | * Extract the list of ready callbacks, disabling to prevent | |
953 | * races with call_rcu() from interrupt handlers. | |
954 | */ | |
955 | local_irq_save(flags); | |
956 | list = rdp->nxtlist; | |
957 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
958 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
959 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
960 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
961 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
962 | rdp->nxttail[count] = &rdp->nxtlist; | |
963 | local_irq_restore(flags); | |
964 | ||
965 | /* Invoke callbacks. */ | |
966 | count = 0; | |
967 | while (list) { | |
968 | next = list->next; | |
969 | prefetch(next); | |
970 | list->func(list); | |
971 | list = next; | |
972 | if (++count >= rdp->blimit) | |
973 | break; | |
974 | } | |
975 | ||
976 | local_irq_save(flags); | |
977 | ||
978 | /* Update count, and requeue any remaining callbacks. */ | |
979 | rdp->qlen -= count; | |
980 | if (list != NULL) { | |
981 | *tail = rdp->nxtlist; | |
982 | rdp->nxtlist = list; | |
983 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
984 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
985 | rdp->nxttail[count] = tail; | |
986 | else | |
987 | break; | |
988 | } | |
989 | ||
990 | /* Reinstate batch limit if we have worked down the excess. */ | |
991 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
992 | rdp->blimit = blimit; | |
993 | ||
994 | local_irq_restore(flags); | |
995 | ||
996 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
997 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
998 | raise_softirq(RCU_SOFTIRQ); | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1003 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1004 | * Also schedule the RCU softirq handler. | |
1005 | * | |
1006 | * This function must be called with hardirqs disabled. It is normally | |
1007 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1008 | * false, there is no point in invoking rcu_check_callbacks(). | |
1009 | */ | |
1010 | void rcu_check_callbacks(int cpu, int user) | |
1011 | { | |
a157229c PM |
1012 | if (!rcu_pending(cpu)) |
1013 | return; /* if nothing for RCU to do. */ | |
64db4cff | 1014 | if (user || |
a6826048 PM |
1015 | (idle_cpu(cpu) && rcu_scheduler_active && |
1016 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1017 | |
1018 | /* | |
1019 | * Get here if this CPU took its interrupt from user | |
1020 | * mode or from the idle loop, and if this is not a | |
1021 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1022 | * a quiescent state, so note it. |
64db4cff PM |
1023 | * |
1024 | * No memory barrier is required here because both | |
d6714c22 PM |
1025 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1026 | * variables that other CPUs neither access nor modify, | |
1027 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1028 | */ |
1029 | ||
d6714c22 PM |
1030 | rcu_sched_qs(cpu); |
1031 | rcu_bh_qs(cpu); | |
64db4cff PM |
1032 | |
1033 | } else if (!in_softirq()) { | |
1034 | ||
1035 | /* | |
1036 | * Get here if this CPU did not take its interrupt from | |
1037 | * softirq, in other words, if it is not interrupting | |
1038 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1039 | * critical section, so note it. |
64db4cff PM |
1040 | */ |
1041 | ||
d6714c22 | 1042 | rcu_bh_qs(cpu); |
64db4cff | 1043 | } |
f41d911f | 1044 | rcu_preempt_check_callbacks(cpu); |
64db4cff PM |
1045 | raise_softirq(RCU_SOFTIRQ); |
1046 | } | |
1047 | ||
1048 | #ifdef CONFIG_SMP | |
1049 | ||
1050 | /* | |
1051 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1052 | * have not yet encountered a quiescent state, using the function specified. | |
1053 | * Returns 1 if the current grace period ends while scanning (possibly | |
1054 | * because we made it end). | |
1055 | */ | |
1056 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, | |
1057 | int (*f)(struct rcu_data *)) | |
1058 | { | |
1059 | unsigned long bit; | |
1060 | int cpu; | |
1061 | unsigned long flags; | |
1062 | unsigned long mask; | |
1063 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
1064 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
1065 | ||
1066 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
