]>
Commit | Line | Data |
---|---|---|
64db4cff PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
18 | * Copyright IBM Corporation, 2008 | |
19 | * | |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | |
23 | * | |
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | |
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | |
26 | * | |
27 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff PM |
29 | */ |
30 | #include <linux/types.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/smp.h> | |
35 | #include <linux/rcupdate.h> | |
36 | #include <linux/interrupt.h> | |
37 | #include <linux/sched.h> | |
c1dc0b9c | 38 | #include <linux/nmi.h> |
64db4cff PM |
39 | #include <asm/atomic.h> |
40 | #include <linux/bitops.h> | |
41 | #include <linux/module.h> | |
42 | #include <linux/completion.h> | |
43 | #include <linux/moduleparam.h> | |
44 | #include <linux/percpu.h> | |
45 | #include <linux/notifier.h> | |
46 | #include <linux/cpu.h> | |
47 | #include <linux/mutex.h> | |
48 | #include <linux/time.h> | |
49 | ||
9f77da9f PM |
50 | #include "rcutree.h" |
51 | ||
64db4cff PM |
52 | #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 | struct rcu_data *rdp; |
111 | ||
f41d911f | 112 | rdp = &per_cpu(rcu_sched_data, cpu); |
b1f77b05 | 113 | rdp->passed_quiesc_completed = rdp->completed; |
c3422bea PM |
114 | barrier(); |
115 | rdp->passed_quiesc = 1; | |
116 | rcu_preempt_note_context_switch(cpu); | |
b1f77b05 IM |
117 | } |
118 | ||
d6714c22 | 119 | void rcu_bh_qs(int cpu) |
b1f77b05 | 120 | { |
f41d911f PM |
121 | struct rcu_data *rdp; |
122 | ||
f41d911f | 123 | rdp = &per_cpu(rcu_bh_data, cpu); |
b1f77b05 | 124 | rdp->passed_quiesc_completed = rdp->completed; |
c3422bea PM |
125 | barrier(); |
126 | rdp->passed_quiesc = 1; | |
b1f77b05 | 127 | } |
64db4cff PM |
128 | |
129 | #ifdef CONFIG_NO_HZ | |
90a4d2c0 PM |
130 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
131 | .dynticks_nesting = 1, | |
132 | .dynticks = 1, | |
133 | }; | |
64db4cff PM |
134 | #endif /* #ifdef CONFIG_NO_HZ */ |
135 | ||
136 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
137 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
138 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
139 | ||
140 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); | |
a157229c | 141 | static int rcu_pending(int cpu); |
64db4cff PM |
142 | |
143 | /* | |
d6714c22 | 144 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 145 | */ |
d6714c22 | 146 | long rcu_batches_completed_sched(void) |
64db4cff | 147 | { |
d6714c22 | 148 | return rcu_sched_state.completed; |
64db4cff | 149 | } |
d6714c22 | 150 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
151 | |
152 | /* | |
153 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
154 | */ | |
155 | long rcu_batches_completed_bh(void) | |
156 | { | |
157 | return rcu_bh_state.completed; | |
158 | } | |
159 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
160 | ||
161 | /* | |
162 | * Does the CPU have callbacks ready to be invoked? | |
163 | */ | |
164 | static int | |
165 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
166 | { | |
167 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Does the current CPU require a yet-as-unscheduled grace period? | |
172 | */ | |
173 | static int | |
174 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
175 | { | |
176 | /* ACCESS_ONCE() because we are accessing outside of lock. */ | |
177 | return *rdp->nxttail[RCU_DONE_TAIL] && | |
178 | ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum); | |
179 | } | |
180 | ||
181 | /* | |
182 | * Return the root node of the specified rcu_state structure. | |
183 | */ | |
184 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
185 | { | |
186 | return &rsp->node[0]; | |
187 | } | |
188 | ||
189 | #ifdef CONFIG_SMP | |
190 | ||
191 | /* | |
192 | * If the specified CPU is offline, tell the caller that it is in | |
193 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
194 | * Grace periods can end up waiting on an offline CPU when that | |
195 | * CPU is in the process of coming online -- it will be added to the | |
196 | * rcu_node bitmasks before it actually makes it online. The same thing | |
197 | * can happen while a CPU is in the process of coming online. Because this | |
198 | * race is quite rare, we check for it after detecting that the grace | |
199 | * period has been delayed rather than checking each and every CPU | |
200 | * each and every time we start a new grace period. | |
201 | */ | |
202 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
203 | { | |
204 | /* | |
205 | * If the CPU is offline, it is in a quiescent state. We can | |
206 | * trust its state not to change because interrupts are disabled. | |
207 | */ | |
208 | if (cpu_is_offline(rdp->cpu)) { | |
209 | rdp->offline_fqs++; | |
210 | return 1; | |
211 | } | |
212 | ||
f41d911f PM |
213 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
214 | if (rdp->preemptable) | |
215 | return 0; | |
216 | ||
64db4cff PM |
217 | /* The CPU is online, so send it a reschedule IPI. */ |
218 | if (rdp->cpu != smp_processor_id()) | |
219 | smp_send_reschedule(rdp->cpu); | |
220 | else | |
221 | set_need_resched(); | |
222 | rdp->resched_ipi++; | |
223 | return 0; | |
224 | } | |
225 | ||
226 | #endif /* #ifdef CONFIG_SMP */ | |
227 | ||
228 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
229 | |
230 | /** | |
231 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
232 | * | |
233 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
234 | * read-side critical sections can occur. (Though RCU read-side | |
235 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
236 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
237 | */ | |
238 | void rcu_enter_nohz(void) | |
239 | { | |
240 | unsigned long flags; | |
241 | struct rcu_dynticks *rdtp; | |
242 | ||
243 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
244 | local_irq_save(flags); | |
245 | rdtp = &__get_cpu_var(rcu_dynticks); | |
246 | rdtp->dynticks++; | |
247 | rdtp->dynticks_nesting--; | |
86848966 | 248 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
249 | local_irq_restore(flags); |
250 | } | |
251 | ||
252 | /* | |
253 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
254 | * | |
255 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
256 | * read-side critical sections normally occur. | |
257 | */ | |
258 | void rcu_exit_nohz(void) | |
259 | { | |
260 | unsigned long flags; | |
261 | struct rcu_dynticks *rdtp; | |
262 | ||
263 | local_irq_save(flags); | |
264 | rdtp = &__get_cpu_var(rcu_dynticks); | |
265 | rdtp->dynticks++; | |
266 | rdtp->dynticks_nesting++; | |
86848966 | 267 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
268 | local_irq_restore(flags); |
269 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | |
270 | } | |
271 | ||
272 | /** | |
273 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
274 | * | |
275 | * If the CPU was idle with dynamic ticks active, and there is no | |
276 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
