Root/
Source at commit b13e7eb172b6f08e5fc22da162bdde5fcde201b5 created 11 years 11 months ago. By Maarten ter Huurne, fbcon: Add 6x10 font | |
---|---|
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 <linux/nmi.h> |
39 | #include <linux/atomic.h> |
40 | #include <linux/bitops.h> |
41 | #include <linux/export.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 | #include <linux/kernel_stat.h> |
50 | #include <linux/wait.h> |
51 | #include <linux/kthread.h> |
52 | #include <linux/prefetch.h> |
53 | |
54 | #include "rcutree.h" |
55 | #include <trace/events/rcu.h> |
56 | |
57 | #include "rcu.h" |
58 | |
59 | /* Data structures. */ |
60 | |
61 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
62 | |
63 | #define RCU_STATE_INITIALIZER(structname) { \ |
64 | .level = { &structname##_state.node[0] }, \ |
65 | .levelcnt = { \ |
66 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ |
67 | NUM_RCU_LVL_1, \ |
68 | NUM_RCU_LVL_2, \ |
69 | NUM_RCU_LVL_3, \ |
70 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ |
71 | }, \ |
72 | .fqs_state = RCU_GP_IDLE, \ |
73 | .gpnum = -300, \ |
74 | .completed = -300, \ |
75 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
76 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ |
77 | .n_force_qs = 0, \ |
78 | .n_force_qs_ngp = 0, \ |
79 | .name = #structname, \ |
80 | } |
81 | |
82 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
83 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
84 | |
85 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
86 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
87 | |
88 | static struct rcu_state *rcu_state; |
89 | |
90 | /* |
91 | * The rcu_scheduler_active variable transitions from zero to one just |
92 | * before the first task is spawned. So when this variable is zero, RCU |
93 | * can assume that there is but one task, allowing RCU to (for example) |
94 | * optimized synchronize_sched() to a simple barrier(). When this variable |
95 | * is one, RCU must actually do all the hard work required to detect real |
96 | * grace periods. This variable is also used to suppress boot-time false |
97 | * positives from lockdep-RCU error checking. |
98 | */ |
99 | int rcu_scheduler_active __read_mostly; |
100 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); |
101 | |
102 | /* |
103 | * The rcu_scheduler_fully_active variable transitions from zero to one |
104 | * during the early_initcall() processing, which is after the scheduler |
105 | * is capable of creating new tasks. So RCU processing (for example, |
106 | * creating tasks for RCU priority boosting) must be delayed until after |
107 | * rcu_scheduler_fully_active transitions from zero to one. We also |
108 | * currently delay invocation of any RCU callbacks until after this point. |
109 | * |
110 | * It might later prove better for people registering RCU callbacks during |
111 | * early boot to take responsibility for these callbacks, but one step at |
112 | * a time. |
113 | */ |
114 | static int rcu_scheduler_fully_active __read_mostly; |
115 | |
116 | #ifdef CONFIG_RCU_BOOST |
117 | |
118 | /* |
119 | * Control variables for per-CPU and per-rcu_node kthreads. These |
120 | * handle all flavors of RCU. |
121 | */ |
122 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); |
123 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
124 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
125 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
126 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
127 | |
128 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
129 | |
130 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
131 | static void invoke_rcu_core(void); |
132 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); |
133 | |
134 | /* |
135 | * Track the rcutorture test sequence number and the update version |
136 | * number within a given test. The rcutorture_testseq is incremented |
137 | * on every rcutorture module load and unload, so has an odd value |
138 | * when a test is running. The rcutorture_vernum is set to zero |
139 | * when rcutorture starts and is incremented on each rcutorture update. |
140 | * These variables enable correlating rcutorture output with the |
141 | * RCU tracing information. |
142 | */ |
143 | unsigned long rcutorture_testseq; |
144 | unsigned long rcutorture_vernum; |
145 | |
146 | /* |
147 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s |
148 | * permit this function to be invoked without holding the root rcu_node |
149 | * structure's ->lock, but of course results can be subject to change. |
150 | */ |
151 | static int rcu_gp_in_progress(struct rcu_state *rsp) |
152 | { |
153 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); |
154 | } |
155 | |
156 | /* |
157 | * Note a quiescent state. Because we do not need to know |
158 | * how many quiescent states passed, just if there was at least |
159 | * one since the start of the grace period, this just sets a flag. |
160 | * The caller must have disabled preemption. |
161 | */ |
162 | void rcu_sched_qs(int cpu) |
163 | { |
164 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
165 | |
166 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
167 | barrier(); |
168 | if (rdp->passed_quiesce == 0) |
169 | trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); |
170 | rdp->passed_quiesce = 1; |
171 | } |
172 | |
173 | void rcu_bh_qs(int cpu) |
174 | { |
175 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
176 | |
177 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
178 | barrier(); |
179 | if (rdp->passed_quiesce == 0) |
180 | trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); |
181 | rdp->passed_quiesce = 1; |
182 | } |
183 | |
184 | /* |
185 | * Note a context switch. This is a quiescent state for RCU-sched, |
186 | * and requires special handling for preemptible RCU. |
187 | * The caller must have disabled preemption. |
188 | */ |
189 | void rcu_note_context_switch(int cpu) |
190 | { |
191 | trace_rcu_utilization("Start context switch"); |
192 | rcu_sched_qs(cpu); |
193 | rcu_preempt_note_context_switch(cpu); |
194 | trace_rcu_utilization("End context switch"); |
195 | } |
196 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
197 | |
198 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
199 | .dynticks_nesting = DYNTICK_TASK_NESTING, |
200 | .dynticks = ATOMIC_INIT(1), |
201 | }; |
202 | |
203 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
204 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
205 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ |
206 | |
207 | module_param(blimit, int, 0); |
208 | module_param(qhimark, int, 0); |
209 | module_param(qlowmark, int, 0); |
210 | |
211 | int rcu_cpu_stall_suppress __read_mostly; |
212 | module_param(rcu_cpu_stall_suppress, int, 0644); |
213 | |
214 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
215 | static int rcu_pending(int cpu); |
216 | |
217 | /* |
218 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
219 | */ |
220 | long rcu_batches_completed_sched(void) |
221 | { |
222 | return rcu_sched_state.completed; |
223 | } |
224 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
225 | |
226 | /* |
227 | * Return the number of RCU BH batches processed thus far for debug & stats. |
228 | */ |
229 | long rcu_batches_completed_bh(void) |
230 | { |
231 | return rcu_bh_state.completed; |
232 | } |
233 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
234 | |
235 | /* |
236 | * Force a quiescent state for RCU BH. |
237 | */ |
238 | void rcu_bh_force_quiescent_state(void) |
239 | { |
240 | force_quiescent_state(&rcu_bh_state, 0); |
241 | } |
242 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); |
243 | |
244 | /* |
245 | * Record the number of times rcutorture tests have been initiated and |
246 | * terminated. This information allows the debugfs tracing stats to be |
247 | * correlated to the rcutorture messages, even when the rcutorture module |
248 | * is being repeatedly loaded and unloaded. In other words, we cannot |
249 | * store this state in rcutorture itself. |
250 | */ |
251 | void rcutorture_record_test_transition(void) |
252 | { |
253 | rcutorture_testseq++; |
254 | rcutorture_vernum = 0; |
255 | } |
256 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); |
257 | |
258 | /* |
259 | * Record the number of writer passes through the current rcutorture test. |
260 | * This is also used to correlate debugfs tracing stats with the rcutorture |
261 | * messages. |
262 | */ |
263 | void rcutorture_record_progress(unsigned long vernum) |
264 | { |
265 | rcutorture_vernum++; |
266 | } |
267 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); |
268 | |
269 | /* |
270 | * Force a quiescent state for RCU-sched. |
271 | */ |
272 | void rcu_sched_force_quiescent_state(void) |
273 | { |
274 | force_quiescent_state(&rcu_sched_state, 0); |
275 | } |
276 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); |
277 | |
278 | /* |
279 | * Does the CPU have callbacks ready to be invoked? |
280 | */ |
281 | static int |
282 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) |
283 | { |
284 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; |
285 | } |
286 | |
287 | /* |
288 | * Does the current CPU require a yet-as-unscheduled grace period? |
289 | */ |
290 | static int |
291 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
292 | { |
293 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
294 | } |
295 | |
296 | /* |
297 | * Return the root node of the specified rcu_state structure. |
298 | */ |
299 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) |
300 | { |
301 | return &rsp->node[0]; |
302 | } |
303 | |
304 | #ifdef CONFIG_SMP |
305 | |
306 | /* |
307 | * If the specified CPU is offline, tell the caller that it is in |
308 | * a quiescent state. Otherwise, whack it with a reschedule IPI. |
309 | * Grace periods can end up waiting on an offline CPU when that |
310 | * CPU is in the process of coming online -- it will be added to the |
311 | * rcu_node bitmasks before it actually makes it online. The same thing |
312 | * can happen while a CPU is in the process of coming online. Because this |
313 | * race is quite rare, we check for it after detecting that the grace |
314 | * period has been delayed rather than checking each and every CPU |
315 | * each and every time we start a new grace period. |
316 | */ |
317 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) |
318 | { |
