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> | |
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 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
53 | static struct lock_class_key rcu_lock_key; | |
54 | struct lockdep_map rcu_lock_map = | |
55 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | |
56 | EXPORT_SYMBOL_GPL(rcu_lock_map); | |
57 | #endif | |
58 | ||
59 | /* Data structures. */ | |
60 | ||
61 | #define RCU_STATE_INITIALIZER(name) { \ | |
62 | .level = { &name.node[0] }, \ | |
63 | .levelcnt = { \ | |
64 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
65 | NUM_RCU_LVL_1, \ | |
66 | NUM_RCU_LVL_2, \ | |
67 | NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \ | |
68 | }, \ | |
69 | .signaled = RCU_SIGNAL_INIT, \ | |
70 | .gpnum = -300, \ | |
71 | .completed = -300, \ | |
72 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ | |
73 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ | |
74 | .n_force_qs = 0, \ | |
75 | .n_force_qs_ngp = 0, \ | |
76 | } | |
77 | ||
d6714c22 PM |
78 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
79 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 80 | |
6258c4fb IM |
81 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
82 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 83 | |
f41d911f | 84 | extern long rcu_batches_completed_sched(void); |
dd5d19ba | 85 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp); |
f41d911f PM |
86 | static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, |
87 | struct rcu_node *rnp, unsigned long flags); | |
88 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags); | |
c935a331 | 89 | #ifdef CONFIG_HOTPLUG_CPU |
33f76148 | 90 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp); |
c935a331 | 91 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
f41d911f PM |
92 | static void __rcu_process_callbacks(struct rcu_state *rsp, |
93 | struct rcu_data *rdp); | |
94 | static void __call_rcu(struct rcu_head *head, | |
95 | void (*func)(struct rcu_head *rcu), | |
96 | struct rcu_state *rsp); | |
97 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp); | |
98 | static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp, | |
99 | int preemptable); | |
100 | ||
101 | #include "rcutree_plugin.h" | |
102 | ||
b1f77b05 | 103 | /* |
d6714c22 | 104 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 105 | * how many quiescent states passed, just if there was at least |
d6714c22 | 106 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 107 | */ |
d6714c22 | 108 | void rcu_sched_qs(int cpu) |
b1f77b05 | 109 | { |
f41d911f PM |
110 | unsigned long flags; |
111 | struct rcu_data *rdp; | |
112 | ||
113 | local_irq_save(flags); | |
114 | rdp = &per_cpu(rcu_sched_data, cpu); | |
b1f77b05 IM |
115 | rdp->passed_quiesc = 1; |
116 | rdp->passed_quiesc_completed = rdp->completed; | |
f41d911f PM |
117 | rcu_preempt_qs(cpu); |
118 | local_irq_restore(flags); | |
b1f77b05 IM |
119 | } |
120 | ||
d6714c22 | 121 | void rcu_bh_qs(int cpu) |
b1f77b05 | 122 | { |
f41d911f PM |
123 | unsigned long flags; |
124 | struct rcu_data *rdp; | |
125 | ||
126 | local_irq_save(flags); | |
127 | rdp = &per_cpu(rcu_bh_data, cpu); | |
b1f77b05 IM |
128 | rdp->passed_quiesc = 1; |
129 | rdp->passed_quiesc_completed = rdp->completed; | |
f41d911f | 130 | local_irq_restore(flags); |
b1f77b05 | 131 | } |
64db4cff PM |
132 | |
133 | #ifdef CONFIG_NO_HZ | |
90a4d2c0 PM |
134 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
135 | .dynticks_nesting = 1, | |
136 | .dynticks = 1, | |
137 | }; | |
64db4cff PM |
138 | #endif /* #ifdef CONFIG_NO_HZ */ |
139 | ||
140 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
141 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
142 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
143 | ||
144 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); | |
a157229c | 145 | static int rcu_pending(int cpu); |
64db4cff PM |
146 | |
147 | /* | |
d6714c22 | 148 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 149 | */ |
d6714c22 | 150 | long rcu_batches_completed_sched(void) |
64db4cff | 151 | { |
d6714c22 | 152 | return rcu_sched_state.completed; |
64db4cff | 153 | } |
d6714c22 | 154 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
155 | |
156 | /* | |
157 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
158 | */ | |
159 | long rcu_batches_completed_bh(void) | |
160 | { | |
161 | return rcu_bh_state.completed; | |
162 | } | |
163 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
164 | ||
165 | /* | |
166 | * Does the CPU have callbacks ready to be invoked? | |
167 | */ | |
168 | static int | |
169 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
170 | { | |
171 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Does the current CPU require a yet-as-unscheduled grace period? | |
176 | */ | |
177 | static int | |
178 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
179 | { | |
180 | /* ACCESS_ONCE() because we are accessing outside of lock. */ | |
181 | return *rdp->nxttail[RCU_DONE_TAIL] && | |
182 | ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum); | |
183 | } | |
184 | ||
185 | /* | |
186 | * Return the root node of the specified rcu_state structure. | |
187 | */ | |
188 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
189 | { | |
190 | return &rsp->node[0]; | |
191 | } | |
192 | ||
193 | #ifdef CONFIG_SMP | |
194 | ||
195 | /* | |
196 | * If the specified CPU is offline, tell the caller that it is in | |
197 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
198 | * Grace periods can end up waiting on an offline CPU when that | |
199 | * CPU is in the process of coming online -- it will be added to the | |
200 | * rcu_node bitmasks before it actually makes it online. The same thing | |
201 | * can happen while a CPU is in the process of coming online. Because this | |
202 | * race is quite rare, we check for it after detecting that the grace | |
203 | * period has been delayed rather than checking each and every CPU | |
204 | * each and every time we start a new grace period. | |
205 | */ | |
206 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
207 | { | |
208 | /* | |
209 | * If the CPU is offline, it is in a quiescent state. We can | |
210 | * trust its state not to change because interrupts are disabled. | |
211 | */ | |
212 | if (cpu_is_offline(rdp->cpu)) { | |
213 | rdp->offline_fqs++; | |
214 | return 1; | |
215 | } | |
216 | ||
f41d911f PM |
217 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
218 | if (rdp->preemptable) | |
219 | return 0; | |
220 | ||
64db4cff PM |
221 | /* The CPU is online, so send it a reschedule IPI. */ |
222 | if (rdp->cpu != smp_processor_id()) | |
223 | smp_send_reschedule(rdp->cpu); | |
224 | else | |
225 | set_need_resched(); | |
226 | rdp->resched_ipi++; | |
227 | return 0; | |
228 | } | |
229 | ||
230 | #endif /* #ifdef CONFIG_SMP */ | |
231 | ||
232 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
233 | |
234 | /** | |
235 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
236 | * | |
237 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
238 | * read-side critical sections can occur. (Though RCU read-side | |
239 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
240 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
241 | */ | |
242 | void rcu_enter_nohz(void) | |
243 | { | |
244 | unsigned long flags; | |
245 | struct rcu_dynticks *rdtp; | |
246 | ||
247 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
