Merge tag 'kconfig-v5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy...
[linux-2.6-block.git] / kernel / irq / manage.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
4  * Copyright (C) 2005-2006 Thomas Gleixner
5  *
6  * This file contains driver APIs to the irq subsystem.
7  */
8
9 #define pr_fmt(fmt) "genirq: " fmt
10
11 #include <linux/irq.h>
12 #include <linux/kthread.h>
13 #include <linux/module.h>
14 #include <linux/random.h>
15 #include <linux/interrupt.h>
16 #include <linux/irqdomain.h>
17 #include <linux/slab.h>
18 #include <linux/sched.h>
19 #include <linux/sched/rt.h>
20 #include <linux/sched/task.h>
21 #include <uapi/linux/sched/types.h>
22 #include <linux/task_work.h>
23
24 #include "internals.h"
25
26 #ifdef CONFIG_IRQ_FORCED_THREADING
27 __read_mostly bool force_irqthreads;
28 EXPORT_SYMBOL_GPL(force_irqthreads);
29
30 static int __init setup_forced_irqthreads(char *arg)
31 {
32         force_irqthreads = true;
33         return 0;
34 }
35 early_param("threadirqs", setup_forced_irqthreads);
36 #endif
37
38 static void __synchronize_hardirq(struct irq_desc *desc, bool sync_chip)
39 {
40         struct irq_data *irqd = irq_desc_get_irq_data(desc);
41         bool inprogress;
42
43         do {
44                 unsigned long flags;
45
46                 /*
47                  * Wait until we're out of the critical section.  This might
48                  * give the wrong answer due to the lack of memory barriers.
49                  */
50                 while (irqd_irq_inprogress(&desc->irq_data))
51                         cpu_relax();
52
53                 /* Ok, that indicated we're done: double-check carefully. */
54                 raw_spin_lock_irqsave(&desc->lock, flags);
55                 inprogress = irqd_irq_inprogress(&desc->irq_data);
56
57                 /*
58                  * If requested and supported, check at the chip whether it
59                  * is in flight at the hardware level, i.e. already pending
60                  * in a CPU and waiting for service and acknowledge.
61                  */
62                 if (!inprogress && sync_chip) {
63                         /*
64                          * Ignore the return code. inprogress is only updated
65                          * when the chip supports it.
66                          */
67                         __irq_get_irqchip_state(irqd, IRQCHIP_STATE_ACTIVE,
68                                                 &inprogress);
69                 }
70                 raw_spin_unlock_irqrestore(&desc->lock, flags);
71
72                 /* Oops, that failed? */
73         } while (inprogress);
74 }
75
76 /**
77  *      synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
78  *      @irq: interrupt number to wait for
79  *
80  *      This function waits for any pending hard IRQ handlers for this
81  *      interrupt to complete before returning. If you use this
82  *      function while holding a resource the IRQ handler may need you
83  *      will deadlock. It does not take associated threaded handlers
84  *      into account.
85  *
86  *      Do not use this for shutdown scenarios where you must be sure
87  *      that all parts (hardirq and threaded handler) have completed.
88  *
89  *      Returns: false if a threaded handler is active.
90  *
91  *      This function may be called - with care - from IRQ context.
92  *
93  *      It does not check whether there is an interrupt in flight at the
94  *      hardware level, but not serviced yet, as this might deadlock when
95  *      called with interrupts disabled and the target CPU of the interrupt
96  *      is the current CPU.
97  */
98 bool synchronize_hardirq(unsigned int irq)
99 {
100         struct irq_desc *desc = irq_to_desc(irq);
101
102         if (desc) {
103                 __synchronize_hardirq(desc, false);
104                 return !atomic_read(&desc->threads_active);
105         }
106
107         return true;
108 }
109 EXPORT_SYMBOL(synchronize_hardirq);
110
111 /**
112  *      synchronize_irq - wait for pending IRQ handlers (on other CPUs)
113  *      @irq: interrupt number to wait for
114  *
115  *      This function waits for any pending IRQ handlers for this interrupt
116  *      to complete before returning. If you use this function while
117  *      holding a resource the IRQ handler may need you will deadlock.
118  *
119  *      Can only be called from preemptible code as it might sleep when
120  *      an interrupt thread is associated to @irq.
121  *
122  *      It optionally makes sure (when the irq chip supports that method)
123  *      that the interrupt is not pending in any CPU and waiting for
124  *      service.
125  */
126 void synchronize_irq(unsigned int irq)
127 {
128         struct irq_desc *desc = irq_to_desc(irq);
129
130         if (desc) {
131                 __synchronize_hardirq(desc, true);
132                 /*
133                  * We made sure that no hardirq handler is
134                  * running. Now verify that no threaded handlers are
135                  * active.
136                  */
137                 wait_event(desc->wait_for_threads,
138                            !atomic_read(&desc->threads_active));
139         }
140 }
141 EXPORT_SYMBOL(synchronize_irq);
142
143 #ifdef CONFIG_SMP
144 cpumask_var_t irq_default_affinity;
145
146 static bool __irq_can_set_affinity(struct irq_desc *desc)
147 {
148         if (!desc || !irqd_can_balance(&desc->irq_data) ||
149             !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
150                 return false;
151         return true;
152 }
153
154 /**
155  *      irq_can_set_affinity - Check if the affinity of a given irq can be set
156  *      @irq:           Interrupt to check
157  *
158  */
159 int irq_can_set_affinity(unsigned int irq)
160 {
161         return __irq_can_set_affinity(irq_to_desc(irq));
162 }
163
164 /**
165  * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
166  * @irq:        Interrupt to check
167  *
168  * Like irq_can_set_affinity() above, but additionally checks for the
169  * AFFINITY_MANAGED flag.
170  */
171 bool irq_can_set_affinity_usr(unsigned int irq)
172 {
173         struct irq_desc *desc = irq_to_desc(irq);
174
175         return __irq_can_set_affinity(desc) &&
176                 !irqd_affinity_is_managed(&desc->irq_data);
177 }
178
179 /**
180  *      irq_set_thread_affinity - Notify irq threads to adjust affinity
181  *      @desc:          irq descriptor which has affitnity changed
182  *
183  *      We just set IRQTF_AFFINITY and delegate the affinity setting
184  *      to the interrupt thread itself. We can not call
185  *      set_cpus_allowed_ptr() here as we hold desc->lock and this
186  *      code can be called from hard interrupt context.