319 | /* |
320 | * If the CPU is offline, it is in a quiescent state. We can |
321 | * trust its state not to change because interrupts are disabled. |
322 | */ |
323 | if (cpu_is_offline(rdp->cpu)) { |
324 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
325 | rdp->offline_fqs++; |
326 | return 1; |
327 | } |
328 | |
329 | /* |
330 | * The CPU is online, so send it a reschedule IPI. This forces |
331 | * it through the scheduler, and (inefficiently) also handles cases |
332 | * where idle loops fail to inform RCU about the CPU being idle. |
333 | */ |
334 | if (rdp->cpu != smp_processor_id()) |
335 | smp_send_reschedule(rdp->cpu); |
336 | else |
337 | set_need_resched(); |
338 | rdp->resched_ipi++; |
339 | return 0; |
340 | } |
341 | |
342 | #endif /* #ifdef CONFIG_SMP */ |
343 | |
344 | /* |
345 | * rcu_idle_enter_common - inform RCU that current CPU is moving towards idle |
346 | * |
347 | * If the new value of the ->dynticks_nesting counter now is zero, |
348 | * we really have entered idle, and must do the appropriate accounting. |
349 | * The caller must have disabled interrupts. |
350 | */ |
351 | static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval) |
352 | { |
353 | trace_rcu_dyntick("Start", oldval, 0); |
354 | if (!is_idle_task(current)) { |
355 | struct task_struct *idle = idle_task(smp_processor_id()); |
356 | |
357 | trace_rcu_dyntick("Error on entry: not idle task", oldval, 0); |
358 | ftrace_dump(DUMP_ALL); |
359 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
360 | current->pid, current->comm, |
361 | idle->pid, idle->comm); /* must be idle task! */ |
362 | } |
363 | rcu_prepare_for_idle(smp_processor_id()); |
364 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
365 | smp_mb__before_atomic_inc(); /* See above. */ |
366 | atomic_inc(&rdtp->dynticks); |
367 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ |
368 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); |
369 | } |
370 | |
371 | /** |
372 | * rcu_idle_enter - inform RCU that current CPU is entering idle |
373 | * |
374 | * Enter idle mode, in other words, -leave- the mode in which RCU |
375 | * read-side critical sections can occur. (Though RCU read-side |
376 | * critical sections can occur in irq handlers in idle, a possibility |
377 | * handled by irq_enter() and irq_exit().) |
378 | * |
379 | * We crowbar the ->dynticks_nesting field to zero to allow for |
380 | * the possibility of usermode upcalls having messed up our count |
381 | * of interrupt nesting level during the prior busy period. |
382 | */ |
383 | void rcu_idle_enter(void) |
384 | { |
385 | unsigned long flags; |
386 | long long oldval; |
387 | struct rcu_dynticks *rdtp; |
388 | |
389 | local_irq_save(flags); |
390 | rdtp = &__get_cpu_var(rcu_dynticks); |
391 | oldval = rdtp->dynticks_nesting; |
392 | rdtp->dynticks_nesting = 0; |
393 | rcu_idle_enter_common(rdtp, oldval); |
394 | local_irq_restore(flags); |
395 | } |
396 | |
397 | /** |
398 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle |
399 | * |
400 | * Exit from an interrupt handler, which might possibly result in entering |
401 | * idle mode, in other words, leaving the mode in which read-side critical |
402 | * sections can occur. |
403 | * |
404 | * This code assumes that the idle loop never does anything that might |
405 | * result in unbalanced calls to irq_enter() and irq_exit(). If your |
406 | * architecture violates this assumption, RCU will give you what you |
407 | * deserve, good and hard. But very infrequently and irreproducibly. |
408 | * |
409 | * Use things like work queues to work around this limitation. |
410 | * |
411 | * You have been warned. |
412 | */ |
413 | void rcu_irq_exit(void) |
414 | { |
415 | unsigned long flags; |
416 | long long oldval; |
417 | struct rcu_dynticks *rdtp; |
418 | |
419 | local_irq_save(flags); |
420 | rdtp = &__get_cpu_var(rcu_dynticks); |
421 | oldval = rdtp->dynticks_nesting; |
422 | rdtp->dynticks_nesting--; |
423 | WARN_ON_ONCE(rdtp->dynticks_nesting < 0); |
424 | if (rdtp->dynticks_nesting) |
425 | trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting); |
426 | else |
427 | rcu_idle_enter_common(rdtp, oldval); |
428 | local_irq_restore(flags); |
429 | } |
430 | |
431 | /* |
432 | * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle |
433 | * |
434 | * If the new value of the ->dynticks_nesting counter was previously zero, |
435 | * we really have exited idle, and must do the appropriate accounting. |
436 | * The caller must have disabled interrupts. |
437 | */ |
438 | static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) |
439 | { |
440 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ |
441 | atomic_inc(&rdtp->dynticks); |
442 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ |
443 | smp_mb__after_atomic_inc(); /* See above. */ |
444 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); |
445 | rcu_cleanup_after_idle(smp_processor_id()); |
446 | trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); |
447 | if (!is_idle_task(current)) { |
448 | struct task_struct *idle = idle_task(smp_processor_id()); |
449 | |
450 | trace_rcu_dyntick("Error on exit: not idle task", |
451 | oldval, rdtp->dynticks_nesting); |
452 | ftrace_dump(DUMP_ALL); |
453 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
454 | current->pid, current->comm, |
455 | idle->pid, idle->comm); /* must be idle task! */ |
456 | } |
457 | } |
458 | |
459 | /** |
460 | * rcu_idle_exit - inform RCU that current CPU is leaving idle |
461 | * |
462 | * Exit idle mode, in other words, -enter- the mode in which RCU |
463 | * read-side critical sections can occur. |
464 | * |
465 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NESTING to |
466 | * allow for the possibility of usermode upcalls messing up our count |
467 | * of interrupt nesting level during the busy period that is just |
468 | * now starting. |
469 | */ |
470 | void rcu_idle_exit(void) |
471 | { |
472 | unsigned long flags; |
473 | struct rcu_dynticks *rdtp; |
474 | long long oldval; |
475 | |
476 | local_irq_save(flags); |
477 | rdtp = &__get_cpu_var(rcu_dynticks); |
478 | oldval = rdtp->dynticks_nesting; |
479 | WARN_ON_ONCE(oldval != 0); |
480 | rdtp->dynticks_nesting = DYNTICK_TASK_NESTING; |
481 | rcu_idle_exit_common(rdtp, oldval); |
482 | local_irq_restore(flags); |
483 | } |
484 | |
485 | /** |
486 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
487 | * |
488 | * Enter an interrupt handler, which might possibly result in exiting |
489 | * idle mode, in other words, entering the mode in which read-side critical |
490 | * sections can occur. |
491 | * |
492 | * Note that the Linux kernel is fully capable of entering an interrupt |
493 | * handler that it never exits, for example when doing upcalls to |
494 | * user mode! This code assumes that the idle loop never does upcalls to |
495 | * user mode. If your architecture does do upcalls from the idle loop (or |
496 | * does anything else that results in unbalanced calls to the irq_enter() |
497 | * and irq_exit() functions), RCU will give you what you deserve, good |
498 | * and hard. But very infrequently and irreproducibly. |
499 | * |
500 | * Use things like work queues to work around this limitation. |
501 | * |
502 | * You have been warned. |
503 | */ |
504 | void rcu_irq_enter(void) |
505 | { |
506 | unsigned long flags; |
507 | struct rcu_dynticks *rdtp; |
508 | long long oldval; |
509 | |
510 | local_irq_save(flags); |
511 | rdtp = &__get_cpu_var(rcu_dynticks); |
512 | oldval = rdtp->dynticks_nesting; |
513 | rdtp->dynticks_nesting++; |
514 | WARN_ON_ONCE(rdtp->dynticks_nesting == 0); |
515 | if (oldval) |
516 | trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); |
517 | else |
518 | rcu_idle_exit_common(rdtp, oldval); |
519 | local_irq_restore(flags); |
520 | } |
521 | |
522 | /** |
523 | * rcu_nmi_enter - inform RCU of entry to NMI context |
524 | * |
525 | * If the CPU was idle with dynamic ticks active, and there is no |
526 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
527 | * RCU grace-period handling know that the CPU is active. |
528 | */ |
529 | void rcu_nmi_enter(void) |
530 | { |
531 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
532 | |
533 | if (rdtp->dynticks_nmi_nesting == 0 && |
534 | (atomic_read(&rdtp->dynticks) & 0x1)) |
535 | return; |
536 | rdtp->dynticks_nmi_nesting++; |
537 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ |
538 | atomic_inc(&rdtp->dynticks); |
539 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ |
540 | smp_mb__after_atomic_inc(); /* See above. */ |
541 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); |
542 | } |
543 | |
544 | /** |
545 | * rcu_nmi_exit - inform RCU of exit from NMI context |
546 | * |
547 | * If the CPU was idle with dynamic ticks active, and there is no |
548 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
549 | * RCU grace-period handling know that the CPU is no longer active. |
550 | */ |
551 | void rcu_nmi_exit(void) |
552 | { |
553 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
554 | |
555 | if (rdtp->dynticks_nmi_nesting == 0 || |
556 | --rdtp->dynticks_nmi_nesting != 0) |
557 | return; |
558 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
559 | smp_mb__before_atomic_inc(); /* See above. */ |
560 | atomic_inc(&rdtp->dynticks); |
561 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ |
562 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); |
563 | } |
564 | |
565 | #ifdef CONFIG_PROVE_RCU |
566 | |
567 | /** |
568 | * rcu_is_cpu_idle - see if RCU thinks that the current CPU is idle |
569 | * |
570 | * If the current CPU is in its idle loop and is neither in an interrupt |
571 | * or NMI handler, return true. |
572 | */ |
573 | int rcu_is_cpu_idle(void) |
574 | { |
575 | int ret; |
576 | |
577 | preempt_disable(); |
578 | ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0; |
579 | preempt_enable(); |
580 | return ret; |
581 | } |
582 | EXPORT_SYMBOL(rcu_is_cpu_idle); |
583 | |
584 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
585 | |
586 | /** |
587 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
588 | * |
589 | * If the current CPU is idle or running at a first-level (not nested) |
590 | * interrupt from idle, return true. The caller must have at least |
591 | * disabled preemption. |
592 | */ |
593 | int rcu_is_cpu_rrupt_from_idle(void) |
594 | { |
595 | return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1; |
