genirq: Remove the irq argument from check_irq_resend()
[linux-2.6-block.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 #define pr_fmt(fmt) "genirq: " fmt
11
12 #include <linux/irq.h>
13 #include <linux/kthread.h>
14 #include <linux/module.h>
15 #include <linux/random.h>
16 #include <linux/interrupt.h>
17 #include <linux/slab.h>
18 #include <linux/sched.h>
19 #include <linux/sched/rt.h>
20 #include <linux/task_work.h>
21
22 #include "internals.h"
23
24 #ifdef CONFIG_IRQ_FORCED_THREADING
25 __read_mostly bool force_irqthreads;
26
27 static int __init setup_forced_irqthreads(char *arg)
28 {
29         force_irqthreads = true;
30         return 0;
31 }
32 early_param("threadirqs", setup_forced_irqthreads);
33 #endif
34
35 static void __synchronize_hardirq(struct irq_desc *desc)
36 {
37         bool inprogress;
38
39         do {
40                 unsigned long flags;
41
42                 /*
43                  * Wait until we're out of the critical section.  This might
44                  * give the wrong answer due to the lack of memory barriers.
45                  */
46                 while (irqd_irq_inprogress(&desc->irq_data))
47                         cpu_relax();
48
49                 /* Ok, that indicated we're done: double-check carefully. */
50                 raw_spin_lock_irqsave(&desc->lock, flags);
51                 inprogress = irqd_irq_inprogress(&desc->irq_data);
52                 raw_spin_unlock_irqrestore(&desc->lock, flags);
53
54                 /* Oops, that failed? */
55         } while (inprogress);
56 }
57
58 /**
59  *      synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
60  *      @irq: interrupt number to wait for
61  *
62  *      This function waits for any pending hard IRQ handlers for this
63  *      interrupt to complete before returning. If you use this
64  *      function while holding a resource the IRQ handler may need you
65  *      will deadlock. It does not take associated threaded handlers
66  *      into account.
67  *
68  *      Do not use this for shutdown scenarios where you must be sure
69  *      that all parts (hardirq and threaded handler) have completed.
70  *
71  *      Returns: false if a threaded handler is active.
72  *
73  *      This function may be called - with care - from IRQ context.
74  */
75 bool synchronize_hardirq(unsigned int irq)
76 {
77         struct irq_desc *desc = irq_to_desc(irq);
78
79         if (desc) {
80                 __synchronize_hardirq(desc);
81                 return !atomic_read(&desc->threads_active);
82         }
83
84         return true;
85 }
86 EXPORT_SYMBOL(synchronize_hardirq);
87
88 /**
89  *      synchronize_irq - wait for pending IRQ handlers (on other CPUs)
90  *      @irq: interrupt number to wait for
91  *
92  *      This function waits for any pending IRQ handlers for this interrupt
93  *      to complete before returning. If you use this function while
94  *      holding a resource the IRQ handler may need you will deadlock.
95  *
96  *      This function may be called - with care - from IRQ context.
97  */
98 void synchronize_irq(unsigned int irq)
99 {
100         struct irq_desc *desc = irq_to_desc(irq);
101
102         if (desc) {
103                 __synchronize_hardirq(desc);
104                 /*
105                  * We made sure that no hardirq handler is
106                  * running. Now verify that no threaded handlers are
107                  * active.
108                  */
109                 wait_event(desc->wait_for_threads,
110                            !atomic_read(&desc->threads_active));
111         }
112 }
113 EXPORT_SYMBOL(synchronize_irq);
114
115 #ifdef CONFIG_SMP
116 cpumask_var_t irq_default_affinity;
117
118 /**
119  *      irq_can_set_affinity - Check if the affinity of a given irq can be set
120  *      @irq:           Interrupt to check
121  *
122  */
123 int irq_can_set_affinity(unsigned int irq)
124 {
125         struct irq_desc *desc = irq_to_desc(irq);
126
127         if (!desc || !irqd_can_balance(&desc->irq_data) ||
128             !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
129                 return 0;
130
131         return 1;
132 }
133
134 /**
135  *      irq_set_thread_affinity - Notify irq threads to adjust affinity
136  *      @desc:          irq descriptor which has affitnity changed
137  *
138  *      We just set IRQTF_AFFINITY and delegate the affinity setting
139  *      to the interrupt thread itself. We can not call
140  *      set_cpus_allowed_ptr() here as we hold desc->lock and this
141  *      code can be called from hard interrupt context.
142  */
143 void irq_set_thread_affinity(struct irq_desc *desc)
144 {
145         struct irqaction *action = desc->action;
146
147         while (action) {
148                 if (action->thread)
149                         set_bit(IRQTF_AFFINITY, &action->thread_flags);
150                 action = action->next;
151         }
152 }
153
154 #ifdef CONFIG_GENERIC_PENDING_IRQ
155 static inline bool irq_can_move_pcntxt(struct irq_data *data)
156 {
157         return irqd_can_move_in_process_context(data);
158 }
159 static inline bool irq_move_pending(struct irq_data *data)
160 {
161         return irqd_is_setaffinity_pending(data);
162 }
163 static inline void
164 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
165 {
166         cpumask_copy(desc->pending_mask, mask);
167 }
168 static inline void
169 irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
170 {
171         cpumask_copy(mask, desc->pending_mask);
172 }
173 #else
174 static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
175 static inline bool irq_move_pending(struct irq_data *data) { return false; }
176 static inline void
177 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
178 static inline void
179 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
180 #endif
181
182 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
183                         bool force)
184 {
185         struct irq_desc *desc = irq_data_to_desc(data);
186         struct irq_chip *chip = irq_data_get_irq_chip(data);
187         int ret;
188
189         ret = chip->irq_set_affinity(data, mask, force);
190         switch (ret) {
191         case IRQ_SET_MASK_OK:
192         case IRQ_SET_MASK_OK_DONE:
193                 cpumask_copy(data->affinity, mask);
194         case IRQ_SET_MASK_OK_NOCOPY:
195                 irq_set_thread_affinity(desc);
196                 ret = 0;
197         }
198
199         return ret;
200 }
201
202 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
203                             bool force)
204 {
205         struct irq_chip *chip = irq_data_get_irq_chip(data);
206         struct irq_desc *desc = irq_data_to_desc(data);
207         int ret = 0;
208
209         if (!chip || !chip->irq_set_affinity)
210                 return -EINVAL;
211
212         if (irq_can_move_pcntxt(data)) {
213                 ret = irq_do_set_affinity(data, mask, force);
214         } else {
215                 irqd_set_move_pending(data);
216                 irq_copy_pending(desc, mask);
217         }
218
219         if (desc->affinity_notify) {
220                 kref_get(&desc->affinity_notify->kref);
221                 schedule_work(&desc->affinity_notify->work);
222         }
223         irqd_set(data, IRQD_AFFINITY_SET);
224
225         return ret;
226 }
227
228 int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
229 {
230         struct irq_desc *desc = irq_to_desc(irq);
231         unsigned long flags;
232         int ret;
233
234         if (!desc)
235                 return -EINVAL;
236
237         raw_spin_lock_irqsave(&desc->lock, flags);
238         ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
239         raw_spin_unlock_irqrestore(&desc->lock, flags);
240         return ret;
241 }
242
243 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
244 {
245         unsigned long flags;
246         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
247
248         if (!desc)
249                 return -EINVAL;
250         desc->affinity_hint = m;
251         irq_put_desc_unlock(desc, flags);
252         /* set the initial affinity to prevent every interrupt being on CPU0 */
253         if (m)
254                 __irq_set_affinity(irq, m, false);
255         return 0;
256 }
257 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
258
259 /**
260  *      irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
261  *      @irq: interrupt number to set affinity
262  *      @vcpu_info: vCPU specific data
263  *
264  *      This function uses the vCPU specific data to set the vCPU
265  *      affinity for an irq. The vCPU specific data is passed from
266  *      outside, such as KVM. One example code path is as below:
267  *      KVM -> IOMMU -> irq_set_vcpu_affinity().
