Remove argument from open_softirq which is always NULL
[linux-2.6.git] / kernel / softirq.c
1 /*
2  *      linux/kernel/softirq.c
3  *
4  *      Copyright (C) 1992 Linus Torvalds
5  *
6  *      Distribute under GPLv2.
7  *
8  *      Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
9  */
10
11 #include <linux/module.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/interrupt.h>
14 #include <linux/init.h>
15 #include <linux/mm.h>
16 #include <linux/notifier.h>
17 #include <linux/percpu.h>
18 #include <linux/cpu.h>
19 #include <linux/freezer.h>
20 #include <linux/kthread.h>
21 #include <linux/rcupdate.h>
22 #include <linux/smp.h>
23 #include <linux/tick.h>
24
25 #include <asm/irq.h>
26 /*
27    - No shared variables, all the data are CPU local.
28    - If a softirq needs serialization, let it serialize itself
29      by its own spinlocks.
30    - Even if softirq is serialized, only local cpu is marked for
31      execution. Hence, we get something sort of weak cpu binding.
32      Though it is still not clear, will it result in better locality
33      or will not.
34
35    Examples:
36    - NET RX softirq. It is multithreaded and does not require
37      any global serialization.
38    - NET TX softirq. It kicks software netdevice queues, hence
39      it is logically serialized per device, but this serialization
40      is invisible to common code.
41    - Tasklets: serialized wrt itself.
42  */
43
44 #ifndef __ARCH_IRQ_STAT
45 irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
46 EXPORT_SYMBOL(irq_stat);
47 #endif
48
49 static struct softirq_action softirq_vec[32] __cacheline_aligned_in_smp;
50
51 static DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
52
53 /*
54  * we cannot loop indefinitely here to avoid userspace starvation,
55  * but we also don't want to introduce a worst case 1/HZ latency
56  * to the pending events, so lets the scheduler to balance
57  * the softirq load for us.
58  */
59 static inline void wakeup_softirqd(void)
60 {
61         /* Interrupts are disabled: no need to stop preemption */
62         struct task_struct *tsk = __get_cpu_var(ksoftirqd);
63
64         if (tsk && tsk->state != TASK_RUNNING)
65                 wake_up_process(tsk);
66 }
67
68 /*
69  * This one is for softirq.c-internal use,
70  * where hardirqs are disabled legitimately:
71  */
72 #ifdef CONFIG_TRACE_IRQFLAGS
73 static void __local_bh_disable(unsigned long ip)
74 {
75         unsigned long flags;
76
77         WARN_ON_ONCE(in_irq());
78
79         raw_local_irq_save(flags);
80         add_preempt_count(SOFTIRQ_OFFSET);
81         /*
82          * Were softirqs turned off above:
83          */
84         if (softirq_count() == SOFTIRQ_OFFSET)
85                 trace_softirqs_off(ip);
86         raw_local_irq_restore(flags);
87 }
88 #else /* !CONFIG_TRACE_IRQFLAGS */
89 static inline void __local_bh_disable(unsigned long ip)
90 {
91         add_preempt_count(SOFTIRQ_OFFSET);
92         barrier();
93 }
94 #endif /* CONFIG_TRACE_IRQFLAGS */
95
96 void local_bh_disable(void)
97 {
98         __local_bh_disable((unsigned long)__builtin_return_address(0));
99 }
100
101 EXPORT_SYMBOL(local_bh_disable);
102
103 void __local_bh_enable(void)
104 {
105         WARN_ON_ONCE(in_irq());
106
107         /*
108          * softirqs should never be enabled by __local_bh_enable(),
109          * it always nests inside local_bh_enable() sections:
110          */
111         WARN_ON_ONCE(softirq_count() == SOFTIRQ_OFFSET);
112
113         sub_preempt_count(SOFTIRQ_OFFSET);
114 }
115 EXPORT_SYMBOL_GPL(__local_bh_enable);
116
117 /*
118  * Special-case - softirqs can safely be enabled in
119  * cond_resched_softirq(), or by __do_softirq(),
120  * without processing still-pending softirqs:
121  */
122 void _local_bh_enable(void)
123 {
124         WARN_ON_ONCE(in_irq());
125         WARN_ON_ONCE(!irqs_disabled());
126
127         if (softirq_count() == SOFTIRQ_OFFSET)
128                 trace_softirqs_on((unsigned long)__builtin_return_address(0));
129         sub_preempt_count(SOFTIRQ_OFFSET);