596 | } |
597 | |
598 | #ifdef CONFIG_SMP |
599 | |
600 | /* |
601 | * Snapshot the specified CPU's dynticks counter so that we can later |
602 | * credit them with an implicit quiescent state. Return 1 if this CPU |
603 | * is in dynticks idle mode, which is an extended quiescent state. |
604 | */ |
605 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
606 | { |
607 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
608 | return (rdp->dynticks_snap & 0x1) == 0; |
609 | } |
610 | |
611 | /* |
612 | * Return true if the specified CPU has passed through a quiescent |
613 | * state by virtue of being in or having passed through an dynticks |
614 | * idle state since the last call to dyntick_save_progress_counter() |
615 | * for this same CPU. |
616 | */ |
617 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
618 | { |
619 | unsigned int curr; |
620 | unsigned int snap; |
621 | |
622 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
623 | snap = (unsigned int)rdp->dynticks_snap; |
624 | |
625 | /* |
626 | * If the CPU passed through or entered a dynticks idle phase with |
627 | * no active irq/NMI handlers, then we can safely pretend that the CPU |
628 | * already acknowledged the request to pass through a quiescent |
629 | * state. Either way, that CPU cannot possibly be in an RCU |
630 | * read-side critical section that started before the beginning |
631 | * of the current RCU grace period. |
632 | */ |
633 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
634 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
635 | rdp->dynticks_fqs++; |
636 | return 1; |
637 | } |
638 | |
639 | /* Go check for the CPU being offline. */ |
640 | return rcu_implicit_offline_qs(rdp); |
641 | } |
642 | |
643 | #endif /* #ifdef CONFIG_SMP */ |
644 | |
645 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
646 | { |
647 | rsp->gp_start = jiffies; |
648 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; |
649 | } |
650 | |
651 | static void print_other_cpu_stall(struct rcu_state *rsp) |
652 | { |
653 | int cpu; |
654 | long delta; |
655 | unsigned long flags; |
656 | int ndetected; |
657 | struct rcu_node *rnp = rcu_get_root(rsp); |
658 | |
659 | /* Only let one CPU complain about others per time interval. */ |
660 | |
661 | raw_spin_lock_irqsave(&rnp->lock, flags); |
662 | delta = jiffies - rsp->jiffies_stall; |
663 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
664 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
665 | return; |
666 | } |
667 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
668 | |
669 | /* |
670 | * Now rat on any tasks that got kicked up to the root rcu_node |
671 | * due to CPU offlining. |
672 | */ |
673 | ndetected = rcu_print_task_stall(rnp); |
674 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
675 | |
676 | /* |
677 | * OK, time to rat on our buddy... |
678 | * See Documentation/RCU/stallwarn.txt for info on how to debug |
679 | * RCU CPU stall warnings. |
680 | */ |
681 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
682 | rsp->name); |
683 | rcu_for_each_leaf_node(rsp, rnp) { |
684 | raw_spin_lock_irqsave(&rnp->lock, flags); |
685 | ndetected += rcu_print_task_stall(rnp); |
686 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
687 | if (rnp->qsmask == 0) |
688 | continue; |
689 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
690 | if (rnp->qsmask & (1UL << cpu)) { |
691 | printk(" %d", rnp->grplo + cpu); |
692 | ndetected++; |
693 | } |
694 | } |
695 | printk("} (detected by %d, t=%ld jiffies)\n", |
696 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
697 | if (ndetected == 0) |
698 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); |
699 | else if (!trigger_all_cpu_backtrace()) |
700 | dump_stack(); |
701 | |
702 | /* If so configured, complain about tasks blocking the grace period. */ |
703 | |
704 | rcu_print_detail_task_stall(rsp); |
705 | |
706 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
707 | } |
708 | |
709 | static void print_cpu_stall(struct rcu_state *rsp) |
710 | { |
711 | unsigned long flags; |
712 | struct rcu_node *rnp = rcu_get_root(rsp); |
713 | |
714 | /* |
715 | * OK, time to rat on ourselves... |
716 | * See Documentation/RCU/stallwarn.txt for info on how to debug |
717 | * RCU CPU stall warnings. |
718 | */ |
719 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
720 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); |
721 | if (!trigger_all_cpu_backtrace()) |
722 | dump_stack(); |
723 | |
724 | raw_spin_lock_irqsave(&rnp->lock, flags); |
725 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
726 | rsp->jiffies_stall = |
727 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
728 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
729 | |
730 | set_need_resched(); /* kick ourselves to get things going. */ |
731 | } |
732 | |
733 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) |
734 | { |
735 | unsigned long j; |
736 | unsigned long js; |
737 | struct rcu_node *rnp; |
738 | |
739 | if (rcu_cpu_stall_suppress) |
740 | return; |
741 | j = ACCESS_ONCE(jiffies); |
742 | js = ACCESS_ONCE(rsp->jiffies_stall); |
743 | rnp = rdp->mynode; |
744 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
745 | |
746 | /* We haven't checked in, so go dump stack. */ |
747 | print_cpu_stall(rsp); |
748 | |
749 | } else if (rcu_gp_in_progress(rsp) && |
750 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { |
751 | |
752 | /* They had a few time units to dump stack, so complain. */ |
753 | print_other_cpu_stall(rsp); |
754 | } |
755 | } |
756 | |
757 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
758 | { |
759 | rcu_cpu_stall_suppress = 1; |
760 | return NOTIFY_DONE; |
761 | } |
762 | |
763 | /** |
764 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period |
765 | * |
766 | * Set the stall-warning timeout way off into the future, thus preventing |
767 | * any RCU CPU stall-warning messages from appearing in the current set of |
768 | * RCU grace periods. |
769 | * |
770 | * The caller must disable hard irqs. |
771 | */ |
772 | void rcu_cpu_stall_reset(void) |
773 | { |
774 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; |
775 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; |
776 | rcu_preempt_stall_reset(); |
777 | } |
778 | |
779 | static struct notifier_block rcu_panic_block = { |
780 | .notifier_call = rcu_panic, |
781 | }; |
782 | |
783 | static void __init check_cpu_stall_init(void) |
784 | { |
785 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); |
786 | } |
787 | |
788 | /* |
789 | * Update CPU-local rcu_data state to record the newly noticed grace period. |
790 | * This is used both when we started the grace period and when we notice |
791 | * that someone else started the grace period. The caller must hold the |
792 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, |
793 | * and must have irqs disabled. |
794 | */ |
795 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
796 | { |
797 | if (rdp->gpnum != rnp->gpnum) { |
798 | /* |
799 | * If the current grace period is waiting for this CPU, |
800 | * set up to detect a quiescent state, otherwise don't |
801 | * go looking for one. |
802 | */ |
803 | rdp->gpnum = rnp->gpnum; |
804 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
805 | if (rnp->qsmask & rdp->grpmask) { |
806 | rdp->qs_pending = 1; |
807 | rdp->passed_quiesce = 0; |
808 | } else |
809 | rdp->qs_pending = 0; |
810 | } |
811 | } |
812 | |
813 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
814 | { |
815 | unsigned long flags; |
816 | struct rcu_node *rnp; |
817 | |
818 | local_irq_save(flags); |
819 | rnp = rdp->mynode; |
820 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ |
821 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
822 | local_irq_restore(flags); |
823 | return; |
824 | } |
825 | __note_new_gpnum(rsp, rnp, rdp); |
826 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
827 | } |
828 | |
829 | /* |
830 | * Did someone else start a new RCU grace period start since we last |
831 | * checked? Update local state appropriately if so. Must be called |
832 | * on the CPU corresponding to rdp. |
833 | */ |
834 | static int |
835 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) |
836 | { |
837 | unsigned long flags; |
838 | int ret = 0; |
839 | |
840 | local_irq_save(flags); |
841 | if (rdp->gpnum != rsp->gpnum) { |
842 | note_new_gpnum(rsp, rdp); |
843 | ret = 1; |
844 | } |
845 | local_irq_restore(flags); |
846 | return ret; |
847 | } |
848 | |
849 | /* |
850 | * Advance this CPU's callbacks, but only if the current grace period |
851 | * has ended. This may be called only from the CPU to whom the rdp |
852 | * belongs. In addition, the corresponding leaf rcu_node structure's |
853 | * ->lock must be held by the caller, with irqs disabled. |
854 | */ |
855 | static void |
856 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
857 | { |
858 | /* Did another grace period end? */ |
859 | if (rdp->completed != rnp->completed) { |
860 | |
861 | /* Advance callbacks. No harm if list empty. */ |
862 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; |
863 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; |
864 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
865 | |
866 | /* Remember that we saw this grace-period completion. */ |
867 | rdp->completed = rnp->completed; |
868 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
869 | |
870 | /* |
871 | * If we were in an extended quiescent state, we may have |
872 | * missed some grace periods that others CPUs handled on |
873 | * our behalf. Catch up with this state to avoid noting |
874 | * spurious new grace periods. If another grace period |
875 | * has started, then rnp->gpnum will have advanced, so |
876 | * we will detect this later on. |
877 | */ |
878 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
879 | rdp->gpnum = rdp->completed; |
880 | |
881 | /* |
882 | * If RCU does not need a quiescent state from this CPU, |
883 | * then make sure that this CPU doesn't go looking for one. |
884 | */ |
885 | if ((rnp->qsmask & rdp->grpmask) == 0) |
886 | rdp->qs_pending = 0; |
887 | } |
888 | } |
889 | |
890 | /* |
891 | * Advance this CPU's callbacks, but only if the current grace period |
892 | * has ended. This may be called only from the CPU to whom the rdp |
893 | * belongs. |
894 | */ |
895 | static void |
896 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) |
897 | { |
898 | unsigned long flags; |