268  */
269 int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
270 {
271         unsigned long flags;
272         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
273         struct irq_data *data;
274         struct irq_chip *chip;
275         int ret = -ENOSYS;
276
277         if (!desc)
278                 return -EINVAL;
279
280         data = irq_desc_get_irq_data(desc);
281         chip = irq_data_get_irq_chip(data);
282         if (chip && chip->irq_set_vcpu_affinity)
283                 ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
284         irq_put_desc_unlock(desc, flags);
285
286         return ret;
287 }
288 EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
289
290 static void irq_affinity_notify(struct work_struct *work)
291 {
292         struct irq_affinity_notify *notify =
293                 container_of(work, struct irq_affinity_notify, work);
294         struct irq_desc *desc = irq_to_desc(notify->irq);
295         cpumask_var_t cpumask;
296         unsigned long flags;
297
298         if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
299                 goto out;
300
301         raw_spin_lock_irqsave(&desc->lock, flags);
302         if (irq_move_pending(&desc->irq_data))
303                 irq_get_pending(cpumask, desc);
304         else
305                 cpumask_copy(cpumask, desc->irq_data.affinity);
306         raw_spin_unlock_irqrestore(&desc->lock, flags);
307
308         notify->notify(notify, cpumask);
309
310         free_cpumask_var(cpumask);
311 out:
312         kref_put(&notify->kref, notify->release);
313 }
314
315 /**
316  *      irq_set_affinity_notifier - control notification of IRQ affinity changes
317  *      @irq:           Interrupt for which to enable/disable notification
318  *      @notify:        Context for notification, or %NULL to disable
319  *                      notification.  Function pointers must be initialised;
320  *                      the other fields will be initialised by this function.
321  *
322  *      Must be called in process context.  Notification may only be enabled
323  *      after the IRQ is allocated and must be disabled before the IRQ is
324  *      freed using free_irq().
325  */
326 int
327 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
328 {
329         struct irq_desc *desc = irq_to_desc(irq);
330         struct irq_affinity_notify *old_notify;
331         unsigned long flags;
332
333         /* The release function is promised process context */
334         might_sleep();
335
336         if (!desc)
337                 return -EINVAL;
338
339         /* Complete initialisation of *notify */
340         if (notify) {
341                 notify->irq = irq;
342                 kref_init(&notify->kref);
343                 INIT_WORK(&notify->work, irq_affinity_notify);
344         }
345
346         raw_spin_lock_irqsave(&desc->lock, flags);
347         old_notify = desc->affinity_notify;
348         desc->affinity_notify = notify;
349         raw_spin_unlock_irqrestore(&desc->lock, flags);
350
351         if (old_notify)
352                 kref_put(&old_notify->kref, old_notify->release);
353
354         return 0;
355 }
356 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
357
358 #ifndef CONFIG_AUTO_IRQ_AFFINITY
359 /*
360  * Generic version of the affinity autoselector.
361  */
362 static int
363 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
364 {
365         struct cpumask *set = irq_default_affinity;
366         int node = irq_desc_get_node(desc);
367
368         /* Excludes PER_CPU and NO_BALANCE interrupts */
369         if (!irq_can_set_affinity(irq))
370                 return 0;
371
372         /*
373          * Preserve an userspace affinity setup, but make sure that
374          * one of the targets is online.
375          */
376         if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
377                 if (cpumask_intersects(desc->irq_data.affinity,
378                                        cpu_online_mask))
379                         set = desc->irq_data.affinity;
380                 else
381                         irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
382         }
383
384         cpumask_and(mask, cpu_online_mask, set);
385         if (node != NUMA_NO_NODE) {
386                 const struct cpumask *nodemask = cpumask_of_node(node);
387
388                 /* make sure at least one of the cpus in nodemask is online */
389                 if (cpumask_intersects(mask, nodemask))
390                         cpumask_and(mask, mask, nodemask);
391         }
392         irq_do_set_affinity(&desc->irq_data, mask, false);
393         return 0;
394 }
395 #else
396 static inline int
397 setup_affinity(unsigned int irq, struct irq_desc *d, struct cpumask *mask)
398 {
399         return irq_select_affinity(irq);
400 }
401 #endif
402
403 /*
404  * Called when affinity is set via /proc/irq
405  */
406 int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
407 {
408         struct irq_desc *desc = irq_to_desc(irq);
409         unsigned long flags;
410         int ret;
411
412         raw_spin_lock_irqsave(&desc->lock, flags);
413         ret = setup_affinity(irq, desc, mask);
414         raw_spin_unlock_irqrestore(&desc->lock, flags);
415         return ret;
416 }
417
418 #else
419 static inline int
420 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
421 {
422         return 0;
423 }
424 #endif
425
426 void __disable_irq(struct irq_desc *desc, unsigned int irq)
427 {
428         if (!desc->depth++)
429                 irq_disable(desc);
430 }
431
432 static int __disable_irq_nosync(unsigned int irq)
433 {
434         unsigned long flags;
435         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
436
437         if (!desc)
438                 return -EINVAL;
439         __disable_irq(desc, irq);
440         irq_put_desc_busunlock(desc, flags);
441         return 0;
442 }
443
444 /**
445  *      disable_irq_nosync - disable an irq without waiting
446  *      @irq: Interrupt to disable
447  *
448  *      Disable the selected interrupt line.  Disables and Enables are
449  *      nested.
450  *      Unlike disable_irq(), this function does not ensure existing
451  *      instances of the IRQ handler have completed before returning.
452  *
453  *      This function may be called from IRQ context.
454  */
455 void disable_irq_nosync(unsigned int irq)
456 {
457         __disable_irq_nosync(irq);
458 }
459 EXPORT_SYMBOL(disable_irq_nosync);
460
461 /**
462  *      disable_irq - disable an irq and wait for completion
463  *      @irq: Interrupt to disable
464  *
465  *      Disable the selected interrupt line.  Enables and Disables are
466  *      nested.
467  *      This function waits for any pending IRQ handlers for this interrupt
468  *      to complete before returning. If you use this function while
469  *      holding a resource the IRQ handler may need you will deadlock.
470  *
471  *      This function may be called - with care - from IRQ context.
472  */
473 void disable_irq(unsigned int irq)
474 {
475         if (!__disable_irq_nosync(irq))
476                 synchronize_irq(irq);
477 }
478 EXPORT_SYMBOL(disable_irq);
479
480 /**
481  *      disable_hardirq - disables an irq and waits for hardirq completion
482  *      @irq: Interrupt to disable
483  *
484  *      Disable the selected interrupt line.  Enables and Disables are
485  *      nested.
486  *      This function waits for any pending hard IRQ handlers for this
487  *      interrupt to complete before returning. If you use this function while
488  *      holding a resource the hard IRQ handler may need you will deadlock.
489  *
490  *      When used to optimistically disable an interrupt from atomic context
491  *      the return value must be checked.
492  *
493  *      Returns: false if a threaded handler is active.
494  *
495  *      This function may be called - with care - from IRQ context.