187  */
188 void irq_set_thread_affinity(struct irq_desc *desc)
189 {
190         struct irqaction *action;
191
192         for_each_action_of_desc(desc, action)
193                 if (action->thread)
194                         set_bit(IRQTF_AFFINITY, &action->thread_flags);
195 }
196
197 static void irq_validate_effective_affinity(struct irq_data *data)
198 {
199 #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
200         const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
201         struct irq_chip *chip = irq_data_get_irq_chip(data);
202
203         if (!cpumask_empty(m))
204                 return;
205         pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
206                      chip->name, data->irq);
207 #endif
208 }
209
210 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
211                         bool force)
212 {
213         struct irq_desc *desc = irq_data_to_desc(data);
214         struct irq_chip *chip = irq_data_get_irq_chip(data);
215         int ret;
216
217         if (!chip || !chip->irq_set_affinity)
218                 return -EINVAL;
219
220         ret = chip->irq_set_affinity(data, mask, force);
221         switch (ret) {
222         case IRQ_SET_MASK_OK:
223         case IRQ_SET_MASK_OK_DONE:
224                 cpumask_copy(desc->irq_common_data.affinity, mask);
225                 /* fall through */
226         case IRQ_SET_MASK_OK_NOCOPY:
227                 irq_validate_effective_affinity(data);
228                 irq_set_thread_affinity(desc);
229                 ret = 0;
230         }
231
232         return ret;
233 }
234
235 #ifdef CONFIG_GENERIC_PENDING_IRQ
236 static inline int irq_set_affinity_pending(struct irq_data *data,
237                                            const struct cpumask *dest)
238 {
239         struct irq_desc *desc = irq_data_to_desc(data);
240
241         irqd_set_move_pending(data);
242         irq_copy_pending(desc, dest);
243         return 0;
244 }
245 #else
246 static inline int irq_set_affinity_pending(struct irq_data *data,
247                                            const struct cpumask *dest)
248 {
249         return -EBUSY;
250 }
251 #endif
252
253 static int irq_try_set_affinity(struct irq_data *data,
254                                 const struct cpumask *dest, bool force)
255 {
256         int ret = irq_do_set_affinity(data, dest, force);
257
258         /*
259          * In case that the underlying vector management is busy and the
260          * architecture supports the generic pending mechanism then utilize
261          * this to avoid returning an error to user space.
262          */
263         if (ret == -EBUSY && !force)
264                 ret = irq_set_affinity_pending(data, dest);
265         return ret;
266 }
267
268 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
269                             bool force)
270 {
271         struct irq_chip *chip = irq_data_get_irq_chip(data);
272         struct irq_desc *desc = irq_data_to_desc(data);
273         int ret = 0;
274
275         if (!chip || !chip->irq_set_affinity)
276                 return -EINVAL;
277
278         if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
279                 ret = irq_try_set_affinity(data, mask, force);
280         } else {
281                 irqd_set_move_pending(data);
282                 irq_copy_pending(desc, mask);
283         }
284
285         if (desc->affinity_notify) {
286                 kref_get(&desc->affinity_notify->kref);
287                 schedule_work(&desc->affinity_notify->work);
288         }
289         irqd_set(data, IRQD_AFFINITY_SET);
290
291         return ret;
292 }
293
294 int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
295 {
296         struct irq_desc *desc = irq_to_desc(irq);
297         unsigned long flags;
298         int ret;
299
300         if (!desc)
301                 return -EINVAL;
302
303         raw_spin_lock_irqsave(&desc->lock, flags);
304         ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
305         raw_spin_unlock_irqrestore(&desc->lock, flags);
306         return ret;
307 }
308
309 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
310 {
311         unsigned long flags;
312         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
313
314         if (!desc)
315                 return -EINVAL;
316         desc->affinity_hint = m;
317         irq_put_desc_unlock(desc, flags);
318         /* set the initial affinity to prevent every interrupt being on CPU0 */
319         if (m)
320                 __irq_set_affinity(irq, m, false);
321         return 0;
322 }
323 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
324
325 static void irq_affinity_notify(struct work_struct *work)
326 {
327         struct irq_affinity_notify *notify =
328                 container_of(work, struct irq_affinity_notify, work);
329         struct irq_desc *desc = irq_to_desc(notify->irq);
330         cpumask_var_t cpumask;
331         unsigned long flags;
332
333         if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
334                 goto out;
335
336         raw_spin_lock_irqsave(&desc->lock, flags);
337         if (irq_move_pending(&desc->irq_data))
338                 irq_get_pending(cpumask, desc);
339         else
340                 cpumask_copy(cpumask, desc->irq_common_data.affinity);
341         raw_spin_unlock_irqrestore(&desc->lock, flags);
342
343         notify->notify(notify, cpumask);
344
345         free_cpumask_var(cpumask);
346 out:
347         kref_put(&notify->kref, notify->release);
348 }
349
350 /**
351  *      irq_set_affinity_notifier - control notification of IRQ affinity changes
352  *      @irq:           Interrupt for which to enable/disable notification
353  *      @notify:        Context for notification, or %NULL to disable
354  *                      notification.  Function pointers must be initialised;
355  *                      the other fields will be initialised by this function.
356  *
357  *      Must be called in process context.  Notification may only be enabled
358  *      after the IRQ is allocated and must be disabled before the IRQ is
359  *      freed using free_irq().
360  */
361 int
362 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
363 {
364         struct irq_desc *desc = irq_to_desc(irq);
365         struct irq_affinity_notify *old_notify;
366         unsigned long flags;
367
368         /* The release function is promised process context */
369         might_sleep();
370
371         if (!desc || desc->istate & IRQS_NMI)
372                 return -EINVAL;
373
374         /* Complete initialisation of *notify */
375         if (notify) {
376                 notify->irq = irq;
377                 kref_init(&notify->kref);
378                 INIT_WORK(&notify->work, irq_affinity_notify);
379         }
380
381         raw_spin_lock_irqsave(&desc->lock, flags);
382         old_notify = desc->affinity_notify;
383         desc->affinity_notify = notify;
384         raw_spin_unlock_irqrestore(&desc->lock, flags);
385
386         if (old_notify) {
387                 cancel_work_sync(&old_notify->work);
388                 kref_put(&old_notify->kref, old_notify->release);
389         }
390
391         return 0;
392 }
393 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
394
395 #ifndef CONFIG_AUTO_IRQ_AFFINITY
396 /*
397  * Generic version of the affinity autoselector.
