diff options
author | Ingo Molnar <mingo@elte.hu> | 2009-01-12 05:32:03 -0500 |
---|---|---|
committer | Ingo Molnar <mingo@elte.hu> | 2009-01-12 05:32:03 -0500 |
commit | e3ee1e123183ca9847e74b7b8e2694c9e3b817a6 (patch) | |
tree | 652a84674ed05eaa46a813de2223af0bd0168a5a /kernel/rcutree.c | |
parent | 5762ba1873b0bb9faa631aaa02f533c2b9837f82 (diff) | |
parent | c59765042f53a79a7a65585042ff463b69cb248c (diff) |
Merge commit 'v2.6.29-rc1' into timers/hrtimers
Conflicts:
kernel/time/tick-common.c
Diffstat (limited to 'kernel/rcutree.c')
-rw-r--r-- | kernel/rcutree.c | 1532 |
1 files changed, 1532 insertions, 0 deletions
diff --git a/kernel/rcutree.c b/kernel/rcutree.c new file mode 100644 index 000000000000..f2d8638e6c60 --- /dev/null +++ b/kernel/rcutree.c | |||
@@ -0,0 +1,1532 @@ | |||
1 | /* | ||
2 | * Read-Copy Update mechanism for mutual exclusion | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify | ||
5 | * it under the terms of the GNU General Public License as published by | ||
6 | * the Free Software Foundation; either version 2 of the License, or | ||
7 | * (at your option) any later version. | ||
8 | * | ||
9 | * This program is distributed in the hope that it will be useful, | ||
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
12 | * GNU General Public License for more details. | ||
13 | * | ||
14 | * You should have received a copy of the GNU General Public License | ||
15 | * along with this program; if not, write to the Free Software | ||
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | ||
17 | * | ||
18 | * Copyright IBM Corporation, 2008 | ||
19 | * | ||
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | ||
21 | * Manfred Spraul <manfred@colorfullife.com> | ||
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | ||
23 | * | ||
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | ||
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | ||
26 | * | ||
27 | * For detailed explanation of Read-Copy Update mechanism see - | ||
28 | * Documentation/RCU | ||
29 | */ | ||
30 | #include <linux/types.h> | ||
31 | #include <linux/kernel.h> | ||
32 | #include <linux/init.h> | ||
33 | #include <linux/spinlock.h> | ||
34 | #include <linux/smp.h> | ||
35 | #include <linux/rcupdate.h> | ||
36 | #include <linux/interrupt.h> | ||
37 | #include <linux/sched.h> | ||
38 | #include <asm/atomic.h> | ||
39 | #include <linux/bitops.h> | ||
40 | #include <linux/module.h> | ||
41 | #include <linux/completion.h> | ||
42 | #include <linux/moduleparam.h> | ||
43 | #include <linux/percpu.h> | ||
44 | #include <linux/notifier.h> | ||
45 | #include <linux/cpu.h> | ||
46 | #include <linux/mutex.h> | ||
47 | #include <linux/time.h> | ||
48 | |||
49 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | ||
50 | static struct lock_class_key rcu_lock_key; | ||
51 | struct lockdep_map rcu_lock_map = | ||
52 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | ||
53 | EXPORT_SYMBOL_GPL(rcu_lock_map); | ||
54 | #endif | ||
55 | |||
56 | /* Data structures. */ | ||
57 | |||
58 | #define RCU_STATE_INITIALIZER(name) { \ | ||
59 | .level = { &name.node[0] }, \ | ||
60 | .levelcnt = { \ | ||
61 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | ||
62 | NUM_RCU_LVL_1, \ | ||
63 | NUM_RCU_LVL_2, \ | ||
64 | NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \ | ||
65 | }, \ | ||
66 | .signaled = RCU_SIGNAL_INIT, \ | ||
67 | .gpnum = -300, \ | ||
68 | .completed = -300, \ | ||
69 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ | ||
70 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ | ||
71 | .n_force_qs = 0, \ | ||
72 | .n_force_qs_ngp = 0, \ | ||
73 | } | ||
74 | |||
75 | struct rcu_state rcu_state = RCU_STATE_INITIALIZER(rcu_state); | ||
76 | DEFINE_PER_CPU(struct rcu_data, rcu_data); | ||
77 | |||
78 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); | ||
79 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | ||
80 | |||
81 | #ifdef CONFIG_NO_HZ | ||
82 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | ||
83 | .dynticks_nesting = 1, | ||
84 | .dynticks = 1, | ||
85 | }; | ||
86 | #endif /* #ifdef CONFIG_NO_HZ */ | ||
87 | |||
88 | static int blimit = 10; /* Maximum callbacks per softirq. */ | ||
89 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | ||
90 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | ||
91 | |||
92 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); | ||
93 | |||
94 | /* | ||
95 | * Return the number of RCU batches processed thus far for debug & stats. | ||
96 | */ | ||
97 | long rcu_batches_completed(void) | ||
98 | { | ||
99 | return rcu_state.completed; | ||
100 | } | ||
101 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | ||
102 | |||
103 | /* | ||
104 | * Return the number of RCU BH batches processed thus far for debug & stats. | ||
105 | */ | ||
106 | long rcu_batches_completed_bh(void) | ||
107 | { | ||
108 | return rcu_bh_state.completed; | ||
109 | } | ||
110 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | ||
111 | |||
112 | /* | ||
113 | * Does the CPU have callbacks ready to be invoked? | ||
114 | */ | ||
115 | static int | ||
116 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | ||
117 | { | ||
118 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | ||
119 | } | ||
120 | |||
121 | /* | ||
122 | * Does the current CPU require a yet-as-unscheduled grace period? | ||
123 | */ | ||
124 | static int | ||
125 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | ||
126 | { | ||
127 | /* ACCESS_ONCE() because we are accessing outside of lock. */ | ||
128 | return *rdp->nxttail[RCU_DONE_TAIL] && | ||
129 | ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum); | ||
130 | } | ||
131 | |||
132 | /* | ||
133 | * Return the root node of the specified rcu_state structure. | ||
134 | */ | ||
135 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | ||
136 | { | ||
137 | return &rsp->node[0]; | ||
138 | } | ||
139 | |||
140 | #ifdef CONFIG_SMP | ||
141 | |||
142 | /* | ||
143 | * If the specified CPU is offline, tell the caller that it is in | ||
144 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | ||
145 | * Grace periods can end up waiting on an offline CPU when that | ||
146 | * CPU is in the process of coming online -- it will be added to the | ||
147 | * rcu_node bitmasks before it actually makes it online. The same thing | ||
148 | * can happen while a CPU is in the process of coming online. Because this | ||
149 | * race is quite rare, we check for it after detecting that the grace | ||
150 | * period has been delayed rather than checking each and every CPU | ||
151 | * each and every time we start a new grace period. | ||
152 | */ | ||
153 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | ||
154 | { | ||
155 | /* | ||
156 | * If the CPU is offline, it is in a quiescent state. We can | ||
157 | * trust its state not to change because interrupts are disabled. | ||
158 | */ | ||
159 | if (cpu_is_offline(rdp->cpu)) { | ||
160 | rdp->offline_fqs++; | ||
161 | return 1; | ||
162 | } | ||
163 | |||
164 | /* The CPU is online, so send it a reschedule IPI. */ | ||
165 | if (rdp->cpu != smp_processor_id()) | ||
166 | smp_send_reschedule(rdp->cpu); | ||
167 | else | ||
168 | set_need_resched(); | ||
169 | rdp->resched_ipi++; | ||
170 | return 0; | ||
171 | } | ||
172 | |||
173 | #endif /* #ifdef CONFIG_SMP */ | ||
174 | |||
175 | #ifdef CONFIG_NO_HZ | ||
176 | static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5); | ||
177 | |||
178 | /** | ||
