nvme: fix Kconfig description for BLK_DEV_NVME_SCSI
[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 static int __irq_can_set_affinity(struct irq_desc *desc)
119 {
120         if (!desc || !irqd_can_balance(&desc->irq_data) ||
121             !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
122                 return 0;
123         return 1;
124 }
125
126 /**
127  *      irq_can_set_affinity - Check if the affinity of a given irq can be set
128  *      @irq:           Interrupt to check
129  *
130  */
131 int irq_can_set_affinity(unsigned int irq)
132 {
133         return __irq_can_set_affinity(irq_to_desc(irq));
134 }
135
136 /**
137  *      irq_set_thread_affinity - Notify irq threads to adjust affinity
138  *      @desc:          irq descriptor which has affitnity changed
139  *
140  *      We just set IRQTF_AFFINITY and delegate the affinity setting
141  *      to the interrupt thread itself. We can not call
142  *      set_cpus_allowed_ptr() here as we hold desc->lock and this
143  *      code can be called from hard interrupt context.
144  */
145 void irq_set_thread_affinity(struct irq_desc *desc)
146 {
147         struct irqaction *action = desc->action;
148
149         while (action) {
150                 if (action->thread)
151                         set_bit(IRQTF_AFFINITY, &action->thread_flags);
152                 action = action->next;
153         }
154 }
155
156 #ifdef CONFIG_GENERIC_PENDING_IRQ
157 static inline bool irq_can_move_pcntxt(struct irq_data *data)
158 {
159         return irqd_can_move_in_process_context(data);
160 }
161 static inline bool irq_move_pending(struct irq_data *data)
162 {
163         return irqd_is_setaffinity_pending(data);
164 }
165 static inline void
166 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
167 {
168         cpumask_copy(desc->pending_mask, mask);
169 }
170 static inline void
171 irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
172 {
173         cpumask_copy(mask, desc->pending_mask);
174 }
175 #else
176 static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
177 static inline bool irq_move_pending(struct irq_data *data) { return false; }
178 static inline void
179 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
180 static inline void
181 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
182 #endif
183
184 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
185                         bool force)
186 {
187         struct irq_desc *desc = irq_data_to_desc(data);
188         struct irq_chip *chip = irq_data_get_irq_chip(data);
189         int ret;
190
191         ret = chip->irq_set_affinity(data, mask, force);
192         switch (ret) {
193         case IRQ_SET_MASK_OK:
194         case IRQ_SET_MASK_OK_DONE:
195                 cpumask_copy(desc->irq_common_data.affinity, mask);
196         case IRQ_SET_MASK_OK_NOCOPY:
197                 irq_set_thread_affinity(desc);
198                 ret = 0;
199         }
200
201         return ret;
202 }
203
204 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
205                             bool force)
206 {
207         struct irq_chip *chip = irq_data_get_irq_chip(data);
208         struct irq_desc *desc = irq_data_to_desc(data);
209         int ret = 0;
210
211         if (!chip || !chip->irq_set_affinity)
212                 return -EINVAL;
213
214         if (irq_can_move_pcntxt(data)) {
215                 ret = irq_do_set_affinity(data, mask, force);
216         } else {
217                 irqd_set_move_pending(data);
218                 irq_copy_pending(desc, mask);
219         }
220
221         if (desc->affinity_notify) {
222                 kref_get(&desc->affinity_notify->kref);
223                 schedule_work(&desc->affinity_notify->work);
224         }
225         irqd_set(data, IRQD_AFFINITY_SET);
226
227         return ret;
228 }
229
230 int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
231 {
232         struct irq_desc *desc = irq_to_desc(irq);
233         unsigned long flags;
234         int ret;
235
236         if (!desc)
237                 return -EINVAL;
238
239         raw_spin_lock_irqsave(&desc->lock, flags);
240         ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
241         raw_spin_unlock_irqrestore(&desc->lock, flags);
242         return ret;
243 }
244
245 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
246 {
247         unsigned long flags;
248         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
249
250         if (!desc)
251                 return -EINVAL;
252         desc->affinity_hint = m;
253         irq_put_desc_unlock(desc, flags);
254         /* set the initial affinity to prevent every interrupt being on CPU0 */
255         if (m)
256                 __irq_set_affinity(irq, m, false);
257         return 0;
258 }
259 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
260
261 static void irq_affinity_notify(struct work_struct *work)
262 {
263         struct irq_affinity_notify *notify =
264                 container_of(work, struct irq_affinity_notify, work);
265         struct irq_desc *desc = irq_to_desc(notify->irq);
266         cpumask_var_t cpumask;
267         unsigned long flags;
268
269         if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
270                 goto out;
271
272         raw_spin_lock_irqsave(&desc->lock, flags);
273         if (irq_move_pending(&desc->irq_data))
274                 irq_get_pending(cpumask, desc);
275         else
276                 cpumask_copy(cpumask, desc->irq_common_data.affinity);
277         raw_spin_unlock_irqrestore(&desc->lock, flags);
278
279         notify->notify(notify, cpumask);
280
281         free_cpumask_var(cpumask);
282 out:
283         kref_put(&notify->kref, notify->release);
284 }
285
286 /**
287  *      irq_set_affinity_notifier - control notification of IRQ affinity changes
288  *      @irq:           Interrupt for which to enable/disable notification
289  *      @notify:        Context for notification, or %NULL to disable
290  *                      notification.  Function pointers must be initialised;
291  *                      the other fields will be initialised by this function.
292  *
293  *      Must be called in process context.  Notification may only be enabled
294  *      after the IRQ is allocated and must be disabled before the IRQ is
295  *      freed using free_irq().
296  */
297 int
298 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
299 {
300         struct irq_desc *desc = irq_to_desc(irq);
301         struct irq_affinity_notify *old_notify;
302         unsigned long flags;
303
304         /* The release function is promised process context */
305         might_sleep();
306
307         if (!desc)
308                 return -EINVAL;
309
310         /* Complete initialisation of *notify */
311         if (notify) {
312                 notify->irq = irq;
313                 kref_init(&notify->kref);
314                 INIT_WORK(&notify->work, irq_affinity_notify);
315         }
316
317         raw_spin_lock_irqsave(&desc->lock, flags);
318         old_notify = desc->affinity_notify;
319         desc->affinity_notify = notify;
320         raw_spin_unlock_irqrestore(&desc->lock, flags);
321
322         if (old_notify)
323                 kref_put(&old_notify->kref, old_notify->release);
324
325         return 0;
326 }
327 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
328
329 #ifndef CONFIG_AUTO_IRQ_AFFINITY
330 /*
331  * Generic version of the affinity autoselector.
332  */
333 static int setup_affinity(struct irq_desc *desc, struct cpumask *mask)
334 {
335         struct cpumask *set = irq_default_affinity;
336         int node = irq_desc_get_node(desc);
337
338         /* Excludes PER_CPU and NO_BALANCE interrupts */
339         if (!__irq_can_set_affinity(desc))
340                 return 0;
341
342         /*
343          * Preserve an userspace affinity setup, but make sure that
344          * one of the targets is online.