277 | * RCU grace-period handling know that the CPU is active. | |
278 | */ | |
279 | void rcu_nmi_enter(void) | |
280 | { | |
281 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
282 | ||
283 | if (rdtp->dynticks & 0x1) | |
284 | return; | |
285 | rdtp->dynticks_nmi++; | |
86848966 | 286 | WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1)); |
64db4cff PM |
287 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
288 | } | |
289 | ||
290 | /** | |
291 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
292 | * | |
293 | * If the CPU was idle with dynamic ticks active, and there is no | |
294 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
295 | * RCU grace-period handling know that the CPU is no longer active. | |
296 | */ | |
297 | void rcu_nmi_exit(void) | |
298 | { | |
299 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
300 | ||
301 | if (rdtp->dynticks & 0x1) | |
302 | return; | |
303 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
304 | rdtp->dynticks_nmi++; | |
86848966 | 305 | WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1); |
64db4cff PM |
306 | } |
307 | ||
308 | /** | |
309 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
310 | * | |
311 | * If the CPU was idle with dynamic ticks active, this updates the | |
312 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
313 | */ | |
314 | void rcu_irq_enter(void) | |
315 | { | |
316 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
317 | ||
318 | if (rdtp->dynticks_nesting++) | |
319 | return; | |
320 | rdtp->dynticks++; | |
86848966 | 321 | WARN_ON_ONCE(!(rdtp->dynticks & 0x1)); |
64db4cff PM |
322 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
323 | } | |
324 | ||
325 | /** | |
326 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
327 | * | |
328 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
329 | * to put let the RCU handling be aware that the CPU is going back to idle | |
330 | * with no ticks. | |
331 | */ | |
332 | void rcu_irq_exit(void) | |
333 | { | |
334 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
335 | ||
336 | if (--rdtp->dynticks_nesting) | |
337 | return; | |
338 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | |
339 | rdtp->dynticks++; | |
86848966 | 340 | WARN_ON_ONCE(rdtp->dynticks & 0x1); |
64db4cff PM |
341 | |
342 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
d6714c22 | 343 | if (__get_cpu_var(rcu_sched_data).nxtlist || |
64db4cff PM |
344 | __get_cpu_var(rcu_bh_data).nxtlist) |
345 | set_need_resched(); | |
346 | } | |
347 | ||
348 | /* | |
349 | * Record the specified "completed" value, which is later used to validate | |
350 | * dynticks counter manipulations. Specify "rsp->completed - 1" to | |
351 | * unconditionally invalidate any future dynticks manipulations (which is | |
352 | * useful at the beginning of a grace period). | |
353 | */ | |
354 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
355 | { | |
356 | rsp->dynticks_completed = comp; | |
357 | } | |
358 | ||
359 | #ifdef CONFIG_SMP | |
360 | ||
361 | /* | |
362 | * Recall the previously recorded value of the completion for dynticks. | |
363 | */ | |
364 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
365 | { | |
366 | return rsp->dynticks_completed; | |
367 | } | |
368 | ||
369 | /* | |
370 | * Snapshot the specified CPU's dynticks counter so that we can later | |
371 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
372 | * is already in a quiescent state courtesy of dynticks idle mode. | |
373 | */ | |
374 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
375 | { | |
376 | int ret; | |
377 | int snap; | |
378 | int snap_nmi; | |
379 | ||
380 | snap = rdp->dynticks->dynticks; | |
381 | snap_nmi = rdp->dynticks->dynticks_nmi; | |
382 | smp_mb(); /* Order sampling of snap with end of grace period. */ | |
383 | rdp->dynticks_snap = snap; | |
384 | rdp->dynticks_nmi_snap = snap_nmi; | |
385 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | |
386 | if (ret) | |
387 | rdp->dynticks_fqs++; | |
388 | return ret; | |
389 | } | |
390 | ||
391 | /* | |
392 | * Return true if the specified CPU has passed through a quiescent | |
393 | * state by virtue of being in or having passed through an dynticks | |
394 | * idle state since the last call to dyntick_save_progress_counter() | |
395 | * for this same CPU. | |
396 | */ | |
397 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
398 | { | |
399 | long curr; | |
400 | long curr_nmi; | |
401 | long snap; | |
402 | long snap_nmi; | |
403 | ||
404 | curr = rdp->dynticks->dynticks; | |
405 | snap = rdp->dynticks_snap; | |
406 | curr_nmi = rdp->dynticks->dynticks_nmi; | |
407 | snap_nmi = rdp->dynticks_nmi_snap; | |
408 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | |
409 | ||
410 | /* | |
411 | * If the CPU passed through or entered a dynticks idle phase with | |
412 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
413 | * already acknowledged the request to pass through a quiescent | |
414 | * state. Either way, that CPU cannot possibly be in an RCU | |
415 | * read-side critical section that started before the beginning | |
416 | * of the current RCU grace period. | |
417 | */ | |
418 | if ((curr != snap || (curr & 0x1) == 0) && | |
419 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | |
420 | rdp->dynticks_fqs++; | |
421 | return 1; | |
422 | } | |
423 | ||
424 | /* Go check for the CPU being offline. */ | |
425 | return rcu_implicit_offline_qs(rdp); | |
426 | } | |
427 | ||
428 | #endif /* #ifdef CONFIG_SMP */ | |
429 | ||
430 | #else /* #ifdef CONFIG_NO_HZ */ | |
431 | ||
432 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | |
433 | { | |
434 | } | |
435 | ||
436 | #ifdef CONFIG_SMP | |
437 | ||
438 | /* | |
439 | * If there are no dynticks, then the only way that a CPU can passively | |
440 | * be in a quiescent state is to be offline. Unlike dynticks idle, which | |
441 | * is a point in time during the prior (already finished) grace period, | |
442 | * an offline CPU is always in a quiescent state, and thus can be | |
443 | * unconditionally applied. So just return the current value of completed. | |
444 | */ | |
445 | static long dyntick_recall_completed(struct rcu_state *rsp) | |
446 | { | |
447 | return rsp->completed; | |
448 | } | |
449 | ||
450 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
451 | { | |
452 | return 0; | |
453 | } | |
454 | ||
455 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
456 | { | |
457 | return rcu_implicit_offline_qs(rdp); | |
458 | } | |
459 | ||
460 | #endif /* #ifdef CONFIG_SMP */ | |
461 | ||
462 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
463 | ||
464 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
465 | ||
466 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
467 | { | |
468 | rsp->gp_start = jiffies; | |
469 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
470 | } | |
471 | ||
472 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
473 | { | |
474 | int cpu; | |
475 | long delta; | |
476 | unsigned long flags; | |
477 | struct rcu_node *rnp = rcu_get_root(rsp); | |
