]> git.karo-electronics.de Git - karo-tx-linux.git/blob - kernel/irq/manage.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[karo-tx-linux.git] / kernel / irq / manage.c
1 /*
2  * linux/kernel/irq/manage.c
3  *
4  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
5  * Copyright (C) 2005-2006 Thomas Gleixner
6  *
7  * This file contains driver APIs to the irq subsystem.
8  */
9
10 #include <linux/irq.h>
11 #include <linux/kthread.h>
12 #include <linux/module.h>
13 #include <linux/random.h>
14 #include <linux/interrupt.h>
15 #include <linux/slab.h>
16 #include <linux/sched.h>
17
18 #include "internals.h"
19
20 #ifdef CONFIG_IRQ_FORCED_THREADING
21 __read_mostly bool force_irqthreads;
22
23 static int __init setup_forced_irqthreads(char *arg)
24 {
25         force_irqthreads = true;
26         return 0;
27 }
28 early_param("threadirqs", setup_forced_irqthreads);
29 #endif
30
31 /**
32  *      synchronize_irq - wait for pending IRQ handlers (on other CPUs)
33  *      @irq: interrupt number to wait for
34  *
35  *      This function waits for any pending IRQ handlers for this interrupt
36  *      to complete before returning. If you use this function while
37  *      holding a resource the IRQ handler may need you will deadlock.
38  *
39  *      This function may be called - with care - from IRQ context.
40  */
41 void synchronize_irq(unsigned int irq)
42 {
43         struct irq_desc *desc = irq_to_desc(irq);
44         bool inprogress;
45
46         if (!desc)
47                 return;
48
49         do {
50                 unsigned long flags;
51
52                 /*
53                  * Wait until we're out of the critical section.  This might
54                  * give the wrong answer due to the lack of memory barriers.
55                  */
56                 while (irqd_irq_inprogress(&desc->irq_data))
57                         cpu_relax();
58
59                 /* Ok, that indicated we're done: double-check carefully. */
60                 raw_spin_lock_irqsave(&desc->lock, flags);
61                 inprogress = irqd_irq_inprogress(&desc->irq_data);
62                 raw_spin_unlock_irqrestore(&desc->lock, flags);
63
64                 /* Oops, that failed? */
65         } while (inprogress);
66
67         /*
68          * We made sure that no hardirq handler is running. Now verify
69          * that no threaded handlers are active.
70          */
71         wait_event(desc->wait_for_threads, !atomic_read(&desc->threads_active));
72 }
73 EXPORT_SYMBOL(synchronize_irq);
74
75 #ifdef CONFIG_SMP
76 cpumask_var_t irq_default_affinity;
77
78 /**
79  *      irq_can_set_affinity - Check if the affinity of a given irq can be set
80  *      @irq:           Interrupt to check
81  *
82  */
83 int irq_can_set_affinity(unsigned int irq)
84 {
85         struct irq_desc *desc = irq_to_desc(irq);
86
87         if (!desc || !irqd_can_balance(&desc->irq_data) ||
88             !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
89                 return 0;
90
91         return 1;
92 }
93
94 /**
95  *      irq_set_thread_affinity - Notify irq threads to adjust affinity
96  *      @desc:          irq descriptor which has affitnity changed
97  *
98  *      We just set IRQTF_AFFINITY and delegate the affinity setting
99  *      to the interrupt thread itself. We can not call
100  *      set_cpus_allowed_ptr() here as we hold desc->lock and this
101  *      code can be called from hard interrupt context.
102  */
103 void irq_set_thread_affinity(struct irq_desc *desc)
104 {
105         struct irqaction *action = desc->action;
106
107         while (action) {
108                 if (action->thread)
109                         set_bit(IRQTF_AFFINITY, &action->thread_flags);
110                 action = action->next;
111         }
112 }
113
114 #ifdef CONFIG_GENERIC_PENDING_IRQ
115 static inline bool irq_can_move_pcntxt(struct irq_data *data)
116 {
117         return irqd_can_move_in_process_context(data);
118 }
119 static inline bool irq_move_pending(struct irq_data *data)
120 {
121         return irqd_is_setaffinity_pending(data);
122 }
123 static inline void
124 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
125 {
126         cpumask_copy(desc->pending_mask, mask);
127 }
128 static inline void
129 irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
130 {
131         cpumask_copy(mask, desc->pending_mask);
132 }
133 #else
134 static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
135 static inline bool irq_move_pending(struct irq_data *data) { return false; }
136 static inline void
137 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
138 static inline void
139 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
140 #endif
141
142 int __irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask)
143 {
144         struct irq_chip *chip = irq_data_get_irq_chip(data);
145         struct irq_desc *desc = irq_data_to_desc(data);
146         int ret = 0;
147
148         if (!chip || !chip->irq_set_affinity)
149                 return -EINVAL;
150
151         if (irq_can_move_pcntxt(data)) {
152                 ret = chip->irq_set_affinity(data, mask, false);
153                 switch (ret) {
154                 case IRQ_SET_MASK_OK:
155                         cpumask_copy(data->affinity, mask);
156                 case IRQ_SET_MASK_OK_NOCOPY:
157                         irq_set_thread_affinity(desc);
158                         ret = 0;
159                 }
160         } else {
161                 irqd_set_move_pending(data);
162                 irq_copy_pending(desc, mask);
163         }
164
165         if (desc->affinity_notify) {
166                 kref_get(&desc->affinity_notify->kref);
167                 schedule_work(&desc->affinity_notify->work);
168         }
169         irqd_set(data, IRQD_AFFINITY_SET);
170
171         return ret;
172 }
173
174 /**
175  *      irq_set_affinity - Set the irq affinity of a given irq
176  *      @irq:           Interrupt to set affinity
177  *      @mask:          cpumask
178  *
179  */
180 int irq_set_affinity(unsigned int irq, const struct cpumask *mask)
181 {
182         struct irq_desc *desc = irq_to_desc(irq);
183         unsigned long flags;
184         int ret;
185
186         if (!desc)