1067 | mask = 0; | |
1068 | spin_lock_irqsave(&rnp_cur->lock, flags); | |
1069 | if (rsp->completed != lastcomp) { | |
1070 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1071 | return 1; | |
1072 | } | |
1073 | if (rnp_cur->qsmask == 0) { | |
1074 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1075 | continue; | |
1076 | } | |
1077 | cpu = rnp_cur->grplo; | |
1078 | bit = 1; | |
1079 | for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { | |
1080 | if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) | |
1081 | mask |= bit; | |
1082 | } | |
1083 | if (mask != 0 && rsp->completed == lastcomp) { | |
1084 | ||
1085 | /* cpu_quiet_msk() releases rnp_cur->lock. */ | |
1086 | cpu_quiet_msk(mask, rsp, rnp_cur, flags); | |
1087 | continue; | |
1088 | } | |
1089 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1090 | } | |
1091 | return 0; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * Force quiescent states on reluctant CPUs, and also detect which | |
1096 | * CPUs are in dyntick-idle mode. | |
1097 | */ | |
1098 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1099 | { | |
1100 | unsigned long flags; | |
1101 | long lastcomp; | |
64db4cff PM |
1102 | struct rcu_node *rnp = rcu_get_root(rsp); |
1103 | u8 signaled; | |
1104 | ||
1105 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) | |
1106 | return; /* No grace period in progress, nothing to force. */ | |
1107 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | |
1108 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | |
1109 | return; /* Someone else is already on the job. */ | |
1110 | } | |
1111 | if (relaxed && | |
ef631b0c | 1112 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
64db4cff PM |
1113 | goto unlock_ret; /* no emergency and done recently. */ |
1114 | rsp->n_force_qs++; | |
1115 | spin_lock(&rnp->lock); | |
1116 | lastcomp = rsp->completed; | |
1117 | signaled = rsp->signaled; | |
1118 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
1119 | if (lastcomp == rsp->gpnum) { |
1120 | rsp->n_force_qs_ngp++; | |
1121 | spin_unlock(&rnp->lock); | |
1122 | goto unlock_ret; /* no GP in progress, time updated. */ | |
1123 | } | |
1124 | spin_unlock(&rnp->lock); | |
1125 | switch (signaled) { | |
1126 | case RCU_GP_INIT: | |
1127 | ||
1128 | break; /* grace period still initializing, ignore. */ | |
1129 | ||
1130 | case RCU_SAVE_DYNTICK: | |
1131 | ||
1132 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) | |
1133 | break; /* So gcc recognizes the dead code. */ | |
1134 | ||
1135 | /* Record dyntick-idle state. */ | |
1136 | if (rcu_process_dyntick(rsp, lastcomp, | |
1137 | dyntick_save_progress_counter)) | |
1138 | goto unlock_ret; | |
1139 | ||
1140 | /* Update state, record completion counter. */ | |
1141 | spin_lock(&rnp->lock); | |
1142 | if (lastcomp == rsp->completed) { | |
1143 | rsp->signaled = RCU_FORCE_QS; | |
1144 | dyntick_record_completed(rsp, lastcomp); | |
1145 | } | |
1146 | spin_unlock(&rnp->lock); | |
1147 | break; | |
1148 | ||
1149 | case RCU_FORCE_QS: | |
1150 | ||
1151 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1152 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), | |
1153 | rcu_implicit_dynticks_qs)) | |
1154 | goto unlock_ret; | |
1155 | ||
1156 | /* Leave state in case more forcing is required. */ | |
1157 | ||
1158 | break; | |
1159 | } | |
1160 | unlock_ret: | |
1161 | spin_unlock_irqrestore(&rsp->fqslock, flags); | |
1162 | } | |
1163 | ||
1164 | #else /* #ifdef CONFIG_SMP */ | |
1165 | ||
1166 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1167 | { | |
1168 | set_need_resched(); | |
1169 | } | |
1170 | ||
1171 | #endif /* #else #ifdef CONFIG_SMP */ | |
1172 | ||
1173 | /* | |
1174 | * This does the RCU processing work from softirq context for the | |
1175 | * specified rcu_state and rcu_data structures. This may be called | |
1176 | * only from the CPU to whom the rdp belongs. | |
1177 | */ | |
1178 | static void | |
1179 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1180 | { | |
1181 | unsigned long flags; | |
1182 | ||
2e597558 PM |
1183 | WARN_ON_ONCE(rdp->beenonline == 0); |
1184 | ||
64db4cff PM |
1185 | /* |
1186 | * If an RCU GP has gone long enough, go check for dyntick | |
1187 | * idle CPUs and, if needed, send resched IPIs. | |