398  */
399 int irq_setup_affinity(struct irq_desc *desc)
400 {
401         struct cpumask *set = irq_default_affinity;
402         int ret, node = irq_desc_get_node(desc);
403         static DEFINE_RAW_SPINLOCK(mask_lock);
404         static struct cpumask mask;
405
406         /* Excludes PER_CPU and NO_BALANCE interrupts */
407         if (!__irq_can_set_affinity(desc))
408                 return 0;
409
410         raw_spin_lock(&mask_lock);
411         /*
412          * Preserve the managed affinity setting and a userspace affinity
413          * setup, but make sure that one of the targets is online.
414          */
415         if (irqd_affinity_is_managed(&desc->irq_data) ||
416             irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
417                 if (cpumask_intersects(desc->irq_common_data.affinity,
418                                        cpu_online_mask))
419                         set = desc->irq_common_data.affinity;
420                 else
421                         irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
422         }
423
424         cpumask_and(&mask, cpu_online_mask, set);
425         if (cpumask_empty(&mask))
426                 cpumask_copy(&mask, cpu_online_mask);
427
428         if (node != NUMA_NO_NODE) {
429                 const struct cpumask *nodemask = cpumask_of_node(node);
430
431                 /* make sure at least one of the cpus in nodemask is online */
432                 if (cpumask_intersects(&mask, nodemask))
433                         cpumask_and(&mask, &mask, nodemask);
434         }
435         ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
436         raw_spin_unlock(&mask_lock);
437         return ret;
438 }
439 #else
440 /* Wrapper for ALPHA specific affinity selector magic */
441 int irq_setup_affinity(struct irq_desc *desc)
442 {
443         return irq_select_affinity(irq_desc_get_irq(desc));
444 }
445 #endif
446
447 /*
448  * Called when a bogus affinity is set via /proc/irq
449  */
450 int irq_select_affinity_usr(unsigned int irq)
451 {
452         struct irq_desc *desc = irq_to_desc(irq);
453         unsigned long flags;
454         int ret;
455
456         raw_spin_lock_irqsave(&desc->lock, flags);
457         ret = irq_setup_affinity(desc);
458         raw_spin_unlock_irqrestore(&desc->lock, flags);
459         return ret;
460 }
461 #endif
462
463 /**
464  *      irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
465  *      @irq: interrupt number to set affinity
466  *      @vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
467  *                  specific data for percpu_devid interrupts
468  *
469  *      This function uses the vCPU specific data to set the vCPU
470  *      affinity for an irq. The vCPU specific data is passed from
471  *      outside, such as KVM. One example code path is as below:
472  *      KVM -> IOMMU -> irq_set_vcpu_affinity().
473  */
474 int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
475 {
476         unsigned long flags;
477         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
478         struct irq_data *data;
479         struct irq_chip *chip;
480         int ret = -ENOSYS;
481
482         if (!desc)
483                 return -EINVAL;
484
485         data = irq_desc_get_irq_data(desc);
486         do {
487                 chip = irq_data_get_irq_chip(data);
488                 if (chip && chip->irq_set_vcpu_affinity)
489                         break;
490 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
491                 data = data->parent_data;
492 #else
493                 data = NULL;
494 #endif
495         } while (data);
496
497         if (data)
498                 ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
499         irq_put_desc_unlock(desc, flags);
500
501         return ret;
502 }
503 EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
504
505 void __disable_irq(struct irq_desc *desc)
506 {
507         if (!desc->depth++)
508                 irq_disable(desc);
509 }
510
511 static int __disable_irq_nosync(unsigned int irq)
512 {
513         unsigned long flags;
514         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
515
516         if (!desc)
517                 return -EINVAL;
518         __disable_irq(desc);
519         irq_put_desc_busunlock(desc, flags);
520         return 0;
521 }
522
523 /**
524  *      disable_irq_nosync - disable an irq without waiting
525  *      @irq: Interrupt to disable
526  *
527  *      Disable the selected interrupt line.  Disables and Enables are
528  *      nested.
529  *      Unlike disable_irq(), this function does not ensure existing
530  *      instances of the IRQ handler have completed before returning.
531  *
532  *      This function may be called from IRQ context.
533  */
534 void disable_irq_nosync(unsigned int irq)
535 {
536         __disable_irq_nosync(irq);
537 }
538 EXPORT_SYMBOL(disable_irq_nosync);
539
540 /**
541  *      disable_irq - disable an irq and wait for completion
542  *      @irq: Interrupt to disable
543  *
544  *      Disable the selected interrupt line.  Enables and Disables are
545  *      nested.
546  *      This function waits for any pending IRQ handlers for this interrupt
547  *      to complete before returning. If you use this function while
548  *      holding a resource the IRQ handler may need you will deadlock.
549  *
550  *      This function may be called - with care - from IRQ context.
551  */
552 void disable_irq(unsigned int irq)
553 {
554         if (!__disable_irq_nosync(irq))
555                 synchronize_irq(irq);
556 }
557 EXPORT_SYMBOL(disable_irq);
558
559 /**
560  *      disable_hardirq - disables an irq and waits for hardirq completion
561  *      @irq: Interrupt to disable
562  *
563  *      Disable the selected interrupt line.  Enables and Disables are
564  *      nested.
565  *      This function waits for any pending hard IRQ handlers for this
566  *      interrupt to complete before returning. If you use this function while
567  *      holding a resource the hard IRQ handler may need you will deadlock.
568  *
569  *      When used to optimistically disable an interrupt from atomic context
570  *      the return value must be checked.
571  *
572  *      Returns: false if a threaded handler is active.
573  *
574  *      This function may be called - with care - from IRQ context.
575  */
576 bool disable_hardirq(unsigned int irq)
577 {
578         if (!__disable_irq_nosync(irq))
579                 return synchronize_hardirq(irq);
580
581         return false;
582 }
583 EXPORT_SYMBOL_GPL(disable_hardirq);
584
585 /**
586  *      disable_nmi_nosync - disable an nmi without waiting
587  *      @irq: Interrupt to disable
588  *
589  *      Disable the selected interrupt line. Disables and enables are
590  *      nested.
591  *      The interrupt to disable must have been requested through request_nmi.
592  *      Unlike disable_nmi(), this function does not ensure existing
593  *      instances of the IRQ handler have completed before returning.
594  */
595 void disable_nmi_nosync(unsigned int irq)
596 {
597         disable_irq_nosync(irq);
598 }
599
600 void __enable_irq(struct irq_desc *desc)
601 {
602         switch (desc->depth) {
603         case 0:
604  err_out:
605                 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
606                      irq_desc_get_irq(desc));
607                 break;
608         case 1: {
609                 if (desc->istate & IRQS_SUSPENDED)
610                         goto err_out;
611                 /* Prevent probing on this irq: */
612                 irq_settings_set_noprobe(desc);
613                 /*
614                  * Call irq_startup() not irq_enable() here because the
615                  * interrupt might be marked NOAUTOEN. So irq_startup()
616                  * needs to be invoked when it gets enabled the first
617                  * time. If it was already started up, then irq_startup()
618                  * will invoke irq_enable() under the hood.