496  */
497 bool disable_hardirq(unsigned int irq)
498 {
499         if (!__disable_irq_nosync(irq))
500                 return synchronize_hardirq(irq);
501
502         return false;
503 }
504 EXPORT_SYMBOL_GPL(disable_hardirq);
505
506 void __enable_irq(struct irq_desc *desc, unsigned int irq)
507 {
508         switch (desc->depth) {
509         case 0:
510  err_out:
511                 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
512                 break;
513         case 1: {
514                 if (desc->istate & IRQS_SUSPENDED)
515                         goto err_out;
516                 /* Prevent probing on this irq: */
517                 irq_settings_set_noprobe(desc);
518                 irq_enable(desc);
519                 check_irq_resend(desc);
520                 /* fall-through */
521         }
522         default:
523                 desc->depth--;
524         }
525 }
526
527 /**
528  *      enable_irq - enable handling of an irq
529  *      @irq: Interrupt to enable
530  *
531  *      Undoes the effect of one call to disable_irq().  If this
532  *      matches the last disable, processing of interrupts on this
533  *      IRQ line is re-enabled.
534  *
535  *      This function may be called from IRQ context only when
536  *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
537  */
538 void enable_irq(unsigned int irq)
539 {
540         unsigned long flags;
541         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
542
543         if (!desc)
544                 return;
545         if (WARN(!desc->irq_data.chip,
546                  KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
547                 goto out;
548
549         __enable_irq(desc, irq);
550 out:
551         irq_put_desc_busunlock(desc, flags);
552 }
553 EXPORT_SYMBOL(enable_irq);
554
555 static int set_irq_wake_real(unsigned int irq, unsigned int on)
556 {
557         struct irq_desc *desc = irq_to_desc(irq);
558         int ret = -ENXIO;
559
560         if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
561                 return 0;
562
563         if (desc->irq_data.chip->irq_set_wake)
564                 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
565
566         return ret;
567 }
568
569 /**
570  *      irq_set_irq_wake - control irq power management wakeup
571  *      @irq:   interrupt to control
572  *      @on:    enable/disable power management wakeup
573  *
574  *      Enable/disable power management wakeup mode, which is
575  *      disabled by default.  Enables and disables must match,
576  *      just as they match for non-wakeup mode support.
577  *
578  *      Wakeup mode lets this IRQ wake the system from sleep
579  *      states like "suspend to RAM".
580  */
581 int irq_set_irq_wake(unsigned int irq, unsigned int on)
582 {
583         unsigned long flags;
584         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
585         int ret = 0;
586
587         if (!desc)
588                 return -EINVAL;
589
590         /* wakeup-capable irqs can be shared between drivers that
591          * don't need to have the same sleep mode behaviors.
592          */
593         if (on) {
594                 if (desc->wake_depth++ == 0) {
595                         ret = set_irq_wake_real(irq, on);
596                         if (ret)
597                                 desc->wake_depth = 0;
598                         else
599                                 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
600                 }
601         } else {
602                 if (desc->wake_depth == 0) {
603                         WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
604                 } else if (--desc->wake_depth == 0) {
605                         ret = set_irq_wake_real(irq, on);
606                         if (ret)
607                                 desc->wake_depth = 1;
608                         else
609                                 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
610                 }
611         }
612         irq_put_desc_busunlock(desc, flags);
613         return ret;
614 }
615 EXPORT_SYMBOL(irq_set_irq_wake);
616
617 /*
618  * Internal function that tells the architecture code whether a
619  * particular irq has been exclusively allocated or is available
620  * for driver use.
621  */
622 int can_request_irq(unsigned int irq, unsigned long irqflags)
623 {
624         unsigned long flags;
625         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
626         int canrequest = 0;
627
628         if (!desc)
629                 return 0;
630
631         if (irq_settings_can_request(desc)) {
632                 if (!desc->action ||
633                     irqflags & desc->action->flags & IRQF_SHARED)
634                         canrequest = 1;
635         }
636         irq_put_desc_unlock(desc, flags);
637         return canrequest;
638 }
639
640 int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
641                       unsigned long flags)
642 {
643         struct irq_chip *chip = desc->irq_data.chip;
644         int ret, unmask = 0;
645
646         if (!chip || !chip->irq_set_type) {
647                 /*
648                  * IRQF_TRIGGER_* but the PIC does not support multiple
649                  * flow-types?
650                  */
651                 pr_debug("No set_type function for IRQ %d (%s)\n", irq,
652                          chip ? (chip->name ? : "unknown") : "unknown");
653                 return 0;
654         }
655
656         flags &= IRQ_TYPE_SENSE_MASK;
657
658         if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
659                 if (!irqd_irq_masked(&desc->irq_data))
660                         mask_irq(desc);
661                 if (!irqd_irq_disabled(&desc->irq_data))
662                         unmask = 1;
663         }
664
665         /* caller masked out all except trigger mode flags */
666         ret = chip->irq_set_type(&desc->irq_data, flags);
667
668         switch (ret) {
669         case IRQ_SET_MASK_OK:
670         case IRQ_SET_MASK_OK_DONE:
671                 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
672                 irqd_set(&desc->irq_data, flags);
673
674         case IRQ_SET_MASK_OK_NOCOPY:
675                 flags = irqd_get_trigger_type(&desc->irq_data);
676                 irq_settings_set_trigger_mask(desc, flags);
677                 irqd_clear(&desc->irq_data, IRQD_LEVEL);
678                 irq_settings_clr_level(desc);
679                 if (flags & IRQ_TYPE_LEVEL_MASK) {
680                         irq_settings_set_level(desc);
681                         irqd_set(&desc->irq_data, IRQD_LEVEL);
682                 }
683
684                 ret = 0;
685                 break;
686         default:
687                 pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
688                        flags, irq, chip->irq_set_type);
689         }
690         if (unmask)
691                 unmask_irq(desc);
692         return ret;
693 }
694
695 #ifdef CONFIG_HARDIRQS_SW_RESEND
696 int irq_set_parent(int irq, int parent_irq)
697 {
698         unsigned long flags;
699         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
700
701         if (!desc)
702                 return -EINVAL;
703
704         desc->parent_irq = parent_irq;
705
706         irq_put_desc_unlock(desc, flags);
707         return 0;
708 }
709 #endif
710
711 /*
712  * Default primary interrupt handler for threaded interrupts. Is
713  * assigned as primary handler when request_threaded_irq is called
714  * with handler == NULL. Useful for oneshot interrupts.
715  */
716 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
717 {
718         return IRQ_WAKE_THREAD;
719 }
720
721 /*
722  * Primary handler for nested threaded interrupts. Should never be
723  * called.
724  */
725 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
726 {
727         WARN(1, "Primary handler called for nested irq %d\n", irq);
728         return IRQ_NONE;
729 }
730
731 static int irq_wait_for_interrupt(struct irqaction *action)
732 {
733         set_current_state(TASK_INTERRUPTIBLE);
734
735         while (!kthread_should_stop()) {
736
737                 if (test_and_clear_bit(IRQTF_RUNTHREAD,
738                                        &action->thread_flags)) {
739                         __set_current_state(TASK_RUNNING);
740                         return 0;
741                 }
742                 schedule();
743                 set_current_state(TASK_INTERRUPTIBLE);
744         }
745         __set_current_state(TASK_RUNNING);
746         return -1;
747 }
748
749 /*
750  * Oneshot interrupts keep the irq line masked until the threaded
751  * handler finished. unmask if the interrupt has not been disabled and
752  * is marked MASKED.
