]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/softirq.c
irq: Remove IRQ_EXIT_OFFSET workaround
[mirror_ubuntu-bionic-kernel.git] / kernel / softirq.c
CommitLineData
1da177e4
LT
1/*
2 * linux/kernel/softirq.c
3 *
4 * Copyright (C) 1992 Linus Torvalds
5 *
b10db7f0
PM
6 * Distribute under GPLv2.
7 *
8 * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
54514a70
DM
9 *
10 * Remote softirq infrastructure is by Jens Axboe.
1da177e4
LT
11 */
12
9984de1a 13#include <linux/export.h>
1da177e4
LT
14#include <linux/kernel_stat.h>
15#include <linux/interrupt.h>
16#include <linux/init.h>
17#include <linux/mm.h>
18#include <linux/notifier.h>
19#include <linux/percpu.h>
20#include <linux/cpu.h>
83144186 21#include <linux/freezer.h>
1da177e4
LT
22#include <linux/kthread.h>
23#include <linux/rcupdate.h>
7e49fcce 24#include <linux/ftrace.h>
78eef01b 25#include <linux/smp.h>
3e339b5d 26#include <linux/smpboot.h>
79bf2bb3 27#include <linux/tick.h>
a0e39ed3
HC
28
29#define CREATE_TRACE_POINTS
ad8d75ff 30#include <trace/events/irq.h>
1da177e4
LT
31
32#include <asm/irq.h>
33/*
34 - No shared variables, all the data are CPU local.
35 - If a softirq needs serialization, let it serialize itself
36 by its own spinlocks.
37 - Even if softirq is serialized, only local cpu is marked for
38 execution. Hence, we get something sort of weak cpu binding.
39 Though it is still not clear, will it result in better locality
40 or will not.
41
42 Examples:
43 - NET RX softirq. It is multithreaded and does not require
44 any global serialization.
45 - NET TX softirq. It kicks software netdevice queues, hence
46 it is logically serialized per device, but this serialization
47 is invisible to common code.
48 - Tasklets: serialized wrt itself.
49 */
50
51#ifndef __ARCH_IRQ_STAT
52irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
53EXPORT_SYMBOL(irq_stat);
54#endif
55
978b0116 56static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
1da177e4 57
4dd53d89 58DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
1da177e4 59
5d592b44 60char *softirq_to_name[NR_SOFTIRQS] = {
5dd4de58 61 "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL",
09223371 62 "TASKLET", "SCHED", "HRTIMER", "RCU"
5d592b44
JB
63};
64
1da177e4
LT
65/*
66 * we cannot loop indefinitely here to avoid userspace starvation,
67 * but we also don't want to introduce a worst case 1/HZ latency
68 * to the pending events, so lets the scheduler to balance
69 * the softirq load for us.
70 */
676cb02d 71static void wakeup_softirqd(void)
1da177e4
LT
72{
73 /* Interrupts are disabled: no need to stop preemption */
909ea964 74 struct task_struct *tsk = __this_cpu_read(ksoftirqd);
1da177e4
LT
75
76 if (tsk && tsk->state != TASK_RUNNING)
77 wake_up_process(tsk);
78}
79
75e1056f
VP
80/*
81 * preempt_count and SOFTIRQ_OFFSET usage:
82 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
83 * softirq processing.
84 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
85 * on local_bh_disable or local_bh_enable.
86 * This lets us distinguish between whether we are currently processing
87 * softirq and whether we just have bh disabled.
88 */
89
de30a2b3
IM
90/*
91 * This one is for softirq.c-internal use,
92 * where hardirqs are disabled legitimately:
93 */
3c829c36 94#ifdef CONFIG_TRACE_IRQFLAGS
75e1056f 95static void __local_bh_disable(unsigned long ip, unsigned int cnt)
de30a2b3
IM
96{
97 unsigned long flags;
98
99 WARN_ON_ONCE(in_irq());
100
101 raw_local_irq_save(flags);
7e49fcce
SR
102 /*
103 * The preempt tracer hooks into add_preempt_count and will break
104 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
105 * is set and before current->softirq_enabled is cleared.
106 * We must manually increment preempt_count here and manually
107 * call the trace_preempt_off later.