345          */
346         if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
347                 if (cpumask_intersects(desc->irq_common_data.affinity,
348                                        cpu_online_mask))
349                         set = desc->irq_common_data.affinity;
350                 else
351                         irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
352         }
353
354         cpumask_and(mask, cpu_online_mask, set);
355         if (node != NUMA_NO_NODE) {
356                 const struct cpumask *nodemask = cpumask_of_node(node);
357
358                 /* make sure at least one of the cpus in nodemask is online */
359                 if (cpumask_intersects(mask, nodemask))
360                         cpumask_and(mask, mask, nodemask);
361         }
362         irq_do_set_affinity(&desc->irq_data, mask, false);
363         return 0;
364 }
365 #else
366 /* Wrapper for ALPHA specific affinity selector magic */
367 static inline int setup_affinity(struct irq_desc *d, struct cpumask *mask)
368 {
369         return irq_select_affinity(irq_desc_get_irq(d));
370 }
371 #endif
372
373 /*
374  * Called when affinity is set via /proc/irq
375  */
376 int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
377 {
378         struct irq_desc *desc = irq_to_desc(irq);
379         unsigned long flags;
380         int ret;
381
382         raw_spin_lock_irqsave(&desc->lock, flags);
383         ret = setup_affinity(desc, mask);
384         raw_spin_unlock_irqrestore(&desc->lock, flags);
385         return ret;
386 }
387
388 #else
389 static inline int
390 setup_affinity(struct irq_desc *desc, struct cpumask *mask)
391 {
392         return 0;
393 }
394 #endif
395
396 /**
397  *      irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
398  *      @irq: interrupt number to set affinity
399  *      @vcpu_info: vCPU specific data
400  *
401  *      This function uses the vCPU specific data to set the vCPU
402  *      affinity for an irq. The vCPU specific data is passed from
403  *      outside, such as KVM. One example code path is as below:
404  *      KVM -> IOMMU -> irq_set_vcpu_affinity().
405  */
406 int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
407 {
408         unsigned long flags;
409         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
410         struct irq_data *data;
411         struct irq_chip *chip;
412         int ret = -ENOSYS;
413
414         if (!desc)
415                 return -EINVAL;
416
417         data = irq_desc_get_irq_data(desc);
418         chip = irq_data_get_irq_chip(data);
419         if (chip && chip->irq_set_vcpu_affinity)
420                 ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
421         irq_put_desc_unlock(desc, flags);
422
423         return ret;
424 }
425 EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
426
427 void __disable_irq(struct irq_desc *desc)
428 {
429         if (!desc->depth++)
430                 irq_disable(desc);
431 }
432
433 static int __disable_irq_nosync(unsigned int irq)
434 {
435         unsigned long flags;
436         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
437
438         if (!desc)
439                 return -EINVAL;
440         __disable_irq(desc);
441         irq_put_desc_busunlock(desc, flags);
442         return 0;
443 }
444
445 /**
446  *      disable_irq_nosync - disable an irq without waiting
447  *      @irq: Interrupt to disable
448  *
449  *      Disable the selected interrupt line.  Disables and Enables are
450  *      nested.
451  *      Unlike disable_irq(), this function does not ensure existing
452  *      instances of the IRQ handler have completed before returning.
453  *
454  *      This function may be called from IRQ context.
455  */
456 void disable_irq_nosync(unsigned int irq)
457 {
458         __disable_irq_nosync(irq);
459 }
460 EXPORT_SYMBOL(disable_irq_nosync);
461
462 /**
463  *      disable_irq - disable an irq and wait for completion
464  *      @irq: Interrupt to disable
465  *
466  *      Disable the selected interrupt line.  Enables and Disables are
467  *      nested.
468  *      This function waits for any pending IRQ handlers for this interrupt
469  *      to complete before returning. If you use this function while
470  *      holding a resource the IRQ handler may need you will deadlock.
471  *
472  *      This function may be called - with care - from IRQ context.
473  */
474 void disable_irq(unsigned int irq)
475 {
476         if (!__disable_irq_nosync(irq))
477                 synchronize_irq(irq);
478 }
479 EXPORT_SYMBOL(disable_irq);
480
481 /**
482  *      disable_hardirq - disables an irq and waits for hardirq completion
483  *      @irq: Interrupt to disable
484  *
485  *      Disable the selected interrupt line.  Enables and Disables are
486  *      nested.
487  *      This function waits for any pending hard IRQ handlers for this
488  *      interrupt to complete before returning. If you use this function while
489  *      holding a resource the hard IRQ handler may need you will deadlock.
490  *
491  *      When used to optimistically disable an interrupt from atomic context
492  *      the return value must be checked.
493  *
494  *      Returns: false if a threaded handler is active.
495  *
496  *      This function may be called - with care - from IRQ context.
497  */
498 bool disable_hardirq(unsigned int irq)
499 {
500         if (!__disable_irq_nosync(irq))
501                 return synchronize_hardirq(irq);
502
503         return false;
504 }
505 EXPORT_SYMBOL_GPL(disable_hardirq);
506
507 void __enable_irq(struct irq_desc *desc)
508 {
509         switch (desc->depth) {
510         case 0:
511  err_out:
512                 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
513                      irq_desc_get_irq(desc));
514                 break;
515         case 1: {
516                 if (desc->istate & IRQS_SUSPENDED)
517                         goto err_out;
518                 /* Prevent probing on this irq: */
519                 irq_settings_set_noprobe(desc);
520                 irq_enable(desc);
521                 check_irq_resend(desc);
522                 /* fall-through */
523         }
524         default:
525                 desc->depth--;
526         }
527 }
528
529 /**
530  *      enable_irq - enable handling of an irq
531  *      @irq: Interrupt to enable
532  *
533  *      Undoes the effect of one call to disable_irq().  If this
534  *      matches the last disable, processing of interrupts on this
535  *      IRQ line is re-enabled.
536  *
537  *      This function may be called from IRQ context only when
538  *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
539  */
540 void enable_irq(unsigned int irq)
541 {
542         unsigned long flags;
543         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
544
545         if (!desc)
546                 return;
547         if (WARN(!desc->irq_data.chip,
548                  KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
549                 goto out;
550
551         __enable_irq(desc);
552 out:
553         irq_put_desc_busunlock(desc, flags);
554 }
555 EXPORT_SYMBOL(enable_irq);
556
557 static int set_irq_wake_real(unsigned int irq, unsigned int on)
558 {
559         struct irq_desc *desc = irq_to_desc(irq);
560         int ret = -ENXIO;
561
562         if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
563                 return 0;
564
565         if (desc->irq_data.chip->irq_set_wake)
566                 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
567
568         return ret;
569 }
570
571 /**
572  *      irq_set_irq_wake - control irq power management wakeup
573  *      @irq:   interrupt to control
574  *      @on:    enable/disable power management wakeup
575  *
576  *      Enable/disable power management wakeup mode, which is
577  *      disabled by default.  Enables and disables must match,
578  *      just as they match for non-wakeup mode support.
579  *
580  *      Wakeup mode lets this IRQ wake the system from sleep
581  *      states like "suspend to RAM".
582  */
583 int irq_set_irq_wake(unsigned int irq, unsigned int on)
584 {
585         unsigned long flags;
586         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
587         int ret = 0;
588
589         if (!desc)
590                 return -EINVAL;
591
592         /* wakeup-capable irqs can be shared between drivers that
593          * don't need to have the same sleep mode behaviors.
594          */
595         if (on) {
596                 if (desc->wake_depth++ == 0) {
597                         ret = set_irq_wake_real(irq, on);
598                         if (ret)
599                                 desc->wake_depth = 0;
600                         else
601                                 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
602                 }
603         } else {
604                 if (desc->wake_depth == 0) {
605                         WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
606                 } else if (--desc->wake_depth == 0) {
607                         ret = set_irq_wake_real(irq, on);
608                         if (ret)
609                                 desc->wake_depth = 1;
610                         else
611                                 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
612                 }
613         }
614         irq_put_desc_busunlock(desc, flags);
615         return ret;
616 }
617 EXPORT_SYMBOL(irq_set_irq_wake);
618
619 /*
620  * Internal function that tells the architecture code whether a
621  * particular irq has been exclusively allocated or is available
622  * for driver use.