753  */
754 static void irq_finalize_oneshot(struct irq_desc *desc,
755                                  struct irqaction *action)
756 {
757         if (!(desc->istate & IRQS_ONESHOT))
758                 return;
759 again:
760         chip_bus_lock(desc);
761         raw_spin_lock_irq(&desc->lock);
762
763         /*
764          * Implausible though it may be we need to protect us against
765          * the following scenario:
766          *
767          * The thread is faster done than the hard interrupt handler
768          * on the other CPU. If we unmask the irq line then the
769          * interrupt can come in again and masks the line, leaves due
770          * to IRQS_INPROGRESS and the irq line is masked forever.
771          *
772          * This also serializes the state of shared oneshot handlers
773          * versus "desc->threads_onehsot |= action->thread_mask;" in
774          * irq_wake_thread(). See the comment there which explains the
775          * serialization.
776          */
777         if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
778                 raw_spin_unlock_irq(&desc->lock);
779                 chip_bus_sync_unlock(desc);
780                 cpu_relax();
781                 goto again;
782         }
783
784         /*
785          * Now check again, whether the thread should run. Otherwise
786          * we would clear the threads_oneshot bit of this thread which
787          * was just set.
788          */
789         if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
790                 goto out_unlock;
791
792         desc->threads_oneshot &= ~action->thread_mask;
793
794         if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
795             irqd_irq_masked(&desc->irq_data))
796                 unmask_threaded_irq(desc);
797
798 out_unlock:
799         raw_spin_unlock_irq(&desc->lock);
800         chip_bus_sync_unlock(desc);
801 }
802
803 #ifdef CONFIG_SMP
804 /*
805  * Check whether we need to change the affinity of the interrupt thread.
806  */
807 static void
808 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
809 {
810         cpumask_var_t mask;
811         bool valid = true;
812
813         if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
814                 return;
815
816         /*
817          * In case we are out of memory we set IRQTF_AFFINITY again and
818          * try again next time
819          */
820         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
821                 set_bit(IRQTF_AFFINITY, &action->thread_flags);
822                 return;
823         }
824
825         raw_spin_lock_irq(&desc->lock);
826         /*
827          * This code is triggered unconditionally. Check the affinity
828          * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
829          */
830         if (desc->irq_data.affinity)
831                 cpumask_copy(mask, desc->irq_data.affinity);
832         else
833                 valid = false;
834         raw_spin_unlock_irq(&desc->lock);
835
836         if (valid)
837                 set_cpus_allowed_ptr(current, mask);
838         free_cpumask_var(mask);
839 }
840 #else
841 static inline void
842 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
843 #endif
844
845 /*
846  * Interrupts which are not explicitely requested as threaded
847  * interrupts rely on the implicit bh/preempt disable of the hard irq
848  * context. So we need to disable bh here to avoid deadlocks and other
849  * side effects.
850  */
851 static irqreturn_t
852 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
853 {
854         irqreturn_t ret;
855
856         local_bh_disable();
857         ret = action->thread_fn(action->irq, action->dev_id);
858         irq_finalize_oneshot(desc, action);
859         local_bh_enable();
860         return ret;
861 }
862
863 /*
864  * Interrupts explicitly requested as threaded interrupts want to be
865  * preemtible - many of them need to sleep and wait for slow busses to
866  * complete.
867  */
868 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
869                 struct irqaction *action)
870 {
871         irqreturn_t ret;
872
873         ret = action->thread_fn(action->irq, action->dev_id);
874         irq_finalize_oneshot(desc, action);
875         return ret;
876 }
877
878 static void wake_threads_waitq(struct irq_desc *desc)
879 {
880         if (atomic_dec_and_test(&desc->threads_active))
881                 wake_up(&desc->wait_for_threads);
882 }
883
884 static void irq_thread_dtor(struct callback_head *unused)
885 {
886         struct task_struct *tsk = current;
887         struct irq_desc *desc;
888         struct irqaction *action;
889
890         if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
891                 return;
892
893         action = kthread_data(tsk);
894
895         pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
896                tsk->comm, tsk->pid, action->irq);
897
898
899         desc = irq_to_desc(action->irq);
900         /*
901          * If IRQTF_RUNTHREAD is set, we need to decrement
902          * desc->threads_active and wake possible waiters.
903          */
904         if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
905                 wake_threads_waitq(desc);
906
907         /* Prevent a stale desc->threads_oneshot */
908         irq_finalize_oneshot(desc, action);
909 }
910
911 /*
912  * Interrupt handler thread
913  */
914 static int irq_thread(void *data)
915 {
916         struct callback_head on_exit_work;
917         struct irqaction *action = data;
918         struct irq_desc *desc = irq_to_desc(action->irq);
919         irqreturn_t (*handler_fn)(struct irq_desc *desc,
920                         struct irqaction *action);
921
922         if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
923                                         &action->thread_flags))
924                 handler_fn = irq_forced_thread_fn;
925         else
926                 handler_fn = irq_thread_fn;
927
928         init_task_work(&on_exit_work, irq_thread_dtor);
929         task_work_add(current, &on_exit_work, false);
930
931         irq_thread_check_affinity(desc, action);
932
933         while (!irq_wait_for_interrupt(action)) {
934                 irqreturn_t action_ret;
935
936                 irq_thread_check_affinity(desc, action);
937
938                 action_ret = handler_fn(desc, action);
939                 if (action_ret == IRQ_HANDLED)
940                         atomic_inc(&desc->threads_handled);
941
942                 wake_threads_waitq(desc);
943         }
944
945         /*
946          * This is the regular exit path. __free_irq() is stopping the
947          * thread via kthread_stop() after calling
948          * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
949          * oneshot mask bit can be set. We cannot verify that as we
950          * cannot touch the oneshot mask at this point anymore as
951          * __setup_irq() might have given out currents thread_mask
952          * again.
953          */
954         task_work_cancel(current, irq_thread_dtor);
955         return 0;
956 }
957
958 /**
959  *      irq_wake_thread - wake the irq thread for the action identified by dev_id
960  *      @irq:           Interrupt line
961  *      @dev_id:        Device identity for which the thread should be woken
962  *
963  */
964 void irq_wake_thread(unsigned int irq, void *dev_id)
965 {
966         struct irq_desc *desc = irq_to_desc(irq);
967         struct irqaction *action;
968         unsigned long flags;
969
970         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
971                 return;
972
973         raw_spin_lock_irqsave(&desc->lock, flags);
974         for (action = desc->action; action; action = action->next) {
975                 if (action->dev_id == dev_id) {
976                         if (action->thread)
977                                 __irq_wake_thread(desc, action);
978                         break;
979                 }
980         }
981         raw_spin_unlock_irqrestore(&desc->lock, flags);
982 }
983 EXPORT_SYMBOL_GPL(irq_wake_thread);
984
985 static void irq_setup_forced_threading(struct irqaction *new)
986 {
987         if (!force_irqthreads)
988                 return;
989         if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
990                 return;
991
992         new->flags |= IRQF_ONESHOT;
993
994         if (!new->thread_fn) {
995                 set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
996                 new->thread_fn = new->handler;
997                 new->handler = irq_default_primary_handler;
998         }
999 }
1000
1001 static int irq_request_resources(struct irq_desc *desc)
1002 {
1003         struct irq_data *d = &desc->irq_data;
1004         struct irq_chip *c = d->chip;
1005
1006         return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1007 }
1008
1009 static void irq_release_resources(struct irq_desc *desc)
1010 {
1011         struct irq_data *d = &desc->irq_data;
1012         struct irq_chip *c = d->chip;
1013
1014         if (c->irq_release_resources)
1015                 c->irq_release_resources(d);
1016 }
1017
1018 /*
1019  * Internal function to register an irqaction - typically used to
1020  * allocate special interrupts that are part of the architecture.