108 */
75e1056f 109 preempt_count() += cnt;
de30a2b3
IM
110 /*
111 * Were softirqs turned off above:
112 */
75e1056f 113 if (softirq_count() == cnt)
de30a2b3
IM
114 trace_softirqs_off(ip);
115 raw_local_irq_restore(flags);
7e49fcce 116
75e1056f 117 if (preempt_count() == cnt)
7e49fcce 118 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
de30a2b3 119}
3c829c36 120#else /* !CONFIG_TRACE_IRQFLAGS */
75e1056f 121static inline void __local_bh_disable(unsigned long ip, unsigned int cnt)
3c829c36 122{
75e1056f 123 add_preempt_count(cnt);
3c829c36
TC
124 barrier();
125}
126#endif /* CONFIG_TRACE_IRQFLAGS */
de30a2b3
IM
127
128void local_bh_disable(void)
129{
75e1056f
VP
130 __local_bh_disable((unsigned long)__builtin_return_address(0),
131 SOFTIRQ_DISABLE_OFFSET);
de30a2b3
IM
132}
133
134EXPORT_SYMBOL(local_bh_disable);
135
75e1056f
VP
136static void __local_bh_enable(unsigned int cnt)
137{
138 WARN_ON_ONCE(in_irq());
139 WARN_ON_ONCE(!irqs_disabled());
140
141 if (softirq_count() == cnt)
142 trace_softirqs_on((unsigned long)__builtin_return_address(0));
143 sub_preempt_count(cnt);
144}
145
de30a2b3
IM
146/*
147 * Special-case - softirqs can safely be enabled in
148 * cond_resched_softirq(), or by __do_softirq(),
149 * without processing still-pending softirqs:
150 */
151void _local_bh_enable(void)
152{
75e1056f 153 __local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
de30a2b3
IM
154}
155
156EXPORT_SYMBOL(_local_bh_enable);
157
0f476b6d 158static inline void _local_bh_enable_ip(unsigned long ip)
de30a2b3 159{
0f476b6d 160 WARN_ON_ONCE(in_irq() || irqs_disabled());
3c829c36 161#ifdef CONFIG_TRACE_IRQFLAGS
0f476b6d 162 local_irq_disable();
3c829c36 163#endif
de30a2b3
IM
164 /*
165 * Are softirqs going to be turned on now:
166 */
75e1056f 167 if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
0f476b6d 168 trace_softirqs_on(ip);
de30a2b3
IM
169 /*
170 * Keep preemption disabled until we are done with
171 * softirq processing:
172 */
75e1056f 173 sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1);
de30a2b3
IM
174
175 if (unlikely(!in_interrupt() && local_softirq_pending()))
176 do_softirq();
177
178 dec_preempt_count();
3c829c36 179#ifdef CONFIG_TRACE_IRQFLAGS
0f476b6d 180 local_irq_enable();
3c829c36 181#endif
de30a2b3
IM
182 preempt_check_resched();
183}
0f476b6d
JB
184
185void local_bh_enable(void)
186{
187 _local_bh_enable_ip((unsigned long)__builtin_return_address(0));
188}
de30a2b3
IM
189EXPORT_SYMBOL(local_bh_enable);
190
191void local_bh_enable_ip(unsigned long ip)
192{
0f476b6d 193 _local_bh_enable_ip(ip);
de30a2b3
IM
194}
195EXPORT_SYMBOL(local_bh_enable_ip);
196
1da177e4
LT
197/*
198 * We restart softirq processing MAX_SOFTIRQ_RESTART times,
199 * and we fall back to softirqd after that.
200 *
201 * This number has been established via experimentation.
202 * The two things to balance is latency against fairness -
203 * we want to handle softirqs as soon as possible, but they
204 * should not be able to lock up the box.