619                  */
620                 irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
621                 break;
622         }
623         default:
624                 desc->depth--;
625         }
626 }
627
628 /**
629  *      enable_irq - enable handling of an irq
630  *      @irq: Interrupt to enable
631  *
632  *      Undoes the effect of one call to disable_irq().  If this
633  *      matches the last disable, processing of interrupts on this
634  *      IRQ line is re-enabled.
635  *
636  *      This function may be called from IRQ context only when
637  *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
638  */
639 void enable_irq(unsigned int irq)
640 {
641         unsigned long flags;
642         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
643
644         if (!desc)
645                 return;
646         if (WARN(!desc->irq_data.chip,
647                  KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
648                 goto out;
649
650         __enable_irq(desc);
651 out:
652         irq_put_desc_busunlock(desc, flags);
653 }
654 EXPORT_SYMBOL(enable_irq);
655
656 /**
657  *      enable_nmi - enable handling of an nmi
658  *      @irq: Interrupt to enable
659  *
660  *      The interrupt to enable must have been requested through request_nmi.
661  *      Undoes the effect of one call to disable_nmi(). If this
662  *      matches the last disable, processing of interrupts on this
663  *      IRQ line is re-enabled.
664  */
665 void enable_nmi(unsigned int irq)
666 {
667         enable_irq(irq);
668 }
669
670 static int set_irq_wake_real(unsigned int irq, unsigned int on)
671 {
672         struct irq_desc *desc = irq_to_desc(irq);
673         int ret = -ENXIO;
674
675         if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
676                 return 0;
677
678         if (desc->irq_data.chip->irq_set_wake)
679                 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
680
681         return ret;
682 }
683
684 /**
685  *      irq_set_irq_wake - control irq power management wakeup
686  *      @irq:   interrupt to control
687  *      @on:    enable/disable power management wakeup
688  *
689  *      Enable/disable power management wakeup mode, which is
690  *      disabled by default.  Enables and disables must match,
691  *      just as they match for non-wakeup mode support.
692  *
693  *      Wakeup mode lets this IRQ wake the system from sleep
694  *      states like "suspend to RAM".
695  */
696 int irq_set_irq_wake(unsigned int irq, unsigned int on)
697 {
698         unsigned long flags;
699         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
700         int ret = 0;
701
702         if (!desc)
703                 return -EINVAL;
704
705         /* Don't use NMIs as wake up interrupts please */
706         if (desc->istate & IRQS_NMI) {
707                 ret = -EINVAL;
708                 goto out_unlock;
709         }
710
711         /* wakeup-capable irqs can be shared between drivers that
712          * don't need to have the same sleep mode behaviors.
713          */
714         if (on) {
715                 if (desc->wake_depth++ == 0) {
716                         ret = set_irq_wake_real(irq, on);
717                         if (ret)
718                                 desc->wake_depth = 0;
719                         else
720                                 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
721                 }
722         } else {
723                 if (desc->wake_depth == 0) {
724                         WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
725                 } else if (--desc->wake_depth == 0) {
726                         ret = set_irq_wake_real(irq, on);
727                         if (ret)
728                                 desc->wake_depth = 1;
729                         else
730                                 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
731                 }
732         }
733
734 out_unlock:
735         irq_put_desc_busunlock(desc, flags);
736         return ret;
737 }
738 EXPORT_SYMBOL(irq_set_irq_wake);
739
740 /*
741  * Internal function that tells the architecture code whether a
742  * particular irq has been exclusively allocated or is available
743  * for driver use.
744  */
745 int can_request_irq(unsigned int irq, unsigned long irqflags)
746 {
747         unsigned long flags;
748         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
749         int canrequest = 0;
750
751         if (!desc)
752                 return 0;
753
754         if (irq_settings_can_request(desc)) {
755                 if (!desc->action ||
756                     irqflags & desc->action->flags & IRQF_SHARED)
757                         canrequest = 1;
758         }
759         irq_put_desc_unlock(desc, flags);
760         return canrequest;
761 }
762
763 int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
764 {
765         struct irq_chip *chip = desc->irq_data.chip;
766         int ret, unmask = 0;
767
768         if (!chip || !chip->irq_set_type) {
769                 /*
770                  * IRQF_TRIGGER_* but the PIC does not support multiple
771                  * flow-types?
772                  */
773                 pr_debug("No set_type function for IRQ %d (%s)\n",
774                          irq_desc_get_irq(desc),
775                          chip ? (chip->name ? : "unknown") : "unknown");
776                 return 0;
777         }
778
779         if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
780                 if (!irqd_irq_masked(&desc->irq_data))
781                         mask_irq(desc);
782                 if (!irqd_irq_disabled(&desc->irq_data))
783                         unmask = 1;
784         }
785
786         /* Mask all flags except trigger mode */
787         flags &= IRQ_TYPE_SENSE_MASK;
788         ret = chip->irq_set_type(&desc->irq_data, flags);
789
790         switch (ret) {
791         case IRQ_SET_MASK_OK:
792         case IRQ_SET_MASK_OK_DONE:
793                 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
794                 irqd_set(&desc->irq_data, flags);
795                 /* fall through */
796
797         case IRQ_SET_MASK_OK_NOCOPY:
798                 flags = irqd_get_trigger_type(&desc->irq_data);
799                 irq_settings_set_trigger_mask(desc, flags);
800                 irqd_clear(&desc->irq_data, IRQD_LEVEL);
801                 irq_settings_clr_level(desc);
802                 if (flags & IRQ_TYPE_LEVEL_MASK) {
803                         irq_settings_set_level(desc);
804                         irqd_set(&desc->irq_data, IRQD_LEVEL);
805                 }
806
807                 ret = 0;
808                 break;
809         default:
810                 pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
811                        flags, irq_desc_get_irq(desc), chip->irq_set_type);
812         }
813         if (unmask)
814                 unmask_irq(desc);
815         return ret;
816 }
817
818 #ifdef CONFIG_HARDIRQS_SW_RESEND
819 int irq_set_parent(int irq, int parent_irq)
820 {
821         unsigned long flags;
822         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
823
824         if (!desc)
825                 return -EINVAL;
826
827         desc->parent_irq = parent_irq;
828
829         irq_put_desc_unlock(desc, flags);
830         return 0;
831 }
832 EXPORT_SYMBOL_GPL(irq_set_parent);
833 #endif
834
835 /*
836  * Default primary interrupt handler for threaded interrupts. Is
837  * assigned as primary handler when request_threaded_irq is called
838  * with handler == NULL. Useful for oneshot interrupts.