130 }
131
132 EXPORT_SYMBOL(_local_bh_enable);
133
134 void local_bh_enable(void)
135 {
136 #ifdef CONFIG_TRACE_IRQFLAGS
137         unsigned long flags;
138
139         WARN_ON_ONCE(in_irq());
140 #endif
141         WARN_ON_ONCE(irqs_disabled());
142
143 #ifdef CONFIG_TRACE_IRQFLAGS
144         local_irq_save(flags);
145 #endif
146         /*
147          * Are softirqs going to be turned on now:
148          */
149         if (softirq_count() == SOFTIRQ_OFFSET)
150                 trace_softirqs_on((unsigned long)__builtin_return_address(0));
151         /*
152          * Keep preemption disabled until we are done with
153          * softirq processing:
154          */
155         sub_preempt_count(SOFTIRQ_OFFSET - 1);
156
157         if (unlikely(!in_interrupt() && local_softirq_pending()))
158                 do_softirq();
159
160         dec_preempt_count();
161 #ifdef CONFIG_TRACE_IRQFLAGS
162         local_irq_restore(flags);
163 #endif
164         preempt_check_resched();
165 }
166 EXPORT_SYMBOL(local_bh_enable);
167
168 void local_bh_enable_ip(unsigned long ip)
169 {
170 #ifdef CONFIG_TRACE_IRQFLAGS
171         unsigned long flags;
172
173         WARN_ON_ONCE(in_irq());
174
175         local_irq_save(flags);
176 #endif
177         /*
178          * Are softirqs going to be turned on now:
179          */
180         if (softirq_count() == SOFTIRQ_OFFSET)
181                 trace_softirqs_on(ip);
182         /*
183          * Keep preemption disabled until we are done with
184          * softirq processing:
185          */
186         sub_preempt_count(SOFTIRQ_OFFSET - 1);
187
188         if (unlikely(!in_interrupt() && local_softirq_pending()))
189                 do_softirq();
190
191         dec_preempt_count();
192 #ifdef CONFIG_TRACE_IRQFLAGS
193         local_irq_restore(flags);
194 #endif
195         preempt_check_resched();
196 }
197 EXPORT_SYMBOL(local_bh_enable_ip);
198
199 /*
200  * We restart softirq processing MAX_SOFTIRQ_RESTART times,
201  * and we fall back to softirqd after that.
202  *
203  * This number has been established via experimentation.
204  * The two things to balance is latency against fairness -
205  * we want to handle softirqs as soon as possible, but they
206  * should not be able to lock up the box.
207  */
208 #define MAX_SOFTIRQ_RESTART 10
209
210 asmlinkage void __do_softirq(void)
211 {
212         struct softirq_action *h;
213         __u32 pending;
214         int max_restart = MAX_SOFTIRQ_RESTART;
215         int cpu;
216
217         pending = local_softirq_pending();
218         account_system_vtime(current);
219
220         __local_bh_disable((unsigned long)__builtin_return_address(0));
221         trace_softirq_enter();
222
223         cpu = smp_processor_id();
224 restart:
225         /* Reset the pending bitmask before enabling irqs */
226         set_softirq_pending(0);
227
228         local_irq_enable();
229
230         h = softirq_vec;
231
232         do {
233                 if (pending & 1) {
234                         h->action(h);
235                         rcu_bh_qsctr_inc(cpu);
236                 }
237                 h++;
238                 pending >>= 1;
239         } while (pending);
240
241         local_irq_disable();
242
243         pending = local_softirq_pending();
244         if (pending && --max_restart)
245                 goto restart;
246
247         if (pending)
248                 wakeup_softirqd();
249
250         trace_softirq_exit();
251
252         account_system_vtime(current);
253         _local_bh_enable();
254 }
255
256 #ifndef __ARCH_HAS_DO_SOFTIRQ
257
258 asmlinkage void do_softirq(void)
259 {
260         __u32 pending;
261         unsigned long flags;
262
263         if (in_interrupt())
264                 return;
265
266         local_irq_save(flags);
267
268         pending = local_softirq_pending();
269
270         if (pending)
271                 __do_softirq();
272
273         local_irq_restore(flags);
274 }
275
276 #endif
277
278 /*
279  * Enter an interrupt context.