1188 | */ | |
ef631b0c | 1189 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1190 | force_quiescent_state(rsp, 1); |
1191 | ||
1192 | /* | |
1193 | * Advance callbacks in response to end of earlier grace | |
1194 | * period that some other CPU ended. | |
1195 | */ | |
1196 | rcu_process_gp_end(rsp, rdp); | |
1197 | ||
1198 | /* Update RCU state based on any recent quiescent states. */ | |
1199 | rcu_check_quiescent_state(rsp, rdp); | |
1200 | ||
1201 | /* Does this CPU require a not-yet-started grace period? */ | |
1202 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1203 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | |
1204 | rcu_start_gp(rsp, flags); /* releases above lock */ | |
1205 | } | |
1206 | ||
1207 | /* If there are callbacks ready, invoke them. */ | |
1208 | rcu_do_batch(rdp); | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * Do softirq processing for the current CPU. | |
1213 | */ | |
1214 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1215 | { | |
1216 | /* | |
1217 | * Memory references from any prior RCU read-side critical sections | |
1218 | * executed by the interrupted code must be seen before any RCU | |
1219 | * grace-period manipulations below. | |
1220 | */ | |
1221 | smp_mb(); /* See above block comment. */ | |
1222 | ||
d6714c22 PM |
1223 | __rcu_process_callbacks(&rcu_sched_state, |
1224 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1225 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1226 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1227 | |
1228 | /* | |
1229 | * Memory references from any later RCU read-side critical sections | |
1230 | * executed by the interrupted code must be seen after any RCU | |
1231 | * grace-period manipulations above. | |
1232 | */ | |
1233 | smp_mb(); /* See above block comment. */ | |
1234 | } | |
1235 | ||
1236 | static void | |
1237 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1238 | struct rcu_state *rsp) | |
1239 | { | |
1240 | unsigned long flags; | |
1241 | struct rcu_data *rdp; | |
1242 | ||
1243 | head->func = func; | |
1244 | head->next = NULL; | |
1245 | ||
1246 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1247 | ||
1248 | /* | |
1249 | * Opportunistically note grace-period endings and beginnings. | |
1250 | * Note that we might see a beginning right after we see an | |
1251 | * end, but never vice versa, since this CPU has to pass through | |
1252 | * a quiescent state betweentimes. | |
1253 | */ | |
1254 | local_irq_save(flags); | |
1255 | rdp = rsp->rda[smp_processor_id()]; | |
1256 | rcu_process_gp_end(rsp, rdp); | |
1257 | check_for_new_grace_period(rsp, rdp); | |
1258 | ||
1259 | /* Add the callback to our list. */ | |
1260 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1261 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1262 | ||
1263 | /* Start a new grace period if one not already started. */ | |
1264 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { | |
1265 | unsigned long nestflag; | |
1266 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1267 | ||
1268 | spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1269 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | |
1270 | } | |
1271 | ||
1272 | /* Force the grace period if too many callbacks or too long waiting. */ | |
1273 | if (unlikely(++rdp->qlen > qhimark)) { | |
1274 | rdp->blimit = LONG_MAX; | |
1275 | force_quiescent_state(rsp, 0); | |
ef631b0c | 1276 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1277 | force_quiescent_state(rsp, 1); |
1278 | local_irq_restore(flags); | |
1279 | } | |
1280 | ||
1281 | /* | |
d6714c22 PM |
1282 | * Queue an RCU-sched callback for invocation after a grace period. |
1283 | */ | |
1284 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1285 | { | |
1286 | __call_rcu(head, func, &rcu_sched_state); | |
1287 | } | |
1288 | EXPORT_SYMBOL_GPL(call_rcu_sched); | |
1289 | ||
64db4cff PM |
1290 | /* |
1291 | * Queue an RCU for invocation after a quicker grace period. | |
1292 | */ | |
1293 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1294 | { | |
1295 | __call_rcu(head, func, &rcu_bh_state); | |
1296 | } | |
1297 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1298 | ||
1299 | /* | |
1300 | * Check to see if there is any immediate RCU-related work to be done | |