205 */
206#define MAX_SOFTIRQ_RESTART 10
207
208asmlinkage void __do_softirq(void)
209{
210 struct softirq_action *h;
211 __u32 pending;
212 int max_restart = MAX_SOFTIRQ_RESTART;
213 int cpu;
907aed48
MG
214 unsigned long old_flags = current->flags;
215
216 /*
217 * Mask out PF_MEMALLOC s current task context is borrowed for the
218 * softirq. A softirq handled such as network RX might set PF_MEMALLOC
219 * again if the socket is related to swap
220 */
221 current->flags &= ~PF_MEMALLOC;
1da177e4
LT
222
223 pending = local_softirq_pending();
6a61671b 224 account_irq_enter_time(current);
829035fd 225
75e1056f
VP
226 __local_bh_disable((unsigned long)__builtin_return_address(0),
227 SOFTIRQ_OFFSET);
d820ac4c 228 lockdep_softirq_enter();
1da177e4 229
1da177e4
LT
230 cpu = smp_processor_id();
231restart:
232 /* Reset the pending bitmask before enabling irqs */
3f74478b 233 set_softirq_pending(0);
1da177e4 234
c70f5d66 235 local_irq_enable();
1da177e4
LT
236
237 h = softirq_vec;
238
239 do {
240 if (pending & 1) {
f4bc6bb2 241 unsigned int vec_nr = h - softirq_vec;
8e85b4b5
TG
242 int prev_count = preempt_count();
243
f4bc6bb2
TG
244 kstat_incr_softirqs_this_cpu(vec_nr);
245
246 trace_softirq_entry(vec_nr);
1da177e4 247 h->action(h);
f4bc6bb2 248 trace_softirq_exit(vec_nr);
8e85b4b5 249 if (unlikely(prev_count != preempt_count())) {
f4bc6bb2 250 printk(KERN_ERR "huh, entered softirq %u %s %p"
8e85b4b5 251 "with preempt_count %08x,"
f4bc6bb2
TG
252 " exited with %08x?\n", vec_nr,
253 softirq_to_name[vec_nr], h->action,
254 prev_count, preempt_count());
8e85b4b5
TG
255 preempt_count() = prev_count;
256 }
257
d6714c22 258 rcu_bh_qs(cpu);
1da177e4
LT
259 }
260 h++;
261 pending >>= 1;
262 } while (pending);
263
c70f5d66 264 local_irq_disable();
1da177e4
LT
265
266 pending = local_softirq_pending();
267 if (pending && --max_restart)
268 goto restart;
269
270 if (pending)
271 wakeup_softirqd();
272
d820ac4c 273 lockdep_softirq_exit();
829035fd 274
6a61671b 275 account_irq_exit_time(current);
75e1056f 276 __local_bh_enable(SOFTIRQ_OFFSET);
907aed48 277 tsk_restore_flags(current, old_flags, PF_MEMALLOC);
1da177e4
LT
278}
279
280#ifndef __ARCH_HAS_DO_SOFTIRQ
281
282asmlinkage void do_softirq(void)
283{
284 __u32 pending;
285 unsigned long flags;
286
287 if (in_interrupt())
288 return;
289
290 local_irq_save(flags);
291
292 pending = local_softirq_pending();
293
294 if (pending)
295 __do_softirq();
296
297 local_irq_restore(flags);
298}
299
1da177e4
LT
300#endif
301
dde4b2b5
IM
302/*
303 * Enter an interrupt context.
304 */
305void irq_enter(void)
306{
6378ddb5 307 int cpu = smp_processor_id();
719254fa 308
64db4cff 309 rcu_irq_enter();
0a8a2e78 310 if (is_idle_task(current) && !in_interrupt()) {
d267f87f
VP
311 /*
312 * Prevent raise_softirq from needlessly waking up ksoftirqd
313 * here, as softirq will be serviced on return from interrupt.
314 */
315 local_bh_disable();
719254fa 316 tick_check_idle(cpu);
d267f87f
VP
317 _local_bh_enable();
318 }
319
320 __irq_enter();
dde4b2b5
IM
321}
322
8d32a307
TG
323static inline void invoke_softirq(void)
324{
facd8b80 325 if (!force_irqthreads)
8d32a307 326 __do_softirq();
facd8b80 327 else
8d32a307
TG
328 wakeup_softirqd();
329}
1da177e4
LT
330
331/*
332 * Exit an interrupt context. Process softirqs if needed and possible:
333 */
334void irq_exit(void)
335{
74eed016
TG
336#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
337 unsigned long flags;
338
339 local_irq_save(flags);
340#else
341 WARN_ON_ONCE(!irqs_disabled());
342#endif
343
6a61671b 344 account_irq_exit_time(current);
de30a2b3 345 trace_hardirq_exit();
4d4c4e24 346 sub_preempt_count(HARDIRQ_OFFSET);
1da177e4
LT
347 if (!in_interrupt() && local_softirq_pending())
348 invoke_softirq();
79bf2bb3
TG
349
350#ifdef CONFIG_NO_HZ
351 /* Make sure that timer wheel updates are propagated */
64db4cff 352 if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched())
280f0677 353 tick_nohz_irq_exit();
79bf2bb3 354#endif
416eb33c 355 rcu_irq_exit();
74eed016
TG
356#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
357 local_irq_restore(flags);
358#endif
1da177e4
LT
359}
360
361/*
362 * This function must run with irqs disabled!