179 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | ||
180 | * | ||
181 | * Enter nohz mode, in other words, -leave- the mode in which RCU | ||
182 | * read-side critical sections can occur. (Though RCU read-side | ||
183 | * critical sections can occur in irq handlers in nohz mode, a possibility | ||
184 | * handled by rcu_irq_enter() and rcu_irq_exit()). | ||
185 | */ | ||
186 | void rcu_enter_nohz(void) | ||
187 | { | ||
188 | unsigned long flags; | ||
189 | struct rcu_dynticks *rdtp; | ||
190 | |||
191 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | ||
192 | local_irq_save(flags); | ||
193 | rdtp = &__get_cpu_var(rcu_dynticks); | ||
194 | rdtp->dynticks++; | ||
195 | rdtp->dynticks_nesting--; | ||
196 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); | ||
197 | local_irq_restore(flags); | ||
198 | } | ||
199 | |||
200 | /* | ||
201 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | ||
202 | * | ||
203 | * Exit nohz mode, in other words, -enter- the mode in which RCU | ||
204 | * read-side critical sections normally occur. | ||
205 | */ | ||
206 | void rcu_exit_nohz(void) | ||
207 | { | ||
208 | unsigned long flags; | ||
209 | struct rcu_dynticks *rdtp; | ||
210 | |||
211 | local_irq_save(flags); | ||
212 | rdtp = &__get_cpu_var(rcu_dynticks); | ||
213 | rdtp->dynticks++; | ||
214 | rdtp->dynticks_nesting++; | ||
215 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); | ||
216 | local_irq_restore(flags); | ||
217 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | ||
218 | } | ||
219 | |||
220 | /** | ||
221 | * rcu_nmi_enter - inform RCU of entry to NMI context | ||
222 | * | ||
223 | * If the CPU was idle with dynamic ticks active, and there is no | ||
224 | * irq handler running, this updates rdtp->dynticks_nmi to let the | ||
225 | * RCU grace-period handling know that the CPU is active. | ||
226 | */ | ||
227 | void rcu_nmi_enter(void) | ||
228 | { | ||
229 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | ||
230 | |||
231 | if (rdtp->dynticks & 0x1) | ||
232 | return; | ||
233 | rdtp->dynticks_nmi++; | ||
234 | WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs); | ||
235 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | ||
236 | } | ||
237 | |||
238 | /** | ||
239 | * rcu_nmi_exit - inform RCU of exit from NMI context | ||
240 | * | ||
241 | * If the CPU was idle with dynamic ticks active, and there is no | ||
242 | * irq handler running, this updates rdtp->dynticks_nmi to let the | ||
243 | * RCU grace-period handling know that the CPU is no longer active. | ||
244 | */ | ||
245 | void rcu_nmi_exit(void) | ||
246 | { | ||
247 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | ||
248 | |||
249 | if (rdtp->dynticks & 0x1) | ||
250 | return; | ||
251 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | ||
252 | rdtp->dynticks_nmi++; | ||
253 | WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs); | ||
254 | } | ||
255 | |||
256 | /** | ||
257 | * rcu_irq_enter - inform RCU of entry to hard irq context | ||
258 | * | ||
259 | * If the CPU was idle with dynamic ticks active, this updates the | ||
260 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | ||
261 | */ | ||
262 | void rcu_irq_enter(void) | ||
263 | { | ||
264 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | ||
265 | |||
266 | if (rdtp->dynticks_nesting++) | ||
267 | return; | ||
268 | rdtp->dynticks++; | ||
269 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); | ||
270 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ | ||
271 | } | ||
272 | |||
273 | /** | ||
274 | * rcu_irq_exit - inform RCU of exit from hard irq context | ||
275 | * | ||
276 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | ||
277 | * to put let the RCU handling be aware that the CPU is going back to idle | ||
278 | * with no ticks. | ||
279 | */ | ||
280 | void rcu_irq_exit(void) | ||
281 | { | ||
282 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | ||
283 | |||
284 | if (--rdtp->dynticks_nesting) | ||
285 | return; | ||
286 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ | ||
287 | rdtp->dynticks++; | ||
288 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); | ||
289 | |||
290 | /* If the interrupt queued a callback, get out of dyntick mode. */ | ||
291 | if (__get_cpu_var(rcu_data).nxtlist || | ||
292 | __get_cpu_var(rcu_bh_data).nxtlist) | ||
293 | set_need_resched(); | ||
294 | } | ||
295 | |||
296 | /* | ||
297 | * Record the specified "completed" value, which is later used to validate | ||
298 | * dynticks counter manipulations. Specify "rsp->completed - 1" to | ||
299 | * unconditionally invalidate any future dynticks manipulations (which is | ||
300 | * useful at the beginning of a grace period). | ||
301 | */ | ||
302 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | ||
303 | { | ||
304 | rsp->dynticks_completed = comp; | ||
305 | } | ||
306 | |||
307 | #ifdef CONFIG_SMP | ||
308 | |||
309 | /* | ||
310 | * Recall the previously recorded value of the completion for dynticks. | ||
311 | */ | ||
312 | static long dyntick_recall_completed(struct rcu_state *rsp) | ||
313 | { | ||
314 | return rsp->dynticks_completed; | ||
315 | } | ||
316 | |||
317 | /* | ||
318 | * Snapshot the specified CPU's dynticks counter so that we can later | ||
319 | * credit them with an implicit quiescent state. Return 1 if this CPU | ||
320 | * is already in a quiescent state courtesy of dynticks idle mode. | ||
321 | */ | ||
322 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | ||
323 | { | ||
324 | int ret; | ||
325 | int snap; | ||
326 | int snap_nmi; | ||
327 | |||
328 | snap = rdp->dynticks->dynticks; | ||
329 | snap_nmi = rdp->dynticks->dynticks_nmi; | ||
330 | smp_mb(); /* Order sampling of snap with end of grace period. */ | ||
331 | rdp->dynticks_snap = snap; | ||
332 | rdp->dynticks_nmi_snap = snap_nmi; | ||
333 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); | ||
334 | if (ret) | ||
335 | rdp->dynticks_fqs++; | ||
336 | return ret; | ||
337 | } | ||
338 | |||
339 | /* | ||
340 | * Return true if the specified CPU has passed through a quiescent | ||
341 | * state by virtue of being in or having passed through an dynticks | ||
342 | * idle state since the last call to dyntick_save_progress_counter() | ||
343 | * for this same CPU. | ||
344 | */ | ||
345 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | ||
346 | { | ||
347 | long curr; | ||
348 | long curr_nmi; | ||
349 | long snap; | ||
350 | long snap_nmi; | ||
351 | |||
352 | curr = rdp->dynticks->dynticks; | ||
353 | snap = rdp->dynticks_snap; | ||
354 | curr_nmi = rdp->dynticks->dynticks_nmi; | ||
355 | snap_nmi = rdp->dynticks_nmi_snap; | ||
356 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ | ||
357 | |||
358 | /* | ||
359 | * If the CPU passed through or entered a dynticks idle phase with | ||
360 | * no active irq/NMI handlers, then we can safely pretend that the CPU | ||
361 | * already acknowledged the request to pass through a quiescent | ||
362 | * state. Either way, that CPU cannot possibly be in an RCU | ||
363 | * read-side critical section that started before the beginning | ||
364 | * of the current RCU grace period. | ||
365 | */ | ||
366 | if ((curr != snap || (curr & 0x1) == 0) && | ||
367 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { | ||
368 | rdp->dynticks_fqs++; | ||
369 | return 1; | ||
370 | } | ||
371 | |||
372 | /* Go check for the CPU being offline. */ | ||
373 | return rcu_implicit_offline_qs(rdp); | ||
374 | } | ||
375 | |||
376 | #endif /* #ifdef CONFIG_SMP */ | ||