839  */
840 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
841 {
842         return IRQ_WAKE_THREAD;
843 }
844
845 /*
846  * Primary handler for nested threaded interrupts. Should never be
847  * called.
848  */
849 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
850 {
851         WARN(1, "Primary handler called for nested irq %d\n", irq);
852         return IRQ_NONE;
853 }
854
855 static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
856 {
857         WARN(1, "Secondary action handler called for irq %d\n", irq);
858         return IRQ_NONE;
859 }
860
861 static int irq_wait_for_interrupt(struct irqaction *action)
862 {
863         for (;;) {
864                 set_current_state(TASK_INTERRUPTIBLE);
865
866                 if (kthread_should_stop()) {
867                         /* may need to run one last time */
868                         if (test_and_clear_bit(IRQTF_RUNTHREAD,
869                                                &action->thread_flags)) {
870                                 __set_current_state(TASK_RUNNING);
871                                 return 0;
872                         }
873                         __set_current_state(TASK_RUNNING);
874                         return -1;
875                 }
876
877                 if (test_and_clear_bit(IRQTF_RUNTHREAD,
878                                        &action->thread_flags)) {
879                         __set_current_state(TASK_RUNNING);
880                         return 0;
881                 }
882                 schedule();
883         }
884 }
885
886 /*
887  * Oneshot interrupts keep the irq line masked until the threaded
888  * handler finished. unmask if the interrupt has not been disabled and
889  * is marked MASKED.
890  */
891 static void irq_finalize_oneshot(struct irq_desc *desc,
892                                  struct irqaction *action)
893 {
894         if (!(desc->istate & IRQS_ONESHOT) ||
895             action->handler == irq_forced_secondary_handler)
896                 return;
897 again:
898         chip_bus_lock(desc);
899         raw_spin_lock_irq(&desc->lock);
900
901         /*
902          * Implausible though it may be we need to protect us against
903          * the following scenario:
904          *
905          * The thread is faster done than the hard interrupt handler
906          * on the other CPU. If we unmask the irq line then the
907          * interrupt can come in again and masks the line, leaves due
908          * to IRQS_INPROGRESS and the irq line is masked forever.
909          *
910          * This also serializes the state of shared oneshot handlers
911          * versus "desc->threads_onehsot |= action->thread_mask;" in
912          * irq_wake_thread(). See the comment there which explains the
913          * serialization.
914          */
915         if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
916                 raw_spin_unlock_irq(&desc->lock);
917                 chip_bus_sync_unlock(desc);
918                 cpu_relax();
919                 goto again;
920         }
921
922         /*
923          * Now check again, whether the thread should run. Otherwise
924          * we would clear the threads_oneshot bit of this thread which
925          * was just set.
926          */
927         if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
928                 goto out_unlock;
929
930         desc->threads_oneshot &= ~action->thread_mask;
931
932         if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
933             irqd_irq_masked(&desc->irq_data))
934                 unmask_threaded_irq(desc);
935
936 out_unlock:
937         raw_spin_unlock_irq(&desc->lock);
938         chip_bus_sync_unlock(desc);
939 }
940
941 #ifdef CONFIG_SMP
942 /*
943  * Check whether we need to change the affinity of the interrupt thread.
944  */
945 static void
946 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
947 {
948         cpumask_var_t mask;
949         bool valid = true;
950
951         if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
952                 return;
953
954         /*
955          * In case we are out of memory we set IRQTF_AFFINITY again and
956          * try again next time
957          */
958         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
959                 set_bit(IRQTF_AFFINITY, &action->thread_flags);
960                 return;
961         }
962
963         raw_spin_lock_irq(&desc->lock);
964         /*
965          * This code is triggered unconditionally. Check the affinity
966          * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
967          */
968         if (cpumask_available(desc->irq_common_data.affinity)) {
969                 const struct cpumask *m;
970
971                 m = irq_data_get_effective_affinity_mask(&desc->irq_data);
972                 cpumask_copy(mask, m);
973         } else {
974                 valid = false;
975         }
976         raw_spin_unlock_irq(&desc->lock);
977
978         if (valid)
979                 set_cpus_allowed_ptr(current, mask);
980         free_cpumask_var(mask);
981 }
982 #else
983 static inline void
984 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
985 #endif
986
987 /*
988  * Interrupts which are not explicitly requested as threaded
989  * interrupts rely on the implicit bh/preempt disable of the hard irq
990  * context. So we need to disable bh here to avoid deadlocks and other
991  * side effects.
992  */
993 static irqreturn_t
994 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
995 {
996         irqreturn_t ret;
997
998         local_bh_disable();
999         ret = action->thread_fn(action->irq, action->dev_id);
1000         if (ret == IRQ_HANDLED)
1001                 atomic_inc(&desc->threads_handled);
1002
1003         irq_finalize_oneshot(desc, action);
1004         local_bh_enable();
1005         return ret;
1006 }
1007
1008 /*
1009  * Interrupts explicitly requested as threaded interrupts want to be
1010  * preemtible - many of them need to sleep and wait for slow busses to
1011  * complete.
1012  */
1013 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1014                 struct irqaction *action)
1015 {
1016         irqreturn_t ret;
1017
1018         ret = action->thread_fn(action->irq, action->dev_id);
1019         if (ret == IRQ_HANDLED)
1020                 atomic_inc(&desc->threads_handled);
1021
1022         irq_finalize_oneshot(desc, action);
1023         return ret;
1024 }
1025
1026 static void wake_threads_waitq(struct irq_desc *desc)
1027 {
1028         if (atomic_dec_and_test(&desc->threads_active))
1029                 wake_up(&desc->wait_for_threads);
1030 }
1031
1032 static void irq_thread_dtor(struct callback_head *unused)
1033 {
1034         struct task_struct *tsk = current;
1035         struct irq_desc *desc;
1036         struct irqaction *action;
1037
1038         if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1039                 return;
1040
1041         action = kthread_data(tsk);
1042
1043         pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1044                tsk->comm, tsk->pid, action->irq);
1045
1046
1047         desc = irq_to_desc(action->irq);
1048         /*
1049          * If IRQTF_RUNTHREAD is set, we need to decrement
1050          * desc->threads_active and wake possible waiters.