280  */
281 void irq_enter(void)
282 {
283 #ifdef CONFIG_NO_HZ
284         int cpu = smp_processor_id();
285         if (idle_cpu(cpu) && !in_interrupt())
286                 tick_nohz_stop_idle(cpu);
287 #endif
288         __irq_enter();
289 #ifdef CONFIG_NO_HZ
290         if (idle_cpu(cpu))
291                 tick_nohz_update_jiffies();
292 #endif
293 }
294
295 #ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED
296 # define invoke_softirq()       __do_softirq()
297 #else
298 # define invoke_softirq()       do_softirq()
299 #endif
300
301 /*
302  * Exit an interrupt context. Process softirqs if needed and possible:
303  */
304 void irq_exit(void)
305 {
306         account_system_vtime(current);
307         trace_hardirq_exit();
308         sub_preempt_count(IRQ_EXIT_OFFSET);
309         if (!in_interrupt() && local_softirq_pending())
310                 invoke_softirq();
311
312 #ifdef CONFIG_NO_HZ
313         /* Make sure that timer wheel updates are propagated */
314         if (!in_interrupt() && idle_cpu(smp_processor_id()) && !need_resched())
315                 tick_nohz_stop_sched_tick();
316         rcu_irq_exit();
317 #endif
318         preempt_enable_no_resched();
319 }
320
321 /*
322  * This function must run with irqs disabled!
323  */
324 inline void raise_softirq_irqoff(unsigned int nr)
325 {
326         __raise_softirq_irqoff(nr);
327
328         /*
329          * If we're in an interrupt or softirq, we're done
330          * (this also catches softirq-disabled code). We will
331          * actually run the softirq once we return from
332          * the irq or softirq.
333          *
334          * Otherwise we wake up ksoftirqd to make sure we
335          * schedule the softirq soon.
336          */
337         if (!in_interrupt())
338                 wakeup_softirqd();
339 }
340
341 void raise_softirq(unsigned int nr)
342 {
343         unsigned long flags;
344
345         local_irq_save(flags);
346         raise_softirq_irqoff(nr);
347         local_irq_restore(flags);
348 }
349
350 void open_softirq(int nr, void (*action)(struct softirq_action *))
351 {
352         softirq_vec[nr].action = action;
353 }
354
355 /* Tasklets */
356 struct tasklet_head
357 {
358         struct tasklet_struct *head;
359         struct tasklet_struct **tail;
360 };
361
362 /* Some compilers disobey section attribute on statics when not
363    initialized -- RR */
364 static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec) = { NULL };
365 static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec) = { NULL };
366
367 void __tasklet_schedule(struct tasklet_struct *t)
368 {
369         unsigned long flags;
370
371         local_irq_save(flags);
372         t->next = NULL;
373         *__get_cpu_var(tasklet_vec).tail = t;
374         __get_cpu_var(tasklet_vec).tail = &(t->next);
375         raise_softirq_irqoff(TASKLET_SOFTIRQ);
376         local_irq_restore(flags);
377 }
378
379 EXPORT_SYMBOL(__tasklet_schedule);
380
381 void __tasklet_hi_schedule(struct tasklet_struct *t)
382 {
383         unsigned long flags;
384
385         local_irq_save(flags);
386         t->next = NULL;
387         *__get_cpu_var(tasklet_hi_vec).tail = t;
388         __get_cpu_var(tasklet_hi_vec).tail = &(t->next);
389         raise_softirq_irqoff(HI_SOFTIRQ);
390         local_irq_restore(flags);
391 }
392
393 EXPORT_SYMBOL(__tasklet_hi_schedule);
394
395 static void tasklet_action(struct softirq_action *a)
396 {
397         struct tasklet_struct *list;
398
399         local_irq_disable();
400         list = __get_cpu_var(tasklet_vec).head;
401         __get_cpu_var(tasklet_vec).head = NULL;
402         __get_cpu_var(tasklet_vec).tail = &__get_cpu_var(tasklet_vec).head;
403         local_irq_enable();
404
405         while (list) {
406                 struct tasklet_struct *t = list;
407
408                 list = list->next;
409
410                 if (tasklet_trylock(t)) {
411                         if (!atomic_read(&t->count)) {
412                                 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
413                                         BUG();
414                                 t->func(t->data);
415                                 tasklet_unlock(t);
416                                 continue;
417                         }
418                         tasklet_unlock(t);
419                 }
420
421                 local_irq_disable();
422                 t->next = NULL;
423                 *__get_cpu_var(tasklet_vec).tail = t;
424                 __get_cpu_var(tasklet_vec).tail = &(t->next);
425                 __raise_softirq_irqoff(TASKLET_SOFTIRQ);