377 | |||
378 | #else /* #ifdef CONFIG_NO_HZ */ | ||
379 | |||
380 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) | ||
381 | { | ||
382 | } | ||
383 | |||
384 | #ifdef CONFIG_SMP | ||
385 | |||
386 | /* | ||
387 | * If there are no dynticks, then the only way that a CPU can passively | ||
388 | * be in a quiescent state is to be offline. Unlike dynticks idle, which | ||
389 | * is a point in time during the prior (already finished) grace period, | ||
390 | * an offline CPU is always in a quiescent state, and thus can be | ||
391 | * unconditionally applied. So just return the current value of completed. | ||
392 | */ | ||
393 | static long dyntick_recall_completed(struct rcu_state *rsp) | ||
394 | { | ||
395 | return rsp->completed; | ||
396 | } | ||
397 | |||
398 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | ||
399 | { | ||
400 | return 0; | ||
401 | } | ||
402 | |||
403 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | ||
404 | { | ||
405 | return rcu_implicit_offline_qs(rdp); | ||
406 | } | ||
407 | |||
408 | #endif /* #ifdef CONFIG_SMP */ | ||
409 | |||
410 | #endif /* #else #ifdef CONFIG_NO_HZ */ | ||
411 | |||
412 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | ||
413 | |||
414 | static void record_gp_stall_check_time(struct rcu_state *rsp) | ||
415 | { | ||
416 | rsp->gp_start = jiffies; | ||
417 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | ||
418 | } | ||
419 | |||
420 | static void print_other_cpu_stall(struct rcu_state *rsp) | ||
421 | { | ||
422 | int cpu; | ||
423 | long delta; | ||
424 | unsigned long flags; | ||
425 | struct rcu_node *rnp = rcu_get_root(rsp); | ||
426 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | ||
427 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | ||
428 | |||
429 | /* Only let one CPU complain about others per time interval. */ | ||
430 | |||
431 | spin_lock_irqsave(&rnp->lock, flags); | ||
432 | delta = jiffies - rsp->jiffies_stall; | ||
433 | if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { | ||
434 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
435 | return; | ||
436 | } | ||
437 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | ||
438 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
439 | |||
440 | /* OK, time to rat on our buddy... */ | ||
441 | |||
442 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); | ||
443 | for (; rnp_cur < rnp_end; rnp_cur++) { | ||
444 | if (rnp_cur->qsmask == 0) | ||
445 | continue; | ||
446 | for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) | ||
447 | if (rnp_cur->qsmask & (1UL << cpu)) | ||
448 | printk(" %d", rnp_cur->grplo + cpu); | ||
449 | } | ||
450 | printk(" (detected by %d, t=%ld jiffies)\n", | ||
451 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); | ||
452 | force_quiescent_state(rsp, 0); /* Kick them all. */ | ||
453 | } | ||
454 | |||
455 | static void print_cpu_stall(struct rcu_state *rsp) | ||
456 | { | ||
457 | unsigned long flags; | ||
458 | struct rcu_node *rnp = rcu_get_root(rsp); | ||
459 | |||
460 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", | ||
461 | smp_processor_id(), jiffies - rsp->gp_start); | ||
462 | dump_stack(); | ||
463 | spin_lock_irqsave(&rnp->lock, flags); | ||
464 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) | ||
465 | rsp->jiffies_stall = | ||
466 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | ||
467 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
468 | set_need_resched(); /* kick ourselves to get things going. */ | ||
469 | } | ||
470 | |||
471 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | ||
472 | { | ||
473 | long delta; | ||
474 | struct rcu_node *rnp; | ||
475 | |||
476 | delta = jiffies - rsp->jiffies_stall; | ||
477 | rnp = rdp->mynode; | ||
478 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { | ||
479 | |||
480 | /* We haven't checked in, so go dump stack. */ | ||
481 | print_cpu_stall(rsp); | ||
482 | |||
483 | } else if (rsp->gpnum != rsp->completed && | ||
484 | delta >= RCU_STALL_RAT_DELAY) { | ||
485 | |||
486 | /* They had two time units to dump stack, so complain. */ | ||
487 | print_other_cpu_stall(rsp); | ||
488 | } | ||
489 | } | ||
490 | |||
491 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | ||
492 | |||
493 | static void record_gp_stall_check_time(struct rcu_state *rsp) | ||
494 | { | ||
495 | } | ||
496 | |||
497 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | ||
498 | { | ||
499 | } | ||
500 | |||
501 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | ||
502 | |||
503 | /* | ||
504 | * Update CPU-local rcu_data state to record the newly noticed grace period. | ||
505 | * This is used both when we started the grace period and when we notice | ||
506 | * that someone else started the grace period. | ||
507 | */ | ||
508 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) | ||
509 | { | ||
510 | rdp->qs_pending = 1; | ||
511 | rdp->passed_quiesc = 0; | ||
512 | rdp->gpnum = rsp->gpnum; | ||
513 | rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending + | ||
514 | RCU_JIFFIES_TILL_FORCE_QS; | ||
515 | } | ||
516 | |||
517 | /* | ||
518 | * Did someone else start a new RCU grace period start since we last | ||
519 | * checked? Update local state appropriately if so. Must be called | ||
520 | * on the CPU corresponding to rdp. | ||
521 | */ | ||
522 | static int | ||
523 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | ||
524 | { | ||
525 | unsigned long flags; | ||
526 | int ret = 0; | ||
527 | |||
528 | local_irq_save(flags); | ||
529 | if (rdp->gpnum != rsp->gpnum) { | ||
530 | note_new_gpnum(rsp, rdp); | ||
531 | ret = 1; | ||
532 | } | ||
533 | local_irq_restore(flags); | ||
534 | return ret; | ||
535 | } | ||
536 | |||
537 | /* | ||
538 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | ||
539 | * in preparation for detecting the next grace period. The caller must hold | ||
540 | * the root node's ->lock, which is released before return. Hard irqs must | ||
541 | * be disabled. | ||
542 | */ | ||
543 | static void | ||
544 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | ||
545 | __releases(rcu_get_root(rsp)->lock) | ||
546 | { | ||
547 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | ||
548 | struct rcu_node *rnp = rcu_get_root(rsp); | ||
549 | struct rcu_node *rnp_cur; | ||
550 | struct rcu_node *rnp_end; | ||
551 | |||
552 | if (!cpu_needs_another_gp(rsp, rdp)) { | ||
553 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
554 | return; | ||
555 | } | ||
556 | |||
557 | /* Advance to a new grace period and initialize state. */ | ||
558 | rsp->gpnum++; | ||
559 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | ||
560 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | ||
561 | rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending + | ||
562 | RCU_JIFFIES_TILL_FORCE_QS; | ||
563 | record_gp_stall_check_time(rsp); | ||
564 | dyntick_record_completed(rsp, rsp->completed - 1); | ||
565 | note_new_gpnum(rsp, rdp); | ||
566 | |||
567 | /* | ||
568 | * Because we are first, we know that all our callbacks will | ||
569 | * be covered by this upcoming grace period, even the ones | ||
570 | * that were registered arbitrarily recently. | ||
571 | */ | ||
572 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | ||
573 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | ||
574 | |||
575 | /* Special-case the common single-level case. */ | ||
576 | if (NUM_RCU_NODES == 1) { | ||
577 | rnp->qsmask = rnp->qsmaskinit; | ||