1051          */
1052         if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1053                 wake_threads_waitq(desc);
1054
1055         /* Prevent a stale desc->threads_oneshot */
1056         irq_finalize_oneshot(desc, action);
1057 }
1058
1059 static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1060 {
1061         struct irqaction *secondary = action->secondary;
1062
1063         if (WARN_ON_ONCE(!secondary))
1064                 return;
1065
1066         raw_spin_lock_irq(&desc->lock);
1067         __irq_wake_thread(desc, secondary);
1068         raw_spin_unlock_irq(&desc->lock);
1069 }
1070
1071 /*
1072  * Interrupt handler thread
1073  */
1074 static int irq_thread(void *data)
1075 {
1076         struct callback_head on_exit_work;
1077         struct irqaction *action = data;
1078         struct irq_desc *desc = irq_to_desc(action->irq);
1079         irqreturn_t (*handler_fn)(struct irq_desc *desc,
1080                         struct irqaction *action);
1081
1082         if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
1083                                         &action->thread_flags))
1084                 handler_fn = irq_forced_thread_fn;
1085         else
1086                 handler_fn = irq_thread_fn;
1087
1088         init_task_work(&on_exit_work, irq_thread_dtor);
1089         task_work_add(current, &on_exit_work, false);
1090
1091         irq_thread_check_affinity(desc, action);
1092
1093         while (!irq_wait_for_interrupt(action)) {
1094                 irqreturn_t action_ret;
1095
1096                 irq_thread_check_affinity(desc, action);
1097
1098                 action_ret = handler_fn(desc, action);
1099                 if (action_ret == IRQ_WAKE_THREAD)
1100                         irq_wake_secondary(desc, action);
1101
1102                 wake_threads_waitq(desc);
1103         }
1104
1105         /*
1106          * This is the regular exit path. __free_irq() is stopping the
1107          * thread via kthread_stop() after calling
1108          * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1109          * oneshot mask bit can be set.
1110          */
1111         task_work_cancel(current, irq_thread_dtor);
1112         return 0;
1113 }
1114
1115 /**
1116  *      irq_wake_thread - wake the irq thread for the action identified by dev_id
1117  *      @irq:           Interrupt line
1118  *      @dev_id:        Device identity for which the thread should be woken
1119  *
1120  */
1121 void irq_wake_thread(unsigned int irq, void *dev_id)
1122 {
1123         struct irq_desc *desc = irq_to_desc(irq);
1124         struct irqaction *action;
1125         unsigned long flags;
1126
1127         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1128                 return;
1129
1130         raw_spin_lock_irqsave(&desc->lock, flags);
1131         for_each_action_of_desc(desc, action) {
1132                 if (action->dev_id == dev_id) {
1133                         if (action->thread)
1134                                 __irq_wake_thread(desc, action);
1135                         break;
1136                 }
1137         }
1138         raw_spin_unlock_irqrestore(&desc->lock, flags);
1139 }
1140 EXPORT_SYMBOL_GPL(irq_wake_thread);
1141
1142 static int irq_setup_forced_threading(struct irqaction *new)
1143 {
1144         if (!force_irqthreads)
1145                 return 0;
1146         if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1147                 return 0;
1148
1149         /*
1150          * No further action required for interrupts which are requested as
1151          * threaded interrupts already
1152          */
1153         if (new->handler == irq_default_primary_handler)
1154                 return 0;
1155
1156         new->flags |= IRQF_ONESHOT;
1157
1158         /*
1159          * Handle the case where we have a real primary handler and a
1160          * thread handler. We force thread them as well by creating a
1161          * secondary action.
1162          */
1163         if (new->handler && new->thread_fn) {
1164                 /* Allocate the secondary action */
1165                 new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1166                 if (!new->secondary)
1167                         return -ENOMEM;
1168                 new->secondary->handler = irq_forced_secondary_handler;
1169                 new->secondary->thread_fn = new->thread_fn;
1170                 new->secondary->dev_id = new->dev_id;
1171                 new->secondary->irq = new->irq;
1172                 new->secondary->name = new->name;
1173         }
1174         /* Deal with the primary handler */
1175         set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1176         new->thread_fn = new->handler;
1177         new->handler = irq_default_primary_handler;
1178         return 0;
1179 }
1180
1181 static int irq_request_resources(struct irq_desc *desc)
1182 {
1183         struct irq_data *d = &desc->irq_data;
1184         struct irq_chip *c = d->chip;
1185
1186         return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1187 }
1188
1189 static void irq_release_resources(struct irq_desc *desc)
1190 {
1191         struct irq_data *d = &desc->irq_data;
1192         struct irq_chip *c = d->chip;
1193
1194         if (c->irq_release_resources)
1195                 c->irq_release_resources(d);
1196 }
1197
1198 static bool irq_supports_nmi(struct irq_desc *desc)
1199 {
1200         struct irq_data *d = irq_desc_get_irq_data(desc);
1201
1202 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1203         /* Only IRQs directly managed by the root irqchip can be set as NMI */
1204         if (d->parent_data)
1205                 return false;
1206 #endif
1207         /* Don't support NMIs for chips behind a slow bus */
1208         if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1209                 return false;
1210
1211         return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1212 }
1213
1214 static int irq_nmi_setup(struct irq_desc *desc)
1215 {
1216         struct irq_data *d = irq_desc_get_irq_data(desc);
1217         struct irq_chip *c = d->chip;
1218
1219         return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1220 }
1221
1222 static void irq_nmi_teardown(struct irq_desc *desc)
1223 {
1224         struct irq_data *d = irq_desc_get_irq_data(desc);
1225         struct irq_chip *c = d->chip;
1226
1227         if (c->irq_nmi_teardown)
1228                 c->irq_nmi_teardown(d);
1229 }
1230
1231 static int
1232 setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1233 {
1234         struct task_struct *t;
1235         struct sched_param param = {
1236                 .sched_priority = MAX_USER_RT_PRIO/2,
1237         };
1238
1239         if (!secondary) {
1240                 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1241                                    new->name);
1242         } else {
1243                 t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1244                                    new->name);
1245                 param.sched_priority -= 1;
1246         }
1247
1248         if (IS_ERR(t))
1249                 return PTR_ERR(t);
1250
1251         sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1252
1253         /*
1254          * We keep the reference to the task struct even if
1255          * the thread dies to avoid that the interrupt code
1256          * references an already freed task_struct.
1257          */
1258         get_task_struct(t);
1259         new->thread = t;
1260         /*
1261          * Tell the thread to set its affinity. This is
1262          * important for shared interrupt handlers as we do
1263          * not invoke setup_affinity() for the secondary
1264          * handlers as everything is already set up. Even for
1265          * interrupts marked with IRQF_NO_BALANCE this is
1266          * correct as we want the thread to move to the cpu(s)
1267          * on which the requesting code placed the interrupt.