187                 return -EINVAL;
188
189         raw_spin_lock_irqsave(&desc->lock, flags);
190         ret =  __irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask);
191         raw_spin_unlock_irqrestore(&desc->lock, flags);
192         return ret;
193 }
194
195 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
196 {
197         unsigned long flags;
198         struct irq_desc *desc = irq_get_desc_lock(irq, &flags);
199
200         if (!desc)
201                 return -EINVAL;
202         desc->affinity_hint = m;
203         irq_put_desc_unlock(desc, flags);
204         return 0;
205 }
206 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
207
208 static void irq_affinity_notify(struct work_struct *work)
209 {
210         struct irq_affinity_notify *notify =
211                 container_of(work, struct irq_affinity_notify, work);
212         struct irq_desc *desc = irq_to_desc(notify->irq);
213         cpumask_var_t cpumask;
214         unsigned long flags;
215
216         if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
217                 goto out;
218
219         raw_spin_lock_irqsave(&desc->lock, flags);
220         if (irq_move_pending(&desc->irq_data))
221                 irq_get_pending(cpumask, desc);
222         else
223                 cpumask_copy(cpumask, desc->irq_data.affinity);
224         raw_spin_unlock_irqrestore(&desc->lock, flags);
225
226         notify->notify(notify, cpumask);
227
228         free_cpumask_var(cpumask);
229 out:
230         kref_put(&notify->kref, notify->release);
231 }
232
233 /**
234  *      irq_set_affinity_notifier - control notification of IRQ affinity changes
235  *      @irq:           Interrupt for which to enable/disable notification
236  *      @notify:        Context for notification, or %NULL to disable
237  *                      notification.  Function pointers must be initialised;
238  *                      the other fields will be initialised by this function.
239  *
240  *      Must be called in process context.  Notification may only be enabled
241  *      after the IRQ is allocated and must be disabled before the IRQ is
242  *      freed using free_irq().
243  */
244 int
245 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
246 {
247         struct irq_desc *desc = irq_to_desc(irq);
248         struct irq_affinity_notify *old_notify;
249         unsigned long flags;
250
251         /* The release function is promised process context */
252         might_sleep();
253
254         if (!desc)
255                 return -EINVAL;
256
257         /* Complete initialisation of *notify */
258         if (notify) {
259                 notify->irq = irq;
260                 kref_init(&notify->kref);
261                 INIT_WORK(&notify->work, irq_affinity_notify);
262         }
263
264         raw_spin_lock_irqsave(&desc->lock, flags);
265         old_notify = desc->affinity_notify;
266         desc->affinity_notify = notify;
267         raw_spin_unlock_irqrestore(&desc->lock, flags);
268
269         if (old_notify)
270                 kref_put(&old_notify->kref, old_notify->release);
271
272         return 0;
273 }
274 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
275
276 #ifndef CONFIG_AUTO_IRQ_AFFINITY
277 /*
278  * Generic version of the affinity autoselector.
279  */
280 static int
281 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
282 {
283         struct irq_chip *chip = irq_desc_get_chip(desc);
284         struct cpumask *set = irq_default_affinity;
285         int ret;
286
287         /* Excludes PER_CPU and NO_BALANCE interrupts */
288         if (!irq_can_set_affinity(irq))
289                 return 0;
290
291         /*
292          * Preserve an userspace affinity setup, but make sure that
293          * one of the targets is online.
294          */
295         if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
296                 if (cpumask_intersects(desc->irq_data.affinity,
297                                        cpu_online_mask))
298                         set = desc->irq_data.affinity;
299                 else
300                         irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
301         }
302
303         cpumask_and(mask, cpu_online_mask, set);
304         ret = chip->irq_set_affinity(&desc->irq_data, mask, false);
305         switch (ret) {
306         case IRQ_SET_MASK_OK:
307                 cpumask_copy(desc->irq_data.affinity, mask);
308         case IRQ_SET_MASK_OK_NOCOPY:
309                 irq_set_thread_affinity(desc);
310         }
311         return 0;
312 }
313 #else
314 static inline int
315 setup_affinity(unsigned int irq, struct irq_desc *d, struct cpumask *mask)
316 {
317         return irq_select_affinity(irq);
318 }
319 #endif
320
321 /*
322  * Called when affinity is set via /proc/irq
323  */
324 int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
325 {
326         struct irq_desc *desc = irq_to_desc(irq);
327         unsigned long flags;
328         int ret;
329
330         raw_spin_lock_irqsave(&desc->lock, flags);
331         ret = setup_affinity(irq, desc, mask);
332         raw_spin_unlock_irqrestore(&desc->lock, flags);
333         return ret;
334 }
335
336 #else
337 static inline int
338 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
339 {
340         return 0;
341 }
342 #endif
343
344 void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
345 {
346         if (suspend) {
347                 if (!desc->action || (desc->action->flags & IRQF_NO_SUSPEND))
348                         return;
349                 desc->istate |= IRQS_SUSPENDED;
350         }
351
352         if (!desc->depth++)
353                 irq_disable(desc);
354 }
355
356 static int __disable_irq_nosync(unsigned int irq)
357 {
358         unsigned long flags;
359         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags);
360
361         if (!desc)
362                 return -EINVAL;
363         __disable_irq(desc, irq, false);
364         irq_put_desc_busunlock(desc, flags);
365         return 0;
366 }
367
368 /**
369  *      disable_irq_nosync - disable an irq without waiting
370  *      @irq: Interrupt to disable
371  *
372  *      Disable the selected interrupt line.  Disables and Enables are
373  *      nested.