248 | local_irq_save(flags); | |
249 | rdtp = &__get_cpu_var(rcu_dynticks); | |
250 | rdtp->dynticks++; | |
251 | rdtp->dynticks_nesting--; | |
86848966 | 252 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
253 | local_irq_restore(flags); |
254 | } | |
255 | ||
256 | /* | |
257 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
258 | * | |
259 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
260 | * read-side critical sections normally occur. | |
261 | */ | |
262 | void rcu_exit_nohz(void) | |
263 | { | |
264 | unsigned long flags; | |
265 | struct rcu_dynticks *rdtp; | |
266 | ||
267 | local_irq_save(flags); | |
268 | rdtp = &__get_cpu_var(rcu_dynticks); | |
269 | rdtp->dynticks++; | |
270 | rdtp->dynticks_nesting++; | |
86848966 | 271 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
272 | local_irq_restore(flags); |
273 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
274 | } | |
275 | ||
276 | /** | |
277 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
278 | * | |
279 | * If the CPU was idle with dynamic ticks active, and there is no | |
280 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
281 | * RCU grace-period handling know that the CPU is active. | |
282 | */ | |
283 | void rcu_nmi_enter(void) | |
284 | { | |
285 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
286 | ||
287 | if (rdtp->dynticks & 0x1) | |
288 | return; | |
289 | rdtp->dynticks_nmi++; | |
86848966 | 290 | WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1)); |
64db4cff PM |
291 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
292 | } | |
293 | ||
294 | /** | |
295 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
296 | * | |
297 | * If the CPU was idle with dynamic ticks active, and there is no | |
298 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
299 | * RCU grace-period handling know that the CPU is no longer active. | |
300 | */ | |
301 | void rcu_nmi_exit(void) | |
302 | { | |
303 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
304 | ||
305 | if (rdtp->dynticks & 0x1) | |
306 | return; | |
307 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
308 | rdtp->dynticks_nmi++; | |
86848966 | 309 | WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1); |
64db4cff PM |
310 | } |
311 | ||
312 | /** | |
313 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
314 | * | |
315 | * If the CPU was idle with dynamic ticks active, this updates the | |
316 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
317 | */ | |
318 | void rcu_irq_enter(void) | |
319 | { | |
320 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
321 | ||
322 | if (rdtp->dynticks_nesting++) | |
323 | return; | |
324 | rdtp->dynticks++; | |
86848966 | 325 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
326 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
327 | } | |
328 | ||
329 | /** | |
330 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
331 | * | |
332 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
333 | * to put let the RCU handling be aware that the CPU is going back to idle | |
334 | * with no ticks. | |
335 | */ | |
336 | void rcu_irq_exit(void) | |
337 | { | |
338 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
339 | ||
340 | if (--rdtp->dynticks_nesting) | |
341 | return; | |
342 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
343 | rdtp->dynticks++; | |
86848966 | 344 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
345 | |
346 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 347 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
348 | __get_cpu_var(rcu_bh_data).nxtlist) |
349 | set_need_resched(); | |
350 | } | |
351 | ||
352 | /* | |
353 | * Record the specified "completed" value, which is later used to validate | |
354 | * dynticks counter manipulations. Specify "rsp->completed - 1" to | |
355 | * unconditionally invalidate any future dynticks manipulations (which is | |
356 | * useful at the beginning of a grace period). | |
357 | */ | |
358 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
359 | { | |
360 | rsp->dynticks_completed = comp; | |
361 | } | |
362 | ||
363 | #ifdef CONFIG_SMP | |
364 | ||
365 | /* | |
366 | * Recall the previously recorded value of the completion for dynticks. | |
367 | */ | |
368 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
369 | { | |
370 | return rsp->dynticks_completed; | |
371 | } | |
372 | ||
373 | /* | |
374 | * Snapshot the specified CPU's dynticks counter so that we can later | |
375 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
376 | * is already in a quiescent state courtesy of dynticks idle mode. | |
377 | */ | |
378 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
379 | { | |
380 | int ret; | |
381 | int snap; | |
382 | int snap_nmi; | |
383 | ||
384 | snap = rdp->dynticks->dynticks; | |
385 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
386 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
387 | rdp->dynticks_snap = snap; | |
388 | rdp->dynticks_nmi_snap = snap_nmi; | |
389 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
390 | if (ret) | |
391 | rdp->dynticks_fqs++; | |
392 | return ret; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Return true if the specified CPU has passed through a quiescent | |
397 | * state by virtue of being in or having passed through an dynticks | |
398 | * idle state since the last call to dyntick_save_progress_counter() | |
399 | * for this same CPU. | |
400 | */ | |
401 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
402 | { | |
403 | long curr; | |
404 | long curr_nmi; | |
405 | long snap; | |
406 | long snap_nmi; | |
407 | ||
408 | curr = rdp->dynticks->dynticks; | |
409 | snap = rdp->dynticks_snap; | |
410 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
411 | snap_nmi = rdp->dynticks_nmi_snap; | |
412 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
413 | ||
414 | /* | |
415 | * If the CPU passed through or entered a dynticks idle phase with | |
416 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
417 | * already acknowledged the request to pass through a quiescent | |
418 | * state. Either way, that CPU cannot possibly be in an RCU | |
419 | * read-side critical section that started before the beginning | |
420 | * of the current RCU grace period. | |
421 | */ | |
422 | if ((curr != snap || (curr & 0x1) == 0) && | |
423 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
424 | rdp->dynticks_fqs++; | |
425 | return 1; | |
426 | } | |
427 | ||
428 | /* Go check for the CPU being offline. */ | |
429 | return rcu_implicit_offline_qs(rdp); | |
430 | } | |
431 | ||
432 | #endif /* #ifdef CONFIG_SMP */ | |
433 | ||
434 | #else /* #ifdef CONFIG_NO_HZ */ | |
435 | ||
436 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
437 | { | |
438 | } | |
439 | ||
440 | #ifdef CONFIG_SMP | |
441 | ||
442 | /* | |
443 | * If there are no dynticks, then the only way that a CPU can passively | |
444 | * be in a quiescent state is to be offline. Unlike dynticks idle, which | |
445 | * is a point in time during the prior (already finished) grace period, | |
446 | * an offline CPU is always in a quiescent state, and thus can be | |
447 | * unconditionally applied. So just return the current value of completed. | |
448 | */ | |
449 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
450 | { | |
451 | return rsp->completed; | |
452 | } | |
453 | ||
454 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