363 */
7ad5b3a5 364inline void raise_softirq_irqoff(unsigned int nr)
1da177e4
LT
365{
366 __raise_softirq_irqoff(nr);
367
368 /*
369 * If we're in an interrupt or softirq, we're done
370 * (this also catches softirq-disabled code). We will
371 * actually run the softirq once we return from
372 * the irq or softirq.
373 *
374 * Otherwise we wake up ksoftirqd to make sure we
375 * schedule the softirq soon.
376 */
377 if (!in_interrupt())
378 wakeup_softirqd();
379}
380
7ad5b3a5 381void raise_softirq(unsigned int nr)
1da177e4
LT
382{
383 unsigned long flags;
384
385 local_irq_save(flags);
386 raise_softirq_irqoff(nr);
387 local_irq_restore(flags);
388}
389
f069686e
SR
390void __raise_softirq_irqoff(unsigned int nr)
391{
392 trace_softirq_raise(nr);
393 or_softirq_pending(1UL << nr);
394}
395
962cf36c 396void open_softirq(int nr, void (*action)(struct softirq_action *))
1da177e4 397{
1da177e4
LT
398 softirq_vec[nr].action = action;
399}
400
9ba5f005
PZ
401/*
402 * Tasklets
403 */
1da177e4
LT
404struct tasklet_head
405{
48f20a9a
OJ
406 struct tasklet_struct *head;
407 struct tasklet_struct **tail;
1da177e4
LT
408};
409
4620b49f
VN
410static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
411static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
1da177e4 412
7ad5b3a5 413void __tasklet_schedule(struct tasklet_struct *t)
1da177e4
LT
414{
415 unsigned long flags;
416
417 local_irq_save(flags);
48f20a9a 418 t->next = NULL;
909ea964
CL
419 *__this_cpu_read(tasklet_vec.tail) = t;
420 __this_cpu_write(tasklet_vec.tail, &(t->next));
1da177e4
LT
421 raise_softirq_irqoff(TASKLET_SOFTIRQ);
422 local_irq_restore(flags);
423}
424
425EXPORT_SYMBOL(__tasklet_schedule);
426
7ad5b3a5 427void __tasklet_hi_schedule(struct tasklet_struct *t)
1da177e4
LT
428{
429 unsigned long flags;
430
431 local_irq_save(flags);
48f20a9a 432 t->next = NULL;
909ea964
CL
433 *__this_cpu_read(tasklet_hi_vec.tail) = t;
434 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
1da177e4
LT
435 raise_softirq_irqoff(HI_SOFTIRQ);
436 local_irq_restore(flags);
437}
438
439EXPORT_SYMBOL(__tasklet_hi_schedule);
440
7c692cba
VN
441void __tasklet_hi_schedule_first(struct tasklet_struct *t)
442{
443 BUG_ON(!irqs_disabled());
444
909ea964
CL
445 t->next = __this_cpu_read(tasklet_hi_vec.head);
446 __this_cpu_write(tasklet_hi_vec.head, t);
7c692cba
VN
447 __raise_softirq_irqoff(HI_SOFTIRQ);
448}
449
450EXPORT_SYMBOL(__tasklet_hi_schedule_first);
451
1da177e4
LT
452static void tasklet_action(struct softirq_action *a)
453{
454 struct tasklet_struct *list;
455
456 local_irq_disable();
909ea964
CL
457 list = __this_cpu_read(tasklet_vec.head);
458 __this_cpu_write(tasklet_vec.head, NULL);
459 __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head);
1da177e4
LT
460 local_irq_enable();
461
462 while (list) {
463 struct tasklet_struct *t = list;
464
465 list = list->next;
466
467 if (tasklet_trylock(t)) {
468 if (!atomic_read(&t->count)) {
469 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
470 BUG();
471 t->func(t->data);
472 tasklet_unlock(t);
473 continue;
474 }
475 tasklet_unlock(t);
476 }
477
478 local_irq_disable();
48f20a9a 479 t->next = NULL;
909ea964
CL
480 *__this_cpu_read(tasklet_vec.tail) = t;
481 __this_cpu_write(tasklet_vec.tail, &(t->next));
1da177e4
LT
482 __raise_softirq_irqoff(TASKLET_SOFTIRQ);
483 local_irq_enable();
484 }
485}
486
487static void tasklet_hi_action(struct softirq_action *a)
488{
489 struct tasklet_struct *list;
490
491 local_irq_disable();
909ea964
CL
492 list = __this_cpu_read(tasklet_hi_vec.head);
493 __this_cpu_write(tasklet_hi_vec.head, NULL);
494 __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head);
1da177e4
LT
495 local_irq_enable();
496
497 while (list) {
498 struct tasklet_struct *t = list;
499
500 list = list->next;
501
502 if (tasklet_trylock(t)) {