374  *      Unlike disable_irq(), this function does not ensure existing
375  *      instances of the IRQ handler have completed before returning.
376  *
377  *      This function may be called from IRQ context.
378  */
379 void disable_irq_nosync(unsigned int irq)
380 {
381         __disable_irq_nosync(irq);
382 }
383 EXPORT_SYMBOL(disable_irq_nosync);
384
385 /**
386  *      disable_irq - disable an irq and wait for completion
387  *      @irq: Interrupt to disable
388  *
389  *      Disable the selected interrupt line.  Enables and Disables are
390  *      nested.
391  *      This function waits for any pending IRQ handlers for this interrupt
392  *      to complete before returning. If you use this function while
393  *      holding a resource the IRQ handler may need you will deadlock.
394  *
395  *      This function may be called - with care - from IRQ context.
396  */
397 void disable_irq(unsigned int irq)
398 {
399         if (!__disable_irq_nosync(irq))
400                 synchronize_irq(irq);
401 }
402 EXPORT_SYMBOL(disable_irq);
403
404 void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
405 {
406         if (resume) {
407                 if (!(desc->istate & IRQS_SUSPENDED)) {
408                         if (!desc->action)
409                                 return;
410                         if (!(desc->action->flags & IRQF_FORCE_RESUME))
411                                 return;
412                         /* Pretend that it got disabled ! */
413                         desc->depth++;
414                 }
415                 desc->istate &= ~IRQS_SUSPENDED;
416         }
417
418         switch (desc->depth) {
419         case 0:
420  err_out:
421                 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
422                 break;
423         case 1: {
424                 if (desc->istate & IRQS_SUSPENDED)
425                         goto err_out;
426                 /* Prevent probing on this irq: */
427                 irq_settings_set_noprobe(desc);
428                 irq_enable(desc);
429                 check_irq_resend(desc, irq);
430                 /* fall-through */
431         }
432         default:
433                 desc->depth--;
434         }
435 }
436
437 /**
438  *      enable_irq - enable handling of an irq
439  *      @irq: Interrupt to enable
440  *
441  *      Undoes the effect of one call to disable_irq().  If this
442  *      matches the last disable, processing of interrupts on this
443  *      IRQ line is re-enabled.
444  *
445  *      This function may be called from IRQ context only when
446  *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
447  */
448 void enable_irq(unsigned int irq)
449 {
450         unsigned long flags;
451         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags);
452
453         if (!desc)
454                 return;
455         if (WARN(!desc->irq_data.chip,
456                  KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
457                 goto out;
458
459         __enable_irq(desc, irq, false);
460 out:
461         irq_put_desc_busunlock(desc, flags);
462 }
463 EXPORT_SYMBOL(enable_irq);
464
465 static int set_irq_wake_real(unsigned int irq, unsigned int on)
466 {
467         struct irq_desc *desc = irq_to_desc(irq);
468         int ret = -ENXIO;
469
470         if (desc->irq_data.chip->irq_set_wake)
471                 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
472
473         return ret;
474 }
475
476 /**
477  *      irq_set_irq_wake - control irq power management wakeup
478  *      @irq:   interrupt to control
479  *      @on:    enable/disable power management wakeup
480  *
481  *      Enable/disable power management wakeup mode, which is
482  *      disabled by default.  Enables and disables must match,
483  *      just as they match for non-wakeup mode support.
484  *
485  *      Wakeup mode lets this IRQ wake the system from sleep
486  *      states like "suspend to RAM".
487  */
488 int irq_set_irq_wake(unsigned int irq, unsigned int on)
489 {
490         unsigned long flags;
491         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags);
492         int ret = 0;
493
494         /* wakeup-capable irqs can be shared between drivers that
495          * don't need to have the same sleep mode behaviors.
496          */
497         if (on) {
498                 if (desc->wake_depth++ == 0) {
499                         ret = set_irq_wake_real(irq, on);
500                         if (ret)
501                                 desc->wake_depth = 0;
502                         else
503                                 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
504                 }
505         } else {
506                 if (desc->wake_depth == 0) {
507                         WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
508                 } else if (--desc->wake_depth == 0) {
509                         ret = set_irq_wake_real(irq, on);
510                         if (ret)
511                                 desc->wake_depth = 1;
512                         else
513                                 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
514                 }
515         }
516         irq_put_desc_busunlock(desc, flags);
517         return ret;
518 }
519 EXPORT_SYMBOL(irq_set_irq_wake);
520
521 /*
522  * Internal function that tells the architecture code whether a
523  * particular irq has been exclusively allocated or is available
524  * for driver use.
525  */
526 int can_request_irq(unsigned int irq, unsigned long irqflags)
527 {
528         unsigned long flags;
529         struct irq_desc *desc = irq_get_desc_lock(irq, &flags);
530         int canrequest = 0;
531
532         if (!desc)
533                 return 0;
534
535         if (irq_settings_can_request(desc)) {
536                 if (desc->action)
537                         if (irqflags & desc->action->flags & IRQF_SHARED)
538                                 canrequest =1;
539         }
540         irq_put_desc_unlock(desc, flags);
541         return canrequest;
542 }
543
544 int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
545                       unsigned long flags)
546 {
547         struct irq_chip *chip = desc->irq_data.chip;
548         int ret, unmask = 0;
549
550         if (!chip || !chip->irq_set_type) {
551                 /*
552                  * IRQF_TRIGGER_* but the PIC does not support multiple
553                  * flow-types?