455 | { | |
456 | return 0; | |
457 | } | |
458 | ||
459 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
460 | { | |
461 | return rcu_implicit_offline_qs(rdp); | |
462 | } | |
463 | ||
464 | #endif /* #ifdef CONFIG_SMP */ | |
465 | ||
466 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
467 | ||
468 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
469 | ||
470 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
471 | { | |
472 | rsp->gp_start = jiffies; | |
473 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
474 | } | |
475 | ||
476 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
477 | { | |
478 | int cpu; | |
479 | long delta; | |
480 | unsigned long flags; | |
481 | struct rcu_node *rnp = rcu_get_root(rsp); | |
482 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
483 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
484 | ||
485 | /* Only let one CPU complain about others per time interval. */ | |
486 | ||
487 | spin_lock_irqsave(&rnp->lock, flags); | |
488 | delta = jiffies - rsp->jiffies_stall; | |
489 | if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { | |
490 | spin_unlock_irqrestore(&rnp->lock, flags); | |
491 | return; | |
492 | } | |
493 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
494 | spin_unlock_irqrestore(&rnp->lock, flags); | |
495 | ||
496 | /* OK, time to rat on our buddy... */ | |
497 | ||
498 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | |
499 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
f41d911f | 500 | rcu_print_task_stall(rnp); |
64db4cff PM |
501 | if (rnp_cur->qsmask == 0) |
502 | continue; | |
503 | for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) | |
504 | if (rnp_cur->qsmask & (1UL << cpu)) | |
505 | printk(" %d", rnp_cur->grplo + cpu); | |
506 | } | |
507 | printk(" (detected by %d, t=%ld jiffies)\n", | |
508 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | |
c1dc0b9c IM |
509 | trigger_all_cpu_backtrace(); |
510 | ||
64db4cff PM |
511 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
512 | } | |
513 | ||
514 | static void print_cpu_stall(struct rcu_state *rsp) | |
515 | { | |
516 | unsigned long flags; | |
517 | struct rcu_node *rnp = rcu_get_root(rsp); | |
518 | ||
519 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | |
520 | smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
521 | trigger_all_cpu_backtrace(); |
522 | ||
64db4cff PM |
523 | spin_lock_irqsave(&rnp->lock, flags); |
524 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) | |
525 | rsp->jiffies_stall = | |
526 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
527 | spin_unlock_irqrestore(&rnp->lock, flags); | |
c1dc0b9c | 528 | |
64db4cff PM |
529 | set_need_resched(); /* kick ourselves to get things going. */ |
530 | } | |
531 | ||
532 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
533 | { | |
534 | long delta; | |
535 | struct rcu_node *rnp; | |
536 | ||
537 | delta = jiffies - rsp->jiffies_stall; | |
538 | rnp = rdp->mynode; | |
539 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
540 | ||
541 | /* We haven't checked in, so go dump stack. */ | |
542 | print_cpu_stall(rsp); | |
543 | ||
544 | } else if (rsp->gpnum != rsp->completed && | |
545 | delta >= RCU_STALL_RAT_DELAY) { | |
546 | ||
547 | /* They had two time units to dump stack, so complain. */ | |
548 | print_other_cpu_stall(rsp); | |
549 | } | |
550 | } | |
551 | ||
552 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
553 | ||
554 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
555 | { | |
556 | } | |
557 | ||
558 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
559 | { | |
560 | } | |
561 | ||
562 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
563 | ||
564 | /* | |
565 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
566 | * This is used both when we started the grace period and when we notice | |
567 | * that someone else started the grace period. | |
568 | */ | |
569 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) | |
570 | { | |
571 | rdp->qs_pending = 1; | |
572 | rdp->passed_quiesc = 0; | |
573 | rdp->gpnum = rsp->gpnum; | |
64db4cff PM |
574 | } |
575 | ||
576 | /* | |
577 | * Did someone else start a new RCU grace period start since we last | |
578 | * checked? Update local state appropriately if so. Must be called | |
579 | * on the CPU corresponding to rdp. | |
580 | */ | |
581 | static int | |
582 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
583 | { | |
584 | unsigned long flags; | |
585 | int ret = 0; | |
586 | ||
587 | local_irq_save(flags); | |
588 | if (rdp->gpnum != rsp->gpnum) { | |
589 | note_new_gpnum(rsp, rdp); | |
590 | ret = 1; | |
591 | } | |
592 | local_irq_restore(flags); | |
593 | return ret; | |
594 | } | |
595 | ||
596 | /* | |
597 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
598 | * in preparation for detecting the next grace period. The caller must hold | |
599 | * the root node's ->lock, which is released before return. Hard irqs must | |
600 | * be disabled. | |
601 | */ | |
602 | static void | |
603 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
604 | __releases(rcu_get_root(rsp)->lock) | |
605 | { | |
606 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
607 | struct rcu_node *rnp = rcu_get_root(rsp); | |
608 | struct rcu_node *rnp_cur; | |
609 | struct rcu_node *rnp_end; | |
610 | ||
611 | if (!cpu_needs_another_gp(rsp, rdp)) { | |
612 | spin_unlock_irqrestore(&rnp->lock, flags); | |
613 | return; | |
614 | } | |
615 | ||
616 | /* Advance to a new grace period and initialize state. */ | |
617 | rsp->gpnum++; | |
618 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
619 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
620 | record_gp_stall_check_time(rsp); |
621 | dyntick_record_completed(rsp, rsp->completed - 1); | |
622 | note_new_gpnum(rsp, rdp); | |
623 | ||
624 | /* | |
625 | * Because we are first, we know that all our callbacks will | |
626 | * be covered by this upcoming grace period, even the ones | |
627 | * that were registered arbitrarily recently. | |
628 | */ | |
629 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
630 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
631 | ||
632 | /* Special-case the common single-level case. */ | |
633 | if (NUM_RCU_NODES == 1) { | |
634 | rnp->qsmask = rnp->qsmaskinit; | |
c12172c0 | 635 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
64db4cff PM |
636 | spin_unlock_irqrestore(&rnp->lock, flags); |
637 | return; | |
638 | } | |
639 | ||
640 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | |
641 | ||
642 | ||
643 | /* Exclude any concurrent CPU-hotplug operations. */ | |
644 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
645 | ||
646 | /* | |
647 | * Set the quiescent-state-needed bits in all the non-leaf RCU | |
648 | * nodes for all currently online CPUs. This operation relies | |
649 | * on the layout of the hierarchy within the rsp->node[] array. | |
650 | * Note that other CPUs will access only the leaves of the | |
651 | * hierarchy, which still indicate that no grace period is in | |
652 | * progress. In addition, we have excluded CPU-hotplug operations. | |
653 | * | |
654 | * We therefore do not need to hold any locks. Any required | |
655 | * memory barriers will be supplied by the locks guarding the | |
656 | * leaf rcu_nodes in the hierarchy. | |
657 | */ | |
658 | ||
659 | rnp_end = rsp->level[NUM_RCU_LVLS - 1]; | |
660 | for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) | |