578 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ | ||
579 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
580 | return; | ||
581 | } | ||
582 | |||
583 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ | ||
584 | |||
585 | |||
586 | /* Exclude any concurrent CPU-hotplug operations. */ | ||
587 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | ||
588 | |||
589 | /* | ||
590 | * Set the quiescent-state-needed bits in all the non-leaf RCU | ||
591 | * nodes for all currently online CPUs. This operation relies | ||
592 | * on the layout of the hierarchy within the rsp->node[] array. | ||
593 | * Note that other CPUs will access only the leaves of the | ||
594 | * hierarchy, which still indicate that no grace period is in | ||
595 | * progress. In addition, we have excluded CPU-hotplug operations. | ||
596 | * | ||
597 | * We therefore do not need to hold any locks. Any required | ||
598 | * memory barriers will be supplied by the locks guarding the | ||
599 | * leaf rcu_nodes in the hierarchy. | ||
600 | */ | ||
601 | |||
602 | rnp_end = rsp->level[NUM_RCU_LVLS - 1]; | ||
603 | for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) | ||
604 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | ||
605 | |||
606 | /* | ||
607 | * Now set up the leaf nodes. Here we must be careful. First, | ||
608 | * we need to hold the lock in order to exclude other CPUs, which | ||
609 | * might be contending for the leaf nodes' locks. Second, as | ||
610 | * soon as we initialize a given leaf node, its CPUs might run | ||
611 | * up the rest of the hierarchy. We must therefore acquire locks | ||
612 | * for each node that we touch during this stage. (But we still | ||
613 | * are excluding CPU-hotplug operations.) | ||
614 | * | ||
615 | * Note that the grace period cannot complete until we finish | ||
616 | * the initialization process, as there will be at least one | ||
617 | * qsmask bit set in the root node until that time, namely the | ||
618 | * one corresponding to this CPU. | ||
619 | */ | ||
620 | rnp_end = &rsp->node[NUM_RCU_NODES]; | ||
621 | rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | ||
622 | for (; rnp_cur < rnp_end; rnp_cur++) { | ||
623 | spin_lock(&rnp_cur->lock); /* irqs already disabled. */ | ||
624 | rnp_cur->qsmask = rnp_cur->qsmaskinit; | ||
625 | spin_unlock(&rnp_cur->lock); /* irqs already disabled. */ | ||
626 | } | ||
627 | |||
628 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ | ||
629 | spin_unlock_irqrestore(&rsp->onofflock, flags); | ||
630 | } | ||
631 | |||
632 | /* | ||
633 | * Advance this CPU's callbacks, but only if the current grace period | ||
634 | * has ended. This may be called only from the CPU to whom the rdp | ||
635 | * belongs. | ||
636 | */ | ||
637 | static void | ||
638 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | ||
639 | { | ||
640 | long completed_snap; | ||
641 | unsigned long flags; | ||
642 | |||
643 | local_irq_save(flags); | ||
644 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ | ||
645 | |||
646 | /* Did another grace period end? */ | ||
647 | if (rdp->completed != completed_snap) { | ||
648 | |||
649 | /* Advance callbacks. No harm if list empty. */ | ||
650 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | ||
651 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | ||
652 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | ||
653 | |||
654 | /* Remember that we saw this grace-period completion. */ | ||
655 | rdp->completed = completed_snap; | ||
656 | } | ||
657 | local_irq_restore(flags); | ||
658 | } | ||
659 | |||
660 | /* | ||
661 | * Similar to cpu_quiet(), for which it is a helper function. Allows | ||
662 | * a group of CPUs to be quieted at one go, though all the CPUs in the | ||
663 | * group must be represented by the same leaf rcu_node structure. | ||
664 | * That structure's lock must be held upon entry, and it is released | ||
665 | * before return. | ||
666 | */ | ||
667 | static void | ||
668 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, | ||
669 | unsigned long flags) | ||
670 | __releases(rnp->lock) | ||
671 | { | ||
672 | /* Walk up the rcu_node hierarchy. */ | ||
673 | for (;;) { | ||
674 | if (!(rnp->qsmask & mask)) { | ||
675 | |||
676 | /* Our bit has already been cleared, so done. */ | ||
677 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
678 | return; | ||
679 | } | ||
680 | rnp->qsmask &= ~mask; | ||
681 | if (rnp->qsmask != 0) { | ||
682 | |||
683 | /* Other bits still set at this level, so done. */ | ||
684 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
685 | return; | ||
686 | } | ||
687 | mask = rnp->grpmask; | ||
688 | if (rnp->parent == NULL) { | ||
689 | |||
690 | /* No more levels. Exit loop holding root lock. */ | ||
691 | |||
692 | break; | ||
693 | } | ||
694 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
695 | rnp = rnp->parent; | ||
696 | spin_lock_irqsave(&rnp->lock, flags); | ||
697 | } | ||
698 | |||
699 | /* | ||
700 | * Get here if we are the last CPU to pass through a quiescent | ||
701 | * state for this grace period. Clean up and let rcu_start_gp() | ||
702 | * start up the next grace period if one is needed. Note that | ||
703 | * we still hold rnp->lock, as required by rcu_start_gp(), which | ||
704 | * will release it. | ||
705 | */ | ||
706 | rsp->completed = rsp->gpnum; | ||
707 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); | ||
708 | rcu_start_gp(rsp, flags); /* releases rnp->lock. */ | ||
709 | } | ||
710 | |||
711 | /* | ||
712 | * Record a quiescent state for the specified CPU, which must either be | ||
713 | * the current CPU or an offline CPU. The lastcomp argument is used to | ||
714 | * make sure we are still in the grace period of interest. We don't want | ||
715 | * to end the current grace period based on quiescent states detected in | ||
716 | * an earlier grace period! | ||
717 | */ | ||
718 | static void | ||
719 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) | ||
720 | { | ||
721 | unsigned long flags; | ||
722 | unsigned long mask; | ||
723 | struct rcu_node *rnp; | ||
724 | |||
725 | rnp = rdp->mynode; | ||
726 | spin_lock_irqsave(&rnp->lock, flags); | ||
727 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { | ||
728 | |||
729 | /* | ||
730 | * Someone beat us to it for this grace period, so leave. | ||
731 | * The race with GP start is resolved by the fact that we | ||
732 | * hold the leaf rcu_node lock, so that the per-CPU bits | ||
733 | * cannot yet be initialized -- so we would simply find our | ||
734 | * CPU's bit already cleared in cpu_quiet_msk() if this race | ||
735 | * occurred. | ||
736 | */ | ||
737 | rdp->passed_quiesc = 0; /* try again later! */ | ||
738 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
739 | return; | ||
740 | } | ||
741 | mask = rdp->grpmask; | ||
742 | if ((rnp->qsmask & mask) == 0) { | ||
743 | spin_unlock_irqrestore(&rnp->lock, flags); | ||
744 | } else { | ||
745 | rdp->qs_pending = 0; | ||
746 | |||
747 | /* | ||
748 | * This GP can't end until cpu checks in, so all of our | ||
749 | * callbacks can be processed during the next GP. | ||
750 | */ | ||
751 | rdp = rsp->rda[smp_processor_id()]; | ||
752 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | ||
753 | |||
754 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ | ||
755 | } | ||
756 | } | ||
757 | |||
758 | /* | ||
759 | * Check to see if there is a new grace period of which this CPU | ||
760 | * is not yet aware, and if so, set up local rcu_data state for it. | ||