503 if (!atomic_read(&t->count)) {
504 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
505 BUG();
506 t->func(t->data);
507 tasklet_unlock(t);
508 continue;
509 }
510 tasklet_unlock(t);
511 }
512
513 local_irq_disable();
48f20a9a 514 t->next = NULL;
909ea964
CL
515 *__this_cpu_read(tasklet_hi_vec.tail) = t;
516 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
1da177e4
LT
517 __raise_softirq_irqoff(HI_SOFTIRQ);
518 local_irq_enable();
519 }
520}
521
522
523void tasklet_init(struct tasklet_struct *t,
524 void (*func)(unsigned long), unsigned long data)
525{
526 t->next = NULL;
527 t->state = 0;
528 atomic_set(&t->count, 0);
529 t->func = func;
530 t->data = data;
531}
532
533EXPORT_SYMBOL(tasklet_init);
534
535void tasklet_kill(struct tasklet_struct *t)
536{
537 if (in_interrupt())
538 printk("Attempt to kill tasklet from interrupt\n");
539
540 while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
79d381c9 541 do {
1da177e4 542 yield();
79d381c9 543 } while (test_bit(TASKLET_STATE_SCHED, &t->state));
1da177e4
LT
544 }
545 tasklet_unlock_wait(t);
546 clear_bit(TASKLET_STATE_SCHED, &t->state);
547}
548
549EXPORT_SYMBOL(tasklet_kill);
550
9ba5f005
PZ
551/*
552 * tasklet_hrtimer
553 */
554
555/*
b9c30322
PZ
556 * The trampoline is called when the hrtimer expires. It schedules a tasklet
557 * to run __tasklet_hrtimer_trampoline() which in turn will call the intended
558 * hrtimer callback, but from softirq context.
9ba5f005
PZ
559 */
560static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
561{
562 struct tasklet_hrtimer *ttimer =
563 container_of(timer, struct tasklet_hrtimer, timer);
564
b9c30322
PZ
565 tasklet_hi_schedule(&ttimer->tasklet);
566 return HRTIMER_NORESTART;
9ba5f005
PZ
567}
568
569/*
570 * Helper function which calls the hrtimer callback from
571 * tasklet/softirq context
572 */
573static void __tasklet_hrtimer_trampoline(unsigned long data)
574{
575 struct tasklet_hrtimer *ttimer = (void *)data;
576 enum hrtimer_restart restart;
577
578 restart = ttimer->function(&ttimer->timer);
579 if (restart != HRTIMER_NORESTART)
580 hrtimer_restart(&ttimer->timer);
581}
582
583/**
584 * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
585 * @ttimer: tasklet_hrtimer which is initialized
25985edc 586 * @function: hrtimer callback function which gets called from softirq context
9ba5f005
PZ
587 * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
588 * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
589 */
590void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
591 enum hrtimer_restart (*function)(struct hrtimer *),
592 clockid_t which_clock, enum hrtimer_mode mode)
593{
594 hrtimer_init(&ttimer->timer, which_clock, mode);
595 ttimer->timer.function = __hrtimer_tasklet_trampoline;
596 tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
597 (unsigned long)ttimer);
598 ttimer->function = function;
599}
600EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
601
602/*
603 * Remote softirq bits
604 */
605
54514a70
DM
606DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list);
607EXPORT_PER_CPU_SYMBOL(softirq_work_list);
608
609static void __local_trigger(struct call_single_data *cp, int softirq)
610{
611 struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]);
612
613 list_add_tail(&cp->list, head);
614
615 /* Trigger the softirq only if the list was previously empty. */
616 if (head->next == &cp->list)
617 raise_softirq_irqoff(softirq);
618}
619
620#ifdef CONFIG_USE_GENERIC_SMP_HELPERS
621static void remote_softirq_receive(void *data)
622{
623 struct call_single_data *cp = data;
624 unsigned long flags;
625 int softirq;
626
627 softirq = cp->priv;
628
629 local_irq_save(flags);
630 __local_trigger(cp, softirq);
631 local_irq_restore(flags);
632}
633
634static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