661 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | |
662 | ||
663 | /* | |
664 | * Now set up the leaf nodes. Here we must be careful. First, | |
665 | * we need to hold the lock in order to exclude other CPUs, which | |
666 | * might be contending for the leaf nodes' locks. Second, as | |
667 | * soon as we initialize a given leaf node, its CPUs might run | |
668 | * up the rest of the hierarchy. We must therefore acquire locks | |
669 | * for each node that we touch during this stage. (But we still | |
670 | * are excluding CPU-hotplug operations.) | |
671 | * | |
672 | * Note that the grace period cannot complete until we finish | |
673 | * the initialization process, as there will be at least one | |
674 | * qsmask bit set in the root node until that time, namely the | |
675 | * one corresponding to this CPU. | |
676 | */ | |
677 | rnp_end = &rsp->node[NUM_RCU_NODES]; | |
678 | rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
679 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
680 | spin_lock(&rnp_cur->lock); /* irqs already disabled. */ | |
681 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | |
682 | spin_unlock(&rnp_cur->lock); /* irqs already disabled. */ | |
683 | } | |
684 | ||
685 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ | |
686 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
687 | } | |
688 | ||
689 | /* | |
690 | * Advance this CPU's callbacks, but only if the current grace period | |
691 | * has ended. This may be called only from the CPU to whom the rdp | |
692 | * belongs. | |
693 | */ | |
694 | static void | |
695 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
696 | { | |
697 | long completed_snap; | |
698 | unsigned long flags; | |
699 | ||
700 | local_irq_save(flags); | |
701 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ | |
702 | ||
703 | /* Did another grace period end? */ | |
704 | if (rdp->completed != completed_snap) { | |
705 | ||
706 | /* Advance callbacks. No harm if list empty. */ | |
707 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
708 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
709 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
710 | ||
711 | /* Remember that we saw this grace-period completion. */ | |
712 | rdp->completed = completed_snap; | |
713 | } | |
714 | local_irq_restore(flags); | |
715 | } | |
716 | ||
f41d911f PM |
717 | /* |
718 | * Clean up after the prior grace period and let rcu_start_gp() start up | |
719 | * the next grace period if one is needed. Note that the caller must | |
720 | * hold rnp->lock, as required by rcu_start_gp(), which will release it. | |
721 | */ | |
722 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) | |
723 | __releases(rnp->lock) | |
724 | { | |
725 | rsp->completed = rsp->gpnum; | |
726 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); | |
727 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ | |
728 | } | |
729 | ||
64db4cff PM |
730 | /* |
731 | * Similar to cpu_quiet(), for which it is a helper function. Allows | |
732 | * a group of CPUs to be quieted at one go, though all the CPUs in the | |
733 | * group must be represented by the same leaf rcu_node structure. | |
734 | * That structure's lock must be held upon entry, and it is released | |
735 | * before return. | |
736 | */ | |
737 | static void | |
738 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, | |
739 | unsigned long flags) | |
740 | __releases(rnp->lock) | |
741 | { | |
742 | /* Walk up the rcu_node hierarchy. */ | |
743 | for (;;) { | |
744 | if (!(rnp->qsmask & mask)) { | |
745 | ||
746 | /* Our bit has already been cleared, so done. */ | |
747 | spin_unlock_irqrestore(&rnp->lock, flags); | |
748 | return; | |
749 | } | |
750 | rnp->qsmask &= ~mask; | |
f41d911f | 751 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
752 | |
753 | /* Other bits still set at this level, so done. */ | |
754 | spin_unlock_irqrestore(&rnp->lock, flags); | |
755 | return; | |
756 | } | |
757 | mask = rnp->grpmask; | |
758 | if (rnp->parent == NULL) { | |
759 | ||
760 | /* No more levels. Exit loop holding root lock. */ | |
761 | ||
762 | break; | |
763 | } | |
764 | spin_unlock_irqrestore(&rnp->lock, flags); | |
765 | rnp = rnp->parent; | |
766 | spin_lock_irqsave(&rnp->lock, flags); | |
767 | } | |
768 | ||
769 | /* | |
770 | * Get here if we are the last CPU to pass through a quiescent | |
f41d911f PM |
771 | * state for this grace period. Invoke cpu_quiet_msk_finish() |
772 | * to clean up and start the next grace period if one is needed. | |
64db4cff | 773 | */ |
f41d911f | 774 | cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
775 | } |
776 | ||
777 | /* | |
778 | * Record a quiescent state for the specified CPU, which must either be | |
779 | * the current CPU or an offline CPU. The lastcomp argument is used to | |
780 | * make sure we are still in the grace period of interest. We don't want | |
781 | * to end the current grace period based on quiescent states detected in | |
782 | * an earlier grace period! | |
783 | */ | |
784 | static void | |
785 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) | |
786 | { | |
787 | unsigned long flags; | |
788 | unsigned long mask; | |
789 | struct rcu_node *rnp; | |
790 | ||
791 | rnp = rdp->mynode; | |
792 | spin_lock_irqsave(&rnp->lock, flags); | |
793 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { | |
794 | ||
795 | /* | |
796 | * Someone beat us to it for this grace period, so leave. | |
797 | * The race with GP start is resolved by the fact that we | |
798 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
799 | * cannot yet be initialized -- so we would simply find our | |
800 | * CPU's bit already cleared in cpu_quiet_msk() if this race | |
801 | * occurred. | |
802 | */ | |
803 | rdp->passed_quiesc = 0; /* try again later! */ | |
804 | spin_unlock_irqrestore(&rnp->lock, flags); | |
805 | return; | |
806 | } | |
807 | mask = rdp->grpmask; | |
808 | if ((rnp->qsmask & mask) == 0) { | |
809 | spin_unlock_irqrestore(&rnp->lock, flags); | |
810 | } else { | |
811 | rdp->qs_pending = 0; | |
812 | ||
813 | /* | |
814 | * This GP can't end until cpu checks in, so all of our | |
815 | * callbacks can be processed during the next GP. | |
816 | */ | |
817 | rdp = rsp->rda[smp_processor_id()]; | |
818 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
819 | ||
820 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ | |
821 | } | |
822 | } | |
823 | ||
824 | /* | |
825 | * Check to see if there is a new grace period of which this CPU | |
826 | * is not yet aware, and if so, set up local rcu_data state for it. | |
827 | * Otherwise, see if this CPU has just passed through its first | |
828 | * quiescent state for this grace period, and record that fact if so. | |
829 | */ | |
830 | static void | |
831 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
832 | { | |
833 | /* If there is now a new grace period, record and return. */ | |
834 | if (check_for_new_grace_period(rsp, rdp)) | |
835 | return; | |
836 | ||
837 | /* | |
838 | * Does this CPU still need to do its part for current grace period? | |
839 | * If no, return and let the other CPUs do their part as well. | |
840 | */ | |
841 | if (!rdp->qs_pending) | |
842 | return; | |
843 | ||
844 | /* | |
845 | * Was there a quiescent state since the beginning of the grace | |
846 | * period? If no, then exit and wait for the next call. | |
847 | */ | |
848 | if (!rdp->passed_quiesc) | |