1021  */
1022 static int
1023 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1024 {
1025         struct irqaction *old, **old_ptr;
1026         unsigned long flags, thread_mask = 0;
1027         int ret, nested, shared = 0;
1028         cpumask_var_t mask;
1029
1030         if (!desc)
1031                 return -EINVAL;
1032
1033         if (desc->irq_data.chip == &no_irq_chip)
1034                 return -ENOSYS;
1035         if (!try_module_get(desc->owner))
1036                 return -ENODEV;
1037
1038         /*
1039          * Check whether the interrupt nests into another interrupt
1040          * thread.
1041          */
1042         nested = irq_settings_is_nested_thread(desc);
1043         if (nested) {
1044                 if (!new->thread_fn) {
1045                         ret = -EINVAL;
1046                         goto out_mput;
1047                 }
1048                 /*
1049                  * Replace the primary handler which was provided from
1050                  * the driver for non nested interrupt handling by the
1051                  * dummy function which warns when called.
1052                  */
1053                 new->handler = irq_nested_primary_handler;
1054         } else {
1055                 if (irq_settings_can_thread(desc))
1056                         irq_setup_forced_threading(new);
1057         }
1058
1059         /*
1060          * Create a handler thread when a thread function is supplied
1061          * and the interrupt does not nest into another interrupt
1062          * thread.
1063          */
1064         if (new->thread_fn && !nested) {
1065                 struct task_struct *t;
1066                 static const struct sched_param param = {
1067                         .sched_priority = MAX_USER_RT_PRIO/2,
1068                 };
1069
1070                 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1071                                    new->name);
1072                 if (IS_ERR(t)) {
1073                         ret = PTR_ERR(t);
1074                         goto out_mput;
1075                 }
1076
1077                 sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1078
1079                 /*
1080                  * We keep the reference to the task struct even if
1081                  * the thread dies to avoid that the interrupt code
1082                  * references an already freed task_struct.
1083                  */
1084                 get_task_struct(t);
1085                 new->thread = t;
1086                 /*
1087                  * Tell the thread to set its affinity. This is
1088                  * important for shared interrupt handlers as we do
1089                  * not invoke setup_affinity() for the secondary
1090                  * handlers as everything is already set up. Even for
1091                  * interrupts marked with IRQF_NO_BALANCE this is
1092                  * correct as we want the thread to move to the cpu(s)
1093                  * on which the requesting code placed the interrupt.
1094                  */
1095                 set_bit(IRQTF_AFFINITY, &new->thread_flags);
1096         }
1097
1098         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1099                 ret = -ENOMEM;
1100                 goto out_thread;
1101         }
1102
1103         /*
1104          * Drivers are often written to work w/o knowledge about the
1105          * underlying irq chip implementation, so a request for a
1106          * threaded irq without a primary hard irq context handler
1107          * requires the ONESHOT flag to be set. Some irq chips like
1108          * MSI based interrupts are per se one shot safe. Check the
1109          * chip flags, so we can avoid the unmask dance at the end of
1110          * the threaded handler for those.
1111          */
1112         if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1113                 new->flags &= ~IRQF_ONESHOT;
1114
1115         /*
1116          * The following block of code has to be executed atomically
1117          */
1118         raw_spin_lock_irqsave(&desc->lock, flags);
1119         old_ptr = &desc->action;
1120         old = *old_ptr;
1121         if (old) {
1122                 /*
1123                  * Can't share interrupts unless both agree to and are
1124                  * the same type (level, edge, polarity). So both flag
1125                  * fields must have IRQF_SHARED set and the bits which
1126                  * set the trigger type must match. Also all must
1127                  * agree on ONESHOT.
1128                  */
1129                 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1130                     ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1131                     ((old->flags ^ new->flags) & IRQF_ONESHOT))
1132                         goto mismatch;
1133
1134                 /* All handlers must agree on per-cpuness */
1135                 if ((old->flags & IRQF_PERCPU) !=
1136                     (new->flags & IRQF_PERCPU))
1137                         goto mismatch;
1138
1139                 /* add new interrupt at end of irq queue */
1140                 do {
1141                         /*
1142                          * Or all existing action->thread_mask bits,
1143                          * so we can find the next zero bit for this
1144                          * new action.
1145                          */
1146                         thread_mask |= old->thread_mask;
1147                         old_ptr = &old->next;
1148                         old = *old_ptr;
1149                 } while (old);
1150                 shared = 1;
1151         }
1152
1153         /*
1154          * Setup the thread mask for this irqaction for ONESHOT. For
1155          * !ONESHOT irqs the thread mask is 0 so we can avoid a
1156          * conditional in irq_wake_thread().
1157          */
1158         if (new->flags & IRQF_ONESHOT) {
1159                 /*
1160                  * Unlikely to have 32 resp 64 irqs sharing one line,
1161                  * but who knows.
1162                  */
1163                 if (thread_mask == ~0UL) {
1164                         ret = -EBUSY;
1165                         goto out_mask;
1166                 }
1167                 /*
1168                  * The thread_mask for the action is or'ed to
1169                  * desc->thread_active to indicate that the
1170                  * IRQF_ONESHOT thread handler has been woken, but not
1171                  * yet finished. The bit is cleared when a thread
1172                  * completes. When all threads of a shared interrupt
1173                  * line have completed desc->threads_active becomes
1174                  * zero and the interrupt line is unmasked. See
1175                  * handle.c:irq_wake_thread() for further information.
1176                  *
1177                  * If no thread is woken by primary (hard irq context)
1178                  * interrupt handlers, then desc->threads_active is
1179                  * also checked for zero to unmask the irq line in the
1180                  * affected hard irq flow handlers
1181                  * (handle_[fasteoi|level]_irq).
1182                  *
1183                  * The new action gets the first zero bit of
1184                  * thread_mask assigned. See the loop above which or's
1185                  * all existing action->thread_mask bits.
1186                  */
1187                 new->thread_mask = 1 << ffz(thread_mask);
1188
1189         } else if (new->handler == irq_default_primary_handler &&
1190                    !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1191                 /*
1192                  * The interrupt was requested with handler = NULL, so
1193                  * we use the default primary handler for it. But it
1194                  * does not have the oneshot flag set. In combination
1195                  * with level interrupts this is deadly, because the
1196                  * default primary handler just wakes the thread, then
1197                  * the irq lines is reenabled, but the device still
1198                  * has the level irq asserted. Rinse and repeat....
1199                  *
1200                  * While this works for edge type interrupts, we play
1201                  * it safe and reject unconditionally because we can't
1202                  * say for sure which type this interrupt really
1203                  * has. The type flags are unreliable as the
1204                  * underlying chip implementation can override them.