478 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
479 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
480 | ||
481 | /* Only let one CPU complain about others per time interval. */ | |
482 | ||
483 | spin_lock_irqsave(&rnp->lock, flags); | |
484 | delta = jiffies - rsp->jiffies_stall; | |
485 | if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { | |
486 | spin_unlock_irqrestore(&rnp->lock, flags); | |
487 | return; | |
488 | } | |
489 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
490 | spin_unlock_irqrestore(&rnp->lock, flags); | |
491 | ||
492 | /* OK, time to rat on our buddy... */ | |
493 | ||
494 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | |
495 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
f41d911f | 496 | rcu_print_task_stall(rnp); |
64db4cff PM |
497 | if (rnp_cur->qsmask == 0) |
498 | continue; | |
499 | for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) | |
500 | if (rnp_cur->qsmask & (1UL << cpu)) | |
501 | printk(" %d", rnp_cur->grplo + cpu); | |
502 | } | |
503 | printk(" (detected by %d, t=%ld jiffies)\n", | |
504 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | |
c1dc0b9c IM |
505 | trigger_all_cpu_backtrace(); |
506 | ||
64db4cff PM |
507 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
508 | } | |
509 | ||
510 | static void print_cpu_stall(struct rcu_state *rsp) | |
511 | { | |
512 | unsigned long flags; | |
513 | struct rcu_node *rnp = rcu_get_root(rsp); | |
514 | ||
515 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | |
516 | smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
517 | trigger_all_cpu_backtrace(); |
518 | ||
64db4cff PM |
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); | |
c1dc0b9c | 524 | |
64db4cff PM |
525 | set_need_resched(); /* kick ourselves to get things going. */ |
526 | } | |
527 | ||
528 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
529 | { | |
530 | long delta; | |
531 | struct rcu_node *rnp; | |
532 | ||
533 | delta = jiffies - rsp->jiffies_stall; | |
534 | rnp = rdp->mynode; | |
535 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | |
536 | ||
537 | /* We haven't checked in, so go dump stack. */ | |
538 | print_cpu_stall(rsp); | |
539 | ||
540 | } else if (rsp->gpnum != rsp->completed && | |
541 | delta >= RCU_STALL_RAT_DELAY) { | |
542 | ||
543 | /* They had two time units to dump stack, so complain. */ | |
544 | print_other_cpu_stall(rsp); | |
545 | } | |
546 | } | |
547 | ||
548 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
549 | ||
550 | static void record_gp_stall_check_time(struct rcu_state *rsp) | |
551 | { | |
552 | } | |
553 | ||
554 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
555 | { | |
556 | } | |
557 | ||
558 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
559 | ||
560 | /* | |
561 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
562 | * This is used both when we started the grace period and when we notice | |
563 | * that someone else started the grace period. | |
564 | */ | |
565 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) | |
566 | { | |
567 | rdp->qs_pending = 1; | |
568 | rdp->passed_quiesc = 0; | |
569 | rdp->gpnum = rsp->gpnum; | |
64db4cff PM |
570 | } |
571 | ||
572 | /* | |
573 | * Did someone else start a new RCU grace period start since we last | |
574 | * checked? Update local state appropriately if so. Must be called | |
575 | * on the CPU corresponding to rdp. | |
576 | */ | |
577 | static int | |
578 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
579 | { | |
580 | unsigned long flags; | |
581 | int ret = 0; | |
582 | ||
583 | local_irq_save(flags); | |
584 | if (rdp->gpnum != rsp->gpnum) { | |
585 | note_new_gpnum(rsp, rdp); | |
586 | ret = 1; | |
587 | } | |
588 | local_irq_restore(flags); | |
589 | return ret; | |
590 | } | |
591 | ||
592 | /* | |
593 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
594 | * in preparation for detecting the next grace period. The caller must hold | |
595 | * the root node's ->lock, which is released before return. Hard irqs must | |
596 | * be disabled. | |
597 | */ | |
598 | static void | |
599 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
600 | __releases(rcu_get_root(rsp)->lock) | |
601 | { | |
602 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | |
603 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
604 | |
605 | if (!cpu_needs_another_gp(rsp, rdp)) { | |
606 | spin_unlock_irqrestore(&rnp->lock, flags); | |
607 | return; | |
608 | } | |
609 | ||
610 | /* Advance to a new grace period and initialize state. */ | |
611 | rsp->gpnum++; | |
c3422bea | 612 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
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) { | |
b0e165c0 | 629 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 630 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 631 | rnp->gpnum = rsp->gpnum; |
c12172c0 | 632 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
64db4cff PM |
633 | spin_unlock_irqrestore(&rnp->lock, flags); |
634 | return; | |
635 | } | |
636 | ||
637 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | |
638 | ||
639 | ||
640 | /* Exclude any concurrent CPU-hotplug operations. */ | |
641 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
642 | ||
643 | /* | |
b835db1f PM |
644 | * Set the quiescent-state-needed bits in all the rcu_node |
645 | * structures for all currently online CPUs in breadth-first | |
646 | * order, starting from the root rcu_node structure. This | |
647 | * operation relies on the layout of the hierarchy within the | |
648 | * rsp->node[] array. Note that other CPUs will access only | |
649 | * the leaves of the hierarchy, which still indicate that no | |
650 | * grace period is in progress, at least until the corresponding | |
651 | * leaf node has been initialized. In addition, we have excluded | |
652 | * CPU-hotplug operations. | |
64db4cff PM |
653 | * |
654 | * Note that the grace period cannot complete until we finish | |
655 | * the initialization process, as there will be at least one | |
656 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
657 | * one corresponding to this CPU, due to the fact that we have |
658 | * irqs disabled. | |
64db4cff | 659 | */ |
49e29126 PM |
660 | for (rnp = &rsp->node[0]; rnp < &rsp->node[NUM_RCU_NODES]; rnp++) { |
661 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
b0e165c0 | 662 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 663 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 664 | rnp->gpnum = rsp->gpnum; |
49e29126 | 665 | spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
666 | } |
667 | ||
668 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ | |
669 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
670 | } | |
671 | ||
672 | /* | |
673 | * Advance this CPU's callbacks, but only if the current grace period | |
674 | * has ended. This may be called only from the CPU to whom the rdp | |
675 | * belongs. | |
676 | */ | |
677 | static void | |
678 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