635{
636 if (cpu_online(cpu)) {
637 cp->func = remote_softirq_receive;
638 cp->info = cp;
639 cp->flags = 0;
640 cp->priv = softirq;
641
6e275637 642 __smp_call_function_single(cpu, cp, 0);
54514a70
DM
643 return 0;
644 }
645 return 1;
646}
647#else /* CONFIG_USE_GENERIC_SMP_HELPERS */
648static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
649{
650 return 1;
651}
652#endif
653
654/**
655 * __send_remote_softirq - try to schedule softirq work on a remote cpu
656 * @cp: private SMP call function data area
657 * @cpu: the remote cpu
658 * @this_cpu: the currently executing cpu
659 * @softirq: the softirq for the work
660 *
661 * Attempt to schedule softirq work on a remote cpu. If this cannot be
662 * done, the work is instead queued up on the local cpu.
663 *
664 * Interrupts must be disabled.
665 */
666void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq)
667{
668 if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq))
669 __local_trigger(cp, softirq);
670}
671EXPORT_SYMBOL(__send_remote_softirq);
672
673/**
674 * send_remote_softirq - try to schedule softirq work on a remote cpu
675 * @cp: private SMP call function data area
676 * @cpu: the remote cpu
677 * @softirq: the softirq for the work
678 *
679 * Like __send_remote_softirq except that disabling interrupts and
680 * computing the current cpu is done for the caller.
681 */
682void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
683{
684 unsigned long flags;
685 int this_cpu;
686
687 local_irq_save(flags);
688 this_cpu = smp_processor_id();
689 __send_remote_softirq(cp, cpu, this_cpu, softirq);
690 local_irq_restore(flags);
691}
692EXPORT_SYMBOL(send_remote_softirq);
693
694static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self,
695 unsigned long action, void *hcpu)
696{
697 /*
698 * If a CPU goes away, splice its entries to the current CPU
699 * and trigger a run of the softirq
700 */
701 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
702 int cpu = (unsigned long) hcpu;
703 int i;
704
705 local_irq_disable();
706 for (i = 0; i < NR_SOFTIRQS; i++) {
707 struct list_head *head = &per_cpu(softirq_work_list[i], cpu);
708 struct list_head *local_head;
709
710 if (list_empty(head))
711 continue;
712
713 local_head = &__get_cpu_var(softirq_work_list[i]);
714 list_splice_init(head, local_head);
715 raise_softirq_irqoff(i);
716 }
717 local_irq_enable();
718 }
719
720 return NOTIFY_OK;
721}
722
723static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = {
724 .notifier_call = remote_softirq_cpu_notify,
725};
726
1da177e4
LT
727void __init softirq_init(void)
728{
48f20a9a
OJ
729 int cpu;
730
731 for_each_possible_cpu(cpu) {
54514a70
DM
732 int i;
733
48f20a9a
OJ
734 per_cpu(tasklet_vec, cpu).tail =
735 &per_cpu(tasklet_vec, cpu).head;
736 per_cpu(tasklet_hi_vec, cpu).tail =
737 &per_cpu(tasklet_hi_vec, cpu).head;
54514a70
DM
738 for (i = 0; i < NR_SOFTIRQS; i++)
739 INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu));
48f20a9a
OJ
740 }
741
54514a70
DM
742 register_hotcpu_notifier(&remote_softirq_cpu_notifier);
743
962cf36c
CM
744 open_softirq(TASKLET_SOFTIRQ, tasklet_action);
745 open_softirq(HI_SOFTIRQ, tasklet_hi_action);
1da177e4
LT
746}
747
3e339b5d 748static int ksoftirqd_should_run(unsigned int cpu)
1da177e4 749{
3e339b5d
TG
750 return local_softirq_pending();
751}
1da177e4 752
3e339b5d
TG
753static void run_ksoftirqd(unsigned int cpu)
754{
755 local_irq_disable();
756 if (local_softirq_pending()) {
757 __do_softirq();
758 rcu_note_context_switch(cpu);
759 local_irq_enable();
760 cond_resched();
761 return;
1da177e4 762 }
3e339b5d 763 local_irq_enable();
1da177e4
LT
764}
765
766#ifdef CONFIG_HOTPLUG_CPU
767/*
768 * tasklet_kill_immediate is called to remove a tasklet which can already be
769 * scheduled for execution on @cpu.