899 | struct rcu_node *rnp; |
900 | |
901 | local_irq_save(flags); |
902 | rnp = rdp->mynode; |
903 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ |
904 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
905 | local_irq_restore(flags); |
906 | return; |
907 | } |
908 | __rcu_process_gp_end(rsp, rnp, rdp); |
909 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
910 | } |
911 | |
912 | /* |
913 | * Do per-CPU grace-period initialization for running CPU. The caller |
914 | * must hold the lock of the leaf rcu_node structure corresponding to |
915 | * this CPU. |
916 | */ |
917 | static void |
918 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
919 | { |
920 | /* Prior grace period ended, so advance callbacks for current CPU. */ |
921 | __rcu_process_gp_end(rsp, rnp, rdp); |
922 | |
923 | /* |
924 | * Because this CPU just now started the new grace period, we know |
925 | * that all of its callbacks will be covered by this upcoming grace |
926 | * period, even the ones that were registered arbitrarily recently. |
927 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. |
928 | * |
929 | * Other CPUs cannot be sure exactly when the grace period started. |
930 | * Therefore, their recently registered callbacks must pass through |
931 | * an additional RCU_NEXT_READY stage, so that they will be handled |
932 | * by the next RCU grace period. |
933 | */ |
934 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
935 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
936 | |
937 | /* Set state so that this CPU will detect the next quiescent state. */ |
938 | __note_new_gpnum(rsp, rnp, rdp); |
939 | } |
940 | |
941 | /* |
942 | * Start a new RCU grace period if warranted, re-initializing the hierarchy |
943 | * in preparation for detecting the next grace period. The caller must hold |
944 | * the root node's ->lock, which is released before return. Hard irqs must |
945 | * be disabled. |
946 | */ |
947 | static void |
948 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) |
949 | __releases(rcu_get_root(rsp)->lock) |
950 | { |
951 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
952 | struct rcu_node *rnp = rcu_get_root(rsp); |
953 | |
954 | if (!rcu_scheduler_fully_active || |
955 | !cpu_needs_another_gp(rsp, rdp)) { |
956 | /* |
957 | * Either the scheduler hasn't yet spawned the first |
958 | * non-idle task or this CPU does not need another |
959 | * grace period. Either way, don't start a new grace |
960 | * period. |
961 | */ |
962 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
963 | return; |
964 | } |
965 | |
966 | if (rsp->fqs_active) { |
967 | /* |
968 | * This CPU needs a grace period, but force_quiescent_state() |
969 | * is running. Tell it to start one on this CPU's behalf. |
970 | */ |
971 | rsp->fqs_need_gp = 1; |
972 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
973 | return; |
974 | } |
975 | |
976 | /* Advance to a new grace period and initialize state. */ |
977 | rsp->gpnum++; |
978 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
979 | WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT); |
980 | rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
981 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
982 | record_gp_stall_check_time(rsp); |
983 | |
984 | /* Special-case the common single-level case. */ |
985 | if (NUM_RCU_NODES == 1) { |
986 | rcu_preempt_check_blocked_tasks(rnp); |
987 | rnp->qsmask = rnp->qsmaskinit; |
988 | rnp->gpnum = rsp->gpnum; |
989 | rnp->completed = rsp->completed; |
990 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state OK */ |
991 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
992 | rcu_preempt_boost_start_gp(rnp); |
993 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
994 | rnp->level, rnp->grplo, |
995 | rnp->grphi, rnp->qsmask); |
996 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
997 | return; |
998 | } |
999 | |
1000 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
1001 | |
1002 | |
1003 | /* Exclude any concurrent CPU-hotplug operations. */ |
1004 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
1005 | |
1006 | /* |
1007 | * Set the quiescent-state-needed bits in all the rcu_node |
1008 | * structures for all currently online CPUs in breadth-first |
1009 | * order, starting from the root rcu_node structure. This |
1010 | * operation relies on the layout of the hierarchy within the |
1011 | * rsp->node[] array. Note that other CPUs will access only |
1012 | * the leaves of the hierarchy, which still indicate that no |
1013 | * grace period is in progress, at least until the corresponding |
1014 | * leaf node has been initialized. In addition, we have excluded |
1015 | * CPU-hotplug operations. |
1016 | * |
1017 | * Note that the grace period cannot complete until we finish |
1018 | * the initialization process, as there will be at least one |
1019 | * qsmask bit set in the root node until that time, namely the |
1020 | * one corresponding to this CPU, due to the fact that we have |
1021 | * irqs disabled. |
1022 | */ |
1023 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1024 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1025 | rcu_preempt_check_blocked_tasks(rnp); |
1026 | rnp->qsmask = rnp->qsmaskinit; |
1027 | rnp->gpnum = rsp->gpnum; |
1028 | rnp->completed = rsp->completed; |
1029 | if (rnp == rdp->mynode) |
1030 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
1031 | rcu_preempt_boost_start_gp(rnp); |
1032 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
1033 | rnp->level, rnp->grplo, |
1034 | rnp->grphi, rnp->qsmask); |
1035 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1036 | } |
1037 | |
1038 | rnp = rcu_get_root(rsp); |
1039 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1040 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1041 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1042 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
1043 | } |
1044 | |
1045 | /* |
1046 | * Report a full set of quiescent states to the specified rcu_state |
1047 | * data structure. This involves cleaning up after the prior grace |
1048 | * period and letting rcu_start_gp() start up the next grace period |
1049 | * if one is needed. Note that the caller must hold rnp->lock, as |
1050 | * required by rcu_start_gp(), which will release it. |
1051 | */ |
1052 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
1053 | __releases(rcu_get_root(rsp)->lock) |
1054 | { |
1055 | unsigned long gp_duration; |
1056 | struct rcu_node *rnp = rcu_get_root(rsp); |
1057 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
1058 | |
1059 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
1060 | |
1061 | /* |
1062 | * Ensure that all grace-period and pre-grace-period activity |
1063 | * is seen before the assignment to rsp->completed. |
1064 | */ |
1065 | smp_mb(); /* See above block comment. */ |
1066 | gp_duration = jiffies - rsp->gp_start; |
1067 | if (gp_duration > rsp->gp_max) |
1068 | rsp->gp_max = gp_duration; |
1069 | |
1070 | /* |
1071 | * We know the grace period is complete, but to everyone else |
1072 | * it appears to still be ongoing. But it is also the case |
1073 | * that to everyone else it looks like there is nothing that |
1074 | * they can do to advance the grace period. It is therefore |
1075 | * safe for us to drop the lock in order to mark the grace |
1076 | * period as completed in all of the rcu_node structures. |
1077 | * |
1078 | * But if this CPU needs another grace period, it will take |
1079 | * care of this while initializing the next grace period. |
1080 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL |
1081 | * because the callbacks have not yet been advanced: Those |
1082 | * callbacks are waiting on the grace period that just now |
1083 | * completed. |
1084 | */ |
1085 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { |
1086 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1087 | |
1088 | /* |
1089 | * Propagate new ->completed value to rcu_node structures |
1090 | * so that other CPUs don't have to wait until the start |
1091 | * of the next grace period to process their callbacks. |
1092 | */ |
1093 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1094 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1095 | rnp->completed = rsp->gpnum; |
1096 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1097 | } |
1098 | rnp = rcu_get_root(rsp); |
1099 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1100 | } |
1101 | |
1102 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ |
1103 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
1104 | rsp->fqs_state = RCU_GP_IDLE; |
1105 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
1106 | } |
1107 | |
1108 | /* |
1109 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1110 | * Allows quiescent states for a group of CPUs to be reported at one go |
1111 | * to the specified rcu_node structure, though all the CPUs in the group |
1112 | * must be represented by the same rcu_node structure (which need not be |
1113 | * a leaf rcu_node structure, though it often will be). That structure's |
1114 | * lock must be held upon entry, and it is released before return. |
1115 | */ |
1116 | static void |
1117 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1118 | struct rcu_node *rnp, unsigned long flags) |
1119 | __releases(rnp->lock) |
1120 | { |
1121 | struct rcu_node *rnp_c; |
1122 | |
1123 | /* Walk up the rcu_node hierarchy. */ |
1124 | for (;;) { |
1125 | if (!(rnp->qsmask & mask)) { |
1126 | |
1127 | /* Our bit has already been cleared, so done. */ |
1128 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1129 | return; |
1130 | } |
1131 | rnp->qsmask &= ~mask; |
1132 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1133 | mask, rnp->qsmask, rnp->level, |
1134 | rnp->grplo, rnp->grphi, |
1135 | !!rnp->gp_tasks); |
1136 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
1137 | |
1138 | /* Other bits still set at this level, so done. */ |
1139 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1140 | return; |
1141 | } |
1142 | mask = rnp->grpmask; |
1143 | if (rnp->parent == NULL) { |
1144 | |
1145 | /* No more levels. Exit loop holding root lock. */ |
1146 | |
1147 | break; |
1148 | } |
1149 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1150 | rnp_c = rnp; |
1151 | rnp = rnp->parent; |
1152 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1153 | WARN_ON_ONCE(rnp_c->qsmask); |