1301 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1302 | * The checks are in order of increasing expense: checks that can be | |
1303 | * carried out against CPU-local state are performed first. However, | |
1304 | * we must check for CPU stalls first, else we might not get a chance. | |
1305 | */ | |
1306 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1307 | { | |
1308 | rdp->n_rcu_pending++; | |
1309 | ||
1310 | /* Check for CPU stalls, if enabled. */ | |
1311 | check_cpu_stall(rsp, rdp); | |
1312 | ||
1313 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
7ba5c840 PM |
1314 | if (rdp->qs_pending) { |
1315 | rdp->n_rp_qs_pending++; | |
64db4cff | 1316 | return 1; |
7ba5c840 | 1317 | } |
64db4cff PM |
1318 | |
1319 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1320 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1321 | rdp->n_rp_cb_ready++; | |
64db4cff | 1322 | return 1; |
7ba5c840 | 1323 | } |
64db4cff PM |
1324 | |
1325 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1326 | if (cpu_needs_another_gp(rsp, rdp)) { |
1327 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1328 | return 1; |
7ba5c840 | 1329 | } |
64db4cff PM |
1330 | |
1331 | /* Has another RCU grace period completed? */ | |
7ba5c840 PM |
1332 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ |
1333 | rdp->n_rp_gp_completed++; | |
64db4cff | 1334 | return 1; |
7ba5c840 | 1335 | } |
64db4cff PM |
1336 | |
1337 | /* Has a new RCU grace period started? */ | |
7ba5c840 PM |
1338 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ |
1339 | rdp->n_rp_gp_started++; | |
64db4cff | 1340 | return 1; |
7ba5c840 | 1341 | } |
64db4cff PM |
1342 | |
1343 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
1344 | if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && | |
7ba5c840 PM |
1345 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
1346 | rdp->n_rp_need_fqs++; | |
64db4cff | 1347 | return 1; |
7ba5c840 | 1348 | } |
64db4cff PM |
1349 | |
1350 | /* nothing to do */ | |
7ba5c840 | 1351 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1352 | return 0; |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Check to see if there is any immediate RCU-related work to be done | |
1357 | * by the current CPU, returning 1 if so. This function is part of the | |
1358 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1359 | */ | |
a157229c | 1360 | static int rcu_pending(int cpu) |
64db4cff | 1361 | { |
d6714c22 | 1362 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1363 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1364 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1365 | } |
1366 | ||
1367 | /* | |
1368 | * Check to see if any future RCU-related work will need to be done | |
1369 | * by the current CPU, even if none need be done immediately, returning | |
1370 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1371 | * an exported member of the RCU API. | |
1372 | */ | |
1373 | int rcu_needs_cpu(int cpu) | |
1374 | { | |
1375 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1376 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1377 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1378 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1379 | } |
1380 | ||
1381 | /* | |
27569620 PM |
1382 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
1383 | */ | |
1384 | static void __init | |
1385 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
1386 | { | |
1387 | unsigned long flags; | |
1388 | int i; | |
1389 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1390 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1391 | ||
1392 | /* Set up local state, ensuring consistent view of global state. */ | |
1393 | spin_lock_irqsave(&rnp->lock, flags); | |
1394 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | |
1395 | rdp->nxtlist = NULL; | |
1396 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1397 | rdp->nxttail[i] = &rdp->nxtlist; | |
1398 | rdp->qlen = 0; | |
1399 | #ifdef CONFIG_NO_HZ | |
1400 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1401 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1402 | rdp->cpu = cpu; | |
1403 | spin_unlock_irqrestore(&rnp->lock, flags); | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1408 | * offline event can be happening at a given time. Note also that we | |