761 | * Otherwise, see if this CPU has just passed through its first | ||
762 | * quiescent state for this grace period, and record that fact if so. | ||
763 | */ | ||
764 | static void | ||
765 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | ||
766 | { | ||
767 | /* If there is now a new grace period, record and return. */ | ||
768 | if (check_for_new_grace_period(rsp, rdp)) | ||
769 | return; | ||
770 | |||
771 | /* | ||
772 | * Does this CPU still need to do its part for current grace period? | ||
773 | * If no, return and let the other CPUs do their part as well. | ||
774 | */ | ||
775 | if (!rdp->qs_pending) | ||
776 | return; | ||
777 | |||
778 | /* | ||
779 | * Was there a quiescent state since the beginning of the grace | ||
780 | * period? If no, then exit and wait for the next call. | ||
781 | */ | ||
782 | if (!rdp->passed_quiesc) | ||
783 | return; | ||
784 | |||
785 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ | ||
786 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | ||
787 | } | ||
788 | |||
789 | #ifdef CONFIG_HOTPLUG_CPU | ||
790 | |||
791 | /* | ||
792 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | ||
793 | * and move all callbacks from the outgoing CPU to the current one. | ||
794 | */ | ||
795 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | ||
796 | { | ||
797 | int i; | ||
798 | unsigned long flags; | ||
799 | long lastcomp; | ||
800 | unsigned long mask; | ||
801 | struct rcu_data *rdp = rsp->rda[cpu]; | ||
802 | struct rcu_data *rdp_me; | ||
803 | struct rcu_node *rnp; | ||
804 | |||
805 | /* Exclude any attempts to start a new grace period. */ | ||
806 | spin_lock_irqsave(&rsp->onofflock, flags); | ||
807 | |||
808 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | ||
809 | rnp = rdp->mynode; | ||
810 | mask = rdp->grpmask; /* rnp->grplo is constant. */ | ||
811 | do { | ||
812 | spin_lock(&rnp->lock); /* irqs already disabled. */ | ||
813 | rnp->qsmaskinit &= ~mask; | ||
814 | if (rnp->qsmaskinit != 0) { | ||
815 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | ||
816 | break; | ||
817 | } | ||
818 | mask = rnp->grpmask; | ||
819 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | ||
820 | rnp = rnp->parent; | ||
821 | } while (rnp != NULL); | ||
822 | lastcomp = rsp->completed; | ||
823 | |||
824 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | ||
825 | |||
826 | /* Being offline is a quiescent state, so go record it. */ | ||
827 | cpu_quiet(cpu, rsp, rdp, lastcomp); | ||
828 | |||
829 | /* | ||
830 | * Move callbacks from the outgoing CPU to the running CPU. | ||
831 | * Note that the outgoing CPU is now quiscent, so it is now | ||
832 | * (uncharacteristically) safe to access it rcu_data structure. | ||
833 | * Note also that we must carefully retain the order of the | ||
834 | * outgoing CPU's callbacks in order for rcu_barrier() to work | ||
835 | * correctly. Finally, note that we start all the callbacks | ||
836 | * afresh, even those that have passed through a grace period | ||
837 | * and are therefore ready to invoke. The theory is that hotplug | ||
838 | * events are rare, and that if they are frequent enough to | ||
839 | * indefinitely delay callbacks, you have far worse things to | ||
840 | * be worrying about. | ||
841 | */ | ||
842 | rdp_me = rsp->rda[smp_processor_id()]; | ||
843 | if (rdp->nxtlist != NULL) { | ||
844 | *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | ||
845 | rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | ||
846 | rdp->nxtlist = NULL; | ||
847 | for (i = 0; i < RCU_NEXT_SIZE; i++) | ||
848 | rdp->nxttail[i] = &rdp->nxtlist; | ||
849 | rdp_me->qlen += rdp->qlen; | ||
850 | rdp->qlen = 0; | ||
851 | } | ||
852 | local_irq_restore(flags); | ||
853 | } | ||
854 | |||
855 | /* | ||
856 | * Remove the specified CPU from the RCU hierarchy and move any pending | ||
857 | * callbacks that it might have to the current CPU. This code assumes | ||
858 | * that at least one CPU in the system will remain running at all times. | ||
859 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | ||
860 | */ | ||
861 | static void rcu_offline_cpu(int cpu) | ||
862 | { | ||
863 | __rcu_offline_cpu(cpu, &rcu_state); | ||
864 | __rcu_offline_cpu(cpu, &rcu_bh_state); | ||
865 | } | ||
866 | |||
867 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | ||
868 | |||
869 | static void rcu_offline_cpu(int cpu) | ||
870 | { | ||
871 | } | ||
872 | |||
873 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | ||
874 | |||
875 | /* | ||
876 | * Invoke any RCU callbacks that have made it to the end of their grace | ||
877 | * period. Thottle as specified by rdp->blimit. | ||
878 | */ | ||
879 | static void rcu_do_batch(struct rcu_data *rdp) | ||
880 | { | ||
881 | unsigned long flags; | ||
882 | struct rcu_head *next, *list, **tail; | ||
883 | int count; | ||
884 | |||
885 | /* If no callbacks are ready, just return.*/ | ||
886 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | ||
887 | return; | ||
888 | |||
889 | /* | ||
890 | * Extract the list of ready callbacks, disabling to prevent | ||
891 | * races with call_rcu() from interrupt handlers. | ||
892 | */ | ||
893 | local_irq_save(flags); | ||
894 | list = rdp->nxtlist; | ||
895 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | ||
896 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | ||
897 | tail = rdp->nxttail[RCU_DONE_TAIL]; | ||
898 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | ||
899 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | ||
900 | rdp->nxttail[count] = &rdp->nxtlist; | ||
901 | local_irq_restore(flags); | ||
902 | |||
903 | /* Invoke callbacks. */ | ||
904 | count = 0; | ||
905 | while (list) { | ||
906 | next = list->next; | ||
907 | prefetch(next); | ||
908 | list->func(list); | ||
909 | list = next; | ||
910 | if (++count >= rdp->blimit) | ||
911 | break; | ||
912 | } | ||
913 | |||
914 | local_irq_save(flags); | ||
915 | |||
916 | /* Update count, and requeue any remaining callbacks. */ | ||
917 | rdp->qlen -= count; | ||
918 | if (list != NULL) { | ||
919 | *tail = rdp->nxtlist; | ||
920 | rdp->nxtlist = list; | ||
921 | for (count = 0; count < RCU_NEXT_SIZE; count++) | ||
922 | if (&rdp->nxtlist == rdp->nxttail[count]) | ||
923 | rdp->nxttail[count] = tail; | ||
924 | else | ||
925 | break; | ||
926 | } | ||
927 | |||
928 | /* Reinstate batch limit if we have worked down the excess. */ | ||
929 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | ||
930 | rdp->blimit = blimit; | ||
931 | |||
932 | local_irq_restore(flags); | ||
933 | |||
934 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | ||
935 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | ||
936 | raise_softirq(RCU_SOFTIRQ); | ||
937 | } | ||
938 | |||
939 | /* | ||
940 | * Check to see if this CPU is in a non-context-switch quiescent state | ||
941 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | ||
942 | * Also schedule the RCU softirq handler. | ||
943 | * | ||
944 | * This function must be called with hardirqs disabled. It is normally | ||
945 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | ||
946 | * false, there is no point in invoking rcu_check_callbacks(). | ||
947 | */ | ||
948 | void rcu_check_callbacks(int cpu, int user) | ||
949 | { | ||
950 | if (user || | ||
951 | (idle_cpu(cpu) && !in_softirq() && | ||
952 | hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | ||
953 | |||
954 | /* | ||
955 | * Get here if this CPU took its interrupt from user | ||