554                  */
555                 pr_debug("No set_type function for IRQ %d (%s)\n", irq,
556                                 chip ? (chip->name ? : "unknown") : "unknown");
557                 return 0;
558         }
559
560         flags &= IRQ_TYPE_SENSE_MASK;
561
562         if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
563                 if (!irqd_irq_masked(&desc->irq_data))
564                         mask_irq(desc);
565                 if (!irqd_irq_disabled(&desc->irq_data))
566                         unmask = 1;
567         }
568
569         /* caller masked out all except trigger mode flags */
570         ret = chip->irq_set_type(&desc->irq_data, flags);
571
572         switch (ret) {
573         case IRQ_SET_MASK_OK:
574                 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
575                 irqd_set(&desc->irq_data, flags);
576
577         case IRQ_SET_MASK_OK_NOCOPY:
578                 flags = irqd_get_trigger_type(&desc->irq_data);
579                 irq_settings_set_trigger_mask(desc, flags);
580                 irqd_clear(&desc->irq_data, IRQD_LEVEL);
581                 irq_settings_clr_level(desc);
582                 if (flags & IRQ_TYPE_LEVEL_MASK) {
583                         irq_settings_set_level(desc);
584                         irqd_set(&desc->irq_data, IRQD_LEVEL);
585                 }
586
587                 ret = 0;
588                 break;
589         default:
590                 pr_err("setting trigger mode %lu for irq %u failed (%pF)\n",
591                        flags, irq, chip->irq_set_type);
592         }
593         if (unmask)
594                 unmask_irq(desc);
595         return ret;
596 }
597
598 /*
599  * Default primary interrupt handler for threaded interrupts. Is
600  * assigned as primary handler when request_threaded_irq is called
601  * with handler == NULL. Useful for oneshot interrupts.
602  */
603 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
604 {
605         return IRQ_WAKE_THREAD;
606 }
607
608 /*
609  * Primary handler for nested threaded interrupts. Should never be
610  * called.
611  */
612 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
613 {
614         WARN(1, "Primary handler called for nested irq %d\n", irq);
615         return IRQ_NONE;
616 }
617
618 static int irq_wait_for_interrupt(struct irqaction *action)
619 {
620         while (!kthread_should_stop()) {
621                 set_current_state(TASK_INTERRUPTIBLE);
622
623                 if (test_and_clear_bit(IRQTF_RUNTHREAD,
624                                        &action->thread_flags)) {
625                         __set_current_state(TASK_RUNNING);
626                         return 0;
627                 }
628                 schedule();
629         }
630         return -1;
631 }
632
633 /*
634  * Oneshot interrupts keep the irq line masked until the threaded
635  * handler finished. unmask if the interrupt has not been disabled and
636  * is marked MASKED.
637  */
638 static void irq_finalize_oneshot(struct irq_desc *desc,
639                                  struct irqaction *action, bool force)
640 {
641         if (!(desc->istate & IRQS_ONESHOT))
642                 return;
643 again:
644         chip_bus_lock(desc);
645         raw_spin_lock_irq(&desc->lock);
646
647         /*
648          * Implausible though it may be we need to protect us against
649          * the following scenario:
650          *
651          * The thread is faster done than the hard interrupt handler
652          * on the other CPU. If we unmask the irq line then the
653          * interrupt can come in again and masks the line, leaves due
654          * to IRQS_INPROGRESS and the irq line is masked forever.
655          *
656          * This also serializes the state of shared oneshot handlers
657          * versus "desc->threads_onehsot |= action->thread_mask;" in
658          * irq_wake_thread(). See the comment there which explains the
659          * serialization.
660          */
661         if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
662                 raw_spin_unlock_irq(&desc->lock);
663                 chip_bus_sync_unlock(desc);
664                 cpu_relax();
665                 goto again;
666         }
667
668         /*
669          * Now check again, whether the thread should run. Otherwise
670          * we would clear the threads_oneshot bit of this thread which
671          * was just set.
672          */
673         if (!force && test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
674                 goto out_unlock;
675
676         desc->threads_oneshot &= ~action->thread_mask;
677
678         if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
679             irqd_irq_masked(&desc->irq_data))
680                 unmask_irq(desc);
681
682 out_unlock:
683         raw_spin_unlock_irq(&desc->lock);
684         chip_bus_sync_unlock(desc);
685 }
686
687 #ifdef CONFIG_SMP
688 /*
689  * Check whether we need to chasnge the affinity of the interrupt thread.
690  */
691 static void
692 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
693 {
694         cpumask_var_t mask;
695
696         if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
697                 return;
698
699         /*
700          * In case we are out of memory we set IRQTF_AFFINITY again and
701          * try again next time
702          */
703         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
704                 set_bit(IRQTF_AFFINITY, &action->thread_flags);
705                 return;
706         }
707
708         raw_spin_lock_irq(&desc->lock);
709         cpumask_copy(mask, desc->irq_data.affinity);
710         raw_spin_unlock_irq(&desc->lock);
711
712         set_cpus_allowed_ptr(current, mask);
713         free_cpumask_var(mask);
714 }
715 #else
716 static inline void
717 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
718 #endif
719
720 /*
721  * Interrupts which are not explicitely requested as threaded
722  * interrupts rely on the implicit bh/preempt disable of the hard irq
723  * context. So we need to disable bh here to avoid deadlocks and other
724  * side effects.