623  */
624 int can_request_irq(unsigned int irq, unsigned long irqflags)
625 {
626         unsigned long flags;
627         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
628         int canrequest = 0;
629
630         if (!desc)
631                 return 0;
632
633         if (irq_settings_can_request(desc)) {
634                 if (!desc->action ||
635                     irqflags & desc->action->flags & IRQF_SHARED)
636                         canrequest = 1;
637         }
638         irq_put_desc_unlock(desc, flags);
639         return canrequest;
640 }
641
642 int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
643 {
644         struct irq_chip *chip = desc->irq_data.chip;
645         int ret, unmask = 0;
646
647         if (!chip || !chip->irq_set_type) {
648                 /*
649                  * IRQF_TRIGGER_* but the PIC does not support multiple
650                  * flow-types?
651                  */
652                 pr_debug("No set_type function for IRQ %d (%s)\n",
653                          irq_desc_get_irq(desc),
654                          chip ? (chip->name ? : "unknown") : "unknown");
655                 return 0;
656         }
657
658         flags &= IRQ_TYPE_SENSE_MASK;
659
660         if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
661                 if (!irqd_irq_masked(&desc->irq_data))
662                         mask_irq(desc);
663                 if (!irqd_irq_disabled(&desc->irq_data))
664                         unmask = 1;
665         }
666
667         /* caller masked out all except trigger mode flags */
668         ret = chip->irq_set_type(&desc->irq_data, flags);
669
670         switch (ret) {
671         case IRQ_SET_MASK_OK:
672         case IRQ_SET_MASK_OK_DONE:
673                 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
674                 irqd_set(&desc->irq_data, flags);
675
676         case IRQ_SET_MASK_OK_NOCOPY:
677                 flags = irqd_get_trigger_type(&desc->irq_data);
678                 irq_settings_set_trigger_mask(desc, flags);
679                 irqd_clear(&desc->irq_data, IRQD_LEVEL);
680                 irq_settings_clr_level(desc);
681                 if (flags & IRQ_TYPE_LEVEL_MASK) {
682                         irq_settings_set_level(desc);
683                         irqd_set(&desc->irq_data, IRQD_LEVEL);
684                 }
685
686                 ret = 0;
687                 break;
688         default:
689                 pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
690                        flags, irq_desc_get_irq(desc), chip->irq_set_type);
691         }
692         if (unmask)
693                 unmask_irq(desc);
694         return ret;
695 }
696
697 #ifdef CONFIG_HARDIRQS_SW_RESEND
698 int irq_set_parent(int irq, int parent_irq)
699 {
700         unsigned long flags;
701         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
702
703         if (!desc)
704                 return -EINVAL;
705
706         desc->parent_irq = parent_irq;
707
708         irq_put_desc_unlock(desc, flags);
709         return 0;
710 }
711 #endif
712
713 /*
714  * Default primary interrupt handler for threaded interrupts. Is
715  * assigned as primary handler when request_threaded_irq is called
716  * with handler == NULL. Useful for oneshot interrupts.
717  */
718 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
719 {
720         return IRQ_WAKE_THREAD;
721 }
722
723 /*
724  * Primary handler for nested threaded interrupts. Should never be
725  * called.
726  */
727 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
728 {
729         WARN(1, "Primary handler called for nested irq %d\n", irq);
730         return IRQ_NONE;
731 }
732
733 static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
734 {
735         WARN(1, "Secondary action handler called for irq %d\n", irq);
736         return IRQ_NONE;
737 }
738
739 static int irq_wait_for_interrupt(struct irqaction *action)
740 {
741         set_current_state(TASK_INTERRUPTIBLE);
742
743         while (!kthread_should_stop()) {
744
745                 if (test_and_clear_bit(IRQTF_RUNTHREAD,
746                                        &action->thread_flags)) {
747                         __set_current_state(TASK_RUNNING);
748                         return 0;
749                 }
750                 schedule();
751                 set_current_state(TASK_INTERRUPTIBLE);
752         }
753         __set_current_state(TASK_RUNNING);
754         return -1;
755 }
756
757 /*
758  * Oneshot interrupts keep the irq line masked until the threaded
759  * handler finished. unmask if the interrupt has not been disabled and
760  * is marked MASKED.
761  */
762 static void irq_finalize_oneshot(struct irq_desc *desc,
763                                  struct irqaction *action)
764 {
765         if (!(desc->istate & IRQS_ONESHOT) ||
766             action->handler == irq_forced_secondary_handler)
767                 return;
768 again:
769         chip_bus_lock(desc);
770         raw_spin_lock_irq(&desc->lock);
771
772         /*
773          * Implausible though it may be we need to protect us against
774          * the following scenario:
775          *
776          * The thread is faster done than the hard interrupt handler
777          * on the other CPU. If we unmask the irq line then the
778          * interrupt can come in again and masks the line, leaves due
779          * to IRQS_INPROGRESS and the irq line is masked forever.
780          *
781          * This also serializes the state of shared oneshot handlers
782          * versus "desc->threads_onehsot |= action->thread_mask;" in
783          * irq_wake_thread(). See the comment there which explains the
784          * serialization.
785          */
786         if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
787                 raw_spin_unlock_irq(&desc->lock);
788                 chip_bus_sync_unlock(desc);
789                 cpu_relax();
790                 goto again;
791         }
792
793         /*
794          * Now check again, whether the thread should run. Otherwise
795          * we would clear the threads_oneshot bit of this thread which
796          * was just set.
797          */
798         if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
799                 goto out_unlock;
800
801         desc->threads_oneshot &= ~action->thread_mask;
802
803         if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
804             irqd_irq_masked(&desc->irq_data))
805                 unmask_threaded_irq(desc);
806
807 out_unlock:
808         raw_spin_unlock_irq(&desc->lock);
809         chip_bus_sync_unlock(desc);
810 }
811
812 #ifdef CONFIG_SMP
813 /*
814  * Check whether we need to change the affinity of the interrupt thread.
815  */
816 static void
817 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
818 {
819         cpumask_var_t mask;
820         bool valid = true;
821
822         if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
823                 return;
824
825         /*
826          * In case we are out of memory we set IRQTF_AFFINITY again and
827          * try again next time
828          */
829         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
830                 set_bit(IRQTF_AFFINITY, &action->thread_flags);
831                 return;
832         }
833
834         raw_spin_lock_irq(&desc->lock);
835         /*
836          * This code is triggered unconditionally. Check the affinity
837          * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
838          */
839         if (desc->irq_common_data.affinity)
840                 cpumask_copy(mask, desc->irq_common_data.affinity);
841         else
842                 valid = false;
843         raw_spin_unlock_irq(&desc->lock);
844
845         if (valid)
846                 set_cpus_allowed_ptr(current, mask);
847         free_cpumask_var(mask);
848 }
849 #else
850 static inline void
851 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
852 #endif
853
854 /*
855  * Interrupts which are not explicitely requested as threaded
856  * interrupts rely on the implicit bh/preempt disable of the hard irq
857  * context. So we need to disable bh here to avoid deadlocks and other
858  * side effects.