849 | return; | |
850 | ||
851 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ | |
852 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
853 | } | |
854 | ||
855 | #ifdef CONFIG_HOTPLUG_CPU | |
856 | ||
857 | /* | |
858 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
859 | * and move all callbacks from the outgoing CPU to the current one. | |
860 | */ | |
861 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
862 | { | |
863 | int i; | |
864 | unsigned long flags; | |
865 | long lastcomp; | |
866 | unsigned long mask; | |
867 | struct rcu_data *rdp = rsp->rda[cpu]; | |
868 | struct rcu_data *rdp_me; | |
869 | struct rcu_node *rnp; | |
870 | ||
871 | /* Exclude any attempts to start a new grace period. */ | |
872 | spin_lock_irqsave(&rsp->onofflock, flags); | |
873 | ||
874 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
875 | rnp = rdp->mynode; | |
876 | mask = rdp->grpmask; /* rnp->grplo is constant. */ | |
877 | do { | |
878 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
879 | rnp->qsmaskinit &= ~mask; | |
880 | if (rnp->qsmaskinit != 0) { | |
f41d911f | 881 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
882 | break; |
883 | } | |
dd5d19ba | 884 | rcu_preempt_offline_tasks(rsp, rnp); |
64db4cff | 885 | mask = rnp->grpmask; |
f41d911f | 886 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
887 | rnp = rnp->parent; |
888 | } while (rnp != NULL); | |
889 | lastcomp = rsp->completed; | |
890 | ||
891 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
892 | ||
893 | /* Being offline is a quiescent state, so go record it. */ | |
894 | cpu_quiet(cpu, rsp, rdp, lastcomp); | |
895 | ||
896 | /* | |
897 | * Move callbacks from the outgoing CPU to the running CPU. | |
898 | * Note that the outgoing CPU is now quiscent, so it is now | |
d6714c22 | 899 | * (uncharacteristically) safe to access its rcu_data structure. |
64db4cff PM |
900 | * Note also that we must carefully retain the order of the |
901 | * outgoing CPU's callbacks in order for rcu_barrier() to work | |
902 | * correctly. Finally, note that we start all the callbacks | |
903 | * afresh, even those that have passed through a grace period | |
904 | * and are therefore ready to invoke. The theory is that hotplug | |
905 | * events are rare, and that if they are frequent enough to | |
906 | * indefinitely delay callbacks, you have far worse things to | |
907 | * be worrying about. | |
908 | */ | |
909 | rdp_me = rsp->rda[smp_processor_id()]; | |
910 | if (rdp->nxtlist != NULL) { | |
911 | *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
912 | rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
913 | rdp->nxtlist = NULL; | |
914 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
915 | rdp->nxttail[i] = &rdp->nxtlist; | |
916 | rdp_me->qlen += rdp->qlen; | |
917 | rdp->qlen = 0; | |
918 | } | |
919 | local_irq_restore(flags); | |
920 | } | |
921 | ||
922 | /* | |
923 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
924 | * callbacks that it might have to the current CPU. This code assumes | |
925 | * that at least one CPU in the system will remain running at all times. | |
926 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
927 | */ | |
928 | static void rcu_offline_cpu(int cpu) | |
929 | { | |
d6714c22 | 930 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 931 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 932 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
933 | } |
934 | ||
935 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
936 | ||
937 | static void rcu_offline_cpu(int cpu) | |
938 | { | |
939 | } | |
940 | ||
941 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
942 | ||
943 | /* | |
944 | * Invoke any RCU callbacks that have made it to the end of their grace | |
945 | * period. Thottle as specified by rdp->blimit. | |
946 | */ | |
947 | static void rcu_do_batch(struct rcu_data *rdp) | |
948 | { | |
949 | unsigned long flags; | |
950 | struct rcu_head *next, *list, **tail; | |
951 | int count; | |
952 | ||
953 | /* If no callbacks are ready, just return.*/ | |
954 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
955 | return; | |
956 | ||
957 | /* | |
958 | * Extract the list of ready callbacks, disabling to prevent | |
959 | * races with call_rcu() from interrupt handlers. | |
960 | */ | |
961 | local_irq_save(flags); | |
962 | list = rdp->nxtlist; | |
963 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
964 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
965 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
966 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
967 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
968 | rdp->nxttail[count] = &rdp->nxtlist; | |
969 | local_irq_restore(flags); | |
970 | ||
971 | /* Invoke callbacks. */ | |
972 | count = 0; | |
973 | while (list) { | |
974 | next = list->next; | |
975 | prefetch(next); | |
976 | list->func(list); | |
977 | list = next; | |
978 | if (++count >= rdp->blimit) | |
979 | break; | |
980 | } | |
981 | ||
982 | local_irq_save(flags); | |
983 | ||
984 | /* Update count, and requeue any remaining callbacks. */ | |
985 | rdp->qlen -= count; | |
986 | if (list != NULL) { | |
987 | *tail = rdp->nxtlist; | |
988 | rdp->nxtlist = list; | |
989 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
990 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
991 | rdp->nxttail[count] = tail; | |
992 | else | |
993 | break; | |
994 | } | |
995 | ||
996 | /* Reinstate batch limit if we have worked down the excess. */ | |
997 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
998 | rdp->blimit = blimit; | |
999 | ||
1000 | local_irq_restore(flags); | |
1001 | ||
1002 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1003 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
1004 | raise_softirq(RCU_SOFTIRQ); | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1009 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1010 | * Also schedule the RCU softirq handler. | |
1011 | * | |
1012 | * This function must be called with hardirqs disabled. It is normally | |
1013 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1014 | * false, there is no point in invoking rcu_check_callbacks(). | |
1015 | */ | |
1016 | void rcu_check_callbacks(int cpu, int user) | |
1017 | { | |
a157229c PM |
1018 | if (!rcu_pending(cpu)) |
1019 | return; /* if nothing for RCU to do. */ | |
64db4cff | 1020 | if (user || |
a6826048 PM |
1021 | (idle_cpu(cpu) && rcu_scheduler_active && |
1022 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1023 | |
1024 | /* | |
1025 | * Get here if this CPU took its interrupt from user | |
1026 | * mode or from the idle loop, and if this is not a | |
1027 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1028 | * a quiescent state, so note it. |
64db4cff PM |
1029 | * |
1030 | * No memory barrier is required here because both | |
d6714c22 PM |
1031 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1032 | * variables that other CPUs neither access nor modify, | |
1033 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1034 | */ |
1035 | ||
d6714c22 PM |
1036 | rcu_sched_qs(cpu); |
1037 | rcu_bh_qs(cpu); | |
64db4cff PM |
1038 | |
1039 | } else if (!in_softirq()) { | |
1040 | ||
1041 | /* | |
1042 | * Get here if this CPU did not take its interrupt from | |
1043 | * softirq, in other words, if it is not interrupting | |
1044 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1045 | * critical section, so note it. |