725  */
726 static void
727 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
728 {
729         local_bh_disable();
730         action->thread_fn(action->irq, action->dev_id);
731         irq_finalize_oneshot(desc, action, false);
732         local_bh_enable();
733 }
734
735 /*
736  * Interrupts explicitely requested as threaded interupts want to be
737  * preemtible - many of them need to sleep and wait for slow busses to
738  * complete.
739  */
740 static void irq_thread_fn(struct irq_desc *desc, struct irqaction *action)
741 {
742         action->thread_fn(action->irq, action->dev_id);
743         irq_finalize_oneshot(desc, action, false);
744 }
745
746 /*
747  * Interrupt handler thread
748  */
749 static int irq_thread(void *data)
750 {
751         static const struct sched_param param = {
752                 .sched_priority = MAX_USER_RT_PRIO/2,
753         };
754         struct irqaction *action = data;
755         struct irq_desc *desc = irq_to_desc(action->irq);
756         void (*handler_fn)(struct irq_desc *desc, struct irqaction *action);
757         int wake;
758
759         if (force_irqthreads & test_bit(IRQTF_FORCED_THREAD,
760                                         &action->thread_flags))
761                 handler_fn = irq_forced_thread_fn;
762         else
763                 handler_fn = irq_thread_fn;
764
765         sched_setscheduler(current, SCHED_FIFO, &param);
766         current->irqaction = action;
767
768         while (!irq_wait_for_interrupt(action)) {
769
770                 irq_thread_check_affinity(desc, action);
771
772                 atomic_inc(&desc->threads_active);
773
774                 raw_spin_lock_irq(&desc->lock);
775                 if (unlikely(irqd_irq_disabled(&desc->irq_data))) {
776                         /*
777                          * CHECKME: We might need a dedicated
778                          * IRQ_THREAD_PENDING flag here, which
779                          * retriggers the thread in check_irq_resend()
780                          * but AFAICT IRQS_PENDING should be fine as it
781                          * retriggers the interrupt itself --- tglx
782                          */
783                         desc->istate |= IRQS_PENDING;
784                         raw_spin_unlock_irq(&desc->lock);
785                 } else {
786                         raw_spin_unlock_irq(&desc->lock);
787                         handler_fn(desc, action);
788                 }
789
790                 wake = atomic_dec_and_test(&desc->threads_active);
791
792                 if (wake && waitqueue_active(&desc->wait_for_threads))
793                         wake_up(&desc->wait_for_threads);
794         }
795
796         /* Prevent a stale desc->threads_oneshot */
797         irq_finalize_oneshot(desc, action, true);
798
799         /*
800          * Clear irqaction. Otherwise exit_irq_thread() would make
801          * fuzz about an active irq thread going into nirvana.
802          */
803         current->irqaction = NULL;
804         return 0;
805 }
806
807 /*
808  * Called from do_exit()
809  */
810 void exit_irq_thread(void)
811 {
812         struct task_struct *tsk = current;
813         struct irq_desc *desc;
814
815         if (!tsk->irqaction)
816                 return;
817
818         printk(KERN_ERR
819                "exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
820                tsk->comm ? tsk->comm : "", tsk->pid, tsk->irqaction->irq);
821
822         desc = irq_to_desc(tsk->irqaction->irq);
823
824         /*
825          * Prevent a stale desc->threads_oneshot. Must be called
826          * before setting the IRQTF_DIED flag.
827          */
828         irq_finalize_oneshot(desc, tsk->irqaction, true);
829
830         /*
831          * Set the THREAD DIED flag to prevent further wakeups of the
832          * soon to be gone threaded handler.
833          */
834         set_bit(IRQTF_DIED, &tsk->irqaction->flags);
835 }
836
837 static void irq_setup_forced_threading(struct irqaction *new)
838 {
839         if (!force_irqthreads)
840                 return;
841         if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
842                 return;
843
844         new->flags |= IRQF_ONESHOT;
845
846         if (!new->thread_fn) {
847                 set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
848                 new->thread_fn = new->handler;
849                 new->handler = irq_default_primary_handler;
850         }
851 }
852
853 /*
854  * Internal function to register an irqaction - typically used to
855  * allocate special interrupts that are part of the architecture.
856  */
857 static int
858 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
859 {
860         struct irqaction *old, **old_ptr;
861         const char *old_name = NULL;
862         unsigned long flags, thread_mask = 0;
863         int ret, nested, shared = 0;
864         cpumask_var_t mask;
865
866         if (!desc)
867                 return -EINVAL;
868
869         if (desc->irq_data.chip == &no_irq_chip)
870                 return -ENOSYS;
871         /*
872          * Some drivers like serial.c use request_irq() heavily,
873          * so we have to be careful not to interfere with a
874          * running system.
875          */
876         if (new->flags & IRQF_SAMPLE_RANDOM) {
877                 /*
878                  * This function might sleep, we want to call it first,
879                  * outside of the atomic block.
880                  * Yes, this might clear the entropy pool if the wrong
881                  * driver is attempted to be loaded, without actually
882                  * installing a new handler, but is this really a problem,
883                  * only the sysadmin is able to do this.
884                  */
885                 rand_initialize_irq(irq);
886         }
887
888         /*
889          * Check whether the interrupt nests into another interrupt
890          * thread.
891          */
892         nested = irq_settings_is_nested_thread(desc);
893         if (nested) {
894                 if (!new->thread_fn)
895                         return -EINVAL;
896                 /*
897                  * Replace the primary handler which was provided from
898                  * the driver for non nested interrupt handling by the
899                  * dummy function which warns when called.