859  */
860 static irqreturn_t
861 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
862 {
863         irqreturn_t ret;
864
865         local_bh_disable();
866         ret = action->thread_fn(action->irq, action->dev_id);
867         irq_finalize_oneshot(desc, action);
868         local_bh_enable();
869         return ret;
870 }
871
872 /*
873  * Interrupts explicitly requested as threaded interrupts want to be
874  * preemtible - many of them need to sleep and wait for slow busses to
875  * complete.
876  */
877 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
878                 struct irqaction *action)
879 {
880         irqreturn_t ret;
881
882         ret = action->thread_fn(action->irq, action->dev_id);
883         irq_finalize_oneshot(desc, action);
884         return ret;
885 }
886
887 static void wake_threads_waitq(struct irq_desc *desc)
888 {
889         if (atomic_dec_and_test(&desc->threads_active))
890                 wake_up(&desc->wait_for_threads);
891 }
892
893 static void irq_thread_dtor(struct callback_head *unused)
894 {
895         struct task_struct *tsk = current;
896         struct irq_desc *desc;
897         struct irqaction *action;
898
899         if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
900                 return;
901
902         action = kthread_data(tsk);
903
904         pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
905                tsk->comm, tsk->pid, action->irq);
906
907
908         desc = irq_to_desc(action->irq);
909         /*
910          * If IRQTF_RUNTHREAD is set, we need to decrement
911          * desc->threads_active and wake possible waiters.
912          */
913         if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
914                 wake_threads_waitq(desc);
915
916         /* Prevent a stale desc->threads_oneshot */
917         irq_finalize_oneshot(desc, action);
918 }
919
920 static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
921 {
922         struct irqaction *secondary = action->secondary;
923
924         if (WARN_ON_ONCE(!secondary))
925                 return;
926
927         raw_spin_lock_irq(&desc->lock);
928         __irq_wake_thread(desc, secondary);
929         raw_spin_unlock_irq(&desc->lock);
930 }
931
932 /*
933  * Interrupt handler thread
934  */
935 static int irq_thread(void *data)
936 {
937         struct callback_head on_exit_work;
938         struct irqaction *action = data;
939         struct irq_desc *desc = irq_to_desc(action->irq);
940         irqreturn_t (*handler_fn)(struct irq_desc *desc,
941                         struct irqaction *action);
942
943         if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
944                                         &action->thread_flags))
945                 handler_fn = irq_forced_thread_fn;
946         else
947                 handler_fn = irq_thread_fn;
948
949         init_task_work(&on_exit_work, irq_thread_dtor);
950         task_work_add(current, &on_exit_work, false);
951
952         irq_thread_check_affinity(desc, action);
953
954         while (!irq_wait_for_interrupt(action)) {
955                 irqreturn_t action_ret;
956
957                 irq_thread_check_affinity(desc, action);
958
959                 action_ret = handler_fn(desc, action);
960                 if (action_ret == IRQ_HANDLED)
961                         atomic_inc(&desc->threads_handled);
962                 if (action_ret == IRQ_WAKE_THREAD)
963                         irq_wake_secondary(desc, action);
964
965                 wake_threads_waitq(desc);
966         }
967
968         /*
969          * This is the regular exit path. __free_irq() is stopping the
970          * thread via kthread_stop() after calling
971          * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
972          * oneshot mask bit can be set. We cannot verify that as we
973          * cannot touch the oneshot mask at this point anymore as
974          * __setup_irq() might have given out currents thread_mask
975          * again.
976          */
977         task_work_cancel(current, irq_thread_dtor);
978         return 0;
979 }
980
981 /**
982  *      irq_wake_thread - wake the irq thread for the action identified by dev_id
983  *      @irq:           Interrupt line
984  *      @dev_id:        Device identity for which the thread should be woken
985  *
986  */
987 void irq_wake_thread(unsigned int irq, void *dev_id)
988 {
989         struct irq_desc *desc = irq_to_desc(irq);
990         struct irqaction *action;
991         unsigned long flags;
992
993         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
994                 return;
995
996         raw_spin_lock_irqsave(&desc->lock, flags);
997         for (action = desc->action; action; action = action->next) {
998                 if (action->dev_id == dev_id) {
999                         if (action->thread)
1000                                 __irq_wake_thread(desc, action);
1001                         break;
1002                 }
1003         }
1004         raw_spin_unlock_irqrestore(&desc->lock, flags);
1005 }
1006 EXPORT_SYMBOL_GPL(irq_wake_thread);
1007
1008 static int irq_setup_forced_threading(struct irqaction *new)
1009 {
1010         if (!force_irqthreads)
1011                 return 0;
1012         if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1013                 return 0;
1014
1015         new->flags |= IRQF_ONESHOT;
1016
1017         /*
1018          * Handle the case where we have a real primary handler and a
1019          * thread handler. We force thread them as well by creating a
1020          * secondary action.
1021          */
1022         if (new->handler != irq_default_primary_handler && new->thread_fn) {
1023                 /* Allocate the secondary action */
1024                 new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1025                 if (!new->secondary)
1026                         return -ENOMEM;
1027                 new->secondary->handler = irq_forced_secondary_handler;
1028                 new->secondary->thread_fn = new->thread_fn;
1029                 new->secondary->dev_id = new->dev_id;
1030                 new->secondary->irq = new->irq;
1031                 new->secondary->name = new->name;
1032         }
1033         /* Deal with the primary handler */
1034         set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1035         new->thread_fn = new->handler;
1036         new->handler = irq_default_primary_handler;
1037         return 0;
1038 }
1039
1040 static int irq_request_resources(struct irq_desc *desc)
1041 {
1042         struct irq_data *d = &desc->irq_data;
1043         struct irq_chip *c = d->chip;
1044
1045         return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1046 }
1047
1048 static void irq_release_resources(struct irq_desc *desc)
1049 {
1050         struct irq_data *d = &desc->irq_data;
1051         struct irq_chip *c = d->chip;
1052
1053         if (c->irq_release_resources)
1054                 c->irq_release_resources(d);
1055 }
1056
1057 static int
1058 setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1059 {
1060         struct task_struct *t;
1061         struct sched_param param = {
1062                 .sched_priority = MAX_USER_RT_PRIO/2,
1063         };
1064
1065         if (!secondary) {
1066                 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1067                                    new->name);
1068         } else {
1069                 t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1070                                    new->name);
1071                 param.sched_priority -= 1;
1072         }
1073
1074         if (IS_ERR(t))
1075                 return PTR_ERR(t);
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         return 0;
1097 }
1098
1099 /*
1100  * Internal function to register an irqaction - typically used to
1101  * allocate special interrupts that are part of the architecture.
1102  */
1103 static int
1104 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1105 {
1106         struct irqaction *old, **old_ptr;
1107         unsigned long flags, thread_mask = 0;
1108         int ret, nested, shared = 0;
1109         cpumask_var_t mask;
1110
1111         if (!desc)
1112                 return -EINVAL;
1113
1114         if (desc->irq_data.chip == &no_irq_chip)
1115                 return -ENOSYS;
1116         if (!try_module_get(desc->owner))
1117                 return -ENODEV;
1118
1119         new->irq = irq;
1120
1121         /*
1122          * Check whether the interrupt nests into another interrupt
1123          * thread.
1124          */
1125         nested = irq_settings_is_nested_thread(desc);
1126         if (nested) {
1127                 if (!new->thread_fn) {
1128                         ret = -EINVAL;
1129                         goto out_mput;
1130                 }
1131                 /*
1132                  * Replace the primary handler which was provided from
1133                  * the driver for non nested interrupt handling by the
1134                  * dummy function which warns when called.