1409 | * can accept some slop in the rsp->completed access due to the fact | |
1410 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1411 | */ |
e4fa4c97 | 1412 | static void __cpuinit |
f41d911f | 1413 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1414 | { |
1415 | unsigned long flags; | |
64db4cff PM |
1416 | long lastcomp; |
1417 | unsigned long mask; | |
1418 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1419 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1420 | ||
1421 | /* Set up local state, ensuring consistent view of global state. */ | |
1422 | spin_lock_irqsave(&rnp->lock, flags); | |
1423 | lastcomp = rsp->completed; | |
1424 | rdp->completed = lastcomp; | |
1425 | rdp->gpnum = lastcomp; | |
1426 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ | |
1427 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1428 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1429 | rdp->preemptable = preemptable; |
64db4cff | 1430 | rdp->passed_quiesc_completed = lastcomp - 1; |
64db4cff | 1431 | rdp->blimit = blimit; |
64db4cff PM |
1432 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1433 | ||
1434 | /* | |
1435 | * A new grace period might start here. If so, we won't be part | |
1436 | * of it, but that is OK, as we are currently in a quiescent state. | |
1437 | */ | |
1438 | ||
1439 | /* Exclude any attempts to start a new GP on large systems. */ | |
1440 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
1441 | ||
1442 | /* Add CPU to rcu_node bitmasks. */ | |
1443 | rnp = rdp->mynode; | |
1444 | mask = rdp->grpmask; | |
1445 | do { | |
1446 | /* Exclude any attempts to start a new GP on small systems. */ | |
1447 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1448 | rnp->qsmaskinit |= mask; | |
1449 | mask = rnp->grpmask; | |
1450 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | |
1451 | rnp = rnp->parent; | |
1452 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1453 | ||
1454 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
1455 | ||
1456 | /* | |
1457 | * A new grace period might start here. If so, we will be part of | |
1458 | * it, and its gpnum will be greater than ours, so we will | |
1459 | * participate. It is also possible for the gpnum to have been | |
1460 | * incremented before this function was called, and the bitmasks | |
1461 | * to not be filled out until now, in which case we will also | |
1462 | * participate due to our gpnum being behind. | |
1463 | */ | |
1464 | ||
1465 | /* Since it is coming online, the CPU is in a quiescent state. */ | |
1466 | cpu_quiet(cpu, rsp, rdp, lastcomp); | |
1467 | local_irq_restore(flags); | |
1468 | } | |
1469 | ||
1470 | static void __cpuinit rcu_online_cpu(int cpu) | |
1471 | { | |
f41d911f PM |
1472 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1473 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1474 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1475 | } |
1476 | ||
1477 | /* | |
f41d911f | 1478 | * Handle CPU online/offline notification events. |
64db4cff | 1479 | */ |
2e597558 PM |
1480 | int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1481 | unsigned long action, void *hcpu) | |
64db4cff PM |
1482 | { |
1483 | long cpu = (long)hcpu; | |
1484 | ||
1485 | switch (action) { | |
1486 | case CPU_UP_PREPARE: | |
1487 | case CPU_UP_PREPARE_FROZEN: | |
1488 | rcu_online_cpu(cpu); | |
1489 | break; | |
1490 | case CPU_DEAD: | |
1491 | case CPU_DEAD_FROZEN: | |
1492 | case CPU_UP_CANCELED: | |
1493 | case CPU_UP_CANCELED_FROZEN: | |
1494 | rcu_offline_cpu(cpu); | |
1495 | break; | |
1496 | default: | |
1497 | break; | |
1498 | } | |
1499 | return NOTIFY_OK; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Compute the per-level fanout, either using the exact fanout specified | |
1504 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1505 | */ | |
1506 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1507 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1508 | { | |
1509 | int i; | |
1510 | ||
1511 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1512 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1513 | } | |