900                  */
901                 new->handler = irq_nested_primary_handler;
902         } else {
903                 irq_setup_forced_threading(new);
904         }
905
906         /*
907          * Create a handler thread when a thread function is supplied
908          * and the interrupt does not nest into another interrupt
909          * thread.
910          */
911         if (new->thread_fn && !nested) {
912                 struct task_struct *t;
913
914                 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
915                                    new->name);
916                 if (IS_ERR(t))
917                         return PTR_ERR(t);
918                 /*
919                  * We keep the reference to the task struct even if
920                  * the thread dies to avoid that the interrupt code
921                  * references an already freed task_struct.
922                  */
923                 get_task_struct(t);
924                 new->thread = t;
925         }
926
927         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
928                 ret = -ENOMEM;
929                 goto out_thread;
930         }
931
932         /*
933          * The following block of code has to be executed atomically
934          */
935         raw_spin_lock_irqsave(&desc->lock, flags);
936         old_ptr = &desc->action;
937         old = *old_ptr;
938         if (old) {
939                 /*
940                  * Can't share interrupts unless both agree to and are
941                  * the same type (level, edge, polarity). So both flag
942                  * fields must have IRQF_SHARED set and the bits which
943                  * set the trigger type must match. Also all must
944                  * agree on ONESHOT.
945                  */
946                 if (!((old->flags & new->flags) & IRQF_SHARED) ||
947                     ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
948                     ((old->flags ^ new->flags) & IRQF_ONESHOT)) {
949                         old_name = old->name;
950                         goto mismatch;
951                 }
952
953                 /* All handlers must agree on per-cpuness */
954                 if ((old->flags & IRQF_PERCPU) !=
955                     (new->flags & IRQF_PERCPU))
956                         goto mismatch;
957
958                 /* add new interrupt at end of irq queue */
959                 do {
960                         thread_mask |= old->thread_mask;
961                         old_ptr = &old->next;
962                         old = *old_ptr;
963                 } while (old);
964                 shared = 1;
965         }
966
967         /*
968          * Setup the thread mask for this irqaction. Unlikely to have
969          * 32 resp 64 irqs sharing one line, but who knows.
970          */
971         if (new->flags & IRQF_ONESHOT && thread_mask == ~0UL) {
972                 ret = -EBUSY;
973                 goto out_mask;
974         }
975         new->thread_mask = 1 << ffz(thread_mask);
976
977         if (!shared) {
978                 init_waitqueue_head(&desc->wait_for_threads);
979
980                 /* Setup the type (level, edge polarity) if configured: */
981                 if (new->flags & IRQF_TRIGGER_MASK) {
982                         ret = __irq_set_trigger(desc, irq,
983                                         new->flags & IRQF_TRIGGER_MASK);
984
985                         if (ret)
986                                 goto out_mask;
987                 }
988
989                 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
990                                   IRQS_ONESHOT | IRQS_WAITING);
991                 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
992
993                 if (new->flags & IRQF_PERCPU) {
994                         irqd_set(&desc->irq_data, IRQD_PER_CPU);
995                         irq_settings_set_per_cpu(desc);
996                 }
997
998                 if (new->flags & IRQF_ONESHOT)
999                         desc->istate |= IRQS_ONESHOT;
1000
1001                 if (irq_settings_can_autoenable(desc))
1002                         irq_startup(desc);
1003                 else
1004                         /* Undo nested disables: */
1005                         desc->depth = 1;
1006
1007                 /* Exclude IRQ from balancing if requested */
1008                 if (new->flags & IRQF_NOBALANCING) {
1009                         irq_settings_set_no_balancing(desc);
1010                         irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1011                 }
1012
1013                 /* Set default affinity mask once everything is setup */
1014                 setup_affinity(irq, desc, mask);
1015
1016         } else if (new->flags & IRQF_TRIGGER_MASK) {
1017                 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1018                 unsigned int omsk = irq_settings_get_trigger_mask(desc);
1019
1020                 if (nmsk != omsk)
1021                         /* hope the handler works with current  trigger mode */
1022                         pr_warning("IRQ %d uses trigger mode %u; requested %u\n",
1023                                    irq, nmsk, omsk);
1024         }
1025
1026         new->irq = irq;
1027         *old_ptr = new;
1028
1029         /* Reset broken irq detection when installing new handler */
1030         desc->irq_count = 0;
1031         desc->irqs_unhandled = 0;
1032
1033         /*
1034          * Check whether we disabled the irq via the spurious handler
1035          * before. Reenable it and give it another chance.
1036          */
1037         if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1038                 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1039                 __enable_irq(desc, irq, false);
1040         }
1041
1042         raw_spin_unlock_irqrestore(&desc->lock, flags);
1043
1044         /*
1045          * Strictly no need to wake it up, but hung_task complains
1046          * when no hard interrupt wakes the thread up.
1047          */
1048         if (new->thread)
1049                 wake_up_process(new->thread);
1050
1051         register_irq_proc(irq, desc);
1052         new->dir = NULL;
1053         register_handler_proc(irq, new);
1054         free_cpumask_var(mask);
1055
1056         return 0;
1057
1058 mismatch:
1059 #ifdef CONFIG_DEBUG_SHIRQ
1060         if (!(new->flags & IRQF_PROBE_SHARED)) {
1061                 printk(KERN_ERR "IRQ handler type mismatch for IRQ %d\n", irq);
1062                 if (old_name)
1063                         printk(KERN_ERR "current handler: %s\n", old_name);
1064                 dump_stack();
1065         }
1066 #endif
1067         ret = -EBUSY;
1068
1069 out_mask:
1070         raw_spin_unlock_irqrestore(&desc->lock, flags);
1071         free_cpumask_var(mask);
1072
1073 out_thread:
1074         if (new->thread) {
1075                 struct task_struct *t = new->thread;
1076
1077                 new->thread = NULL;
1078                 if (likely(!test_bit(IRQTF_DIED, &new->thread_flags)))
1079                         kthread_stop(t);
1080                 put_task_struct(t);
1081         }
1082         return ret;
1083 }
1084
1085 /**
1086  *      setup_irq - setup an interrupt
1087  *      @irq: Interrupt line to setup
1088  *      @act: irqaction for the interrupt
1089  *
1090  * Used to statically setup interrupts in the early boot process.