956 | * mode or from the idle loop, and if this is not a | ||
957 | * nested interrupt. In this case, the CPU is in | ||
958 | * a quiescent state, so count it. | ||
959 | * | ||
960 | * No memory barrier is required here because both | ||
961 | * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference | ||
962 | * only CPU-local variables that other CPUs neither | ||
963 | * access nor modify, at least not while the corresponding | ||
964 | * CPU is online. | ||
965 | */ | ||
966 | |||
967 | rcu_qsctr_inc(cpu); | ||
968 | rcu_bh_qsctr_inc(cpu); | ||
969 | |||
970 | } else if (!in_softirq()) { | ||
971 | |||
972 | /* | ||
973 | * Get here if this CPU did not take its interrupt from | ||
974 | * softirq, in other words, if it is not interrupting | ||
975 | * a rcu_bh read-side critical section. This is an _bh | ||
976 | * critical section, so count it. | ||
977 | */ | ||
978 | |||
979 | rcu_bh_qsctr_inc(cpu); | ||
980 | } | ||
981 | raise_softirq(RCU_SOFTIRQ); | ||
982 | } | ||
983 | |||
984 | #ifdef CONFIG_SMP | ||
985 | |||
986 | /* | ||
987 | * Scan the leaf rcu_node structures, processing dyntick state for any that | ||
988 | * have not yet encountered a quiescent state, using the function specified. | ||
989 | * Returns 1 if the current grace period ends while scanning (possibly | ||
990 | * because we made it end). | ||
991 | */ | ||
992 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, | ||
993 | int (*f)(struct rcu_data *)) | ||
994 | { | ||
995 | unsigned long bit; | ||
996 | int cpu; | ||
997 | unsigned long flags; | ||
998 | unsigned long mask; | ||
999 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; | ||
1000 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; | ||
1001 | |||
1002 | for (; rnp_cur < rnp_end; rnp_cur++) { | ||
1003 | mask = 0; | ||
1004 | spin_lock_irqsave(&rnp_cur->lock, flags); | ||
1005 | if (rsp->completed != lastcomp) { | ||
1006 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | ||
1007 | return 1; | ||
1008 | } | ||
1009 | if (rnp_cur->qsmask == 0) { | ||
1010 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | ||
1011 | continue; | ||
1012 | } | ||
1013 | cpu = rnp_cur->grplo; | ||
1014 | bit = 1; | ||
1015 | for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { | ||
1016 | if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) | ||
1017 | mask |= bit; | ||
1018 | } | ||
1019 | if (mask != 0 && rsp->completed == lastcomp) { | ||
1020 | |||
1021 | /* cpu_quiet_msk() releases rnp_cur->lock. */ | ||
1022 | cpu_quiet_msk(mask, rsp, rnp_cur, flags); | ||
1023 | continue; | ||
1024 | } | ||
1025 | spin_unlock_irqrestore(&rnp_cur->lock, flags); | ||
1026 | } | ||
1027 | return 0; | ||
1028 | } | ||
1029 | |||
1030 | /* | ||
1031 | * Force quiescent states on reluctant CPUs, and also detect which | ||
1032 | * CPUs are in dyntick-idle mode. | ||
1033 | */ | ||
1034 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | ||
1035 | { | ||
1036 | unsigned long flags; | ||
1037 | long lastcomp; | ||
1038 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; | ||
1039 | struct rcu_node *rnp = rcu_get_root(rsp); | ||
1040 | u8 signaled; | ||
1041 | |||
1042 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) | ||
1043 | return; /* No grace period in progress, nothing to force. */ | ||
1044 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { | ||
1045 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ | ||
1046 | return; /* Someone else is already on the job. */ | ||
1047 | } | ||
1048 | if (relaxed && | ||
1049 | (long)(rsp->jiffies_force_qs - jiffies) >= 0 && | ||
1050 | (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) >= 0) | ||
1051 | goto unlock_ret; /* no emergency and done recently. */ | ||
1052 | rsp->n_force_qs++; | ||
1053 | spin_lock(&rnp->lock); | ||
1054 | lastcomp = rsp->completed; | ||
1055 | signaled = rsp->signaled; | ||
1056 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | ||
1057 | rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending + | ||
1058 | RCU_JIFFIES_TILL_FORCE_QS; | ||
1059 | if (lastcomp == rsp->gpnum) { | ||
1060 | rsp->n_force_qs_ngp++; | ||
1061 | spin_unlock(&rnp->lock); | ||
1062 | goto unlock_ret; /* no GP in progress, time updated. */ | ||
1063 | } | ||
1064 | spin_unlock(&rnp->lock); | ||
1065 | switch (signaled) { | ||
1066 | case RCU_GP_INIT: | ||
1067 | |||
1068 | break; /* grace period still initializing, ignore. */ | ||
1069 | |||
1070 | case RCU_SAVE_DYNTICK: | ||
1071 | |||
1072 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) | ||
1073 | break; /* So gcc recognizes the dead code. */ | ||
1074 | |||
1075 | /* Record dyntick-idle state. */ | ||
1076 | if (rcu_process_dyntick(rsp, lastcomp, | ||
1077 | dyntick_save_progress_counter)) | ||
1078 | goto unlock_ret; | ||
1079 | |||
1080 | /* Update state, record completion counter. */ | ||
1081 | spin_lock(&rnp->lock); | ||
1082 | if (lastcomp == rsp->completed) { | ||
1083 | rsp->signaled = RCU_FORCE_QS; | ||
1084 | dyntick_record_completed(rsp, lastcomp); | ||
1085 | } | ||
1086 | spin_unlock(&rnp->lock); | ||
1087 | break; | ||
1088 | |||
1089 | case RCU_FORCE_QS: | ||
1090 | |||
1091 | /* Check dyntick-idle state, send IPI to laggarts. */ | ||
1092 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), | ||
1093 | rcu_implicit_dynticks_qs)) | ||
1094 | goto unlock_ret; | ||
1095 | |||
1096 | /* Leave state in case more forcing is required. */ | ||
1097 | |||
1098 | break; | ||
1099 | } | ||
1100 | unlock_ret: | ||
1101 | spin_unlock_irqrestore(&rsp->fqslock, flags); | ||
1102 | } | ||
1103 | |||
1104 | #else /* #ifdef CONFIG_SMP */ | ||
1105 | |||
1106 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | ||
1107 | { | ||
1108 | set_need_resched(); | ||
1109 | } | ||
1110 | |||
1111 | #endif /* #else #ifdef CONFIG_SMP */ | ||
1112 | |||
1113 | /* | ||
1114 | * This does the RCU processing work from softirq context for the | ||
1115 | * specified rcu_state and rcu_data structures. This may be called | ||
1116 | * only from the CPU to whom the rdp belongs. | ||
1117 | */ | ||
1118 | static void | ||
1119 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | ||
1120 | { | ||
1121 | unsigned long flags; | ||
1122 | |||
1123 | /* | ||
1124 | * If an RCU GP has gone long enough, go check for dyntick | ||
1125 | * idle CPUs and, if needed, send resched IPIs. | ||
1126 | */ | ||
1127 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 || | ||
1128 | (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0) | ||
1129 | force_quiescent_state(rsp, 1); | ||
1130 | |||
1131 | /* | ||
1132 | * Advance callbacks in response to end of earlier grace | ||
1133 | * period that some other CPU ended. | ||
1134 | */ | ||
1135 | rcu_process_gp_end(rsp, rdp); | ||
1136 | |||
1137 | /* Update RCU state based on any recent quiescent states. */ | ||
1138 | rcu_check_quiescent_state(rsp, rdp); | ||
1139 | |||
1140 | /* Does this CPU require a not-yet-started grace period? */ | ||
1141 | if (cpu_needs_another_gp(rsp, rdp)) { | ||
1142 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); | ||
1143 | rcu_start_gp(rsp, flags); /* releases above lock */ | ||
1144 | } | ||
1145 | |||
1146 | /* If there are callbacks ready, invoke them. */ | ||
1147 | rcu_do_batch(rdp); | ||
1148 | } | ||
1149 | |||
1150 | /* | ||
1151 | * Do softirq processing for the current CPU. | ||
1152 | */ | ||
1153 | static void rcu_process_callbacks(struct softirq_action *unused) | ||
1154 | { | ||
1155 | /* | ||