1135                  */
1136                 new->handler = irq_nested_primary_handler;
1137         } else {
1138                 if (irq_settings_can_thread(desc)) {
1139                         ret = irq_setup_forced_threading(new);
1140                         if (ret)
1141                                 goto out_mput;
1142                 }
1143         }
1144
1145         /*
1146          * Create a handler thread when a thread function is supplied
1147          * and the interrupt does not nest into another interrupt
1148          * thread.
1149          */
1150         if (new->thread_fn && !nested) {
1151                 ret = setup_irq_thread(new, irq, false);
1152                 if (ret)
1153                         goto out_mput;
1154                 if (new->secondary) {
1155                         ret = setup_irq_thread(new->secondary, irq, true);
1156                         if (ret)
1157                                 goto out_thread;
1158                 }
1159         }
1160
1161         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1162                 ret = -ENOMEM;
1163                 goto out_thread;
1164         }
1165
1166         /*
1167          * Drivers are often written to work w/o knowledge about the
1168          * underlying irq chip implementation, so a request for a
1169          * threaded irq without a primary hard irq context handler
1170          * requires the ONESHOT flag to be set. Some irq chips like
1171          * MSI based interrupts are per se one shot safe. Check the
1172          * chip flags, so we can avoid the unmask dance at the end of
1173          * the threaded handler for those.
1174          */
1175         if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1176                 new->flags &= ~IRQF_ONESHOT;
1177
1178         /*
1179          * The following block of code has to be executed atomically
1180          */
1181         raw_spin_lock_irqsave(&desc->lock, flags);
1182         old_ptr = &desc->action;
1183         old = *old_ptr;
1184         if (old) {
1185                 /*
1186                  * Can't share interrupts unless both agree to and are
1187                  * the same type (level, edge, polarity). So both flag
1188                  * fields must have IRQF_SHARED set and the bits which
1189                  * set the trigger type must match. Also all must
1190                  * agree on ONESHOT.
1191                  */
1192                 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1193                     ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1194                     ((old->flags ^ new->flags) & IRQF_ONESHOT))
1195                         goto mismatch;
1196
1197                 /* All handlers must agree on per-cpuness */
1198                 if ((old->flags & IRQF_PERCPU) !=
1199                     (new->flags & IRQF_PERCPU))
1200                         goto mismatch;
1201
1202                 /* add new interrupt at end of irq queue */
1203                 do {
1204                         /*
1205                          * Or all existing action->thread_mask bits,
1206                          * so we can find the next zero bit for this
1207                          * new action.
1208                          */
1209                         thread_mask |= old->thread_mask;
1210                         old_ptr = &old->next;
1211                         old = *old_ptr;
1212                 } while (old);
1213                 shared = 1;
1214         }
1215
1216         /*
1217          * Setup the thread mask for this irqaction for ONESHOT. For
1218          * !ONESHOT irqs the thread mask is 0 so we can avoid a
1219          * conditional in irq_wake_thread().
1220          */
1221         if (new->flags & IRQF_ONESHOT) {
1222                 /*
1223                  * Unlikely to have 32 resp 64 irqs sharing one line,
1224                  * but who knows.
1225                  */
1226                 if (thread_mask == ~0UL) {
1227                         ret = -EBUSY;
1228                         goto out_mask;
1229                 }
1230                 /*
1231                  * The thread_mask for the action is or'ed to
1232                  * desc->thread_active to indicate that the
1233                  * IRQF_ONESHOT thread handler has been woken, but not
1234                  * yet finished. The bit is cleared when a thread
1235                  * completes. When all threads of a shared interrupt
1236                  * line have completed desc->threads_active becomes
1237                  * zero and the interrupt line is unmasked. See
1238                  * handle.c:irq_wake_thread() for further information.
1239                  *
1240                  * If no thread is woken by primary (hard irq context)
1241                  * interrupt handlers, then desc->threads_active is
1242                  * also checked for zero to unmask the irq line in the
1243                  * affected hard irq flow handlers
1244                  * (handle_[fasteoi|level]_irq).
1245                  *
1246                  * The new action gets the first zero bit of
1247                  * thread_mask assigned. See the loop above which or's
1248                  * all existing action->thread_mask bits.
1249                  */
1250                 new->thread_mask = 1 << ffz(thread_mask);
1251
1252         } else if (new->handler == irq_default_primary_handler &&
1253                    !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1254                 /*
1255                  * The interrupt was requested with handler = NULL, so
1256                  * we use the default primary handler for it. But it
1257                  * does not have the oneshot flag set. In combination
1258                  * with level interrupts this is deadly, because the
1259                  * default primary handler just wakes the thread, then
1260                  * the irq lines is reenabled, but the device still
1261                  * has the level irq asserted. Rinse and repeat....
1262                  *
1263                  * While this works for edge type interrupts, we play
1264                  * it safe and reject unconditionally because we can't
1265                  * say for sure which type this interrupt really
1266                  * has. The type flags are unreliable as the
1267                  * underlying chip implementation can override them.
1268                  */
1269                 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1270                        irq);
1271                 ret = -EINVAL;
1272                 goto out_mask;
1273         }
1274
1275         if (!shared) {
1276                 ret = irq_request_resources(desc);
1277                 if (ret) {
1278                         pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1279                                new->name, irq, desc->irq_data.chip->name);
1280                         goto out_mask;
1281                 }
1282
1283                 init_waitqueue_head(&desc->wait_for_threads);
1284
1285                 /* Setup the type (level, edge polarity) if configured: */
1286                 if (new->flags & IRQF_TRIGGER_MASK) {
1287                         ret = __irq_set_trigger(desc,
1288                                                 new->flags & IRQF_TRIGGER_MASK);
1289
1290                         if (ret)
1291                                 goto out_mask;
1292                 }
1293
1294                 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1295                                   IRQS_ONESHOT | IRQS_WAITING);
1296                 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1297
1298                 if (new->flags & IRQF_PERCPU) {
1299                         irqd_set(&desc->irq_data, IRQD_PER_CPU);
1300                         irq_settings_set_per_cpu(desc);
1301                 }
1302
1303                 if (new->flags & IRQF_ONESHOT)
1304                         desc->istate |= IRQS_ONESHOT;
1305
1306                 if (irq_settings_can_autoenable(desc))
1307                         irq_startup(desc, true);
1308                 else
1309                         /* Undo nested disables: */
1310                         desc->depth = 1;
1311
1312                 /* Exclude IRQ from balancing if requested */
1313                 if (new->flags & IRQF_NOBALANCING) {
1314                         irq_settings_set_no_balancing(desc);
1315                         irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1316                 }
1317
1318                 /* Set default affinity mask once everything is setup */
1319                 setup_affinity(desc, mask);
1320
1321         } else if (new->flags & IRQF_TRIGGER_MASK) {
1322                 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1323                 unsigned int omsk = irq_settings_get_trigger_mask(desc);
1324
1325                 if (nmsk != omsk)
1326                         /* hope the handler works with current  trigger mode */
1327                         pr_warning("irq %d uses trigger mode %u; requested %u\n",
1328                                    irq, nmsk, omsk);
1329         }
1330
1331         *old_ptr = new;
1332
1333         irq_pm_install_action(desc, new);
1334
1335         /* Reset broken irq detection when installing new handler */
1336         desc->irq_count = 0;
1337         desc->irqs_unhandled = 0;
1338
1339         /*
1340          * Check whether we disabled the irq via the spurious handler
1341          * before. Reenable it and give it another chance.