1091  */
1092 int setup_irq(unsigned int irq, struct irqaction *act)
1093 {
1094         int retval;
1095         struct irq_desc *desc = irq_to_desc(irq);
1096
1097         chip_bus_lock(desc);
1098         retval = __setup_irq(irq, desc, act);
1099         chip_bus_sync_unlock(desc);
1100
1101         return retval;
1102 }
1103 EXPORT_SYMBOL_GPL(setup_irq);
1104
1105  /*
1106  * Internal function to unregister an irqaction - used to free
1107  * regular and special interrupts that are part of the architecture.
1108  */
1109 static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1110 {
1111         struct irq_desc *desc = irq_to_desc(irq);
1112         struct irqaction *action, **action_ptr;
1113         unsigned long flags;
1114
1115         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1116
1117         if (!desc)
1118                 return NULL;
1119
1120         raw_spin_lock_irqsave(&desc->lock, flags);
1121
1122         /*
1123          * There can be multiple actions per IRQ descriptor, find the right
1124          * one based on the dev_id:
1125          */
1126         action_ptr = &desc->action;
1127         for (;;) {
1128                 action = *action_ptr;
1129
1130                 if (!action) {
1131                         WARN(1, "Trying to free already-free IRQ %d\n", irq);
1132                         raw_spin_unlock_irqrestore(&desc->lock, flags);
1133
1134                         return NULL;
1135                 }
1136
1137                 if (action->dev_id == dev_id)
1138                         break;
1139                 action_ptr = &action->next;
1140         }
1141
1142         /* Found it - now remove it from the list of entries: */
1143         *action_ptr = action->next;
1144
1145         /* Currently used only by UML, might disappear one day: */
1146 #ifdef CONFIG_IRQ_RELEASE_METHOD
1147         if (desc->irq_data.chip->release)
1148                 desc->irq_data.chip->release(irq, dev_id);
1149 #endif
1150
1151         /* If this was the last handler, shut down the IRQ line: */
1152         if (!desc->action)
1153                 irq_shutdown(desc);
1154
1155 #ifdef CONFIG_SMP
1156         /* make sure affinity_hint is cleaned up */
1157         if (WARN_ON_ONCE(desc->affinity_hint))
1158                 desc->affinity_hint = NULL;
1159 #endif
1160
1161         raw_spin_unlock_irqrestore(&desc->lock, flags);
1162
1163         unregister_handler_proc(irq, action);
1164
1165         /* Make sure it's not being used on another CPU: */
1166         synchronize_irq(irq);
1167
1168 #ifdef CONFIG_DEBUG_SHIRQ
1169         /*
1170          * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1171          * event to happen even now it's being freed, so let's make sure that
1172          * is so by doing an extra call to the handler ....
1173          *
1174          * ( We do this after actually deregistering it, to make sure that a
1175          *   'real' IRQ doesn't run in * parallel with our fake. )
1176          */
1177         if (action->flags & IRQF_SHARED) {
1178                 local_irq_save(flags);
1179                 action->handler(irq, dev_id);
1180                 local_irq_restore(flags);
1181         }
1182 #endif
1183
1184         if (action->thread) {
1185                 if (!test_bit(IRQTF_DIED, &action->thread_flags))
1186                         kthread_stop(action->thread);
1187                 put_task_struct(action->thread);
1188         }
1189
1190         return action;
1191 }
1192
1193 /**
1194  *      remove_irq - free an interrupt
1195  *      @irq: Interrupt line to free
1196  *      @act: irqaction for the interrupt
1197  *
1198  * Used to remove interrupts statically setup by the early boot process.
1199  */
1200 void remove_irq(unsigned int irq, struct irqaction *act)
1201 {
1202         __free_irq(irq, act->dev_id);
1203 }
1204 EXPORT_SYMBOL_GPL(remove_irq);
1205
1206 /**
1207  *      free_irq - free an interrupt allocated with request_irq
1208  *      @irq: Interrupt line to free
1209  *      @dev_id: Device identity to free
1210  *
1211  *      Remove an interrupt handler. The handler is removed and if the
1212  *      interrupt line is no longer in use by any driver it is disabled.
1213  *      On a shared IRQ the caller must ensure the interrupt is disabled
1214  *      on the card it drives before calling this function. The function
1215  *      does not return until any executing interrupts for this IRQ
1216  *      have completed.
1217  *
1218  *      This function must not be called from interrupt context.
1219  */
1220 void free_irq(unsigned int irq, void *dev_id)
1221 {
1222         struct irq_desc *desc = irq_to_desc(irq);
1223
1224         if (!desc)
1225                 return;
1226
1227 #ifdef CONFIG_SMP
1228         if (WARN_ON(desc->affinity_notify))
1229                 desc->affinity_notify = NULL;
1230 #endif
1231
1232         chip_bus_lock(desc);
1233         kfree(__free_irq(irq, dev_id));
1234         chip_bus_sync_unlock(desc);
1235 }
1236 EXPORT_SYMBOL(free_irq);
1237
1238 /**
1239  *      request_threaded_irq - allocate an interrupt line
1240  *      @irq: Interrupt line to allocate
1241  *      @handler: Function to be called when the IRQ occurs.