1156 | * Memory references from any prior RCU read-side critical sections | ||
1157 | * executed by the interrupted code must be seen before any RCU | ||
1158 | * grace-period manipulations below. | ||
1159 | */ | ||
1160 | smp_mb(); /* See above block comment. */ | ||
1161 | |||
1162 | __rcu_process_callbacks(&rcu_state, &__get_cpu_var(rcu_data)); | ||
1163 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); | ||
1164 | |||
1165 | /* | ||
1166 | * Memory references from any later RCU read-side critical sections | ||
1167 | * executed by the interrupted code must be seen after any RCU | ||
1168 | * grace-period manipulations above. | ||
1169 | */ | ||
1170 | smp_mb(); /* See above block comment. */ | ||
1171 | } | ||
1172 | |||
1173 | static void | ||
1174 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | ||
1175 | struct rcu_state *rsp) | ||
1176 | { | ||
1177 | unsigned long flags; | ||
1178 | struct rcu_data *rdp; | ||
1179 | |||
1180 | head->func = func; | ||
1181 | head->next = NULL; | ||
1182 | |||
1183 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | ||
1184 | |||
1185 | /* | ||
1186 | * Opportunistically note grace-period endings and beginnings. | ||
1187 | * Note that we might see a beginning right after we see an | ||
1188 | * end, but never vice versa, since this CPU has to pass through | ||
1189 | * a quiescent state betweentimes. | ||
1190 | */ | ||
1191 | local_irq_save(flags); | ||
1192 | rdp = rsp->rda[smp_processor_id()]; | ||
1193 | rcu_process_gp_end(rsp, rdp); | ||
1194 | check_for_new_grace_period(rsp, rdp); | ||
1195 | |||
1196 | /* Add the callback to our list. */ | ||
1197 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | ||
1198 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | ||
1199 | |||
1200 | /* Start a new grace period if one not already started. */ | ||
1201 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { | ||
1202 | unsigned long nestflag; | ||
1203 | struct rcu_node *rnp_root = rcu_get_root(rsp); | ||
1204 | |||
1205 | spin_lock_irqsave(&rnp_root->lock, nestflag); | ||
1206 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ | ||
1207 | } | ||
1208 | |||
1209 | /* Force the grace period if too many callbacks or too long waiting. */ | ||
1210 | if (unlikely(++rdp->qlen > qhimark)) { | ||
1211 | rdp->blimit = LONG_MAX; | ||
1212 | force_quiescent_state(rsp, 0); | ||
1213 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 || | ||
1214 | (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0) | ||
1215 | force_quiescent_state(rsp, 1); | ||
1216 | local_irq_restore(flags); | ||
1217 | } | ||
1218 | |||
1219 | /* | ||
1220 | * Queue an RCU callback for invocation after a grace period. | ||
1221 | */ | ||
1222 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | ||
1223 | { | ||
1224 | __call_rcu(head, func, &rcu_state); | ||
1225 | } | ||
1226 | EXPORT_SYMBOL_GPL(call_rcu); | ||
1227 | |||
1228 | /* | ||
1229 | * Queue an RCU for invocation after a quicker grace period. | ||
1230 | */ | ||
1231 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | ||
1232 | { | ||
1233 | __call_rcu(head, func, &rcu_bh_state); | ||
1234 | } | ||
1235 | EXPORT_SYMBOL_GPL(call_rcu_bh); | ||
1236 | |||
1237 | /* | ||
1238 | * Check to see if there is any immediate RCU-related work to be done | ||
1239 | * by the current CPU, for the specified type of RCU, returning 1 if so. | ||
1240 | * The checks are in order of increasing expense: checks that can be | ||
1241 | * carried out against CPU-local state are performed first. However, | ||
1242 | * we must check for CPU stalls first, else we might not get a chance. | ||
1243 | */ | ||
1244 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | ||
1245 | { | ||
1246 | rdp->n_rcu_pending++; | ||
1247 | |||
1248 | /* Check for CPU stalls, if enabled. */ | ||
1249 | check_cpu_stall(rsp, rdp); | ||
1250 | |||
1251 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | ||
1252 | if (rdp->qs_pending) | ||
1253 | return 1; | ||
1254 | |||
1255 | /* Does this CPU have callbacks ready to invoke? */ | ||
1256 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | ||
1257 | return 1; | ||
1258 | |||
1259 | /* Has RCU gone idle with this CPU needing another grace period? */ | ||
1260 | if (cpu_needs_another_gp(rsp, rdp)) | ||
1261 | return 1; | ||
1262 | |||
1263 | /* Has another RCU grace period completed? */ | ||
1264 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) /* outside of lock */ | ||
1265 | return 1; | ||
1266 | |||
1267 | /* Has a new RCU grace period started? */ | ||
1268 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) /* outside of lock */ | ||
1269 | return 1; | ||
1270 | |||
1271 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | ||
1272 | if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && | ||
1273 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 || | ||
1274 | (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0)) | ||
1275 | return 1; | ||
1276 | |||
1277 | /* nothing to do */ | ||
1278 | return 0; | ||
1279 | } | ||
1280 | |||
1281 | /* | ||
1282 | * Check to see if there is any immediate RCU-related work to be done | ||
1283 | * by the current CPU, returning 1 if so. This function is part of the | ||
1284 | * RCU implementation; it is -not- an exported member of the RCU API. | ||
1285 | */ | ||
1286 | int rcu_pending(int cpu) | ||
1287 | { | ||
1288 | return __rcu_pending(&rcu_state, &per_cpu(rcu_data, cpu)) || | ||
1289 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)); | ||
1290 | } | ||
1291 | |||
1292 | /* | ||
1293 | * Check to see if any future RCU-related work will need to be done | ||
1294 | * by the current CPU, even if none need be done immediately, returning | ||
1295 | * 1 if so. This function is part of the RCU implementation; it is -not- | ||
1296 | * an exported member of the RCU API. | ||
1297 | */ | ||
1298 | int rcu_needs_cpu(int cpu) | ||
1299 | { | ||
1300 | /* RCU callbacks either ready or pending? */ | ||
1301 | return per_cpu(rcu_data, cpu).nxtlist || | ||
1302 | per_cpu(rcu_bh_data, cpu).nxtlist; | ||
1303 | } | ||
1304 | |||
1305 | /* | ||
1306 | * Initialize a CPU's per-CPU RCU data. We take this "scorched earth" | ||
1307 | * approach so that we don't have to worry about how long the CPU has | ||
1308 | * been gone, or whether it ever was online previously. We do trust the | ||
1309 | * ->mynode field, as it is constant for a given struct rcu_data and | ||
1310 | * initialized during early boot. | ||
1311 | * | ||
1312 | * Note that only one online or offline event can be happening at a given | ||
1313 | * time. Note also that we can accept some slop in the rsp->completed | ||
1314 | * access due to the fact that this CPU cannot possibly have any RCU | ||
1315 | * callbacks in flight yet. | ||
1316 | */ | ||
1317 | static void | ||
1318 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) | ||
1319 | { | ||
1320 | unsigned long flags; | ||
1321 | int i; | ||
1322 | long lastcomp; | ||
1323 | unsigned long mask; | ||
1324 | struct rcu_data *rdp = rsp->rda[cpu]; | ||
1325 | struct rcu_node *rnp = rcu_get_root(rsp); | ||
1326 | |||
1327 | /* Set up local state, ensuring consistent view of global state. */ | ||
1328 | spin_lock_irqsave(&rnp->lock, flags); | ||
1329 | lastcomp = rsp->completed; | ||
1330 | rdp->completed = lastcomp; | ||
1331 | rdp->gpnum = lastcomp; | ||
1332 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ | ||
1333 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | ||
1334 | rdp->beenonline = 1; /* We have now been online. */ | ||