679 | { | |
680 | long completed_snap; | |
681 | unsigned long flags; | |
682 | ||
683 | local_irq_save(flags); | |
684 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ | |
685 | ||
686 | /* Did another grace period end? */ | |
687 | if (rdp->completed != completed_snap) { | |
688 | ||
689 | /* Advance callbacks. No harm if list empty. */ | |
690 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
691 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
692 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
693 | ||
694 | /* Remember that we saw this grace-period completion. */ | |
695 | rdp->completed = completed_snap; | |
696 | } | |
697 | local_irq_restore(flags); | |
698 | } | |
699 | ||
f41d911f PM |
700 | /* |
701 | * Clean up after the prior grace period and let rcu_start_gp() start up | |
702 | * the next grace period if one is needed. Note that the caller must | |
703 | * hold rnp->lock, as required by rcu_start_gp(), which will release it. | |
704 | */ | |
705 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) | |
706 | __releases(rnp->lock) | |
707 | { | |
28ecd580 | 708 | WARN_ON_ONCE(rsp->completed == rsp->gpnum); |
f41d911f PM |
709 | rsp->completed = rsp->gpnum; |
710 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); | |
711 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ | |
712 | } | |
713 | ||
64db4cff PM |
714 | /* |
715 | * Similar to cpu_quiet(), for which it is a helper function. Allows | |
716 | * a group of CPUs to be quieted at one go, though all the CPUs in the | |
717 | * group must be represented by the same leaf rcu_node structure. | |
718 | * That structure's lock must be held upon entry, and it is released | |
719 | * before return. | |
720 | */ | |
721 | static void | |
722 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, | |
723 | unsigned long flags) | |
724 | __releases(rnp->lock) | |
725 | { | |
28ecd580 PM |
726 | struct rcu_node *rnp_c; |
727 | ||
64db4cff PM |
728 | /* Walk up the rcu_node hierarchy. */ |
729 | for (;;) { | |
730 | if (!(rnp->qsmask & mask)) { | |
731 | ||
732 | /* Our bit has already been cleared, so done. */ | |
733 | spin_unlock_irqrestore(&rnp->lock, flags); | |
734 | return; | |
735 | } | |
736 | rnp->qsmask &= ~mask; | |
f41d911f | 737 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { |
64db4cff PM |
738 | |
739 | /* Other bits still set at this level, so done. */ | |
740 | spin_unlock_irqrestore(&rnp->lock, flags); | |
741 | return; | |
742 | } | |
743 | mask = rnp->grpmask; | |
744 | if (rnp->parent == NULL) { | |
745 | ||
746 | /* No more levels. Exit loop holding root lock. */ | |
747 | ||
748 | break; | |
749 | } | |
750 | spin_unlock_irqrestore(&rnp->lock, flags); | |
28ecd580 | 751 | rnp_c = rnp; |
64db4cff PM |
752 | rnp = rnp->parent; |
753 | spin_lock_irqsave(&rnp->lock, flags); | |
28ecd580 | 754 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
755 | } |
756 | ||
757 | /* | |
758 | * Get here if we are the last CPU to pass through a quiescent | |
f41d911f PM |
759 | * state for this grace period. Invoke cpu_quiet_msk_finish() |
760 | * to clean up and start the next grace period if one is needed. | |
64db4cff | 761 | */ |
f41d911f | 762 | cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
763 | } |
764 | ||
765 | /* | |
766 | * Record a quiescent state for the specified CPU, which must either be | |
e7d8842e PM |
767 | * the current CPU. The lastcomp argument is used to make sure we are |
768 | * still in the grace period of interest. We don't want to end the current | |
769 | * grace period based on quiescent states detected in an earlier grace | |
770 | * period! | |
64db4cff PM |
771 | */ |
772 | static void | |
773 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) | |
774 | { | |
775 | unsigned long flags; | |
776 | unsigned long mask; | |
777 | struct rcu_node *rnp; | |
778 | ||
779 | rnp = rdp->mynode; | |
780 | spin_lock_irqsave(&rnp->lock, flags); | |
781 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { | |
782 | ||
783 | /* | |
784 | * Someone beat us to it for this grace period, so leave. | |
785 | * The race with GP start is resolved by the fact that we | |
786 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
787 | * cannot yet be initialized -- so we would simply find our | |
788 | * CPU's bit already cleared in cpu_quiet_msk() if this race | |
789 | * occurred. | |
790 | */ | |
791 | rdp->passed_quiesc = 0; /* try again later! */ | |
792 | spin_unlock_irqrestore(&rnp->lock, flags); | |
793 | return; | |
794 | } | |
795 | mask = rdp->grpmask; | |
796 | if ((rnp->qsmask & mask) == 0) { | |
797 | spin_unlock_irqrestore(&rnp->lock, flags); | |
798 | } else { | |
799 | rdp->qs_pending = 0; | |
800 | ||
801 | /* | |
802 | * This GP can't end until cpu checks in, so all of our | |
803 | * callbacks can be processed during the next GP. | |
804 | */ | |
64db4cff PM |
805 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
806 | ||
807 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ | |
808 | } | |
809 | } | |
810 | ||
811 | /* | |
812 | * Check to see if there is a new grace period of which this CPU | |
813 | * is not yet aware, and if so, set up local rcu_data state for it. | |
814 | * Otherwise, see if this CPU has just passed through its first | |
815 | * quiescent state for this grace period, and record that fact if so. | |
816 | */ | |
817 | static void | |
818 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
819 | { | |
820 | /* If there is now a new grace period, record and return. */ | |
821 | if (check_for_new_grace_period(rsp, rdp)) | |
822 | return; | |
823 | ||
824 | /* | |
825 | * Does this CPU still need to do its part for current grace period? | |
826 | * If no, return and let the other CPUs do their part as well. | |
827 | */ | |
828 | if (!rdp->qs_pending) | |
829 | return; | |
830 | ||
831 | /* | |
832 | * Was there a quiescent state since the beginning of the grace | |
833 | * period? If no, then exit and wait for the next call. | |
834 | */ | |
835 | if (!rdp->passed_quiesc) | |
836 | return; | |
837 | ||
838 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ | |
839 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
840 | } | |
841 | ||
842 | #ifdef CONFIG_HOTPLUG_CPU | |
843 | ||
844 | /* | |
845 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
846 | * and move all callbacks from the outgoing CPU to the current one. | |
847 | */ | |
848 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
849 | { | |
850 | int i; | |
851 | unsigned long flags; | |
852 | long lastcomp; | |
853 | unsigned long mask; | |
854 | struct rcu_data *rdp = rsp->rda[cpu]; | |
855 | struct rcu_data *rdp_me; | |
856 | struct rcu_node *rnp; | |
857 | ||
858 | /* Exclude any attempts to start a new grace period. */ | |
859 | spin_lock_irqsave(&rsp->onofflock, flags); | |
860 | ||
861 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 862 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
863 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
864 | do { | |
865 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