1154 | } |
1155 | |
1156 | /* |
1157 | * Get here if we are the last CPU to pass through a quiescent |
1158 | * state for this grace period. Invoke rcu_report_qs_rsp() |
1159 | * to clean up and start the next grace period if one is needed. |
1160 | */ |
1161 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
1162 | } |
1163 | |
1164 | /* |
1165 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1166 | * structure. This must be either called from the specified CPU, or |
1167 | * called when the specified CPU is known to be offline (and when it is |
1168 | * also known that no other CPU is concurrently trying to help the offline |
1169 | * CPU). The lastcomp argument is used to make sure we are still in the |
1170 | * grace period of interest. We don't want to end the current grace period |
1171 | * based on quiescent states detected in an earlier grace period! |
1172 | */ |
1173 | static void |
1174 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
1175 | { |
1176 | unsigned long flags; |
1177 | unsigned long mask; |
1178 | struct rcu_node *rnp; |
1179 | |
1180 | rnp = rdp->mynode; |
1181 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1182 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
1183 | |
1184 | /* |
1185 | * The grace period in which this quiescent state was |
1186 | * recorded has ended, so don't report it upwards. |
1187 | * We will instead need a new quiescent state that lies |
1188 | * within the current grace period. |
1189 | */ |
1190 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1191 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1192 | return; |
1193 | } |
1194 | mask = rdp->grpmask; |
1195 | if ((rnp->qsmask & mask) == 0) { |
1196 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1197 | } else { |
1198 | rdp->qs_pending = 0; |
1199 | |
1200 | /* |
1201 | * This GP can't end until cpu checks in, so all of our |
1202 | * callbacks can be processed during the next GP. |
1203 | */ |
1204 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1205 | |
1206 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
1207 | } |
1208 | } |
1209 | |
1210 | /* |
1211 | * Check to see if there is a new grace period of which this CPU |
1212 | * is not yet aware, and if so, set up local rcu_data state for it. |
1213 | * Otherwise, see if this CPU has just passed through its first |
1214 | * quiescent state for this grace period, and record that fact if so. |
1215 | */ |
1216 | static void |
1217 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) |
1218 | { |
1219 | /* If there is now a new grace period, record and return. */ |
1220 | if (check_for_new_grace_period(rsp, rdp)) |
1221 | return; |
1222 | |
1223 | /* |
1224 | * Does this CPU still need to do its part for current grace period? |
1225 | * If no, return and let the other CPUs do their part as well. |
1226 | */ |
1227 | if (!rdp->qs_pending) |
1228 | return; |
1229 | |
1230 | /* |
1231 | * Was there a quiescent state since the beginning of the grace |
1232 | * period? If no, then exit and wait for the next call. |
1233 | */ |
1234 | if (!rdp->passed_quiesce) |
1235 | return; |
1236 | |
1237 | /* |
1238 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the |
1239 | * judge of that). |
1240 | */ |
1241 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
1242 | } |
1243 | |
1244 | #ifdef CONFIG_HOTPLUG_CPU |
1245 | |
1246 | /* |
1247 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1248 | * Synchronization is not required because this function executes |
1249 | * in stop_machine() context. |
1250 | */ |
1251 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
1252 | { |
1253 | int i; |
1254 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1255 | int receive_cpu = cpumask_any(cpu_online_mask); |
1256 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
1257 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
1258 | |
1259 | if (rdp->nxtlist == NULL) |
1260 | return; /* irqs disabled, so comparison is stable. */ |
1261 | |
1262 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; |
1263 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1264 | receive_rdp->qlen += rdp->qlen; |
1265 | receive_rdp->n_cbs_adopted += rdp->qlen; |
1266 | rdp->n_cbs_orphaned += rdp->qlen; |
1267 | |
1268 | rdp->nxtlist = NULL; |
1269 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
1270 | rdp->nxttail[i] = &rdp->nxtlist; |
1271 | rdp->qlen = 0; |
1272 | } |
1273 | |
1274 | /* |
1275 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy |
1276 | * and move all callbacks from the outgoing CPU to the current one. |
1277 | * There can only be one CPU hotplug operation at a time, so no other |
1278 | * CPU can be attempting to update rcu_cpu_kthread_task. |
1279 | */ |
1280 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) |
1281 | { |
1282 | unsigned long flags; |
1283 | unsigned long mask; |
1284 | int need_report = 0; |
1285 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
1286 | struct rcu_node *rnp; |
1287 | |
1288 | rcu_stop_cpu_kthread(cpu); |
1289 | |
1290 | /* Exclude any attempts to start a new grace period. */ |
1291 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
1292 | |
1293 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ |
1294 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
1295 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1296 | do { |
1297 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1298 | rnp->qsmaskinit &= ~mask; |
1299 | if (rnp->qsmaskinit != 0) { |
1300 | if (rnp != rdp->mynode) |
1301 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1302 | else |
1303 | trace_rcu_grace_period(rsp->name, |
1304 | rnp->gpnum + 1 - |
1305 | !!(rnp->qsmask & mask), |
1306 | "cpuofl"); |
1307 | break; |
1308 | } |
1309 | if (rnp == rdp->mynode) { |
1310 | trace_rcu_grace_period(rsp->name, |
1311 | rnp->gpnum + 1 - |
1312 | !!(rnp->qsmask & mask), |
1313 | "cpuofl"); |
1314 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
1315 | } else |
1316 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1317 | mask = rnp->grpmask; |
1318 | rnp = rnp->parent; |
1319 | } while (rnp != NULL); |
1320 | |
1321 | /* |
1322 | * We still hold the leaf rcu_node structure lock here, and |
1323 | * irqs are still disabled. The reason for this subterfuge is |
1324 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1325 | * held leads to deadlock. |
1326 | */ |
1327 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
1328 | rnp = rdp->mynode; |
1329 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
1330 | rcu_report_unblock_qs_rnp(rnp, flags); |
1331 | else |
1332 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1333 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1334 | rcu_report_exp_rnp(rsp, rnp, true); |
1335 | rcu_node_kthread_setaffinity(rnp, -1); |
1336 | } |
1337 | |
1338 | /* |
1339 | * Remove the specified CPU from the RCU hierarchy and move any pending |
1340 | * callbacks that it might have to the current CPU. This code assumes |
1341 | * that at least one CPU in the system will remain running at all times. |
1342 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. |
1343 | */ |
1344 | static void rcu_offline_cpu(int cpu) |
1345 | { |
1346 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
1347 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
1348 | rcu_preempt_offline_cpu(cpu); |
1349 | } |
1350 | |
1351 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ |
1352 | |
1353 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
1354 | { |
1355 | } |
1356 | |
1357 | static void rcu_offline_cpu(int cpu) |
1358 | { |
1359 | } |
1360 | |
1361 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ |
1362 | |
1363 | /* |
1364 | * Invoke any RCU callbacks that have made it to the end of their grace |
1365 | * period. Thottle as specified by rdp->blimit. |
1366 | */ |
1367 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
1368 | { |
1369 | unsigned long flags; |
1370 | struct rcu_head *next, *list, **tail; |
1371 | int bl, count; |
1372 | |
1373 | /* If no callbacks are ready, just return.*/ |
1374 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
1375 | trace_rcu_batch_start(rsp->name, 0, 0); |
1376 | trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist), |
1377 | need_resched(), is_idle_task(current), |
1378 | rcu_is_callbacks_kthread()); |
1379 | return; |
1380 | } |
1381 | |
1382 | /* |
1383 | * Extract the list of ready callbacks, disabling to prevent |
1384 | * races with call_rcu() from interrupt handlers. |
1385 | */ |
1386 | local_irq_save(flags); |
1387 | bl = rdp->blimit; |
1388 | trace_rcu_batch_start(rsp->name, rdp->qlen, bl); |
1389 | list = rdp->nxtlist; |
1390 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; |
1391 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; |
1392 | tail = rdp->nxttail[RCU_DONE_TAIL]; |
1393 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) |
1394 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) |
1395 | rdp->nxttail[count] = &rdp->nxtlist; |
1396 | local_irq_restore(flags); |
1397 | |
1398 | /* Invoke callbacks. */ |
1399 | count = 0; |
1400 | while (list) { |
1401 | next = list->next; |
1402 | prefetch(next); |
1403 | debug_rcu_head_unqueue(list); |
1404 | __rcu_reclaim(rsp->name, list); |
1405 | list = next; |
1406 | /* Stop only if limit reached and CPU has something to do. */ |
1407 | if (++count >= bl && |
1408 | (need_resched() || |
1409 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) |
1410 | break; |
1411 | } |
1412 | |
1413 | local_irq_save(flags); |
1414 | trace_rcu_batch_end(rsp->name, count, !!list, need_resched(), |
1415 | is_idle_task(current), |
1416 | rcu_is_callbacks_kthread()); |
1417 | |
1418 | /* Update count, and requeue any remaining callbacks. */ |
1419 | rdp->qlen -= count; |
1420 | rdp->n_cbs_invoked += count; |
1421 | if (list != NULL) { |
1422 | *tail = rdp->nxtlist; |
1423 | rdp->nxtlist = list; |
1424 | for (count = 0; count < RCU_NEXT_SIZE; count++) |
1425 | if (&rdp->nxtlist == rdp->nxttail[count]) |
1426 | rdp->nxttail[count] = tail; |
1427 | else |
1428 | break; |
1429 | } |
1430 | |
1431 | /* Reinstate batch limit if we have worked down the excess. */ |
1432 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) |
1433 | rdp->blimit = blimit; |
1434 | |
1435 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1436 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { |
1437 | rdp->qlen_last_fqs_check = 0; |