1342          */
1343         if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1344                 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1345                 __enable_irq(desc);
1346         }
1347
1348         raw_spin_unlock_irqrestore(&desc->lock, flags);
1349
1350         /*
1351          * Strictly no need to wake it up, but hung_task complains
1352          * when no hard interrupt wakes the thread up.
1353          */
1354         if (new->thread)
1355                 wake_up_process(new->thread);
1356         if (new->secondary)
1357                 wake_up_process(new->secondary->thread);
1358
1359         register_irq_proc(irq, desc);
1360         new->dir = NULL;
1361         register_handler_proc(irq, new);
1362         free_cpumask_var(mask);
1363
1364         return 0;
1365
1366 mismatch:
1367         if (!(new->flags & IRQF_PROBE_SHARED)) {
1368                 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1369                        irq, new->flags, new->name, old->flags, old->name);
1370 #ifdef CONFIG_DEBUG_SHIRQ
1371                 dump_stack();
1372 #endif
1373         }
1374         ret = -EBUSY;
1375
1376 out_mask:
1377         raw_spin_unlock_irqrestore(&desc->lock, flags);
1378         free_cpumask_var(mask);
1379
1380 out_thread:
1381         if (new->thread) {
1382                 struct task_struct *t = new->thread;
1383
1384                 new->thread = NULL;
1385                 kthread_stop(t);
1386                 put_task_struct(t);
1387         }
1388         if (new->secondary && new->secondary->thread) {
1389                 struct task_struct *t = new->secondary->thread;
1390
1391                 new->secondary->thread = NULL;
1392                 kthread_stop(t);
1393                 put_task_struct(t);
1394         }
1395 out_mput:
1396         module_put(desc->owner);
1397         return ret;
1398 }
1399
1400 /**
1401  *      setup_irq - setup an interrupt
1402  *      @irq: Interrupt line to setup
1403  *      @act: irqaction for the interrupt
1404  *
1405  * Used to statically setup interrupts in the early boot process.
1406  */
1407 int setup_irq(unsigned int irq, struct irqaction *act)
1408 {
1409         int retval;
1410         struct irq_desc *desc = irq_to_desc(irq);
1411
1412         if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1413                 return -EINVAL;
1414         chip_bus_lock(desc);
1415         retval = __setup_irq(irq, desc, act);
1416         chip_bus_sync_unlock(desc);
1417
1418         return retval;
1419 }
1420 EXPORT_SYMBOL_GPL(setup_irq);
1421
1422 /*
1423  * Internal function to unregister an irqaction - used to free
1424  * regular and special interrupts that are part of the architecture.
1425  */
1426 static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1427 {
1428         struct irq_desc *desc = irq_to_desc(irq);
1429         struct irqaction *action, **action_ptr;
1430         unsigned long flags;
1431
1432         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1433
1434         if (!desc)
1435                 return NULL;
1436
1437         chip_bus_lock(desc);
1438         raw_spin_lock_irqsave(&desc->lock, flags);
1439
1440         /*
1441          * There can be multiple actions per IRQ descriptor, find the right
1442          * one based on the dev_id:
1443          */
1444         action_ptr = &desc->action;
1445         for (;;) {
1446                 action = *action_ptr;
1447
1448                 if (!action) {
1449                         WARN(1, "Trying to free already-free IRQ %d\n", irq);
1450                         raw_spin_unlock_irqrestore(&desc->lock, flags);
1451                         chip_bus_sync_unlock(desc);
1452                         return NULL;
1453                 }
1454
1455                 if (action->dev_id == dev_id)
1456                         break;
1457                 action_ptr = &action->next;
1458         }
1459
1460         /* Found it - now remove it from the list of entries: */
1461         *action_ptr = action->next;
1462
1463         irq_pm_remove_action(desc, action);
1464
1465         /* If this was the last handler, shut down the IRQ line: */
1466         if (!desc->action) {
1467                 irq_settings_clr_disable_unlazy(desc);
1468                 irq_shutdown(desc);
1469                 irq_release_resources(desc);
1470         }
1471
1472 #ifdef CONFIG_SMP
1473         /* make sure affinity_hint is cleaned up */
1474         if (WARN_ON_ONCE(desc->affinity_hint))
1475                 desc->affinity_hint = NULL;
1476 #endif
1477
1478         raw_spin_unlock_irqrestore(&desc->lock, flags);
1479         chip_bus_sync_unlock(desc);
1480
1481         unregister_handler_proc(irq, action);
1482
1483         /* Make sure it's not being used on another CPU: */
1484         synchronize_irq(irq);
1485
1486 #ifdef CONFIG_DEBUG_SHIRQ
1487         /*
1488          * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1489          * event to happen even now it's being freed, so let's make sure that
1490          * is so by doing an extra call to the handler ....
1491          *
1492          * ( We do this after actually deregistering it, to make sure that a
1493          *   'real' IRQ doesn't run in * parallel with our fake. )
1494          */
1495         if (action->flags & IRQF_SHARED) {
1496                 local_irq_save(flags);
1497                 action->handler(irq, dev_id);
1498                 local_irq_restore(flags);
1499         }
1500 #endif
1501
1502         if (action->thread) {
1503                 kthread_stop(action->thread);
1504                 put_task_struct(action->thread);
1505                 if (action->secondary && action->secondary->thread) {
1506                         kthread_stop(action->secondary->thread);
1507                         put_task_struct(action->secondary->thread);
1508                 }
1509         }
1510
1511         module_put(desc->owner);
1512         kfree(action->secondary);
1513         return action;
1514 }
1515
1516 /**
1517  *      remove_irq - free an interrupt
1518  *      @irq: Interrupt line to free
1519  *      @act: irqaction for the interrupt
1520  *
1521  * Used to remove interrupts statically setup by the early boot process.
1522  */
1523 void remove_irq(unsigned int irq, struct irqaction *act)
1524 {
1525         struct irq_desc *desc = irq_to_desc(irq);
1526
1527         if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1528             __free_irq(irq, act->dev_id);
1529 }
1530 EXPORT_SYMBOL_GPL(remove_irq);
1531
1532 /**
1533  *      free_irq - free an interrupt allocated with request_irq
1534  *      @irq: Interrupt line to free
1535  *      @dev_id: Device identity to free
1536  *
1537  *      Remove an interrupt handler. The handler is removed and if the
1538  *      interrupt line is no longer in use by any driver it is disabled.
1539  *      On a shared IRQ the caller must ensure the interrupt is disabled
1540  *      on the card it drives before calling this function. The function
1541  *      does not return until any executing interrupts for this IRQ
1542  *      have completed.
1543  *
1544  *      This function must not be called from interrupt context.
1545  */
1546 void free_irq(unsigned int irq, void *dev_id)
1547 {
1548         struct irq_desc *desc = irq_to_desc(irq);
1549
1550         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1551                 return;
1552
1553 #ifdef CONFIG_SMP
1554         if (WARN_ON(desc->affinity_notify))
1555                 desc->affinity_notify = NULL;
1556 #endif
1557
1558         kfree(__free_irq(irq, dev_id));
1559 }
1560 EXPORT_SYMBOL(free_irq);
1561
1562 /**
1563  *      request_threaded_irq - allocate an interrupt line
1564  *      @irq: Interrupt line to allocate
1565  *      @handler: Function to be called when the IRQ occurs.