426                 local_irq_enable();
427         }
428 }
429
430 static void tasklet_hi_action(struct softirq_action *a)
431 {
432         struct tasklet_struct *list;
433
434         local_irq_disable();
435         list = __get_cpu_var(tasklet_hi_vec).head;
436         __get_cpu_var(tasklet_hi_vec).head = NULL;
437         __get_cpu_var(tasklet_hi_vec).tail = &__get_cpu_var(tasklet_hi_vec).head;
438         local_irq_enable();
439
440         while (list) {
441                 struct tasklet_struct *t = list;
442
443                 list = list->next;
444
445                 if (tasklet_trylock(t)) {
446                         if (!atomic_read(&t->count)) {
447                                 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
448                                         BUG();
449                                 t->func(t->data);
450                                 tasklet_unlock(t);
451                                 continue;
452                         }
453                         tasklet_unlock(t);
454                 }
455
456                 local_irq_disable();
457                 t->next = NULL;
458                 *__get_cpu_var(tasklet_hi_vec).tail = t;
459                 __get_cpu_var(tasklet_hi_vec).tail = &(t->next);
460                 __raise_softirq_irqoff(HI_SOFTIRQ);
461                 local_irq_enable();
462         }
463 }
464
465
466 void tasklet_init(struct tasklet_struct *t,
467                   void (*func)(unsigned long), unsigned long data)
468 {
469         t->next = NULL;
470         t->state = 0;
471         atomic_set(&t->count, 0);
472         t->func = func;
473         t->data = data;
474 }
475
476 EXPORT_SYMBOL(tasklet_init);
477
478 void tasklet_kill(struct tasklet_struct *t)
479 {
480         if (in_interrupt())
481                 printk("Attempt to kill tasklet from interrupt\n");
482
483         while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
484                 do
485                         yield();
486                 while (test_bit(TASKLET_STATE_SCHED, &t->state));
487         }
488         tasklet_unlock_wait(t);
489         clear_bit(TASKLET_STATE_SCHED, &t->state);
490 }
491
492 EXPORT_SYMBOL(tasklet_kill);
493
494 void __init softirq_init(void)
495 {
496         int cpu;
497
498         for_each_possible_cpu(cpu) {
499                 per_cpu(tasklet_vec, cpu).tail =
500                         &per_cpu(tasklet_vec, cpu).head;
501                 per_cpu(tasklet_hi_vec, cpu).tail =
502                         &per_cpu(tasklet_hi_vec, cpu).head;
503         }
504
505         open_softirq(TASKLET_SOFTIRQ, tasklet_action);
506         open_softirq(HI_SOFTIRQ, tasklet_hi_action);
507 }
508
509 static int ksoftirqd(void * __bind_cpu)
510 {
511         set_current_state(TASK_INTERRUPTIBLE);
512
513         while (!kthread_should_stop()) {
514                 preempt_disable();
515                 if (!local_softirq_pending()) {
516                         preempt_enable_no_resched();
517                         schedule();
518                         preempt_disable();
519                 }
520
521                 __set_current_state(TASK_RUNNING);
522
523                 while (local_softirq_pending()) {
524                         /* Preempt disable stops cpu going offline.
525                            If already offline, we'll be on wrong CPU:
526                            don't process */
527                         if (cpu_is_offline((long)__bind_cpu))
528                                 goto wait_to_die;
529                         do_softirq();
530                         preempt_enable_no_resched();
531                         cond_resched();
532                         preempt_disable();
533                 }
534                 preempt_enable();
535                 set_current_state(TASK_INTERRUPTIBLE);
536         }
537         __set_current_state(TASK_RUNNING);
538         return 0;
539
540 wait_to_die:
541         preempt_enable();
542         /* Wait for kthread_stop */
543         set_current_state(TASK_INTERRUPTIBLE);
544         while (!kthread_should_stop()) {
545                 schedule();
546                 set_current_state(TASK_INTERRUPTIBLE);
547         }
548         __set_current_state(TASK_RUNNING);
549         return 0;
550 }
551
552 #ifdef CONFIG_HOTPLUG_CPU
553 /*
554  * tasklet_kill_immediate is called to remove a tasklet which can already be
555  * scheduled for execution on @cpu.
556  *
557  * Unlike tasklet_kill, this function removes the tasklet
558  * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
559  *
560  * When this function is called, @cpu must be in the CPU_DEAD state.