1205                  */
1206                 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1207                        irq);
1208                 ret = -EINVAL;
1209                 goto out_mask;
1210         }
1211
1212         if (!shared) {
1213                 ret = irq_request_resources(desc);
1214                 if (ret) {
1215                         pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1216                                new->name, irq, desc->irq_data.chip->name);
1217                         goto out_mask;
1218                 }
1219
1220                 init_waitqueue_head(&desc->wait_for_threads);
1221
1222                 /* Setup the type (level, edge polarity) if configured: */
1223                 if (new->flags & IRQF_TRIGGER_MASK) {
1224                         ret = __irq_set_trigger(desc, irq,
1225                                         new->flags & IRQF_TRIGGER_MASK);
1226
1227                         if (ret)
1228                                 goto out_mask;
1229                 }
1230
1231                 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1232                                   IRQS_ONESHOT | IRQS_WAITING);
1233                 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1234
1235                 if (new->flags & IRQF_PERCPU) {
1236                         irqd_set(&desc->irq_data, IRQD_PER_CPU);
1237                         irq_settings_set_per_cpu(desc);
1238                 }
1239
1240                 if (new->flags & IRQF_ONESHOT)
1241                         desc->istate |= IRQS_ONESHOT;
1242
1243                 if (irq_settings_can_autoenable(desc))
1244                         irq_startup(desc, true);
1245                 else
1246                         /* Undo nested disables: */
1247                         desc->depth = 1;
1248
1249                 /* Exclude IRQ from balancing if requested */
1250                 if (new->flags & IRQF_NOBALANCING) {
1251                         irq_settings_set_no_balancing(desc);
1252                         irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1253                 }
1254
1255                 /* Set default affinity mask once everything is setup */
1256                 setup_affinity(irq, desc, mask);
1257
1258         } else if (new->flags & IRQF_TRIGGER_MASK) {
1259                 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1260                 unsigned int omsk = irq_settings_get_trigger_mask(desc);
1261
1262                 if (nmsk != omsk)
1263                         /* hope the handler works with current  trigger mode */
1264                         pr_warning("irq %d uses trigger mode %u; requested %u\n",
1265                                    irq, nmsk, omsk);
1266         }
1267
1268         new->irq = irq;
1269         *old_ptr = new;
1270
1271         irq_pm_install_action(desc, new);
1272
1273         /* Reset broken irq detection when installing new handler */
1274         desc->irq_count = 0;
1275         desc->irqs_unhandled = 0;
1276
1277         /*
1278          * Check whether we disabled the irq via the spurious handler
1279          * before. Reenable it and give it another chance.
1280          */
1281         if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1282                 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1283                 __enable_irq(desc, irq);
1284         }
1285
1286         raw_spin_unlock_irqrestore(&desc->lock, flags);
1287
1288         /*
1289          * Strictly no need to wake it up, but hung_task complains
1290          * when no hard interrupt wakes the thread up.
1291          */
1292         if (new->thread)
1293                 wake_up_process(new->thread);
1294
1295         register_irq_proc(irq, desc);
1296         new->dir = NULL;
1297         register_handler_proc(irq, new);
1298         free_cpumask_var(mask);
1299
1300         return 0;
1301
1302 mismatch:
1303         if (!(new->flags & IRQF_PROBE_SHARED)) {
1304                 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1305                        irq, new->flags, new->name, old->flags, old->name);
1306 #ifdef CONFIG_DEBUG_SHIRQ
1307                 dump_stack();
1308 #endif
1309         }
1310         ret = -EBUSY;
1311
1312 out_mask:
1313         raw_spin_unlock_irqrestore(&desc->lock, flags);
1314         free_cpumask_var(mask);
1315
1316 out_thread:
1317         if (new->thread) {
1318                 struct task_struct *t = new->thread;
1319
1320                 new->thread = NULL;
1321                 kthread_stop(t);
1322                 put_task_struct(t);
1323         }
1324 out_mput:
1325         module_put(desc->owner);
1326         return ret;
1327 }
1328
1329 /**
1330  *      setup_irq - setup an interrupt
1331  *      @irq: Interrupt line to setup
1332  *      @act: irqaction for the interrupt
1333  *
1334  * Used to statically setup interrupts in the early boot process.
1335  */
1336 int setup_irq(unsigned int irq, struct irqaction *act)
1337 {
1338         int retval;
1339         struct irq_desc *desc = irq_to_desc(irq);
1340
1341         if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1342                 return -EINVAL;
1343         chip_bus_lock(desc);
1344         retval = __setup_irq(irq, desc, act);
1345         chip_bus_sync_unlock(desc);
1346
1347         return retval;
1348 }
1349 EXPORT_SYMBOL_GPL(setup_irq);
1350
1351 /*
1352  * Internal function to unregister an irqaction - used to free
1353  * regular and special interrupts that are part of the architecture.
1354  */
1355 static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1356 {
1357         struct irq_desc *desc = irq_to_desc(irq);
1358         struct irqaction *action, **action_ptr;
1359         unsigned long flags;
1360
1361         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1362
1363         if (!desc)
1364                 return NULL;
1365
1366         raw_spin_lock_irqsave(&desc->lock, flags);
1367
1368         /*
1369          * There can be multiple actions per IRQ descriptor, find the right
1370          * one based on the dev_id:
1371          */
1372         action_ptr = &desc->action;
1373         for (;;) {
1374                 action = *action_ptr;
1375
1376                 if (!action) {
1377                         WARN(1, "Trying to free already-free IRQ %d\n", irq);
1378                         raw_spin_unlock_irqrestore(&desc->lock, flags);
1379
1380                         return NULL;
1381                 }
1382
1383                 if (action->dev_id == dev_id)
1384                         break;
1385                 action_ptr = &action->next;
1386         }
1387
1388         /* Found it - now remove it from the list of entries: */
1389         *action_ptr = action->next;
1390
1391         irq_pm_remove_action(desc, action);
1392
1393         /* If this was the last handler, shut down the IRQ line: */
1394         if (!desc->action) {
1395                 irq_shutdown(desc);
1396                 irq_release_resources(desc);
1397         }
1398
1399 #ifdef CONFIG_SMP
1400         /* make sure affinity_hint is cleaned up */
1401         if (WARN_ON_ONCE(desc->affinity_hint))
1402                 desc->affinity_hint = NULL;
1403 #endif
1404
1405         raw_spin_unlock_irqrestore(&desc->lock, flags);
1406
1407         unregister_handler_proc(irq, action);
1408
1409         /* Make sure it's not being used on another CPU: */
1410         synchronize_irq(irq);
1411
1412 #ifdef CONFIG_DEBUG_SHIRQ
1413         /*
1414          * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1415          * event to happen even now it's being freed, so let's make sure that
1416          * is so by doing an extra call to the handler ....
1417          *
1418          * ( We do this after actually deregistering it, to make sure that a
1419          *   'real' IRQ doesn't run in * parallel with our fake. )
1420          */
1421         if (action->flags & IRQF_SHARED) {
1422                 local_irq_save(flags);
1423                 action->handler(irq, dev_id);
1424                 local_irq_restore(flags);
1425         }
1426 #endif
1427
1428         if (action->thread) {
1429                 kthread_stop(action->thread);
1430                 put_task_struct(action->thread);
1431         }
1432
1433         module_put(desc->owner);
1434         return action;
1435 }
1436
1437 /**
1438  *      remove_irq - free an interrupt
1439  *      @irq: Interrupt line to free
1440  *      @act: irqaction for the interrupt
1441  *
1442  * Used to remove interrupts statically setup by the early boot process.
1443  */
1444 void remove_irq(unsigned int irq, struct irqaction *act)
1445 {
1446         struct irq_desc *desc = irq_to_desc(irq);
1447
1448         if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1449             __free_irq(irq, act->dev_id);
1450 }
1451 EXPORT_SYMBOL_GPL(remove_irq);
1452
1453 /**
1454  *      free_irq - free an interrupt allocated with request_irq
1455  *      @irq: Interrupt line to free
1456  *      @dev_id: Device identity to free
1457  *
1458  *      Remove an interrupt handler. The handler is removed and if the
1459  *      interrupt line is no longer in use by any driver it is disabled.
1460  *      On a shared IRQ the caller must ensure the interrupt is disabled
1461  *      on the card it drives before calling this function. The function
1462  *      does not return until any executing interrupts for this IRQ
1463  *      have completed.
1464  *
1465  *      This function must not be called from interrupt context.