1268          */
1269         set_bit(IRQTF_AFFINITY, &new->thread_flags);
1270         return 0;
1271 }
1272
1273 /*
1274  * Internal function to register an irqaction - typically used to
1275  * allocate special interrupts that are part of the architecture.
1276  *
1277  * Locking rules:
1278  *
1279  * desc->request_mutex  Provides serialization against a concurrent free_irq()
1280  *   chip_bus_lock      Provides serialization for slow bus operations
1281  *     desc->lock       Provides serialization against hard interrupts
1282  *
1283  * chip_bus_lock and desc->lock are sufficient for all other management and
1284  * interrupt related functions. desc->request_mutex solely serializes
1285  * request/free_irq().
1286  */
1287 static int
1288 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1289 {
1290         struct irqaction *old, **old_ptr;
1291         unsigned long flags, thread_mask = 0;
1292         int ret, nested, shared = 0;
1293
1294         if (!desc)
1295                 return -EINVAL;
1296
1297         if (desc->irq_data.chip == &no_irq_chip)
1298                 return -ENOSYS;
1299         if (!try_module_get(desc->owner))
1300                 return -ENODEV;
1301
1302         new->irq = irq;
1303
1304         /*
1305          * If the trigger type is not specified by the caller,
1306          * then use the default for this interrupt.
1307          */
1308         if (!(new->flags & IRQF_TRIGGER_MASK))
1309                 new->flags |= irqd_get_trigger_type(&desc->irq_data);
1310
1311         /*
1312          * Check whether the interrupt nests into another interrupt
1313          * thread.
1314          */
1315         nested = irq_settings_is_nested_thread(desc);
1316         if (nested) {
1317                 if (!new->thread_fn) {
1318                         ret = -EINVAL;
1319                         goto out_mput;
1320                 }
1321                 /*
1322                  * Replace the primary handler which was provided from
1323                  * the driver for non nested interrupt handling by the
1324                  * dummy function which warns when called.
1325                  */
1326                 new->handler = irq_nested_primary_handler;
1327         } else {
1328                 if (irq_settings_can_thread(desc)) {
1329                         ret = irq_setup_forced_threading(new);
1330                         if (ret)
1331                                 goto out_mput;
1332                 }
1333         }
1334
1335         /*
1336          * Create a handler thread when a thread function is supplied
1337          * and the interrupt does not nest into another interrupt
1338          * thread.
1339          */
1340         if (new->thread_fn && !nested) {
1341                 ret = setup_irq_thread(new, irq, false);
1342                 if (ret)
1343                         goto out_mput;
1344                 if (new->secondary) {
1345                         ret = setup_irq_thread(new->secondary, irq, true);
1346                         if (ret)
1347                                 goto out_thread;
1348                 }
1349         }
1350
1351         /*
1352          * Drivers are often written to work w/o knowledge about the
1353          * underlying irq chip implementation, so a request for a
1354          * threaded irq without a primary hard irq context handler
1355          * requires the ONESHOT flag to be set. Some irq chips like
1356          * MSI based interrupts are per se one shot safe. Check the
1357          * chip flags, so we can avoid the unmask dance at the end of
1358          * the threaded handler for those.
1359          */
1360         if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1361                 new->flags &= ~IRQF_ONESHOT;
1362
1363         /*
1364          * Protects against a concurrent __free_irq() call which might wait
1365          * for synchronize_hardirq() to complete without holding the optional
1366          * chip bus lock and desc->lock. Also protects against handing out
1367          * a recycled oneshot thread_mask bit while it's still in use by
1368          * its previous owner.
1369          */
1370         mutex_lock(&desc->request_mutex);
1371
1372         /*
1373          * Acquire bus lock as the irq_request_resources() callback below
1374          * might rely on the serialization or the magic power management
1375          * functions which are abusing the irq_bus_lock() callback,
1376          */
1377         chip_bus_lock(desc);
1378
1379         /* First installed action requests resources. */
1380         if (!desc->action) {
1381                 ret = irq_request_resources(desc);
1382                 if (ret) {
1383                         pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1384                                new->name, irq, desc->irq_data.chip->name);
1385                         goto out_bus_unlock;
1386                 }
1387         }
1388
1389         /*
1390          * The following block of code has to be executed atomically
1391          * protected against a concurrent interrupt and any of the other
1392          * management calls which are not serialized via
1393          * desc->request_mutex or the optional bus lock.
1394          */
1395         raw_spin_lock_irqsave(&desc->lock, flags);
1396         old_ptr = &desc->action;
1397         old = *old_ptr;
1398         if (old) {
1399                 /*
1400                  * Can't share interrupts unless both agree to and are
1401                  * the same type (level, edge, polarity). So both flag
1402                  * fields must have IRQF_SHARED set and the bits which
1403                  * set the trigger type must match. Also all must
1404                  * agree on ONESHOT.
1405                  * Interrupt lines used for NMIs cannot be shared.
1406                  */
1407                 unsigned int oldtype;
1408
1409                 if (desc->istate & IRQS_NMI) {
1410                         pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1411                                 new->name, irq, desc->irq_data.chip->name);
1412                         ret = -EINVAL;
1413                         goto out_unlock;
1414                 }
1415
1416                 /*
1417                  * If nobody did set the configuration before, inherit
1418                  * the one provided by the requester.
1419                  */
1420                 if (irqd_trigger_type_was_set(&desc->irq_data)) {
1421                         oldtype = irqd_get_trigger_type(&desc->irq_data);
1422                 } else {
1423                         oldtype = new->flags & IRQF_TRIGGER_MASK;
1424                         irqd_set_trigger_type(&desc->irq_data, oldtype);
1425                 }
1426
1427                 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1428                     (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1429                     ((old->flags ^ new->flags) & IRQF_ONESHOT))
1430                         goto mismatch;
1431
1432                 /* All handlers must agree on per-cpuness */
1433                 if ((old->flags & IRQF_PERCPU) !=
1434                     (new->flags & IRQF_PERCPU))
1435                         goto mismatch;
1436
1437                 /* add new interrupt at end of irq queue */
1438                 do {
1439                         /*
1440                          * Or all existing action->thread_mask bits,
1441                          * so we can find the next zero bit for this
1442                          * new action.