770 *
771 * Unlike tasklet_kill, this function removes the tasklet
772 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
773 *
774 * When this function is called, @cpu must be in the CPU_DEAD state.
775 */
776void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
777{
778 struct tasklet_struct **i;
779
780 BUG_ON(cpu_online(cpu));
781 BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
782
783 if (!test_bit(TASKLET_STATE_SCHED, &t->state))
784 return;
785
786 /* CPU is dead, so no lock needed. */
48f20a9a 787 for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
1da177e4
LT
788 if (*i == t) {
789 *i = t->next;
48f20a9a
OJ
790 /* If this was the tail element, move the tail ptr */
791 if (*i == NULL)
792 per_cpu(tasklet_vec, cpu).tail = i;
1da177e4
LT
793 return;
794 }
795 }
796 BUG();
797}
798
799static void takeover_tasklets(unsigned int cpu)
800{
1da177e4
LT
801 /* CPU is dead, so no lock needed. */
802 local_irq_disable();
803
804 /* Find end, append list for that CPU. */
e5e41723 805 if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
909ea964
CL
806 *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
807 this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
e5e41723
CB
808 per_cpu(tasklet_vec, cpu).head = NULL;
809 per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
810 }
1da177e4
LT
811 raise_softirq_irqoff(TASKLET_SOFTIRQ);
812
e5e41723 813 if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
909ea964
CL
814 *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
815 __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
e5e41723
CB
816 per_cpu(tasklet_hi_vec, cpu).head = NULL;
817 per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
818 }
1da177e4
LT
819 raise_softirq_irqoff(HI_SOFTIRQ);
820
821 local_irq_enable();
822}
823#endif /* CONFIG_HOTPLUG_CPU */
824
8c78f307 825static int __cpuinit cpu_callback(struct notifier_block *nfb,
1da177e4
LT
826 unsigned long action,
827 void *hcpu)
828{
1da177e4 829 switch (action) {
1da177e4 830#ifdef CONFIG_HOTPLUG_CPU
1da177e4 831 case CPU_DEAD:
3e339b5d
TG
832 case CPU_DEAD_FROZEN:
833 takeover_tasklets((unsigned long)hcpu);
1da177e4
LT
834 break;
835#endif /* CONFIG_HOTPLUG_CPU */
3e339b5d 836 }
1da177e4
LT
837 return NOTIFY_OK;
838}
839
8c78f307 840static struct notifier_block __cpuinitdata cpu_nfb = {
1da177e4
LT
841 .notifier_call = cpu_callback
842};
843
3e339b5d
TG
844static struct smp_hotplug_thread softirq_threads = {
845 .store = &ksoftirqd,
846 .thread_should_run = ksoftirqd_should_run,
847 .thread_fn = run_ksoftirqd,
848 .thread_comm = "ksoftirqd/%u",
849};
850
7babe8db 851static __init int spawn_ksoftirqd(void)
1da177e4 852{
1da177e4 853 register_cpu_notifier(&cpu_nfb);
3e339b5d
TG
854
855 BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
856
1da177e4
LT
857 return 0;
858}
7babe8db 859early_initcall(spawn_ksoftirqd);
78eef01b 860
43a25632
YL
861/*
862 * [ These __weak aliases are kept in a separate compilation unit, so that
863 * GCC does not inline them incorrectly. ]
864 */
865
866int __init __weak early_irq_init(void)
867{
868 return 0;
869}
870
b683de2b 871#ifdef CONFIG_GENERIC_HARDIRQS
4a046d17
YL
872int __init __weak arch_probe_nr_irqs(void)
873{
b683de2b 874 return NR_IRQS_LEGACY;
4a046d17
YL
875}
876
43a25632
YL
877int __init __weak arch_early_irq_init(void)
878{
879 return 0;
880}
b683de2b 881#endif