1566  *                Primary handler for threaded interrupts
1567  *                If NULL and thread_fn != NULL the default
1568  *                primary handler is installed
1569  *      @thread_fn: Function called from the irq handler thread
1570  *                  If NULL, no irq thread is created
1571  *      @irqflags: Interrupt type flags
1572  *      @devname: An ascii name for the claiming device
1573  *      @dev_id: A cookie passed back to the handler function
1574  *
1575  *      This call allocates interrupt resources and enables the
1576  *      interrupt line and IRQ handling. From the point this
1577  *      call is made your handler function may be invoked. Since
1578  *      your handler function must clear any interrupt the board
1579  *      raises, you must take care both to initialise your hardware
1580  *      and to set up the interrupt handler in the right order.
1581  *
1582  *      If you want to set up a threaded irq handler for your device
1583  *      then you need to supply @handler and @thread_fn. @handler is
1584  *      still called in hard interrupt context and has to check
1585  *      whether the interrupt originates from the device. If yes it
1586  *      needs to disable the interrupt on the device and return
1587  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1588  *      @thread_fn. This split handler design is necessary to support
1589  *      shared interrupts.
1590  *
1591  *      Dev_id must be globally unique. Normally the address of the
1592  *      device data structure is used as the cookie. Since the handler
1593  *      receives this value it makes sense to use it.
1594  *
1595  *      If your interrupt is shared you must pass a non NULL dev_id
1596  *      as this is required when freeing the interrupt.
1597  *
1598  *      Flags:
1599  *
1600  *      IRQF_SHARED             Interrupt is shared
1601  *      IRQF_TRIGGER_*          Specify active edge(s) or level
1602  *
1603  */
1604 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1605                          irq_handler_t thread_fn, unsigned long irqflags,
1606                          const char *devname, void *dev_id)
1607 {
1608         struct irqaction *action;
1609         struct irq_desc *desc;
1610         int retval;
1611
1612         /*
1613          * Sanity-check: shared interrupts must pass in a real dev-ID,
1614          * otherwise we'll have trouble later trying to figure out
1615          * which interrupt is which (messes up the interrupt freeing
1616          * logic etc).
1617          *
1618          * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1619          * it cannot be set along with IRQF_NO_SUSPEND.
1620          */
1621         if (((irqflags & IRQF_SHARED) && !dev_id) ||
1622             (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1623             ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1624                 return -EINVAL;
1625
1626         desc = irq_to_desc(irq);
1627         if (!desc)
1628                 return -EINVAL;
1629
1630         if (!irq_settings_can_request(desc) ||
1631             WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1632                 return -EINVAL;
1633
1634         if (!handler) {
1635                 if (!thread_fn)
1636                         return -EINVAL;
1637                 handler = irq_default_primary_handler;
1638         }
1639
1640         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1641         if (!action)
1642                 return -ENOMEM;
1643
1644         action->handler = handler;
1645         action->thread_fn = thread_fn;
1646         action->flags = irqflags;
1647         action->name = devname;
1648         action->dev_id = dev_id;
1649
1650         chip_bus_lock(desc);
1651         retval = __setup_irq(irq, desc, action);
1652         chip_bus_sync_unlock(desc);
1653
1654         if (retval) {
1655                 kfree(action->secondary);
1656                 kfree(action);
1657         }
1658
1659 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
1660         if (!retval && (irqflags & IRQF_SHARED)) {
1661                 /*
1662                  * It's a shared IRQ -- the driver ought to be prepared for it
1663                  * to happen immediately, so let's make sure....
1664                  * We disable the irq to make sure that a 'real' IRQ doesn't
1665                  * run in parallel with our fake.
1666                  */
1667                 unsigned long flags;
1668
1669                 disable_irq(irq);
1670                 local_irq_save(flags);
1671
1672                 handler(irq, dev_id);
1673
1674                 local_irq_restore(flags);
1675                 enable_irq(irq);
1676         }
1677 #endif
1678         return retval;
1679 }
1680 EXPORT_SYMBOL(request_threaded_irq);
1681
1682 /**
1683  *      request_any_context_irq - allocate an interrupt line
1684  *      @irq: Interrupt line to allocate
1685  *      @handler: Function to be called when the IRQ occurs.
1686  *                Threaded handler for threaded interrupts.
1687  *      @flags: Interrupt type flags
1688  *      @name: An ascii name for the claiming device
1689  *      @dev_id: A cookie passed back to the handler function
1690  *
1691  *      This call allocates interrupt resources and enables the
1692  *      interrupt line and IRQ handling. It selects either a
1693  *      hardirq or threaded handling method depending on the
1694  *      context.
1695  *
1696  *      On failure, it returns a negative value. On success,
1697  *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1698  */
1699 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1700                             unsigned long flags, const char *name, void *dev_id)
1701 {
1702         struct irq_desc *desc = irq_to_desc(irq);
1703         int ret;
1704
1705         if (!desc)
1706                 return -EINVAL;
1707
1708         if (irq_settings_is_nested_thread(desc)) {
1709                 ret = request_threaded_irq(irq, NULL, handler,
1710                                            flags, name, dev_id);
1711                 return !ret ? IRQC_IS_NESTED : ret;
1712         }
1713
1714         ret = request_irq(irq, handler, flags, name, dev_id);
1715         return !ret ? IRQC_IS_HARDIRQ : ret;
1716 }
1717 EXPORT_SYMBOL_GPL(request_any_context_irq);
1718
1719 void enable_percpu_irq(unsigned int irq, unsigned int type)
1720 {
1721         unsigned int cpu = smp_processor_id();
1722         unsigned long flags;
1723         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1724
1725         if (!desc)
1726                 return;
1727
1728         type &= IRQ_TYPE_SENSE_MASK;
1729         if (type != IRQ_TYPE_NONE) {
1730                 int ret;
1731
1732                 ret = __irq_set_trigger(desc, type);
1733
1734                 if (ret) {
1735                         WARN(1, "failed to set type for IRQ%d\n", irq);
1736                         goto out;
1737                 }
1738         }
1739
1740         irq_percpu_enable(desc, cpu);
1741 out:
1742         irq_put_desc_unlock(desc, flags);
1743 }
1744 EXPORT_SYMBOL_GPL(enable_percpu_irq);
1745
1746 /**
1747  * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
1748  * @irq:        Linux irq number to check for
1749  *
1750  * Must be called from a non migratable context. Returns the enable
1751  * state of a per cpu interrupt on the current cpu.
1752  */
1753 bool irq_percpu_is_enabled(unsigned int irq)
1754 {
1755         unsigned int cpu = smp_processor_id();
1756         struct irq_desc *desc;
1757         unsigned long flags;
1758         bool is_enabled;
1759
1760         desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1761         if (!desc)
1762                 return false;
1763
1764         is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
1765         irq_put_desc_unlock(desc, flags);
1766
1767         return is_enabled;
1768 }
1769 EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
1770
1771 void disable_percpu_irq(unsigned int irq)
1772 {
1773         unsigned int cpu = smp_processor_id();
1774         unsigned long flags;
1775         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1776
1777         if (!desc)
1778                 return;
1779
1780         irq_percpu_disable(desc, cpu);
1781         irq_put_desc_unlock(desc, flags);
1782 }
1783 EXPORT_SYMBOL_GPL(disable_percpu_irq);
1784
1785 /*
1786  * Internal function to unregister a percpu irqaction.