1438 | rdp->n_force_qs_snap = rsp->n_force_qs; |
1439 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) |
1440 | rdp->qlen_last_fqs_check = rdp->qlen; |
1441 | |
1442 | local_irq_restore(flags); |
1443 | |
1444 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
1445 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1446 | invoke_rcu_core(); |
1447 | } |
1448 | |
1449 | /* |
1450 | * Check to see if this CPU is in a non-context-switch quiescent state |
1451 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). |
1452 | * Also schedule RCU core processing. |
1453 | * |
1454 | * This function must be called from hardirq context. It is normally |
1455 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1456 | * false, there is no point in invoking rcu_check_callbacks(). |
1457 | */ |
1458 | void rcu_check_callbacks(int cpu, int user) |
1459 | { |
1460 | trace_rcu_utilization("Start scheduler-tick"); |
1461 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
1462 | |
1463 | /* |
1464 | * Get here if this CPU took its interrupt from user |
1465 | * mode or from the idle loop, and if this is not a |
1466 | * nested interrupt. In this case, the CPU is in |
1467 | * a quiescent state, so note it. |
1468 | * |
1469 | * No memory barrier is required here because both |
1470 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1471 | * variables that other CPUs neither access nor modify, |
1472 | * at least not while the corresponding CPU is online. |
1473 | */ |
1474 | |
1475 | rcu_sched_qs(cpu); |
1476 | rcu_bh_qs(cpu); |
1477 | |
1478 | } else if (!in_softirq()) { |
1479 | |
1480 | /* |
1481 | * Get here if this CPU did not take its interrupt from |
1482 | * softirq, in other words, if it is not interrupting |
1483 | * a rcu_bh read-side critical section. This is an _bh |
1484 | * critical section, so note it. |
1485 | */ |
1486 | |
1487 | rcu_bh_qs(cpu); |
1488 | } |
1489 | rcu_preempt_check_callbacks(cpu); |
1490 | if (rcu_pending(cpu)) |
1491 | invoke_rcu_core(); |
1492 | trace_rcu_utilization("End scheduler-tick"); |
1493 | } |
1494 | |
1495 | #ifdef CONFIG_SMP |
1496 | |
1497 | /* |
1498 | * Scan the leaf rcu_node structures, processing dyntick state for any that |
1499 | * have not yet encountered a quiescent state, using the function specified. |
1500 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1501 | * |
1502 | * The caller must have suppressed start of new grace periods. |
1503 | */ |
1504 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
1505 | { |
1506 | unsigned long bit; |
1507 | int cpu; |
1508 | unsigned long flags; |
1509 | unsigned long mask; |
1510 | struct rcu_node *rnp; |
1511 | |
1512 | rcu_for_each_leaf_node(rsp, rnp) { |
1513 | mask = 0; |
1514 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1515 | if (!rcu_gp_in_progress(rsp)) { |
1516 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1517 | return; |
1518 | } |
1519 | if (rnp->qsmask == 0) { |
1520 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
1521 | continue; |
1522 | } |
1523 | cpu = rnp->grplo; |
1524 | bit = 1; |
1525 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
1526 | if ((rnp->qsmask & bit) != 0 && |
1527 | f(per_cpu_ptr(rsp->rda, cpu))) |
1528 | mask |= bit; |
1529 | } |
1530 | if (mask != 0) { |
1531 | |
1532 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1533 | rcu_report_qs_rnp(mask, rsp, rnp, flags); |
1534 | continue; |
1535 | } |
1536 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1537 | } |
1538 | rnp = rcu_get_root(rsp); |
1539 | if (rnp->qsmask == 0) { |
1540 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1541 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ |
1542 | } |
1543 | } |
1544 | |
1545 | /* |
1546 | * Force quiescent states on reluctant CPUs, and also detect which |
1547 | * CPUs are in dyntick-idle mode. |
1548 | */ |
1549 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
1550 | { |
1551 | unsigned long flags; |
1552 | struct rcu_node *rnp = rcu_get_root(rsp); |
1553 | |
1554 | trace_rcu_utilization("Start fqs"); |
1555 | if (!rcu_gp_in_progress(rsp)) { |
1556 | trace_rcu_utilization("End fqs"); |
1557 | return; /* No grace period in progress, nothing to force. */ |
1558 | } |
1559 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
1560 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1561 | trace_rcu_utilization("End fqs"); |
1562 | return; /* Someone else is already on the job. */ |
1563 | } |
1564 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
1565 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
1566 | rsp->n_force_qs++; |
1567 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1568 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
1569 | if(!rcu_gp_in_progress(rsp)) { |
1570 | rsp->n_force_qs_ngp++; |
1571 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1572 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
1573 | } |
1574 | rsp->fqs_active = 1; |
1575 | switch (rsp->fqs_state) { |
1576 | case RCU_GP_IDLE: |
1577 | case RCU_GP_INIT: |
1578 | |
1579 | break; /* grace period idle or initializing, ignore. */ |
1580 | |
1581 | case RCU_SAVE_DYNTICK: |
1582 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1583 | break; /* So gcc recognizes the dead code. */ |
1584 | |
1585 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1586 | |
1587 | /* Record dyntick-idle state. */ |
1588 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1589 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1590 | if (rcu_gp_in_progress(rsp)) |
1591 | rsp->fqs_state = RCU_FORCE_QS; |
1592 | break; |
1593 | |
1594 | case RCU_FORCE_QS: |
1595 | |
1596 | /* Check dyntick-idle state, send IPI to laggarts. */ |
1597 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1598 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
1599 | |
1600 | /* Leave state in case more forcing is required. */ |
1601 | |
1602 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1603 | break; |
1604 | } |
1605 | rsp->fqs_active = 0; |
1606 | if (rsp->fqs_need_gp) { |
1607 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
1608 | rsp->fqs_need_gp = 0; |
1609 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ |
1610 | trace_rcu_utilization("End fqs"); |
1611 | return; |
1612 | } |
1613 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1614 | unlock_fqs_ret: |
1615 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
1616 | trace_rcu_utilization("End fqs"); |
1617 | } |
1618 | |
1619 | #else /* #ifdef CONFIG_SMP */ |
1620 | |
1621 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
1622 | { |
1623 | set_need_resched(); |
1624 | } |
1625 | |
1626 | #endif /* #else #ifdef CONFIG_SMP */ |
1627 | |
1628 | /* |
1629 | * This does the RCU core processing work for the specified rcu_state |
1630 | * and rcu_data structures. This may be called only from the CPU to |
1631 | * whom the rdp belongs. |
1632 | */ |
1633 | static void |
1634 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
1635 | { |
1636 | unsigned long flags; |
1637 | |
1638 | WARN_ON_ONCE(rdp->beenonline == 0); |
1639 | |
1640 | /* |
1641 | * If an RCU GP has gone long enough, go check for dyntick |
1642 | * idle CPUs and, if needed, send resched IPIs. |
1643 | */ |
1644 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
1645 | force_quiescent_state(rsp, 1); |
1646 | |
1647 | /* |
1648 | * Advance callbacks in response to end of earlier grace |
1649 | * period that some other CPU ended. |
1650 | */ |
1651 | rcu_process_gp_end(rsp, rdp); |
1652 | |
1653 | /* Update RCU state based on any recent quiescent states. */ |
1654 | rcu_check_quiescent_state(rsp, rdp); |
1655 | |
1656 | /* Does this CPU require a not-yet-started grace period? */ |
1657 | if (cpu_needs_another_gp(rsp, rdp)) { |
1658 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
1659 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1660 | } |
1661 | |
1662 | /* If there are callbacks ready, invoke them. */ |
1663 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1664 | invoke_rcu_callbacks(rsp, rdp); |
1665 | } |
1666 | |
1667 | /* |
1668 | * Do RCU core processing for the current CPU. |
1669 | */ |
1670 | static void rcu_process_callbacks(struct softirq_action *unused) |
1671 | { |
1672 | trace_rcu_utilization("Start RCU core"); |
1673 | __rcu_process_callbacks(&rcu_sched_state, |
1674 | &__get_cpu_var(rcu_sched_data)); |
1675 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
1676 | rcu_preempt_process_callbacks(); |
1677 | trace_rcu_utilization("End RCU core"); |
1678 | } |
1679 | |
1680 | /* |
1681 | * Schedule RCU callback invocation. If the specified type of RCU |
1682 | * does not support RCU priority boosting, just do a direct call, |
1683 | * otherwise wake up the per-CPU kernel kthread. Note that because we |
1684 | * are running on the current CPU with interrupts disabled, the |
1685 | * rcu_cpu_kthread_task cannot disappear out from under us. |
1686 | */ |
1687 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
1688 | { |
1689 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1690 | return; |
1691 | if (likely(!rsp->boost)) { |
1692 | rcu_do_batch(rsp, rdp); |
1693 | return; |
1694 | } |
1695 | invoke_rcu_callbacks_kthread(); |
1696 | } |
1697 | |
1698 | static void invoke_rcu_core(void) |
1699 | { |
1700 | raise_softirq(RCU_SOFTIRQ); |
1701 | } |
1702 | |
1703 | static void |
1704 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), |
1705 | struct rcu_state *rsp) |
1706 | { |
1707 | unsigned long flags; |
1708 | struct rcu_data *rdp; |
1709 | |
1710 | debug_rcu_head_queue(head); |
1711 | head->func = func; |
1712 | head->next = NULL; |
1713 | |
1714 | smp_mb(); /* Ensure RCU update seen before callback registry. */ |
1715 | |
1716 | /* |
1717 | * Opportunistically note grace-period endings and beginnings. |
1718 | * Note that we might see a beginning right after we see an |
1719 | * end, but never vice versa, since this CPU has to pass through |
1720 | * a quiescent state betweentimes. |
1721 | */ |
1722 | local_irq_save(flags); |
1723 | rdp = this_cpu_ptr(rsp->rda); |
1724 | |
1725 | /* Add the callback to our list. */ |
1726 | *rdp->nxttail[RCU_NEXT_TAIL] = head; |
1727 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; |
1728 | rdp->qlen++; |
1729 | |
1730 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1731 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, |
1732 | rdp->qlen); |
1733 | else |