1335 | rdp->passed_quiesc_completed = lastcomp - 1; | ||
1336 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); | ||
1337 | rdp->nxtlist = NULL; | ||
1338 | for (i = 0; i < RCU_NEXT_SIZE; i++) | ||
1339 | rdp->nxttail[i] = &rdp->nxtlist; | ||
1340 | rdp->qlen = 0; | ||
1341 | rdp->blimit = blimit; | ||
1342 | #ifdef CONFIG_NO_HZ | ||
1343 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | ||
1344 | #endif /* #ifdef CONFIG_NO_HZ */ | ||
1345 | rdp->cpu = cpu; | ||
1346 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | ||
1347 | |||
1348 | /* | ||
1349 | * A new grace period might start here. If so, we won't be part | ||
1350 | * of it, but that is OK, as we are currently in a quiescent state. | ||
1351 | */ | ||
1352 | |||
1353 | /* Exclude any attempts to start a new GP on large systems. */ | ||
1354 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ | ||
1355 | |||
1356 | /* Add CPU to rcu_node bitmasks. */ | ||
1357 | rnp = rdp->mynode; | ||
1358 | mask = rdp->grpmask; | ||
1359 | do { | ||
1360 | /* Exclude any attempts to start a new GP on small systems. */ | ||
1361 | spin_lock(&rnp->lock); /* irqs already disabled. */ | ||
1362 | rnp->qsmaskinit |= mask; | ||
1363 | mask = rnp->grpmask; | ||
1364 | spin_unlock(&rnp->lock); /* irqs already disabled. */ | ||
1365 | rnp = rnp->parent; | ||
1366 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | ||
1367 | |||
1368 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | ||
1369 | |||
1370 | /* | ||
1371 | * A new grace period might start here. If so, we will be part of | ||
1372 | * it, and its gpnum will be greater than ours, so we will | ||
1373 | * participate. It is also possible for the gpnum to have been | ||
1374 | * incremented before this function was called, and the bitmasks | ||
1375 | * to not be filled out until now, in which case we will also | ||
1376 | * participate due to our gpnum being behind. | ||
1377 | */ | ||
1378 | |||
1379 | /* Since it is coming online, the CPU is in a quiescent state. */ | ||
1380 | cpu_quiet(cpu, rsp, rdp, lastcomp); | ||
1381 | local_irq_restore(flags); | ||
1382 | } | ||
1383 | |||
1384 | static void __cpuinit rcu_online_cpu(int cpu) | ||
1385 | { | ||
1386 | rcu_init_percpu_data(cpu, &rcu_state); | ||
1387 | rcu_init_percpu_data(cpu, &rcu_bh_state); | ||
1388 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); | ||
1389 | } | ||
1390 | |||
1391 | /* | ||
1392 | * Handle CPU online/offline notifcation events. | ||
1393 | */ | ||
1394 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, | ||
1395 | unsigned long action, void *hcpu) | ||
1396 | { | ||
1397 | long cpu = (long)hcpu; | ||
1398 | |||
1399 | switch (action) { | ||
1400 | case CPU_UP_PREPARE: | ||
1401 | case CPU_UP_PREPARE_FROZEN: | ||
1402 | rcu_online_cpu(cpu); | ||
1403 | break; | ||
1404 | case CPU_DEAD: | ||
1405 | case CPU_DEAD_FROZEN: | ||
1406 | case CPU_UP_CANCELED: | ||
1407 | case CPU_UP_CANCELED_FROZEN: | ||
1408 | rcu_offline_cpu(cpu); | ||
1409 | break; | ||
1410 | default: | ||
1411 | break; | ||
1412 | } | ||
1413 | return NOTIFY_OK; | ||
1414 | } | ||
1415 | |||
1416 | /* | ||
1417 | * Compute the per-level fanout, either using the exact fanout specified | ||
1418 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | ||
1419 | */ | ||
1420 | #ifdef CONFIG_RCU_FANOUT_EXACT | ||
1421 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | ||
1422 | { | ||
1423 | int i; | ||
1424 | |||
1425 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) | ||
1426 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; | ||
1427 | } | ||
1428 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | ||
1429 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | ||
1430 | { | ||
1431 | int ccur; | ||
1432 | int cprv; | ||
1433 | int i; | ||
1434 | |||
1435 | cprv = NR_CPUS; | ||
1436 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | ||
1437 | ccur = rsp->levelcnt[i]; | ||
1438 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | ||
1439 | cprv = ccur; | ||
1440 | } | ||
1441 | } | ||
1442 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | ||
1443 | |||
1444 | /* | ||
1445 | * Helper function for rcu_init() that initializes one rcu_state structure. | ||
1446 | */ | ||
1447 | static void __init rcu_init_one(struct rcu_state *rsp) | ||
1448 | { | ||
1449 | int cpustride = 1; | ||
1450 | int i; | ||
1451 | int j; | ||
1452 | struct rcu_node *rnp; | ||
1453 | |||
1454 | /* Initialize the level-tracking arrays. */ | ||
1455 | |||
1456 | for (i = 1; i < NUM_RCU_LVLS; i++) | ||
1457 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | ||
1458 | rcu_init_levelspread(rsp); | ||
1459 | |||
1460 | /* Initialize the elements themselves, starting from the leaves. */ | ||
1461 | |||
1462 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | ||
1463 | cpustride *= rsp->levelspread[i]; | ||
1464 | rnp = rsp->level[i]; | ||
1465 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | ||
1466 | spin_lock_init(&rnp->lock); | ||
1467 | rnp->qsmask = 0; | ||
1468 | rnp->qsmaskinit = 0; | ||
1469 | rnp->grplo = j * cpustride; | ||
1470 | rnp->grphi = (j + 1) * cpustride - 1; | ||
1471 | if (rnp->grphi >= NR_CPUS) | ||
1472 | rnp->grphi = NR_CPUS - 1; | ||
1473 | if (i == 0) { | ||
1474 | rnp->grpnum = 0; | ||
1475 | rnp->grpmask = 0; | ||
1476 | rnp->parent = NULL; | ||
1477 | } else { | ||
1478 | rnp->grpnum = j % rsp->levelspread[i - 1]; | ||
1479 | rnp->grpmask = 1UL << rnp->grpnum; | ||
1480 | rnp->parent = rsp->level[i - 1] + | ||
1481 | j / rsp->levelspread[i - 1]; | ||
1482 | } | ||
1483 | rnp->level = i; | ||
1484 | } | ||
1485 | } | ||
1486 | } | ||
1487 | |||
1488 | /* | ||
1489 | * Helper macro for __rcu_init(). To be used nowhere else! | ||
1490 | * Assigns leaf node pointers into each CPU's rcu_data structure. | ||
1491 | */ | ||
1492 | #define RCU_DATA_PTR_INIT(rsp, rcu_data) \ | ||
1493 | do { \ | ||
1494 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ | ||
1495 | j = 0; \ | ||
1496 | for_each_possible_cpu(i) { \ | ||
1497 | if (i > rnp[j].grphi) \ | ||
1498 | j++; \ | ||
1499 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ | ||
1500 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ | ||
1501 | } \ | ||
1502 | } while (0) | ||
1503 | |||
1504 | static struct notifier_block __cpuinitdata rcu_nb = { | ||
1505 | .notifier_call = rcu_cpu_notify, | ||
1506 | }; | ||
1507 | |||
1508 | void __init __rcu_init(void) | ||
1509 | { | ||
1510 | int i; /* All used by RCU_DATA_PTR_INIT(). */ | ||
1511 | int j; | ||
1512 | struct rcu_node *rnp; | ||
1513 | |||
1514 | printk(KERN_WARNING "Experimental hierarchical RCU implementation.\n"); | ||
1515 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | ||
1516 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); | ||
1517 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | ||
1518 | rcu_init_one(&rcu_state); | ||
1519 | RCU_DATA_PTR_INIT(&rcu_state, rcu_data); | ||
1520 | rcu_init_one(&rcu_bh_state); | ||
1521 | RCU_DATA_PTR_INIT(&rcu_bh_state, rcu_bh_data); | ||
1522 | |||
1523 | for_each_online_cpu(i) | ||
1524 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long)i); | ||
1525 | /* Register notifier for non-boot CPUs */ | ||
1526 | register_cpu_notifier(&rcu_nb); | ||
1527 | printk(KERN_WARNING "Experimental hierarchical RCU init done.\n"); | ||
1528 | } | ||
1529 | |||
1530 | module_param(blimit, int, 0); | ||
1531 | module_param(qhimark, int, 0); | ||
1532 | module_param(qlowmark, int, 0); | ||