64db4cff PM |
1046 | */ |
1047 | ||
d6714c22 | 1048 | rcu_bh_qs(cpu); |
64db4cff | 1049 | } |
f41d911f | 1050 | rcu_preempt_check_callbacks(cpu); |
64db4cff PM |
1051 | raise_softirq(RCU_SOFTIRQ); |
1052 | } | |
1053 | ||
1054 | #ifdef CONFIG_SMP | |
1055 | ||
1056 | /* | |
1057 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1058 | * have not yet encountered a quiescent state, using the function specified. | |
1059 | * Returns 1 if the current grace period ends while scanning (possibly | |
1060 | * because we made it end). | |
1061 | */ | |
1062 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, | |
1063 | int (*f)(struct rcu_data *)) | |
1064 | { | |
1065 | unsigned long bit; | |
1066 | int cpu; | |
1067 | unsigned long flags; | |
1068 | unsigned long mask; | |
1069 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
1070 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
1071 | ||
1072 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
1073 | mask = 0; | |
1074 | spin_lock_irqsave(&rnp_cur->lock, flags); | |
1075 | if (rsp->completed != lastcomp) { | |
1076 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1077 | return 1; | |
1078 | } | |
1079 | if (rnp_cur->qsmask == 0) { | |
1080 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1081 | continue; | |
1082 | } | |
1083 | cpu = rnp_cur->grplo; | |
1084 | bit = 1; | |
1085 | for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { | |
1086 | if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) | |
1087 | mask |= bit; | |
1088 | } | |
1089 | if (mask != 0 && rsp->completed == lastcomp) { | |
1090 | ||
1091 | /* cpu_quiet_msk() releases rnp_cur->lock. */ | |
1092 | cpu_quiet_msk(mask, rsp, rnp_cur, flags); | |
1093 | continue; | |
1094 | } | |
1095 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1096 | } | |
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * Force quiescent states on reluctant CPUs, and also detect which | |
1102 | * CPUs are in dyntick-idle mode. | |
1103 | */ | |
1104 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1105 | { | |
1106 | unsigned long flags; | |
1107 | long lastcomp; | |
64db4cff PM |
1108 | struct rcu_node *rnp = rcu_get_root(rsp); |
1109 | u8 signaled; | |
1110 | ||
1111 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) | |
1112 | return; /* No grace period in progress, nothing to force. */ | |
1113 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | |
1114 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | |
1115 | return; /* Someone else is already on the job. */ | |
1116 | } | |
1117 | if (relaxed && | |
ef631b0c | 1118 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
64db4cff PM |
1119 | goto unlock_ret; /* no emergency and done recently. */ |
1120 | rsp->n_force_qs++; | |
1121 | spin_lock(&rnp->lock); | |
1122 | lastcomp = rsp->completed; | |
1123 | signaled = rsp->signaled; | |
1124 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
1125 | if (lastcomp == rsp->gpnum) { |
1126 | rsp->n_force_qs_ngp++; | |
1127 | spin_unlock(&rnp->lock); | |
1128 | goto unlock_ret; /* no GP in progress, time updated. */ | |
1129 | } | |
1130 | spin_unlock(&rnp->lock); | |
1131 | switch (signaled) { | |
1132 | case RCU_GP_INIT: | |
1133 | ||
1134 | break; /* grace period still initializing, ignore. */ | |
1135 | ||
1136 | case RCU_SAVE_DYNTICK: | |
1137 | ||
1138 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) | |
1139 | break; /* So gcc recognizes the dead code. */ | |
1140 | ||
1141 | /* Record dyntick-idle state. */ | |
1142 | if (rcu_process_dyntick(rsp, lastcomp, | |
1143 | dyntick_save_progress_counter)) | |
1144 | goto unlock_ret; | |
1145 | ||
1146 | /* Update state, record completion counter. */ | |
1147 | spin_lock(&rnp->lock); | |
1148 | if (lastcomp == rsp->completed) { | |
1149 | rsp->signaled = RCU_FORCE_QS; | |
1150 | dyntick_record_completed(rsp, lastcomp); | |
1151 | } | |
1152 | spin_unlock(&rnp->lock); | |
1153 | break; | |
1154 | ||
1155 | case RCU_FORCE_QS: | |
1156 | ||
1157 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1158 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), | |
1159 | rcu_implicit_dynticks_qs)) | |
1160 | goto unlock_ret; | |
1161 | ||
1162 | /* Leave state in case more forcing is required. */ | |
1163 | ||
1164 | break; | |
1165 | } | |
1166 | unlock_ret: | |
1167 | spin_unlock_irqrestore(&rsp->fqslock, flags); | |
1168 | } | |
1169 | ||
1170 | #else /* #ifdef CONFIG_SMP */ | |
1171 | ||
1172 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1173 | { | |
1174 | set_need_resched(); | |
1175 | } | |
1176 | ||
1177 | #endif /* #else #ifdef CONFIG_SMP */ | |
1178 | ||
1179 | /* | |
1180 | * This does the RCU processing work from softirq context for the | |
1181 | * specified rcu_state and rcu_data structures. This may be called | |
1182 | * only from the CPU to whom the rdp belongs. | |
1183 | */ | |
1184 | static void | |
1185 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1186 | { | |
1187 | unsigned long flags; | |
1188 | ||
2e597558 PM |
1189 | WARN_ON_ONCE(rdp->beenonline == 0); |
1190 | ||
64db4cff PM |
1191 | /* |
1192 | * If an RCU GP has gone long enough, go check for dyntick | |
1193 | * idle CPUs and, if needed, send resched IPIs. | |
1194 | */ | |
ef631b0c | 1195 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1196 | force_quiescent_state(rsp, 1); |
1197 | ||
1198 | /* | |
1199 | * Advance callbacks in response to end of earlier grace | |
1200 | * period that some other CPU ended. | |
1201 | */ | |
1202 | rcu_process_gp_end(rsp, rdp); | |
1203 | ||
1204 | /* Update RCU state based on any recent quiescent states. */ | |
1205 | rcu_check_quiescent_state(rsp, rdp); | |
1206 | ||
1207 | /* Does this CPU require a not-yet-started grace period? */ | |
1208 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1209 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | |
1210 | rcu_start_gp(rsp, flags); /* releases above lock */ | |
1211 | } | |
1212 | ||
1213 | /* If there are callbacks ready, invoke them. */ | |
1214 | rcu_do_batch(rdp); | |
1215 | } | |
1216 | ||
1217 | /* | |
1218 | * Do softirq processing for the current CPU. | |
1219 | */ | |
1220 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1221 | { | |
1222 | /* | |
1223 | * Memory references from any prior RCU read-side critical sections | |
1224 | * executed by the interrupted code must be seen before any RCU | |
1225 | * grace-period manipulations below. | |
1226 | */ | |
1227 | smp_mb(); /* See above block comment. */ | |
1228 | ||
d6714c22 PM |
1229 | __rcu_process_callbacks(&rcu_sched_state, |
1230 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1231 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1232 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1233 | |
1234 | /* | |
1235 | * Memory references from any later RCU read-side critical sections | |
1236 | * executed by the interrupted code must be seen after any RCU | |
1237 | * grace-period manipulations above. | |
1238 | */ | |
1239 | smp_mb(); /* See above block comment. */ | |
1240 | } | |
1241 | ||
1242 | static void | |
1243 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1244 | struct rcu_state *rsp) | |
1245 | { | |
1246 | unsigned long flags; | |
1247 | struct rcu_data *rdp; | |
1248 | ||
1249 | head->func = func; | |
1250 | head->next = NULL; | |
1251 | ||
1252 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1253 | ||
1254 | /* | |