1466  */
1467 void free_irq(unsigned int irq, void *dev_id)
1468 {
1469         struct irq_desc *desc = irq_to_desc(irq);
1470
1471         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1472                 return;
1473
1474 #ifdef CONFIG_SMP
1475         if (WARN_ON(desc->affinity_notify))
1476                 desc->affinity_notify = NULL;
1477 #endif
1478
1479         chip_bus_lock(desc);
1480         kfree(__free_irq(irq, dev_id));
1481         chip_bus_sync_unlock(desc);
1482 }
1483 EXPORT_SYMBOL(free_irq);
1484
1485 /**
1486  *      request_threaded_irq - allocate an interrupt line
1487  *      @irq: Interrupt line to allocate
1488  *      @handler: Function to be called when the IRQ occurs.
1489  *                Primary handler for threaded interrupts
1490  *                If NULL and thread_fn != NULL the default
1491  *                primary handler is installed
1492  *      @thread_fn: Function called from the irq handler thread
1493  *                  If NULL, no irq thread is created
1494  *      @irqflags: Interrupt type flags
1495  *      @devname: An ascii name for the claiming device
1496  *      @dev_id: A cookie passed back to the handler function
1497  *
1498  *      This call allocates interrupt resources and enables the
1499  *      interrupt line and IRQ handling. From the point this
1500  *      call is made your handler function may be invoked. Since
1501  *      your handler function must clear any interrupt the board
1502  *      raises, you must take care both to initialise your hardware
1503  *      and to set up the interrupt handler in the right order.
1504  *
1505  *      If you want to set up a threaded irq handler for your device
1506  *      then you need to supply @handler and @thread_fn. @handler is
1507  *      still called in hard interrupt context and has to check
1508  *      whether the interrupt originates from the device. If yes it
1509  *      needs to disable the interrupt on the device and return
1510  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1511  *      @thread_fn. This split handler design is necessary to support
1512  *      shared interrupts.
1513  *
1514  *      Dev_id must be globally unique. Normally the address of the
1515  *      device data structure is used as the cookie. Since the handler
1516  *      receives this value it makes sense to use it.
1517  *
1518  *      If your interrupt is shared you must pass a non NULL dev_id
1519  *      as this is required when freeing the interrupt.
1520  *
1521  *      Flags:
1522  *
1523  *      IRQF_SHARED             Interrupt is shared
1524  *      IRQF_TRIGGER_*          Specify active edge(s) or level
1525  *
1526  */
1527 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1528                          irq_handler_t thread_fn, unsigned long irqflags,
1529                          const char *devname, void *dev_id)
1530 {
1531         struct irqaction *action;
1532         struct irq_desc *desc;
1533         int retval;
1534
1535         /*
1536          * Sanity-check: shared interrupts must pass in a real dev-ID,
1537          * otherwise we'll have trouble later trying to figure out
1538          * which interrupt is which (messes up the interrupt freeing
1539          * logic etc).
1540          *
1541          * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1542          * it cannot be set along with IRQF_NO_SUSPEND.
1543          */
1544         if (((irqflags & IRQF_SHARED) && !dev_id) ||
1545             (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1546             ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1547                 return -EINVAL;
1548
1549         desc = irq_to_desc(irq);
1550         if (!desc)
1551                 return -EINVAL;
1552
1553         if (!irq_settings_can_request(desc) ||
1554             WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1555                 return -EINVAL;
1556
1557         if (!handler) {
1558                 if (!thread_fn)
1559                         return -EINVAL;
1560                 handler = irq_default_primary_handler;
1561         }
1562
1563         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1564         if (!action)
1565                 return -ENOMEM;
1566
1567         action->handler = handler;
1568         action->thread_fn = thread_fn;
1569         action->flags = irqflags;
1570         action->name = devname;
1571         action->dev_id = dev_id;
1572
1573         chip_bus_lock(desc);
1574         retval = __setup_irq(irq, desc, action);
1575         chip_bus_sync_unlock(desc);
1576
1577         if (retval)
1578                 kfree(action);
1579
1580 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
1581         if (!retval && (irqflags & IRQF_SHARED)) {
1582                 /*
1583                  * It's a shared IRQ -- the driver ought to be prepared for it
1584                  * to happen immediately, so let's make sure....
1585                  * We disable the irq to make sure that a 'real' IRQ doesn't
1586                  * run in parallel with our fake.
1587                  */
1588                 unsigned long flags;
1589
1590                 disable_irq(irq);
1591                 local_irq_save(flags);
1592
1593                 handler(irq, dev_id);
1594
1595                 local_irq_restore(flags);
1596                 enable_irq(irq);
1597         }
1598 #endif
1599         return retval;
1600 }
1601 EXPORT_SYMBOL(request_threaded_irq);
1602
1603 /**
1604  *      request_any_context_irq - allocate an interrupt line
1605  *      @irq: Interrupt line to allocate
1606  *      @handler: Function to be called when the IRQ occurs.
1607  *                Threaded handler for threaded interrupts.
1608  *      @flags: Interrupt type flags
1609  *      @name: An ascii name for the claiming device
1610  *      @dev_id: A cookie passed back to the handler function
1611  *
1612  *      This call allocates interrupt resources and enables the
1613  *      interrupt line and IRQ handling. It selects either a
1614  *      hardirq or threaded handling method depending on the
1615  *      context.
1616  *
1617  *      On failure, it returns a negative value. On success,
1618  *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1619  */
1620 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1621                             unsigned long flags, const char *name, void *dev_id)
1622 {
1623         struct irq_desc *desc = irq_to_desc(irq);
1624         int ret;
1625
1626         if (!desc)
1627                 return -EINVAL;
1628
1629         if (irq_settings_is_nested_thread(desc)) {
1630                 ret = request_threaded_irq(irq, NULL, handler,
1631                                            flags, name, dev_id);
1632                 return !ret ? IRQC_IS_NESTED : ret;
1633         }
1634
1635         ret = request_irq(irq, handler, flags, name, dev_id);
1636         return !ret ? IRQC_IS_HARDIRQ : ret;
1637 }
1638 EXPORT_SYMBOL_GPL(request_any_context_irq);
1639
1640 void enable_percpu_irq(unsigned int irq, unsigned int type)
1641 {
1642         unsigned int cpu = smp_processor_id();
1643         unsigned long flags;
1644         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1645
1646         if (!desc)
1647                 return;
1648
1649         type &= IRQ_TYPE_SENSE_MASK;
1650         if (type != IRQ_TYPE_NONE) {
1651                 int ret;
1652
1653                 ret = __irq_set_trigger(desc, irq, type);
1654
1655                 if (ret) {
1656                         WARN(1, "failed to set type for IRQ%d\n", irq);
1657                         goto out;
1658                 }
1659         }
1660
1661         irq_percpu_enable(desc, cpu);
1662 out:
1663         irq_put_desc_unlock(desc, flags);
1664 }
1665 EXPORT_SYMBOL_GPL(enable_percpu_irq);
1666
1667 void disable_percpu_irq(unsigned int irq)
1668 {
1669         unsigned int cpu = smp_processor_id();
1670         unsigned long flags;
1671         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1672
1673         if (!desc)
1674                 return;
1675
1676         irq_percpu_disable(desc, cpu);
1677         irq_put_desc_unlock(desc, flags);
1678 }
1679 EXPORT_SYMBOL_GPL(disable_percpu_irq);
1680
1681 /*
1682  * Internal function to unregister a percpu irqaction.