1734 | trace_rcu_callback(rsp->name, head, rdp->qlen); |
1735 | |
1736 | /* If interrupts were disabled, don't dive into RCU core. */ |
1737 | if (irqs_disabled_flags(flags)) { |
1738 | local_irq_restore(flags); |
1739 | return; |
1740 | } |
1741 | |
1742 | /* |
1743 | * Force the grace period if too many callbacks or too long waiting. |
1744 | * Enforce hysteresis, and don't invoke force_quiescent_state() |
1745 | * if some other CPU has recently done so. Also, don't bother |
1746 | * invoking force_quiescent_state() if the newly enqueued callback |
1747 | * is the only one waiting for a grace period to complete. |
1748 | */ |
1749 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
1750 | |
1751 | /* Are we ignoring a completed grace period? */ |
1752 | rcu_process_gp_end(rsp, rdp); |
1753 | check_for_new_grace_period(rsp, rdp); |
1754 | |
1755 | /* Start a new grace period if one not already started. */ |
1756 | if (!rcu_gp_in_progress(rsp)) { |
1757 | unsigned long nestflag; |
1758 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
1759 | |
1760 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); |
1761 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ |
1762 | } else { |
1763 | /* Give the grace period a kick. */ |
1764 | rdp->blimit = LONG_MAX; |
1765 | if (rsp->n_force_qs == rdp->n_force_qs_snap && |
1766 | *rdp->nxttail[RCU_DONE_TAIL] != head) |
1767 | force_quiescent_state(rsp, 0); |
1768 | rdp->n_force_qs_snap = rsp->n_force_qs; |
1769 | rdp->qlen_last_fqs_check = rdp->qlen; |
1770 | } |
1771 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
1772 | force_quiescent_state(rsp, 1); |
1773 | local_irq_restore(flags); |
1774 | } |
1775 | |
1776 | /* |
1777 | * Queue an RCU-sched callback for invocation after a grace period. |
1778 | */ |
1779 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
1780 | { |
1781 | __call_rcu(head, func, &rcu_sched_state); |
1782 | } |
1783 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
1784 | |
1785 | /* |
1786 | * Queue an RCU for invocation after a quicker grace period. |
1787 | */ |
1788 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
1789 | { |
1790 | __call_rcu(head, func, &rcu_bh_state); |
1791 | } |
1792 | EXPORT_SYMBOL_GPL(call_rcu_bh); |
1793 | |
1794 | /** |
1795 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. |
1796 | * |
1797 | * Control will return to the caller some time after a full rcu-sched |
1798 | * grace period has elapsed, in other words after all currently executing |
1799 | * rcu-sched read-side critical sections have completed. These read-side |
1800 | * critical sections are delimited by rcu_read_lock_sched() and |
1801 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), |
1802 | * local_irq_disable(), and so on may be used in place of |
1803 | * rcu_read_lock_sched(). |
1804 | * |
1805 | * This means that all preempt_disable code sequences, including NMI and |
1806 | * hardware-interrupt handlers, in progress on entry will have completed |
1807 | * before this primitive returns. However, this does not guarantee that |
1808 | * softirq handlers will have completed, since in some kernels, these |
1809 | * handlers can run in process context, and can block. |
1810 | * |
1811 | * This primitive provides the guarantees made by the (now removed) |
1812 | * synchronize_kernel() API. In contrast, synchronize_rcu() only |
1813 | * guarantees that rcu_read_lock() sections will have completed. |
1814 | * In "classic RCU", these two guarantees happen to be one and |
1815 | * the same, but can differ in realtime RCU implementations. |
1816 | */ |
1817 | void synchronize_sched(void) |
1818 | { |
1819 | if (rcu_blocking_is_gp()) |
1820 | return; |
1821 | wait_rcu_gp(call_rcu_sched); |
1822 | } |
1823 | EXPORT_SYMBOL_GPL(synchronize_sched); |
1824 | |
1825 | /** |
1826 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. |
1827 | * |
1828 | * Control will return to the caller some time after a full rcu_bh grace |
1829 | * period has elapsed, in other words after all currently executing rcu_bh |
1830 | * read-side critical sections have completed. RCU read-side critical |
1831 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), |
1832 | * and may be nested. |
1833 | */ |
1834 | void synchronize_rcu_bh(void) |
1835 | { |
1836 | if (rcu_blocking_is_gp()) |
1837 | return; |
1838 | wait_rcu_gp(call_rcu_bh); |
1839 | } |
1840 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); |
1841 | |
1842 | /* |
1843 | * Check to see if there is any immediate RCU-related work to be done |
1844 | * by the current CPU, for the specified type of RCU, returning 1 if so. |
1845 | * The checks are in order of increasing expense: checks that can be |
1846 | * carried out against CPU-local state are performed first. However, |
1847 | * we must check for CPU stalls first, else we might not get a chance. |
1848 | */ |
1849 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) |
1850 | { |
1851 | struct rcu_node *rnp = rdp->mynode; |
1852 | |
1853 | rdp->n_rcu_pending++; |
1854 | |
1855 | /* Check for CPU stalls, if enabled. */ |
1856 | check_cpu_stall(rsp, rdp); |
1857 | |
1858 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
1859 | if (rcu_scheduler_fully_active && |
1860 | rdp->qs_pending && !rdp->passed_quiesce) { |
1861 | |
1862 | /* |
1863 | * If force_quiescent_state() coming soon and this CPU |
1864 | * needs a quiescent state, and this is either RCU-sched |
1865 | * or RCU-bh, force a local reschedule. |
1866 | */ |
1867 | rdp->n_rp_qs_pending++; |
1868 | if (!rdp->preemptible && |
1869 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1870 | jiffies)) |
1871 | set_need_resched(); |
1872 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
1873 | rdp->n_rp_report_qs++; |
1874 | return 1; |
1875 | } |
1876 | |
1877 | /* Does this CPU have callbacks ready to invoke? */ |
1878 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1879 | rdp->n_rp_cb_ready++; |
1880 | return 1; |
1881 | } |
1882 | |
1883 | /* Has RCU gone idle with this CPU needing another grace period? */ |
1884 | if (cpu_needs_another_gp(rsp, rdp)) { |
1885 | rdp->n_rp_cpu_needs_gp++; |
1886 | return 1; |
1887 | } |
1888 | |
1889 | /* Has another RCU grace period completed? */ |
1890 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
1891 | rdp->n_rp_gp_completed++; |
1892 | return 1; |
1893 | } |
1894 | |
1895 | /* Has a new RCU grace period started? */ |
1896 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
1897 | rdp->n_rp_gp_started++; |
1898 | return 1; |
1899 | } |
1900 | |
1901 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ |
1902 | if (rcu_gp_in_progress(rsp) && |
1903 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
1904 | rdp->n_rp_need_fqs++; |
1905 | return 1; |
1906 | } |
1907 | |
1908 | /* nothing to do */ |
1909 | rdp->n_rp_need_nothing++; |
1910 | return 0; |
1911 | } |
1912 | |
1913 | /* |
1914 | * Check to see if there is any immediate RCU-related work to be done |
1915 | * by the current CPU, returning 1 if so. This function is part of the |
1916 | * RCU implementation; it is -not- an exported member of the RCU API. |
1917 | */ |
1918 | static int rcu_pending(int cpu) |
1919 | { |
1920 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
1921 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1922 | rcu_preempt_pending(cpu); |
1923 | } |
1924 | |
1925 | /* |
1926 | * Check to see if any future RCU-related work will need to be done |
1927 | * by the current CPU, even if none need be done immediately, returning |
1928 | * 1 if so. |
1929 | */ |
1930 | static int rcu_cpu_has_callbacks(int cpu) |
1931 | { |
1932 | /* RCU callbacks either ready or pending? */ |
1933 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
1934 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1935 | rcu_preempt_needs_cpu(cpu); |
1936 | } |
1937 | |
1938 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1939 | static atomic_t rcu_barrier_cpu_count; |
1940 | static DEFINE_MUTEX(rcu_barrier_mutex); |
1941 | static struct completion rcu_barrier_completion; |
1942 | |
1943 | static void rcu_barrier_callback(struct rcu_head *notused) |
1944 | { |
1945 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1946 | complete(&rcu_barrier_completion); |
1947 | } |
1948 | |
1949 | /* |
1950 | * Called with preemption disabled, and from cross-cpu IRQ context. |
1951 | */ |
1952 | static void rcu_barrier_func(void *type) |
1953 | { |
1954 | int cpu = smp_processor_id(); |
1955 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); |
1956 | void (*call_rcu_func)(struct rcu_head *head, |
1957 | void (*func)(struct rcu_head *head)); |
1958 | |
1959 | atomic_inc(&rcu_barrier_cpu_count); |
1960 | call_rcu_func = type; |
1961 | call_rcu_func(head, rcu_barrier_callback); |
1962 | } |
1963 | |
1964 | /* |
1965 | * Orchestrate the specified type of RCU barrier, waiting for all |
1966 | * RCU callbacks of the specified type to complete. |
1967 | */ |
1968 | static void _rcu_barrier(struct rcu_state *rsp, |
1969 | void (*call_rcu_func)(struct rcu_head *head, |
1970 | void (*func)(struct rcu_head *head))) |
1971 | { |
1972 | BUG_ON(in_interrupt()); |
1973 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
1974 | mutex_lock(&rcu_barrier_mutex); |
1975 | init_completion(&rcu_barrier_completion); |
1976 | /* |
1977 | * Initialize rcu_barrier_cpu_count to 1, then invoke |
1978 | * rcu_barrier_func() on each CPU, so that each CPU also has |
1979 | * incremented rcu_barrier_cpu_count. Only then is it safe to |
1980 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU |
1981 | * might complete its grace period before all of the other CPUs |
1982 | * did their increment, causing this function to return too |
1983 | * early. Note that on_each_cpu() disables irqs, which prevents |
1984 | * any CPUs from coming online or going offline until each online |
1985 | * CPU has queued its RCU-barrier callback. |
1986 | */ |
1987 | atomic_set(&rcu_barrier_cpu_count, 1); |
1988 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); |
1989 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1990 | complete(&rcu_barrier_completion); |
1991 | wait_for_completion(&rcu_barrier_completion); |
1992 | mutex_unlock(&rcu_barrier_mutex); |
1993 | } |
1994 | |
1995 | /** |
1996 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. |
1997 | */ |
1998 | void rcu_barrier_bh(void) |
1999 | { |
2000 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
2001 | } |