561  */
562 void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
563 {
564         struct tasklet_struct **i;
565
566         BUG_ON(cpu_online(cpu));
567         BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
568
569         if (!test_bit(TASKLET_STATE_SCHED, &t->state))
570                 return;
571
572         /* CPU is dead, so no lock needed. */
573         for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
574                 if (*i == t) {
575                         *i = t->next;
576                         /* If this was the tail element, move the tail ptr */
577                         if (*i == NULL)
578                                 per_cpu(tasklet_vec, cpu).tail = i;
579                         return;
580                 }
581         }
582         BUG();
583 }
584
585 static void takeover_tasklets(unsigned int cpu)
586 {
587         /* CPU is dead, so no lock needed. */
588         local_irq_disable();
589
590         /* Find end, append list for that CPU. */
591         if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
592                 *(__get_cpu_var(tasklet_vec).tail) = per_cpu(tasklet_vec, cpu).head;
593                 __get_cpu_var(tasklet_vec).tail = per_cpu(tasklet_vec, cpu).tail;
594                 per_cpu(tasklet_vec, cpu).head = NULL;
595                 per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
596         }
597         raise_softirq_irqoff(TASKLET_SOFTIRQ);
598
599         if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
600                 *__get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).head;
601                 __get_cpu_var(tasklet_hi_vec).tail = per_cpu(tasklet_hi_vec, cpu).tail;
602                 per_cpu(tasklet_hi_vec, cpu).head = NULL;
603                 per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
604         }
605         raise_softirq_irqoff(HI_SOFTIRQ);
606
607         local_irq_enable();
608 }
609 #endif /* CONFIG_HOTPLUG_CPU */
610
611 static int __cpuinit cpu_callback(struct notifier_block *nfb,
612                                   unsigned long action,
613                                   void *hcpu)
614 {
615         int hotcpu = (unsigned long)hcpu;
616         struct task_struct *p;
617
618         switch (action) {
619         case CPU_UP_PREPARE:
620         case CPU_UP_PREPARE_FROZEN:
621                 p = kthread_create(ksoftirqd, hcpu, "ksoftirqd/%d", hotcpu);
622                 if (IS_ERR(p)) {
623                         printk("ksoftirqd for %i failed\n", hotcpu);
624                         return NOTIFY_BAD;
625                 }
626                 kthread_bind(p, hotcpu);
627                 per_cpu(ksoftirqd, hotcpu) = p;
628                 break;
629         case CPU_ONLINE:
630         case CPU_ONLINE_FROZEN:
631                 wake_up_process(per_cpu(ksoftirqd, hotcpu));
632                 break;
633 #ifdef CONFIG_HOTPLUG_CPU
634         case CPU_UP_CANCELED:
635         case CPU_UP_CANCELED_FROZEN:
636                 if (!per_cpu(ksoftirqd, hotcpu))
637                         break;
638                 /* Unbind so it can run.  Fall thru. */
639                 kthread_bind(per_cpu(ksoftirqd, hotcpu),
640                              any_online_cpu(cpu_online_map));
641         case CPU_DEAD:
642         case CPU_DEAD_FROZEN: {
643                 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
644
645                 p = per_cpu(ksoftirqd, hotcpu);
646                 per_cpu(ksoftirqd, hotcpu) = NULL;
647                 sched_setscheduler(p, SCHED_FIFO, &param);
648                 kthread_stop(p);
649                 takeover_tasklets(hotcpu);
650                 break;
651         }
652 #endif /* CONFIG_HOTPLUG_CPU */
653         }
654         return NOTIFY_OK;
655 }
656
657 static struct notifier_block __cpuinitdata cpu_nfb = {
658         .notifier_call = cpu_callback
659 };
660
661 __init int spawn_ksoftirqd(void)
662 {
663         void *cpu = (void *)(long)smp_processor_id();
664         int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
665
666         BUG_ON(err == NOTIFY_BAD);
667         cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
668         register_cpu_notifier(&cpu_nfb);
669         return 0;
670 }
671
672 #ifdef CONFIG_SMP
673 /*
674  * Call a function on all processors
675  */
676 int on_each_cpu(void (*func) (void *info), void *info, int retry, int wait)
677 {
678         int ret = 0;
679
680         preempt_disable();
681         ret = smp_call_function(func, info, retry, wait);
682         local_irq_disable();
683         func(info);
684         local_irq_enable();
685         preempt_enable();
686         return ret;
687 }
688 EXPORT_SYMBOL(on_each_cpu);
689 #endif