1787  */
1788 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1789 {
1790         struct irq_desc *desc = irq_to_desc(irq);
1791         struct irqaction *action;
1792         unsigned long flags;
1793
1794         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1795
1796         if (!desc)
1797                 return NULL;
1798
1799         raw_spin_lock_irqsave(&desc->lock, flags);
1800
1801         action = desc->action;
1802         if (!action || action->percpu_dev_id != dev_id) {
1803                 WARN(1, "Trying to free already-free IRQ %d\n", irq);
1804                 goto bad;
1805         }
1806
1807         if (!cpumask_empty(desc->percpu_enabled)) {
1808                 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1809                      irq, cpumask_first(desc->percpu_enabled));
1810                 goto bad;
1811         }
1812
1813         /* Found it - now remove it from the list of entries: */
1814         desc->action = NULL;
1815
1816         raw_spin_unlock_irqrestore(&desc->lock, flags);
1817
1818         unregister_handler_proc(irq, action);
1819
1820         module_put(desc->owner);
1821         return action;
1822
1823 bad:
1824         raw_spin_unlock_irqrestore(&desc->lock, flags);
1825         return NULL;
1826 }
1827
1828 /**
1829  *      remove_percpu_irq - free a per-cpu interrupt
1830  *      @irq: Interrupt line to free
1831  *      @act: irqaction for the interrupt
1832  *
1833  * Used to remove interrupts statically setup by the early boot process.
1834  */
1835 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1836 {
1837         struct irq_desc *desc = irq_to_desc(irq);
1838
1839         if (desc && irq_settings_is_per_cpu_devid(desc))
1840             __free_percpu_irq(irq, act->percpu_dev_id);
1841 }
1842
1843 /**
1844  *      free_percpu_irq - free an interrupt allocated with request_percpu_irq
1845  *      @irq: Interrupt line to free
1846  *      @dev_id: Device identity to free
1847  *
1848  *      Remove a percpu interrupt handler. The handler is removed, but
1849  *      the interrupt line is not disabled. This must be done on each
1850  *      CPU before calling this function. The function does not return
1851  *      until any executing interrupts for this IRQ have completed.
1852  *
1853  *      This function must not be called from interrupt context.
1854  */
1855 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1856 {
1857         struct irq_desc *desc = irq_to_desc(irq);
1858
1859         if (!desc || !irq_settings_is_per_cpu_devid(desc))
1860                 return;
1861
1862         chip_bus_lock(desc);
1863         kfree(__free_percpu_irq(irq, dev_id));
1864         chip_bus_sync_unlock(desc);
1865 }
1866 EXPORT_SYMBOL_GPL(free_percpu_irq);
1867
1868 /**
1869  *      setup_percpu_irq - setup a per-cpu interrupt
1870  *      @irq: Interrupt line to setup
1871  *      @act: irqaction for the interrupt
1872  *
1873  * Used to statically setup per-cpu interrupts in the early boot process.
1874  */
1875 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1876 {
1877         struct irq_desc *desc = irq_to_desc(irq);
1878         int retval;
1879
1880         if (!desc || !irq_settings_is_per_cpu_devid(desc))
1881                 return -EINVAL;
1882         chip_bus_lock(desc);
1883         retval = __setup_irq(irq, desc, act);
1884         chip_bus_sync_unlock(desc);
1885
1886         return retval;
1887 }
1888
1889 /**
1890  *      request_percpu_irq - allocate a percpu interrupt line
1891  *      @irq: Interrupt line to allocate
1892  *      @handler: Function to be called when the IRQ occurs.
1893  *      @devname: An ascii name for the claiming device
1894  *      @dev_id: A percpu cookie passed back to the handler function
1895  *
1896  *      This call allocates interrupt resources and enables the
1897  *      interrupt on the local CPU. If the interrupt is supposed to be
1898  *      enabled on other CPUs, it has to be done on each CPU using
1899  *      enable_percpu_irq().
1900  *
1901  *      Dev_id must be globally unique. It is a per-cpu variable, and
1902  *      the handler gets called with the interrupted CPU's instance of
1903  *      that variable.
1904  */
1905 int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1906                        const char *devname, void __percpu *dev_id)
1907 {
1908         struct irqaction *action;
1909         struct irq_desc *desc;
1910         int retval;
1911
1912         if (!dev_id)
1913                 return -EINVAL;
1914
1915         desc = irq_to_desc(irq);
1916         if (!desc || !irq_settings_can_request(desc) ||
1917             !irq_settings_is_per_cpu_devid(desc))
1918                 return -EINVAL;
1919
1920         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1921         if (!action)
1922                 return -ENOMEM;
1923
1924         action->handler = handler;
1925         action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1926         action->name = devname;
1927         action->percpu_dev_id = dev_id;
1928
1929         chip_bus_lock(desc);
1930         retval = __setup_irq(irq, desc, action);
1931         chip_bus_sync_unlock(desc);
1932
1933         if (retval)
1934                 kfree(action);
1935
1936         return retval;
1937 }
1938 EXPORT_SYMBOL_GPL(request_percpu_irq);
1939
1940 /**
1941  *      irq_get_irqchip_state - returns the irqchip state of a interrupt.
1942  *      @irq: Interrupt line that is forwarded to a VM
1943  *      @which: One of IRQCHIP_STATE_* the caller wants to know about
1944  *      @state: a pointer to a boolean where the state is to be storeed
1945  *
1946  *      This call snapshots the internal irqchip state of an
1947  *      interrupt, returning into @state the bit corresponding to
1948  *      stage @which
1949  *
1950  *      This function should be called with preemption disabled if the
1951  *      interrupt controller has per-cpu registers.
1952  */
1953 int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1954                           bool *state)
1955 {
1956         struct irq_desc *desc;
1957         struct irq_data *data;
1958         struct irq_chip *chip;
1959         unsigned long flags;
1960         int err = -EINVAL;
1961
1962         desc = irq_get_desc_buslock(irq, &flags, 0);
1963         if (!desc)
1964                 return err;
1965
1966         data = irq_desc_get_irq_data(desc);
1967
1968         do {
1969                 chip = irq_data_get_irq_chip(data);
1970                 if (chip->irq_get_irqchip_state)
1971                         break;
1972 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1973                 data = data->parent_data;
1974 #else
1975                 data = NULL;
1976 #endif
1977         } while (data);
1978
1979         if (data)
1980                 err = chip->irq_get_irqchip_state(data, which, state);
1981
1982         irq_put_desc_busunlock(desc, flags);
1983         return err;
1984 }
1985 EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
1986
1987 /**
1988  *      irq_set_irqchip_state - set the state of a forwarded interrupt.
1989  *      @irq: Interrupt line that is forwarded to a VM
1990  *      @which: State to be restored (one of IRQCHIP_STATE_*)
1991  *      @val: Value corresponding to @which
1992  *
1993  *      This call sets the internal irqchip state of an interrupt,
1994  *      depending on the value of @which.
1995  *
1996  *      This function should be called with preemption disabled if the
1997  *      interrupt controller has per-cpu registers.
1998  */
1999 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2000                           bool val)
2001 {
2002         struct irq_desc *desc;
2003         struct irq_data *data;
2004         struct irq_chip *chip;
2005         unsigned long flags;
2006         int err = -EINVAL;
2007
2008         desc = irq_get_desc_buslock(irq, &flags, 0);
2009         if (!desc)
2010                 return err;
2011
2012         data = irq_desc_get_irq_data(desc);
2013
2014         do {
2015                 chip = irq_data_get_irq_chip(data);
2016                 if (chip->irq_set_irqchip_state)
2017                         break;
2018 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2019                 data = data->parent_data;
2020 #else
2021                 data = NULL;
2022 #endif
2023         } while (data);
2024
2025         if (data)
2026                 err = chip->irq_set_irqchip_state(data, which, val);
2027
2028         irq_put_desc_busunlock(desc, flags);
2029         return err;
2030 }
2031 EXPORT_SYMBOL_GPL(irq_set_irqchip_state);