1242  *                Primary handler for threaded interrupts
1243  *                If NULL and thread_fn != NULL the default
1244  *                primary handler is installed
1245  *      @thread_fn: Function called from the irq handler thread
1246  *                  If NULL, no irq thread is created
1247  *      @irqflags: Interrupt type flags
1248  *      @devname: An ascii name for the claiming device
1249  *      @dev_id: A cookie passed back to the handler function
1250  *
1251  *      This call allocates interrupt resources and enables the
1252  *      interrupt line and IRQ handling. From the point this
1253  *      call is made your handler function may be invoked. Since
1254  *      your handler function must clear any interrupt the board
1255  *      raises, you must take care both to initialise your hardware
1256  *      and to set up the interrupt handler in the right order.
1257  *
1258  *      If you want to set up a threaded irq handler for your device
1259  *      then you need to supply @handler and @thread_fn. @handler ist
1260  *      still called in hard interrupt context and has to check
1261  *      whether the interrupt originates from the device. If yes it
1262  *      needs to disable the interrupt on the device and return
1263  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1264  *      @thread_fn. This split handler design is necessary to support
1265  *      shared interrupts.
1266  *
1267  *      Dev_id must be globally unique. Normally the address of the
1268  *      device data structure is used as the cookie. Since the handler
1269  *      receives this value it makes sense to use it.
1270  *
1271  *      If your interrupt is shared you must pass a non NULL dev_id
1272  *      as this is required when freeing the interrupt.
1273  *
1274  *      Flags:
1275  *
1276  *      IRQF_SHARED             Interrupt is shared
1277  *      IRQF_SAMPLE_RANDOM      The interrupt can be used for entropy
1278  *      IRQF_TRIGGER_*          Specify active edge(s) or level
1279  *
1280  */
1281 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1282                          irq_handler_t thread_fn, unsigned long irqflags,
1283                          const char *devname, void *dev_id)
1284 {
1285         struct irqaction *action;
1286         struct irq_desc *desc;
1287         int retval;
1288
1289         /*
1290          * Sanity-check: shared interrupts must pass in a real dev-ID,
1291          * otherwise we'll have trouble later trying to figure out
1292          * which interrupt is which (messes up the interrupt freeing
1293          * logic etc).
1294          */
1295         if ((irqflags & IRQF_SHARED) && !dev_id)
1296                 return -EINVAL;
1297
1298         desc = irq_to_desc(irq);
1299         if (!desc)
1300                 return -EINVAL;
1301
1302         if (!irq_settings_can_request(desc))
1303                 return -EINVAL;
1304
1305         if (!handler) {
1306                 if (!thread_fn)
1307                         return -EINVAL;
1308                 handler = irq_default_primary_handler;
1309         }
1310
1311         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1312         if (!action)
1313                 return -ENOMEM;
1314
1315         action->handler = handler;
1316         action->thread_fn = thread_fn;
1317         action->flags = irqflags;
1318         action->name = devname;
1319         action->dev_id = dev_id;
1320
1321         chip_bus_lock(desc);
1322         retval = __setup_irq(irq, desc, action);
1323         chip_bus_sync_unlock(desc);
1324
1325         if (retval)
1326                 kfree(action);
1327
1328 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
1329         if (!retval && (irqflags & IRQF_SHARED)) {
1330                 /*
1331                  * It's a shared IRQ -- the driver ought to be prepared for it
1332                  * to happen immediately, so let's make sure....
1333                  * We disable the irq to make sure that a 'real' IRQ doesn't
1334                  * run in parallel with our fake.
1335                  */
1336                 unsigned long flags;
1337
1338                 disable_irq(irq);
1339                 local_irq_save(flags);
1340
1341                 handler(irq, dev_id);
1342
1343                 local_irq_restore(flags);
1344                 enable_irq(irq);
1345         }
1346 #endif
1347         return retval;
1348 }
1349 EXPORT_SYMBOL(request_threaded_irq);
1350
1351 /**
1352  *      request_any_context_irq - allocate an interrupt line
1353  *      @irq: Interrupt line to allocate
1354  *      @handler: Function to be called when the IRQ occurs.
1355  *                Threaded handler for threaded interrupts.
1356  *      @flags: Interrupt type flags
1357  *      @name: An ascii name for the claiming device
1358  *      @dev_id: A cookie passed back to the handler function
1359  *
1360  *      This call allocates interrupt resources and enables the
1361  *      interrupt line and IRQ handling. It selects either a
1362  *      hardirq or threaded handling method depending on the
1363  *      context.
1364  *
1365  *      On failure, it returns a negative value. On success,
1366  *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1367  */
1368 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1369                             unsigned long flags, const char *name, void *dev_id)
1370 {
1371         struct irq_desc *desc = irq_to_desc(irq);
1372         int ret;
1373
1374         if (!desc)
1375                 return -EINVAL;
1376
1377         if (irq_settings_is_nested_thread(desc)) {
1378                 ret = request_threaded_irq(irq, NULL, handler,
1379                                            flags, name, dev_id);
1380                 return !ret ? IRQC_IS_NESTED : ret;
1381         }
1382
1383         ret = request_irq(irq, handler, flags, name, dev_id);
1384         return !ret ? IRQC_IS_HARDIRQ : ret;
1385 }
1386 EXPORT_SYMBOL_GPL(request_any_context_irq);