1255 | * Opportunistically note grace-period endings and beginnings. | |
1256 | * Note that we might see a beginning right after we see an | |
1257 | * end, but never vice versa, since this CPU has to pass through | |
1258 | * a quiescent state betweentimes. | |
1259 | */ | |
1260 | local_irq_save(flags); | |
1261 | rdp = rsp->rda[smp_processor_id()]; | |
1262 | rcu_process_gp_end(rsp, rdp); | |
1263 | check_for_new_grace_period(rsp, rdp); | |
1264 | ||
1265 | /* Add the callback to our list. */ | |
1266 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1267 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1268 | ||
1269 | /* Start a new grace period if one not already started. */ | |
1270 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { | |
1271 | unsigned long nestflag; | |
1272 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1273 | ||
1274 | spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1275 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | |
1276 | } | |
1277 | ||
1278 | /* Force the grace period if too many callbacks or too long waiting. */ | |
1279 | if (unlikely(++rdp->qlen > qhimark)) { | |
1280 | rdp->blimit = LONG_MAX; | |
1281 | force_quiescent_state(rsp, 0); | |
ef631b0c | 1282 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1283 | force_quiescent_state(rsp, 1); |
1284 | local_irq_restore(flags); | |
1285 | } | |
1286 | ||
1287 | /* | |
d6714c22 | 1288 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1289 | */ |
d6714c22 | 1290 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1291 | { |
d6714c22 | 1292 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1293 | } |
d6714c22 | 1294 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1295 | |
1296 | /* | |
1297 | * Queue an RCU for invocation after a quicker grace period. | |
1298 | */ | |
1299 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1300 | { | |
1301 | __call_rcu(head, func, &rcu_bh_state); | |
1302 | } | |
1303 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1304 | ||
1305 | /* | |
1306 | * Check to see if there is any immediate RCU-related work to be done | |
1307 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1308 | * The checks are in order of increasing expense: checks that can be | |
1309 | * carried out against CPU-local state are performed first. However, | |
1310 | * we must check for CPU stalls first, else we might not get a chance. | |
1311 | */ | |
1312 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1313 | { | |
1314 | rdp->n_rcu_pending++; | |
1315 | ||
1316 | /* Check for CPU stalls, if enabled. */ | |
1317 | check_cpu_stall(rsp, rdp); | |
1318 | ||
1319 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
7ba5c840 PM |
1320 | if (rdp->qs_pending) { |
1321 | rdp->n_rp_qs_pending++; | |
64db4cff | 1322 | return 1; |
7ba5c840 | 1323 | } |
64db4cff PM |
1324 | |
1325 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1326 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1327 | rdp->n_rp_cb_ready++; | |
64db4cff | 1328 | return 1; |
7ba5c840 | 1329 | } |
64db4cff PM |
1330 | |
1331 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1332 | if (cpu_needs_another_gp(rsp, rdp)) { |
1333 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1334 | return 1; |
7ba5c840 | 1335 | } |
64db4cff PM |
1336 | |
1337 | /* Has another RCU grace period completed? */ | |
7ba5c840 PM |
1338 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ |
1339 | rdp->n_rp_gp_completed++; | |
64db4cff | 1340 | return 1; |
7ba5c840 | 1341 | } |
64db4cff PM |
1342 | |
1343 | /* Has a new RCU grace period started? */ | |
7ba5c840 PM |
1344 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ |
1345 | rdp->n_rp_gp_started++; | |
64db4cff | 1346 | return 1; |
7ba5c840 | 1347 | } |
64db4cff PM |
1348 | |
1349 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
1350 | if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && | |
7ba5c840 PM |
1351 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
1352 | rdp->n_rp_need_fqs++; | |
64db4cff | 1353 | return 1; |
7ba5c840 | 1354 | } |
64db4cff PM |
1355 | |
1356 | /* nothing to do */ | |
7ba5c840 | 1357 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1358 | return 0; |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * Check to see if there is any immediate RCU-related work to be done | |
1363 | * by the current CPU, returning 1 if so. This function is part of the | |
1364 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1365 | */ | |
a157229c | 1366 | static int rcu_pending(int cpu) |
64db4cff | 1367 | { |
d6714c22 | 1368 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1369 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1370 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1371 | } |
1372 | ||
1373 | /* | |
1374 | * Check to see if any future RCU-related work will need to be done | |
1375 | * by the current CPU, even if none need be done immediately, returning | |
1376 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1377 | * an exported member of the RCU API. | |
1378 | */ | |
1379 | int rcu_needs_cpu(int cpu) | |
1380 | { | |
1381 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1382 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1383 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1384 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1385 | } |
1386 | ||
1387 | /* | |
27569620 | 1388 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1389 | */ |
27569620 PM |
1390 | static void __init |
1391 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1392 | { |
1393 | unsigned long flags; | |
1394 | int i; | |
27569620 PM |
1395 | struct rcu_data *rdp = rsp->rda[cpu]; |
1396 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1397 | ||
1398 | /* Set up local state, ensuring consistent view of global state. */ | |
1399 | spin_lock_irqsave(&rnp->lock, flags); | |
1400 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | |
1401 | rdp->nxtlist = NULL; | |
1402 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1403 | rdp->nxttail[i] = &rdp->nxtlist; | |
1404 | rdp->qlen = 0; | |
1405 | #ifdef CONFIG_NO_HZ | |
1406 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1407 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1408 | rdp->cpu = cpu; | |
1409 | spin_unlock_irqrestore(&rnp->lock, flags); | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1414 | * offline event can be happening at a given time. Note also that we | |
1415 | * can accept some slop in the rsp->completed access due to the fact | |
1416 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1417 | */ |
e4fa4c97 | 1418 | static void __cpuinit |
f41d911f | 1419 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1420 | { |
1421 | unsigned long flags; | |
64db4cff PM |
1422 | long lastcomp; |
1423 | unsigned long mask; | |
1424 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1425 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1426 | ||
1427 | /* Set up local state, ensuring consistent view of global state. */ | |
1428 | spin_lock_irqsave(&rnp->lock, flags); | |
1429 | lastcomp = rsp->completed; | |
1430 | rdp->completed = lastcomp; | |
1431 | rdp->gpnum = lastcomp; | |
1432 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ | |
1433 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1434 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1435 | rdp->preemptable = preemptable; |
64db4cff | 1436 | rdp->passed_quiesc_completed = lastcomp - 1; |