866 | rnp->qsmaskinit &= ~mask; | |
867 | if (rnp->qsmaskinit != 0) { | |
f41d911f | 868 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
869 | break; |
870 | } | |
28ecd580 | 871 | rcu_preempt_offline_tasks(rsp, rnp, rdp); |
64db4cff | 872 | mask = rnp->grpmask; |
f41d911f | 873 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
874 | rnp = rnp->parent; |
875 | } while (rnp != NULL); | |
876 | lastcomp = rsp->completed; | |
877 | ||
878 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
879 | ||
64db4cff PM |
880 | /* |
881 | * Move callbacks from the outgoing CPU to the running CPU. | |
882 | * Note that the outgoing CPU is now quiscent, so it is now | |
d6714c22 | 883 | * (uncharacteristically) safe to access its rcu_data structure. |
64db4cff PM |
884 | * Note also that we must carefully retain the order of the |
885 | * outgoing CPU's callbacks in order for rcu_barrier() to work | |
886 | * correctly. Finally, note that we start all the callbacks | |
887 | * afresh, even those that have passed through a grace period | |
888 | * and are therefore ready to invoke. The theory is that hotplug | |
889 | * events are rare, and that if they are frequent enough to | |
890 | * indefinitely delay callbacks, you have far worse things to | |
891 | * be worrying about. | |
892 | */ | |
893 | rdp_me = rsp->rda[smp_processor_id()]; | |
894 | if (rdp->nxtlist != NULL) { | |
895 | *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
896 | rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
897 | rdp->nxtlist = NULL; | |
898 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
899 | rdp->nxttail[i] = &rdp->nxtlist; | |
900 | rdp_me->qlen += rdp->qlen; | |
901 | rdp->qlen = 0; | |
902 | } | |
903 | local_irq_restore(flags); | |
904 | } | |
905 | ||
906 | /* | |
907 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
908 | * callbacks that it might have to the current CPU. This code assumes | |
909 | * that at least one CPU in the system will remain running at all times. | |
910 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
911 | */ | |
912 | static void rcu_offline_cpu(int cpu) | |
913 | { | |
d6714c22 | 914 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 915 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 916 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
917 | } |
918 | ||
919 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
920 | ||
921 | static void rcu_offline_cpu(int cpu) | |
922 | { | |
923 | } | |
924 | ||
925 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
926 | ||
927 | /* | |
928 | * Invoke any RCU callbacks that have made it to the end of their grace | |
929 | * period. Thottle as specified by rdp->blimit. | |
930 | */ | |
931 | static void rcu_do_batch(struct rcu_data *rdp) | |
932 | { | |
933 | unsigned long flags; | |
934 | struct rcu_head *next, *list, **tail; | |
935 | int count; | |
936 | ||
937 | /* If no callbacks are ready, just return.*/ | |
938 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
939 | return; | |
940 | ||
941 | /* | |
942 | * Extract the list of ready callbacks, disabling to prevent | |
943 | * races with call_rcu() from interrupt handlers. | |
944 | */ | |
945 | local_irq_save(flags); | |
946 | list = rdp->nxtlist; | |
947 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
948 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
949 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
950 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
951 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
952 | rdp->nxttail[count] = &rdp->nxtlist; | |
953 | local_irq_restore(flags); | |
954 | ||
955 | /* Invoke callbacks. */ | |
956 | count = 0; | |
957 | while (list) { | |
958 | next = list->next; | |
959 | prefetch(next); | |
960 | list->func(list); | |
961 | list = next; | |
962 | if (++count >= rdp->blimit) | |
963 | break; | |
964 | } | |
965 | ||
966 | local_irq_save(flags); | |
967 | ||
968 | /* Update count, and requeue any remaining callbacks. */ | |
969 | rdp->qlen -= count; | |
970 | if (list != NULL) { | |
971 | *tail = rdp->nxtlist; | |
972 | rdp->nxtlist = list; | |
973 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
974 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
975 | rdp->nxttail[count] = tail; | |
976 | else | |
977 | break; | |
978 | } | |
979 | ||
980 | /* Reinstate batch limit if we have worked down the excess. */ | |
981 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
982 | rdp->blimit = blimit; | |
983 | ||
984 | local_irq_restore(flags); | |
985 | ||
986 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
987 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
988 | raise_softirq(RCU_SOFTIRQ); | |
989 | } | |
990 | ||
991 | /* | |
992 | * Check to see if this CPU is in a non-context-switch quiescent state | |
993 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
994 | * Also schedule the RCU softirq handler. | |
995 | * | |
996 | * This function must be called with hardirqs disabled. It is normally | |
997 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
998 | * false, there is no point in invoking rcu_check_callbacks(). | |
999 | */ | |
1000 | void rcu_check_callbacks(int cpu, int user) | |
1001 | { | |
a157229c PM |
1002 | if (!rcu_pending(cpu)) |
1003 | return; /* if nothing for RCU to do. */ | |
64db4cff | 1004 | if (user || |
a6826048 PM |
1005 | (idle_cpu(cpu) && rcu_scheduler_active && |
1006 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1007 | |
1008 | /* | |
1009 | * Get here if this CPU took its interrupt from user | |
1010 | * mode or from the idle loop, and if this is not a | |
1011 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1012 | * a quiescent state, so note it. |
64db4cff PM |
1013 | * |
1014 | * No memory barrier is required here because both | |
d6714c22 PM |
1015 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1016 | * variables that other CPUs neither access nor modify, | |
1017 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1018 | */ |
1019 | ||
d6714c22 PM |
1020 | rcu_sched_qs(cpu); |
1021 | rcu_bh_qs(cpu); | |
64db4cff PM |
1022 | |
1023 | } else if (!in_softirq()) { | |
1024 | ||
1025 | /* | |
1026 | * Get here if this CPU did not take its interrupt from | |
1027 | * softirq, in other words, if it is not interrupting | |
1028 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1029 | * critical section, so note it. |
64db4cff PM |
1030 | */ |
1031 | ||
d6714c22 | 1032 | rcu_bh_qs(cpu); |
64db4cff | 1033 | } |
f41d911f | 1034 | rcu_preempt_check_callbacks(cpu); |
64db4cff PM |
1035 | raise_softirq(RCU_SOFTIRQ); |
1036 | } | |
1037 | ||
1038 | #ifdef CONFIG_SMP | |
1039 | ||
1040 | /* | |
1041 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1042 | * have not yet encountered a quiescent state, using the function specified. | |
1043 | * Returns 1 if the current grace period ends while scanning (possibly | |
1044 | * because we made it end). | |
1045 | */ | |
1046 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, | |