1683  */
1684 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1685 {
1686         struct irq_desc *desc = irq_to_desc(irq);
1687         struct irqaction *action;
1688         unsigned long flags;
1689
1690         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1691
1692         if (!desc)
1693                 return NULL;
1694
1695         raw_spin_lock_irqsave(&desc->lock, flags);
1696
1697         action = desc->action;
1698         if (!action || action->percpu_dev_id != dev_id) {
1699                 WARN(1, "Trying to free already-free IRQ %d\n", irq);
1700                 goto bad;
1701         }
1702
1703         if (!cpumask_empty(desc->percpu_enabled)) {
1704                 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1705                      irq, cpumask_first(desc->percpu_enabled));
1706                 goto bad;
1707         }
1708
1709         /* Found it - now remove it from the list of entries: */
1710         desc->action = NULL;
1711
1712         raw_spin_unlock_irqrestore(&desc->lock, flags);
1713
1714         unregister_handler_proc(irq, action);
1715
1716         module_put(desc->owner);
1717         return action;
1718
1719 bad:
1720         raw_spin_unlock_irqrestore(&desc->lock, flags);
1721         return NULL;
1722 }
1723
1724 /**
1725  *      remove_percpu_irq - free a per-cpu interrupt
1726  *      @irq: Interrupt line to free
1727  *      @act: irqaction for the interrupt
1728  *
1729  * Used to remove interrupts statically setup by the early boot process.
1730  */
1731 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1732 {
1733         struct irq_desc *desc = irq_to_desc(irq);
1734
1735         if (desc && irq_settings_is_per_cpu_devid(desc))
1736             __free_percpu_irq(irq, act->percpu_dev_id);
1737 }
1738
1739 /**
1740  *      free_percpu_irq - free an interrupt allocated with request_percpu_irq
1741  *      @irq: Interrupt line to free
1742  *      @dev_id: Device identity to free
1743  *
1744  *      Remove a percpu interrupt handler. The handler is removed, but
1745  *      the interrupt line is not disabled. This must be done on each
1746  *      CPU before calling this function. The function does not return
1747  *      until any executing interrupts for this IRQ have completed.
1748  *
1749  *      This function must not be called from interrupt context.
1750  */
1751 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1752 {
1753         struct irq_desc *desc = irq_to_desc(irq);
1754
1755         if (!desc || !irq_settings_is_per_cpu_devid(desc))
1756                 return;
1757
1758         chip_bus_lock(desc);
1759         kfree(__free_percpu_irq(irq, dev_id));
1760         chip_bus_sync_unlock(desc);
1761 }
1762
1763 /**
1764  *      setup_percpu_irq - setup a per-cpu interrupt
1765  *      @irq: Interrupt line to setup
1766  *      @act: irqaction for the interrupt
1767  *
1768  * Used to statically setup per-cpu interrupts in the early boot process.
1769  */
1770 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1771 {
1772         struct irq_desc *desc = irq_to_desc(irq);
1773         int retval;
1774
1775         if (!desc || !irq_settings_is_per_cpu_devid(desc))
1776                 return -EINVAL;
1777         chip_bus_lock(desc);
1778         retval = __setup_irq(irq, desc, act);
1779         chip_bus_sync_unlock(desc);
1780
1781         return retval;
1782 }
1783
1784 /**
1785  *      request_percpu_irq - allocate a percpu interrupt line
1786  *      @irq: Interrupt line to allocate
1787  *      @handler: Function to be called when the IRQ occurs.
1788  *      @devname: An ascii name for the claiming device
1789  *      @dev_id: A percpu cookie passed back to the handler function
1790  *
1791  *      This call allocates interrupt resources, but doesn't
1792  *      automatically enable the interrupt. It has to be done on each
1793  *      CPU using enable_percpu_irq().
1794  *
1795  *      Dev_id must be globally unique. It is a per-cpu variable, and
1796  *      the handler gets called with the interrupted CPU's instance of
1797  *      that variable.
1798  */
1799 int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1800                        const char *devname, void __percpu *dev_id)
1801 {
1802         struct irqaction *action;
1803         struct irq_desc *desc;
1804         int retval;
1805
1806         if (!dev_id)
1807                 return -EINVAL;
1808
1809         desc = irq_to_desc(irq);
1810         if (!desc || !irq_settings_can_request(desc) ||
1811             !irq_settings_is_per_cpu_devid(desc))
1812                 return -EINVAL;
1813
1814         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1815         if (!action)
1816                 return -ENOMEM;
1817
1818         action->handler = handler;
1819         action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1820         action->name = devname;
1821         action->percpu_dev_id = dev_id;
1822
1823         chip_bus_lock(desc);
1824         retval = __setup_irq(irq, desc, action);
1825         chip_bus_sync_unlock(desc);
1826
1827         if (retval)
1828                 kfree(action);
1829
1830         return retval;
1831 }
1832
1833 /**
1834  *      irq_get_irqchip_state - returns the irqchip state of a interrupt.
1835  *      @irq: Interrupt line that is forwarded to a VM
1836  *      @which: One of IRQCHIP_STATE_* the caller wants to know about
1837  *      @state: a pointer to a boolean where the state is to be storeed
1838  *
1839  *      This call snapshots the internal irqchip state of an
1840  *      interrupt, returning into @state the bit corresponding to
1841  *      stage @which
1842  *
1843  *      This function should be called with preemption disabled if the
1844  *      interrupt controller has per-cpu registers.
1845  */
1846 int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1847                           bool *state)
1848 {
1849         struct irq_desc *desc;
1850         struct irq_data *data;
1851         struct irq_chip *chip;
1852         unsigned long flags;
1853         int err = -EINVAL;
1854
1855         desc = irq_get_desc_buslock(irq, &flags, 0);
1856         if (!desc)
1857                 return err;
1858
1859         data = irq_desc_get_irq_data(desc);
1860
1861         do {
1862                 chip = irq_data_get_irq_chip(data);
1863                 if (chip->irq_get_irqchip_state)
1864                         break;
1865 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1866                 data = data->parent_data;
1867 #else
1868                 data = NULL;
1869 #endif
1870         } while (data);
1871
1872         if (data)
1873                 err = chip->irq_get_irqchip_state(data, which, state);
1874
1875         irq_put_desc_busunlock(desc, flags);
1876         return err;
1877 }
1878
1879 /**
1880  *      irq_set_irqchip_state - set the state of a forwarded interrupt.
1881  *      @irq: Interrupt line that is forwarded to a VM
1882  *      @which: State to be restored (one of IRQCHIP_STATE_*)
1883  *      @val: Value corresponding to @which
1884  *
1885  *      This call sets the internal irqchip state of an interrupt,
1886  *      depending on the value of @which.
1887  *
1888  *      This function should be called with preemption disabled if the
1889  *      interrupt controller has per-cpu registers.
1890  */
1891 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1892                           bool val)
1893 {
1894         struct irq_desc *desc;
1895         struct irq_data *data;
1896         struct irq_chip *chip;
1897         unsigned long flags;
1898         int err = -EINVAL;
1899
1900         desc = irq_get_desc_buslock(irq, &flags, 0);
1901         if (!desc)
1902                 return err;
1903
1904         data = irq_desc_get_irq_data(desc);
1905
1906         do {
1907                 chip = irq_data_get_irq_chip(data);
1908                 if (chip->irq_set_irqchip_state)
1909                         break;
1910 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1911                 data = data->parent_data;
1912 #else
1913                 data = NULL;
1914 #endif
1915         } while (data);
1916
1917         if (data)
1918                 err = chip->irq_set_irqchip_state(data, which, val);
1919
1920         irq_put_desc_busunlock(desc, flags);
1921         return err;
1922 }