1514 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1515 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1516 | { | |
1517 | int ccur; | |
1518 | int cprv; | |
1519 | int i; | |
1520 | ||
1521 | cprv = NR_CPUS; | |
1522 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1523 | ccur = rsp->levelcnt[i]; | |
1524 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1525 | cprv = ccur; | |
1526 | } | |
1527 | } | |
1528 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1529 | ||
1530 | /* | |
1531 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1532 | */ | |
1533 | static void __init rcu_init_one(struct rcu_state *rsp) | |
1534 | { | |
1535 | int cpustride = 1; | |
1536 | int i; | |
1537 | int j; | |
1538 | struct rcu_node *rnp; | |
1539 | ||
1540 | /* Initialize the level-tracking arrays. */ | |
1541 | ||
1542 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1543 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1544 | rcu_init_levelspread(rsp); | |
1545 | ||
1546 | /* Initialize the elements themselves, starting from the leaves. */ | |
1547 | ||
1548 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1549 | cpustride *= rsp->levelspread[i]; | |
1550 | rnp = rsp->level[i]; | |
1551 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1552 | spin_lock_init(&rnp->lock); | |
f41d911f | 1553 | rnp->gpnum = 0; |
64db4cff PM |
1554 | rnp->qsmask = 0; |
1555 | rnp->qsmaskinit = 0; | |
1556 | rnp->grplo = j * cpustride; | |
1557 | rnp->grphi = (j + 1) * cpustride - 1; | |
1558 | if (rnp->grphi >= NR_CPUS) | |
1559 | rnp->grphi = NR_CPUS - 1; | |
1560 | if (i == 0) { | |
1561 | rnp->grpnum = 0; | |
1562 | rnp->grpmask = 0; | |
1563 | rnp->parent = NULL; | |
1564 | } else { | |
1565 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1566 | rnp->grpmask = 1UL << rnp->grpnum; | |
1567 | rnp->parent = rsp->level[i - 1] + | |
1568 | j / rsp->levelspread[i - 1]; | |
1569 | } | |
1570 | rnp->level = i; | |
f41d911f PM |
1571 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1572 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
64db4cff PM |
1573 | } |
1574 | } | |
1575 | } | |
1576 | ||
1577 | /* | |
f41d911f PM |
1578 | * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used |
1579 | * nowhere else! Assigns leaf node pointers into each CPU's rcu_data | |
1580 | * structure. | |
64db4cff | 1581 | */ |
65cf8f86 | 1582 | #define RCU_INIT_FLAVOR(rsp, rcu_data) \ |
64db4cff | 1583 | do { \ |
65cf8f86 | 1584 | rcu_init_one(rsp); \ |
64db4cff PM |
1585 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
1586 | j = 0; \ | |
1587 | for_each_possible_cpu(i) { \ | |
1588 | if (i > rnp[j].grphi) \ | |
1589 | j++; \ | |
1590 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | |
1591 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | |
65cf8f86 | 1592 | rcu_boot_init_percpu_data(i, rsp); \ |
64db4cff PM |
1593 | } \ |
1594 | } while (0) | |
1595 | ||
f41d911f PM |
1596 | #ifdef CONFIG_TREE_PREEMPT_RCU |
1597 | ||
1598 | void __init __rcu_init_preempt(void) | |
1599 | { | |
1600 | int i; /* All used by RCU_INIT_FLAVOR(). */ | |
1601 | int j; | |
1602 | struct rcu_node *rnp; | |
1603 | ||
1604 | RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); | |
1605 | } | |
1606 | ||
1607 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1608 | ||
1609 | void __init __rcu_init_preempt(void) | |
1610 | { | |
1611 | } | |
1612 | ||
1613 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1614 | ||
64db4cff PM |
1615 | void __init __rcu_init(void) |
1616 | { | |
f41d911f | 1617 | int i; /* All used by RCU_INIT_FLAVOR(). */ |
64db4cff PM |
1618 | int j; |
1619 | struct rcu_node *rnp; | |
1620 | ||
f41d911f | 1621 | rcu_bootup_announce(); |
64db4cff PM |
1622 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
1623 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | |
1624 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
65cf8f86 PM |
1625 | RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); |
1626 | RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); | |
f41d911f | 1627 | __rcu_init_preempt(); |
2e597558 | 1628 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
64db4cff PM |
1629 | } |
1630 | ||
1631 | module_param(blimit, int, 0); | |
1632 | module_param(qhimark, int, 0); | |
1633 | module_param(qlowmark, int, 0); |