1047 | int (*f)(struct rcu_data *)) | |
1048 | { | |
1049 | unsigned long bit; | |
1050 | int cpu; | |
1051 | unsigned long flags; | |
1052 | unsigned long mask; | |
1053 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | |
1054 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | |
1055 | ||
1056 | for (; rnp_cur < rnp_end; rnp_cur++) { | |
1057 | mask = 0; | |
1058 | spin_lock_irqsave(&rnp_cur->lock, flags); | |
1059 | if (rsp->completed != lastcomp) { | |
1060 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1061 | return 1; | |
1062 | } | |
1063 | if (rnp_cur->qsmask == 0) { | |
1064 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1065 | continue; | |
1066 | } | |
1067 | cpu = rnp_cur->grplo; | |
1068 | bit = 1; | |
1069 | for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { | |
1070 | if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) | |
1071 | mask |= bit; | |
1072 | } | |
1073 | if (mask != 0 && rsp->completed == lastcomp) { | |
1074 | ||
1075 | /* cpu_quiet_msk() releases rnp_cur->lock. */ | |
1076 | cpu_quiet_msk(mask, rsp, rnp_cur, flags); | |
1077 | continue; | |
1078 | } | |
1079 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | |
1080 | } | |
1081 | return 0; | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Force quiescent states on reluctant CPUs, and also detect which | |
1086 | * CPUs are in dyntick-idle mode. | |
1087 | */ | |
1088 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1089 | { | |
1090 | unsigned long flags; | |
1091 | long lastcomp; | |
64db4cff PM |
1092 | struct rcu_node *rnp = rcu_get_root(rsp); |
1093 | u8 signaled; | |
1094 | ||
1095 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) | |
1096 | return; /* No grace period in progress, nothing to force. */ | |
1097 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | |
1098 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | |
1099 | return; /* Someone else is already on the job. */ | |
1100 | } | |
1101 | if (relaxed && | |
ef631b0c | 1102 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
64db4cff PM |
1103 | goto unlock_ret; /* no emergency and done recently. */ |
1104 | rsp->n_force_qs++; | |
1105 | spin_lock(&rnp->lock); | |
1106 | lastcomp = rsp->completed; | |
1107 | signaled = rsp->signaled; | |
1108 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff PM |
1109 | if (lastcomp == rsp->gpnum) { |
1110 | rsp->n_force_qs_ngp++; | |
1111 | spin_unlock(&rnp->lock); | |
1112 | goto unlock_ret; /* no GP in progress, time updated. */ | |
1113 | } | |
1114 | spin_unlock(&rnp->lock); | |
1115 | switch (signaled) { | |
1116 | case RCU_GP_INIT: | |
1117 | ||
1118 | break; /* grace period still initializing, ignore. */ | |
1119 | ||
1120 | case RCU_SAVE_DYNTICK: | |
1121 | ||
1122 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) | |
1123 | break; /* So gcc recognizes the dead code. */ | |
1124 | ||
1125 | /* Record dyntick-idle state. */ | |
1126 | if (rcu_process_dyntick(rsp, lastcomp, | |
1127 | dyntick_save_progress_counter)) | |
1128 | goto unlock_ret; | |
1129 | ||
1130 | /* Update state, record completion counter. */ | |
1131 | spin_lock(&rnp->lock); | |
1132 | if (lastcomp == rsp->completed) { | |
1133 | rsp->signaled = RCU_FORCE_QS; | |
1134 | dyntick_record_completed(rsp, lastcomp); | |
1135 | } | |
1136 | spin_unlock(&rnp->lock); | |
1137 | break; | |
1138 | ||
1139 | case RCU_FORCE_QS: | |
1140 | ||
1141 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1142 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), | |
1143 | rcu_implicit_dynticks_qs)) | |
1144 | goto unlock_ret; | |
1145 | ||
1146 | /* Leave state in case more forcing is required. */ | |
1147 | ||
1148 | break; | |
1149 | } | |
1150 | unlock_ret: | |
1151 | spin_unlock_irqrestore(&rsp->fqslock, flags); | |
1152 | } | |
1153 | ||
1154 | #else /* #ifdef CONFIG_SMP */ | |
1155 | ||
1156 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1157 | { | |
1158 | set_need_resched(); | |
1159 | } | |
1160 | ||
1161 | #endif /* #else #ifdef CONFIG_SMP */ | |
1162 | ||
1163 | /* | |
1164 | * This does the RCU processing work from softirq context for the | |
1165 | * specified rcu_state and rcu_data structures. This may be called | |
1166 | * only from the CPU to whom the rdp belongs. | |
1167 | */ | |
1168 | static void | |
1169 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1170 | { | |
1171 | unsigned long flags; | |
1172 | ||
2e597558 PM |
1173 | WARN_ON_ONCE(rdp->beenonline == 0); |
1174 | ||
64db4cff PM |
1175 | /* |
1176 | * If an RCU GP has gone long enough, go check for dyntick | |
1177 | * idle CPUs and, if needed, send resched IPIs. | |
1178 | */ | |
ef631b0c | 1179 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1180 | force_quiescent_state(rsp, 1); |
1181 | ||
1182 | /* | |
1183 | * Advance callbacks in response to end of earlier grace | |
1184 | * period that some other CPU ended. | |
1185 | */ | |
1186 | rcu_process_gp_end(rsp, rdp); | |
1187 | ||
1188 | /* Update RCU state based on any recent quiescent states. */ | |
1189 | rcu_check_quiescent_state(rsp, rdp); | |
1190 | ||
1191 | /* Does this CPU require a not-yet-started grace period? */ | |
1192 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1193 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | |
1194 | rcu_start_gp(rsp, flags); /* releases above lock */ | |
1195 | } | |
1196 | ||
1197 | /* If there are callbacks ready, invoke them. */ | |
1198 | rcu_do_batch(rdp); | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Do softirq processing for the current CPU. | |
1203 | */ | |
1204 | static void rcu_process_callbacks(struct softirq_action *unused) | |
1205 | { | |
1206 | /* | |
1207 | * Memory references from any prior RCU read-side critical sections | |
1208 | * executed by the interrupted code must be seen before any RCU | |
1209 | * grace-period manipulations below. | |
1210 | */ | |
1211 | smp_mb(); /* See above block comment. */ | |
1212 | ||
d6714c22 PM |
1213 | __rcu_process_callbacks(&rcu_sched_state, |
1214 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1215 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1216 | rcu_preempt_process_callbacks(); |
64db4cff PM |
1217 | |
1218 | /* | |
1219 | * Memory references from any later RCU read-side critical sections | |
1220 | * executed by the interrupted code must be seen after any RCU | |
1221 | * grace-period manipulations above. | |
1222 | */ | |
1223 | smp_mb(); /* See above block comment. */ | |
1224 | } | |
1225 | ||
1226 | static void | |
1227 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1228 | struct rcu_state *rsp) | |
1229 | { | |
1230 | unsigned long flags; | |
1231 | struct rcu_data *rdp; | |
1232 | ||
1233 | head->func = func; | |
1234 | head->next = NULL; | |
1235 | ||
1236 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1237 | ||
1238 | /* | |
1239 | * Opportunistically note grace-period endings and beginnings. | |
1240 | * Note that we might see a beginning right after we see an | |
1241 | * end, but never vice versa, since this CPU has to pass through | |