2002 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); |
2003 | |
2004 | /** |
2005 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. |
2006 | */ |
2007 | void rcu_barrier_sched(void) |
2008 | { |
2009 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
2010 | } |
2011 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); |
2012 | |
2013 | /* |
2014 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
2015 | */ |
2016 | static void __init |
2017 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) |
2018 | { |
2019 | unsigned long flags; |
2020 | int i; |
2021 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
2022 | struct rcu_node *rnp = rcu_get_root(rsp); |
2023 | |
2024 | /* Set up local state, ensuring consistent view of global state. */ |
2025 | raw_spin_lock_irqsave(&rnp->lock, flags); |
2026 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2027 | rdp->nxtlist = NULL; |
2028 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
2029 | rdp->nxttail[i] = &rdp->nxtlist; |
2030 | rdp->qlen = 0; |
2031 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
2032 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_NESTING); |
2033 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
2034 | rdp->cpu = cpu; |
2035 | rdp->rsp = rsp; |
2036 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
2037 | } |
2038 | |
2039 | /* |
2040 | * Initialize a CPU's per-CPU RCU data. Note that only one online or |
2041 | * offline event can be happening at a given time. Note also that we |
2042 | * can accept some slop in the rsp->completed access due to the fact |
2043 | * that this CPU cannot possibly have any RCU callbacks in flight yet. |
2044 | */ |
2045 | static void __cpuinit |
2046 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
2047 | { |
2048 | unsigned long flags; |
2049 | unsigned long mask; |
2050 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
2051 | struct rcu_node *rnp = rcu_get_root(rsp); |
2052 | |
2053 | /* Set up local state, ensuring consistent view of global state. */ |
2054 | raw_spin_lock_irqsave(&rnp->lock, flags); |
2055 | rdp->beenonline = 1; /* We have now been online. */ |
2056 | rdp->preemptible = preemptible; |
2057 | rdp->qlen_last_fqs_check = 0; |
2058 | rdp->n_force_qs_snap = rsp->n_force_qs; |
2059 | rdp->blimit = blimit; |
2060 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_NESTING; |
2061 | atomic_set(&rdp->dynticks->dynticks, |
2062 | (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); |
2063 | rcu_prepare_for_idle_init(cpu); |
2064 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
2065 | |
2066 | /* |
2067 | * A new grace period might start here. If so, we won't be part |
2068 | * of it, but that is OK, as we are currently in a quiescent state. |
2069 | */ |
2070 | |
2071 | /* Exclude any attempts to start a new GP on large systems. */ |
2072 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
2073 | |
2074 | /* Add CPU to rcu_node bitmasks. */ |
2075 | rnp = rdp->mynode; |
2076 | mask = rdp->grpmask; |
2077 | do { |
2078 | /* Exclude any attempts to start a new GP on small systems. */ |
2079 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
2080 | rnp->qsmaskinit |= mask; |
2081 | mask = rnp->grpmask; |
2082 | if (rnp == rdp->mynode) { |
2083 | /* |
2084 | * If there is a grace period in progress, we will |
2085 | * set up to wait for it next time we run the |
2086 | * RCU core code. |
2087 | */ |
2088 | rdp->gpnum = rnp->completed; |
2089 | rdp->completed = rnp->completed; |
2090 | rdp->passed_quiesce = 0; |
2091 | rdp->qs_pending = 0; |
2092 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
2093 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
2094 | } |
2095 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
2096 | rnp = rnp->parent; |
2097 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); |
2098 | |
2099 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
2100 | } |
2101 | |
2102 | static void __cpuinit rcu_prepare_cpu(int cpu) |
2103 | { |
2104 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2105 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); |
2106 | rcu_preempt_init_percpu_data(cpu); |
2107 | } |
2108 | |
2109 | /* |
2110 | * Handle CPU online/offline notification events. |
2111 | */ |
2112 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2113 | unsigned long action, void *hcpu) |
2114 | { |
2115 | long cpu = (long)hcpu; |
2116 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
2117 | struct rcu_node *rnp = rdp->mynode; |
2118 | |
2119 | trace_rcu_utilization("Start CPU hotplug"); |
2120 | switch (action) { |
2121 | case CPU_UP_PREPARE: |
2122 | case CPU_UP_PREPARE_FROZEN: |
2123 | rcu_prepare_cpu(cpu); |
2124 | rcu_prepare_kthreads(cpu); |
2125 | break; |
2126 | case CPU_ONLINE: |
2127 | case CPU_DOWN_FAILED: |
2128 | rcu_node_kthread_setaffinity(rnp, -1); |
2129 | rcu_cpu_kthread_setrt(cpu, 1); |
2130 | break; |
2131 | case CPU_DOWN_PREPARE: |
2132 | rcu_node_kthread_setaffinity(rnp, cpu); |
2133 | rcu_cpu_kthread_setrt(cpu, 0); |
2134 | break; |
2135 | case CPU_DYING: |
2136 | case CPU_DYING_FROZEN: |
2137 | /* |
2138 | * The whole machine is "stopped" except this CPU, so we can |
2139 | * touch any data without introducing corruption. We send the |
2140 | * dying CPU's callbacks to an arbitrarily chosen online CPU. |
2141 | */ |
2142 | rcu_send_cbs_to_online(&rcu_bh_state); |
2143 | rcu_send_cbs_to_online(&rcu_sched_state); |
2144 | rcu_preempt_send_cbs_to_online(); |
2145 | rcu_cleanup_after_idle(cpu); |
2146 | break; |
2147 | case CPU_DEAD: |
2148 | case CPU_DEAD_FROZEN: |
2149 | case CPU_UP_CANCELED: |
2150 | case CPU_UP_CANCELED_FROZEN: |
2151 | rcu_offline_cpu(cpu); |
2152 | break; |
2153 | default: |
2154 | break; |
2155 | } |
2156 | trace_rcu_utilization("End CPU hotplug"); |
2157 | return NOTIFY_OK; |
2158 | } |
2159 | |
2160 | /* |
2161 | * This function is invoked towards the end of the scheduler's initialization |
2162 | * process. Before this is called, the idle task might contain |
2163 | * RCU read-side critical sections (during which time, this idle |
2164 | * task is booting the system). After this function is called, the |
2165 | * idle tasks are prohibited from containing RCU read-side critical |
2166 | * sections. This function also enables RCU lockdep checking. |
2167 | */ |
2168 | void rcu_scheduler_starting(void) |
2169 | { |
2170 | WARN_ON(num_online_cpus() != 1); |
2171 | WARN_ON(nr_context_switches() > 0); |
2172 | rcu_scheduler_active = 1; |
2173 | } |
2174 | |
2175 | /* |
2176 | * Compute the per-level fanout, either using the exact fanout specified |
2177 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. |
2178 | */ |
2179 | #ifdef CONFIG_RCU_FANOUT_EXACT |
2180 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
2181 | { |
2182 | int i; |
2183 | |
2184 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
2185 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
2186 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
2187 | } |
2188 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ |
2189 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
2190 | { |
2191 | int ccur; |
2192 | int cprv; |
2193 | int i; |
2194 | |
2195 | cprv = NR_CPUS; |
2196 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
2197 | ccur = rsp->levelcnt[i]; |
2198 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; |
2199 | cprv = ccur; |
2200 | } |
2201 | } |
2202 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ |
2203 | |
2204 | /* |
2205 | * Helper function for rcu_init() that initializes one rcu_state structure. |
2206 | */ |
2207 | static void __init rcu_init_one(struct rcu_state *rsp, |
2208 | struct rcu_data __percpu *rda) |
2209 | { |
2210 | static char *buf[] = { "rcu_node_level_0", |
2211 | "rcu_node_level_1", |
2212 | "rcu_node_level_2", |
2213 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ |
2214 | int cpustride = 1; |
2215 | int i; |
2216 | int j; |
2217 | struct rcu_node *rnp; |
2218 | |
2219 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2220 | |
2221 | /* Initialize the level-tracking arrays. */ |
2222 | |
2223 | for (i = 1; i < NUM_RCU_LVLS; i++) |
2224 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; |
2225 | rcu_init_levelspread(rsp); |
2226 | |
2227 | /* Initialize the elements themselves, starting from the leaves. */ |
2228 | |
2229 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
2230 | cpustride *= rsp->levelspread[i]; |
2231 | rnp = rsp->level[i]; |
2232 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { |
2233 | raw_spin_lock_init(&rnp->lock); |
2234 | lockdep_set_class_and_name(&rnp->lock, |
2235 | &rcu_node_class[i], buf[i]); |
2236 | rnp->gpnum = 0; |
2237 | rnp->qsmask = 0; |
2238 | rnp->qsmaskinit = 0; |
2239 | rnp->grplo = j * cpustride; |
2240 | rnp->grphi = (j + 1) * cpustride - 1; |
2241 | if (rnp->grphi >= NR_CPUS) |
2242 | rnp->grphi = NR_CPUS - 1; |
2243 | if (i == 0) { |
2244 | rnp->grpnum = 0; |
2245 | rnp->grpmask = 0; |
2246 | rnp->parent = NULL; |
2247 | } else { |
2248 | rnp->grpnum = j % rsp->levelspread[i - 1]; |
2249 | rnp->grpmask = 1UL << rnp->grpnum; |
2250 | rnp->parent = rsp->level[i - 1] + |
2251 | j / rsp->levelspread[i - 1]; |
2252 | } |
2253 | rnp->level = i; |
2254 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
2255 | } |
2256 | } |
2257 | |
2258 | rsp->rda = rda; |
2259 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2260 | for_each_possible_cpu(i) { |
2261 | while (i > rnp->grphi) |
2262 | rnp++; |
2263 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
2264 | rcu_boot_init_percpu_data(i, rsp); |
2265 | } |
2266 | } |
2267 | |
2268 | void __init rcu_init(void) |
2269 | { |
2270 | int cpu; |
2271 | |
2272 | rcu_bootup_announce(); |
2273 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2274 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); |
2275 | __rcu_init_preempt(); |
2276 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
2277 | |
2278 | /* |
2279 | * We don't need protection against CPU-hotplug here because |
2280 | * this is called early in boot, before either interrupts |
2281 | * or the scheduler are operational. |
2282 | */ |
2283 | cpu_notifier(rcu_cpu_notify, 0); |
2284 | for_each_online_cpu(cpu) |
2285 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); |
2286 | check_cpu_stall_init(); |
2287 | } |
2288 | |
2289 | #include "rcutree_plugin.h" |
2290 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9