64db4cff | 1437 | rdp->blimit = blimit; |
64db4cff PM |
1438 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1439 | ||
1440 | /* | |
1441 | * A new grace period might start here. If so, we won't be part | |
1442 | * of it, but that is OK, as we are currently in a quiescent state. | |
1443 | */ | |
1444 | ||
1445 | /* Exclude any attempts to start a new GP on large systems. */ | |
1446 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
1447 | ||
1448 | /* Add CPU to rcu_node bitmasks. */ | |
1449 | rnp = rdp->mynode; | |
1450 | mask = rdp->grpmask; | |
1451 | do { | |
1452 | /* Exclude any attempts to start a new GP on small systems. */ | |
1453 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1454 | rnp->qsmaskinit |= mask; | |
1455 | mask = rnp->grpmask; | |
1456 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | |
1457 | rnp = rnp->parent; | |
1458 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1459 | ||
1460 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
1461 | ||
1462 | /* | |
1463 | * A new grace period might start here. If so, we will be part of | |
1464 | * it, and its gpnum will be greater than ours, so we will | |
1465 | * participate. It is also possible for the gpnum to have been | |
1466 | * incremented before this function was called, and the bitmasks | |
1467 | * to not be filled out until now, in which case we will also | |
1468 | * participate due to our gpnum being behind. | |
1469 | */ | |
1470 | ||
1471 | /* Since it is coming online, the CPU is in a quiescent state. */ | |
1472 | cpu_quiet(cpu, rsp, rdp, lastcomp); | |
1473 | local_irq_restore(flags); | |
1474 | } | |
1475 | ||
1476 | static void __cpuinit rcu_online_cpu(int cpu) | |
1477 | { | |
f41d911f PM |
1478 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1479 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1480 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1481 | } |
1482 | ||
1483 | /* | |
f41d911f | 1484 | * Handle CPU online/offline notification events. |
64db4cff | 1485 | */ |
2e597558 PM |
1486 | int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1487 | unsigned long action, void *hcpu) | |
64db4cff PM |
1488 | { |
1489 | long cpu = (long)hcpu; | |
1490 | ||
1491 | switch (action) { | |
1492 | case CPU_UP_PREPARE: | |
1493 | case CPU_UP_PREPARE_FROZEN: | |
1494 | rcu_online_cpu(cpu); | |
1495 | break; | |
1496 | case CPU_DEAD: | |
1497 | case CPU_DEAD_FROZEN: | |
1498 | case CPU_UP_CANCELED: | |
1499 | case CPU_UP_CANCELED_FROZEN: | |
1500 | rcu_offline_cpu(cpu); | |
1501 | break; | |
1502 | default: | |
1503 | break; | |
1504 | } | |
1505 | return NOTIFY_OK; | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * Compute the per-level fanout, either using the exact fanout specified | |
1510 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1511 | */ | |
1512 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1513 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1514 | { | |
1515 | int i; | |
1516 | ||
1517 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1518 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1519 | } | |
1520 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1521 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1522 | { | |
1523 | int ccur; | |
1524 | int cprv; | |
1525 | int i; | |
1526 | ||
1527 | cprv = NR_CPUS; | |
1528 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1529 | ccur = rsp->levelcnt[i]; | |
1530 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1531 | cprv = ccur; | |
1532 | } | |
1533 | } | |
1534 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1535 | ||
1536 | /* | |
1537 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1538 | */ | |
1539 | static void __init rcu_init_one(struct rcu_state *rsp) | |
1540 | { | |
1541 | int cpustride = 1; | |
1542 | int i; | |
1543 | int j; | |
1544 | struct rcu_node *rnp; | |
1545 | ||
1546 | /* Initialize the level-tracking arrays. */ | |
1547 | ||
1548 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1549 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1550 | rcu_init_levelspread(rsp); | |
1551 | ||
1552 | /* Initialize the elements themselves, starting from the leaves. */ | |
1553 | ||
1554 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1555 | cpustride *= rsp->levelspread[i]; | |
1556 | rnp = rsp->level[i]; | |
1557 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1558 | spin_lock_init(&rnp->lock); | |
f41d911f | 1559 | rnp->gpnum = 0; |
64db4cff PM |
1560 | rnp->qsmask = 0; |
1561 | rnp->qsmaskinit = 0; | |
1562 | rnp->grplo = j * cpustride; | |
1563 | rnp->grphi = (j + 1) * cpustride - 1; | |
1564 | if (rnp->grphi >= NR_CPUS) | |
1565 | rnp->grphi = NR_CPUS - 1; | |
1566 | if (i == 0) { | |
1567 | rnp->grpnum = 0; | |
1568 | rnp->grpmask = 0; | |
1569 | rnp->parent = NULL; | |
1570 | } else { | |
1571 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1572 | rnp->grpmask = 1UL << rnp->grpnum; | |
1573 | rnp->parent = rsp->level[i - 1] + | |
1574 | j / rsp->levelspread[i - 1]; | |
1575 | } | |
1576 | rnp->level = i; | |
f41d911f PM |
1577 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1578 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
64db4cff PM |
1579 | } |
1580 | } | |
1581 | } | |
1582 | ||
1583 | /* | |
f41d911f PM |
1584 | * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used |
1585 | * nowhere else! Assigns leaf node pointers into each CPU's rcu_data | |
1586 | * structure. | |
64db4cff | 1587 | */ |
65cf8f86 | 1588 | #define RCU_INIT_FLAVOR(rsp, rcu_data) \ |
64db4cff | 1589 | do { \ |
65cf8f86 | 1590 | rcu_init_one(rsp); \ |
64db4cff PM |
1591 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
1592 | j = 0; \ | |
1593 | for_each_possible_cpu(i) { \ | |
1594 | if (i > rnp[j].grphi) \ | |
1595 | j++; \ | |
1596 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | |
1597 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | |
65cf8f86 | 1598 | rcu_boot_init_percpu_data(i, rsp); \ |
64db4cff PM |
1599 | } \ |
1600 | } while (0) | |
1601 | ||
f41d911f PM |
1602 | #ifdef CONFIG_TREE_PREEMPT_RCU |
1603 | ||
1604 | void __init __rcu_init_preempt(void) | |
1605 | { | |
1606 | int i; /* All used by RCU_INIT_FLAVOR(). */ | |
1607 | int j; | |
1608 | struct rcu_node *rnp; | |
1609 | ||
1610 | RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); | |
1611 | } | |
1612 | ||
1613 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1614 | ||
1615 | void __init __rcu_init_preempt(void) | |
1616 | { | |
1617 | } | |
1618 | ||
1619 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
64db4cff PM |
1620 | |
1621 | void __init __rcu_init(void) | |
1622 | { | |
f41d911f | 1623 | int i; /* All used by RCU_INIT_FLAVOR(). */ |
64db4cff PM |
1624 | int j; |
1625 | struct rcu_node *rnp; | |
1626 | ||
f41d911f | 1627 | rcu_bootup_announce(); |
64db4cff PM |
1628 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
1629 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | |
1630 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
65cf8f86 PM |
1631 | RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); |
1632 | RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); | |
f41d911f | 1633 | __rcu_init_preempt(); |
2e597558 | 1634 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
64db4cff PM |
1635 | } |
1636 | ||
1637 | module_param(blimit, int, 0); | |
1638 | module_param(qhimark, int, 0); | |
1639 | module_param(qlowmark, int, 0); |