1242 | * a quiescent state betweentimes. | |
1243 | */ | |
1244 | local_irq_save(flags); | |
1245 | rdp = rsp->rda[smp_processor_id()]; | |
1246 | rcu_process_gp_end(rsp, rdp); | |
1247 | check_for_new_grace_period(rsp, rdp); | |
1248 | ||
1249 | /* Add the callback to our list. */ | |
1250 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1251 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1252 | ||
1253 | /* Start a new grace period if one not already started. */ | |
1254 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { | |
1255 | unsigned long nestflag; | |
1256 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1257 | ||
1258 | spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1259 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | |
1260 | } | |
1261 | ||
1262 | /* Force the grace period if too many callbacks or too long waiting. */ | |
1263 | if (unlikely(++rdp->qlen > qhimark)) { | |
1264 | rdp->blimit = LONG_MAX; | |
1265 | force_quiescent_state(rsp, 0); | |
ef631b0c | 1266 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
64db4cff PM |
1267 | force_quiescent_state(rsp, 1); |
1268 | local_irq_restore(flags); | |
1269 | } | |
1270 | ||
1271 | /* | |
d6714c22 | 1272 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1273 | */ |
d6714c22 | 1274 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1275 | { |
d6714c22 | 1276 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1277 | } |
d6714c22 | 1278 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1279 | |
1280 | /* | |
1281 | * Queue an RCU for invocation after a quicker grace period. | |
1282 | */ | |
1283 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1284 | { | |
1285 | __call_rcu(head, func, &rcu_bh_state); | |
1286 | } | |
1287 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1288 | ||
1289 | /* | |
1290 | * Check to see if there is any immediate RCU-related work to be done | |
1291 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1292 | * The checks are in order of increasing expense: checks that can be | |
1293 | * carried out against CPU-local state are performed first. However, | |
1294 | * we must check for CPU stalls first, else we might not get a chance. | |
1295 | */ | |
1296 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1297 | { | |
1298 | rdp->n_rcu_pending++; | |
1299 | ||
1300 | /* Check for CPU stalls, if enabled. */ | |
1301 | check_cpu_stall(rsp, rdp); | |
1302 | ||
1303 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
7ba5c840 PM |
1304 | if (rdp->qs_pending) { |
1305 | rdp->n_rp_qs_pending++; | |
64db4cff | 1306 | return 1; |
7ba5c840 | 1307 | } |
64db4cff PM |
1308 | |
1309 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1310 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1311 | rdp->n_rp_cb_ready++; | |
64db4cff | 1312 | return 1; |
7ba5c840 | 1313 | } |
64db4cff PM |
1314 | |
1315 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1316 | if (cpu_needs_another_gp(rsp, rdp)) { |
1317 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1318 | return 1; |
7ba5c840 | 1319 | } |
64db4cff PM |
1320 | |
1321 | /* Has another RCU grace period completed? */ | |
7ba5c840 PM |
1322 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ |
1323 | rdp->n_rp_gp_completed++; | |
64db4cff | 1324 | return 1; |
7ba5c840 | 1325 | } |
64db4cff PM |
1326 | |
1327 | /* Has a new RCU grace period started? */ | |
7ba5c840 PM |
1328 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ |
1329 | rdp->n_rp_gp_started++; | |
64db4cff | 1330 | return 1; |
7ba5c840 | 1331 | } |
64db4cff PM |
1332 | |
1333 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
1334 | if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && | |
7ba5c840 PM |
1335 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
1336 | rdp->n_rp_need_fqs++; | |
64db4cff | 1337 | return 1; |
7ba5c840 | 1338 | } |
64db4cff PM |
1339 | |
1340 | /* nothing to do */ | |
7ba5c840 | 1341 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1342 | return 0; |
1343 | } | |
1344 | ||
1345 | /* | |
1346 | * Check to see if there is any immediate RCU-related work to be done | |
1347 | * by the current CPU, returning 1 if so. This function is part of the | |
1348 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1349 | */ | |
a157229c | 1350 | static int rcu_pending(int cpu) |
64db4cff | 1351 | { |
d6714c22 | 1352 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1353 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1354 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1355 | } |
1356 | ||
1357 | /* | |
1358 | * Check to see if any future RCU-related work will need to be done | |
1359 | * by the current CPU, even if none need be done immediately, returning | |
1360 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1361 | * an exported member of the RCU API. | |
1362 | */ | |
1363 | int rcu_needs_cpu(int cpu) | |
1364 | { | |
1365 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1366 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1367 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1368 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1369 | } |
1370 | ||
1371 | /* | |
27569620 | 1372 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1373 | */ |
27569620 PM |
1374 | static void __init |
1375 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1376 | { |
1377 | unsigned long flags; | |
1378 | int i; | |
27569620 PM |
1379 | struct rcu_data *rdp = rsp->rda[cpu]; |
1380 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1381 | ||
1382 | /* Set up local state, ensuring consistent view of global state. */ | |
1383 | spin_lock_irqsave(&rnp->lock, flags); | |
1384 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | |
1385 | rdp->nxtlist = NULL; | |
1386 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1387 | rdp->nxttail[i] = &rdp->nxtlist; | |
1388 | rdp->qlen = 0; | |
1389 | #ifdef CONFIG_NO_HZ | |
1390 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1391 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1392 | rdp->cpu = cpu; | |
1393 | spin_unlock_irqrestore(&rnp->lock, flags); | |
1394 | } | |
1395 | ||
1396 | /* | |
1397 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1398 | * offline event can be happening at a given time. Note also that we | |
1399 | * can accept some slop in the rsp->completed access due to the fact | |
1400 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1401 | */ |
e4fa4c97 | 1402 | static void __cpuinit |
f41d911f | 1403 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
1404 | { |
1405 | unsigned long flags; | |
64db4cff PM |
1406 | long lastcomp; |
1407 | unsigned long mask; | |
1408 | struct rcu_data *rdp = rsp->rda[cpu]; | |
1409 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1410 | ||
1411 | /* Set up local state, ensuring consistent view of global state. */ | |
1412 | spin_lock_irqsave(&rnp->lock, flags); | |
1413 | lastcomp = rsp->completed; | |
1414 | rdp->completed = lastcomp; | |
1415 | rdp->gpnum = lastcomp; | |
1416 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ | |
1417 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
1418 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 1419 | rdp->preemptable = preemptable; |
64db4cff | 1420 | rdp->passed_quiesc_completed = lastcomp - 1; |
64db4cff | 1421 | rdp->blimit = blimit; |
64db4cff PM |
1422 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1423 | ||
1424 | /* | |
1425 | * A new grace period might start here. If so, we won't be part | |
1426 | * of it, but that is OK, as we are currently in a quiescent state. | |
1427 | */ | |
1428 | ||
1429 | /* Exclude any attempts to start a new GP on large systems. */ | |
1430 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | |
1431 | ||
1432 | /* Add CPU to rcu_node bitmasks. */ | |
1433 | rnp = rdp->mynode; | |
1434 | mask = rdp->grpmask; | |
1435 | do { | |
1436 | /* Exclude any attempts to start a new GP on small systems. */ | |
1437 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1438 | rnp->qsmaskinit |= mask; | |
1439 | mask = rnp->grpmask; | |
1440 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | |
1441 | rnp = rnp->parent; | |
1442 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1443 | ||
e7d8842e | 1444 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
1445 | } |
1446 | ||
1447 | static void __cpuinit rcu_online_cpu(int cpu) | |
1448 | { | |
f41d911f PM |
1449 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1450 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1451 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1452 | } |
1453 | ||
1454 | /* | |
f41d911f | 1455 | * Handle CPU online/offline notification events. |
64db4cff | 1456 | */ |
2e597558 PM |
1457 | int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1458 | unsigned long action, void *hcpu) | |
64db4cff PM |
1459 | { |
1460 | long cpu = (long)hcpu; | |
1461 | ||
1462 | switch (action) { | |
1463 | case CPU_UP_PREPARE: | |
1464 | case CPU_UP_PREPARE_FROZEN: | |
1465 | rcu_online_cpu(cpu); | |
1466 | break; | |
1467 | case CPU_DEAD: | |
1468 | case CPU_DEAD_FROZEN: | |
1469 | case CPU_UP_CANCELED: | |
1470 | case CPU_UP_CANCELED_FROZEN: | |
1471 | rcu_offline_cpu(cpu); | |
1472 | break; | |
1473 | default: | |
1474 | break; | |
1475 | } | |
1476 | return NOTIFY_OK; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Compute the per-level fanout, either using the exact fanout specified | |
1481 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
1482 | */ | |
1483 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
1484 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1485 | { | |
1486 | int i; | |
1487 | ||
1488 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | |
1489 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | |
1490 | } | |
1491 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1492 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
1493 | { | |
1494 | int ccur; | |
1495 | int cprv; | |
1496 | int i; | |
1497 | ||
1498 | cprv = NR_CPUS; | |
1499 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1500 | ccur = rsp->levelcnt[i]; | |
1501 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
1502 | cprv = ccur; | |
1503 | } | |
1504 | } | |
1505 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
1506 | ||
1507 | /* | |
1508 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
1509 | */ | |
1510 | static void __init rcu_init_one(struct rcu_state *rsp) | |
1511 | { | |
1512 | int cpustride = 1; | |
1513 | int i; | |
1514 | int j; | |
1515 | struct rcu_node *rnp; | |
1516 | ||
1517 | /* Initialize the level-tracking arrays. */ | |
1518 | ||
1519 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
1520 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
1521 | rcu_init_levelspread(rsp); | |
1522 | ||
1523 | /* Initialize the elements themselves, starting from the leaves. */ | |
1524 | ||
1525 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
1526 | cpustride *= rsp->levelspread[i]; | |
1527 | rnp = rsp->level[i]; | |
1528 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1529 | spin_lock_init(&rnp->lock); | |
f41d911f | 1530 | rnp->gpnum = 0; |
64db4cff PM |
1531 | rnp->qsmask = 0; |
1532 | rnp->qsmaskinit = 0; | |
1533 | rnp->grplo = j * cpustride; | |
1534 | rnp->grphi = (j + 1) * cpustride - 1; | |
1535 | if (rnp->grphi >= NR_CPUS) | |
1536 | rnp->grphi = NR_CPUS - 1; | |
1537 | if (i == 0) { | |
1538 | rnp->grpnum = 0; | |
1539 | rnp->grpmask = 0; | |
1540 | rnp->parent = NULL; | |
1541 | } else { | |
1542 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
1543 | rnp->grpmask = 1UL << rnp->grpnum; | |
1544 | rnp->parent = rsp->level[i - 1] + | |
1545 | j / rsp->levelspread[i - 1]; | |
1546 | } | |
1547 | rnp->level = i; | |
f41d911f PM |
1548 | INIT_LIST_HEAD(&rnp->blocked_tasks[0]); |
1549 | INIT_LIST_HEAD(&rnp->blocked_tasks[1]); | |
64db4cff PM |
1550 | } |
1551 | } | |
1552 | } | |
1553 | ||
1554 | /* | |
f41d911f PM |
1555 | * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used |
1556 | * nowhere else! Assigns leaf node pointers into each CPU's rcu_data | |
1557 | * structure. | |
64db4cff | 1558 | */ |
65cf8f86 | 1559 | #define RCU_INIT_FLAVOR(rsp, rcu_data) \ |
64db4cff | 1560 | do { \ |
65cf8f86 | 1561 | rcu_init_one(rsp); \ |
64db4cff PM |
1562 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
1563 | j = 0; \ | |
1564 | for_each_possible_cpu(i) { \ | |
1565 | if (i > rnp[j].grphi) \ | |
1566 | j++; \ | |
1567 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | |
1568 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | |
65cf8f86 | 1569 | rcu_boot_init_percpu_data(i, rsp); \ |
64db4cff PM |
1570 | } \ |
1571 | } while (0) | |
1572 | ||
f41d911f PM |
1573 | #ifdef CONFIG_TREE_PREEMPT_RCU |
1574 | ||
1575 | void __init __rcu_init_preempt(void) | |
1576 | { | |
1577 | int i; /* All used by RCU_INIT_FLAVOR(). */ | |
1578 | int j; | |
1579 | struct rcu_node *rnp; | |
1580 | ||
1581 | RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); | |
1582 | } | |
1583 | ||
1584 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1585 | ||
1586 | void __init __rcu_init_preempt(void) | |
1587 | { | |
1588 | } | |
1589 | ||
1590 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
64db4cff PM |
1591 | |
1592 | void __init __rcu_init(void) | |
1593 | { | |
f41d911f | 1594 | int i; /* All used by RCU_INIT_FLAVOR(). */ |
64db4cff PM |
1595 | int j; |
1596 | struct rcu_node *rnp; | |
1597 | ||
f41d911f | 1598 | rcu_bootup_announce(); |
64db4cff PM |
1599 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
1600 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | |
1601 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
65cf8f86 PM |
1602 | RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); |
1603 | RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); | |
f41d911f | 1604 | __rcu_init_preempt(); |
2e597558 | 1605 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
64db4cff PM |
1606 | } |
1607 | ||
1608 | module_param(blimit, int, 0); | |
1609 | module_param(qhimark, int, 0); | |
1610 | module_param(qlowmark, int, 0); |