1443                          */
1444                         thread_mask |= old->thread_mask;
1445                         old_ptr = &old->next;
1446                         old = *old_ptr;
1447                 } while (old);
1448                 shared = 1;
1449         }
1450
1451         /*
1452          * Setup the thread mask for this irqaction for ONESHOT. For
1453          * !ONESHOT irqs the thread mask is 0 so we can avoid a
1454          * conditional in irq_wake_thread().
1455          */
1456         if (new->flags & IRQF_ONESHOT) {
1457                 /*
1458                  * Unlikely to have 32 resp 64 irqs sharing one line,
1459                  * but who knows.
1460                  */
1461                 if (thread_mask == ~0UL) {
1462                         ret = -EBUSY;
1463                         goto out_unlock;
1464                 }
1465                 /*
1466                  * The thread_mask for the action is or'ed to
1467                  * desc->thread_active to indicate that the
1468                  * IRQF_ONESHOT thread handler has been woken, but not
1469                  * yet finished. The bit is cleared when a thread
1470                  * completes. When all threads of a shared interrupt
1471                  * line have completed desc->threads_active becomes
1472                  * zero and the interrupt line is unmasked. See
1473                  * handle.c:irq_wake_thread() for further information.
1474                  *
1475                  * If no thread is woken by primary (hard irq context)
1476                  * interrupt handlers, then desc->threads_active is
1477                  * also checked for zero to unmask the irq line in the
1478                  * affected hard irq flow handlers
1479                  * (handle_[fasteoi|level]_irq).
1480                  *
1481                  * The new action gets the first zero bit of
1482                  * thread_mask assigned. See the loop above which or's
1483                  * all existing action->thread_mask bits.
1484                  */
1485                 new->thread_mask = 1UL << ffz(thread_mask);
1486
1487         } else if (new->handler == irq_default_primary_handler &&
1488                    !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1489                 /*
1490                  * The interrupt was requested with handler = NULL, so
1491                  * we use the default primary handler for it. But it
1492                  * does not have the oneshot flag set. In combination
1493                  * with level interrupts this is deadly, because the
1494                  * default primary handler just wakes the thread, then
1495                  * the irq lines is reenabled, but the device still
1496                  * has the level irq asserted. Rinse and repeat....
1497                  *
1498                  * While this works for edge type interrupts, we play
1499                  * it safe and reject unconditionally because we can't
1500                  * say for sure which type this interrupt really
1501                  * has. The type flags are unreliable as the
1502                  * underlying chip implementation can override them.
1503                  */
1504                 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1505                        irq);
1506                 ret = -EINVAL;
1507                 goto out_unlock;
1508         }
1509
1510         if (!shared) {
1511                 init_waitqueue_head(&desc->wait_for_threads);
1512
1513                 /* Setup the type (level, edge polarity) if configured: */
1514                 if (new->flags & IRQF_TRIGGER_MASK) {
1515                         ret = __irq_set_trigger(desc,
1516                                                 new->flags & IRQF_TRIGGER_MASK);
1517
1518                         if (ret)
1519                                 goto out_unlock;
1520                 }
1521
1522                 /*
1523                  * Activate the interrupt. That activation must happen
1524                  * independently of IRQ_NOAUTOEN. request_irq() can fail
1525                  * and the callers are supposed to handle
1526                  * that. enable_irq() of an interrupt requested with
1527                  * IRQ_NOAUTOEN is not supposed to fail. The activation
1528                  * keeps it in shutdown mode, it merily associates
1529                  * resources if necessary and if that's not possible it
1530                  * fails. Interrupts which are in managed shutdown mode
1531                  * will simply ignore that activation request.
1532                  */
1533                 ret = irq_activate(desc);
1534                 if (ret)
1535                         goto out_unlock;
1536
1537                 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1538                                   IRQS_ONESHOT | IRQS_WAITING);
1539                 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1540
1541                 if (new->flags & IRQF_PERCPU) {
1542                         irqd_set(&desc->irq_data, IRQD_PER_CPU);
1543                         irq_settings_set_per_cpu(desc);
1544                 }
1545
1546                 if (new->flags & IRQF_ONESHOT)
1547                         desc->istate |= IRQS_ONESHOT;
1548
1549                 /* Exclude IRQ from balancing if requested */
1550                 if (new->flags & IRQF_NOBALANCING) {
1551                         irq_settings_set_no_balancing(desc);
1552                         irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1553                 }
1554
1555                 if (irq_settings_can_autoenable(desc)) {
1556                         irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1557                 } else {
1558                         /*
1559                          * Shared interrupts do not go well with disabling
1560                          * auto enable. The sharing interrupt might request
1561                          * it while it's still disabled and then wait for
1562                          * interrupts forever.
1563                          */
1564                         WARN_ON_ONCE(new->flags & IRQF_SHARED);
1565                         /* Undo nested disables: */
1566                         desc->depth = 1;
1567                 }
1568
1569         } else if (new->flags & IRQF_TRIGGER_MASK) {
1570                 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1571                 unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1572
1573                 if (nmsk != omsk)
1574                         /* hope the handler works with current  trigger mode */
1575                         pr_warn("irq %d uses trigger mode %u; requested %u\n",
1576                                 irq, omsk, nmsk);
1577         }
1578
1579         *old_ptr = new;
1580
1581         irq_pm_install_action(desc, new);
1582
1583         /* Reset broken irq detection when installing new handler */
1584         desc->irq_count = 0;
1585         desc->irqs_unhandled = 0;
1586
1587         /*
1588          * Check whether we disabled the irq via the spurious handler
1589          * before. Reenable it and give it another chance.
1590          */
1591         if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1592                 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1593                 __enable_irq(desc);
1594         }
1595
1596         raw_spin_unlock_irqrestore(&desc->lock, flags);
1597         chip_bus_sync_unlock(desc);
1598         mutex_unlock(&desc->request_mutex);
1599
1600         irq_setup_timings(desc, new);
1601
1602         /*
1603          * Strictly no need to wake it up, but hung_task complains
1604          * when no hard interrupt wakes the thread up.
1605          */
1606         if (new->thread)
1607                 wake_up_process(new->thread);
1608         if (new->secondary)
1609                 wake_up_process(new->secondary->thread);
1610
1611         register_irq_proc(irq, desc);
1612         new->dir = NULL;
1613         register_handler_proc(irq, new);
1614         return 0;
1615
1616 mismatch:
1617         if (!(new->flags & IRQF_PROBE_SHARED)) {
1618                 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1619                        irq, new->flags, new->name, old->flags, old->name);
1620 #ifdef CONFIG_DEBUG_SHIRQ
1621                 dump_stack();
1622 #endif
1623         }
1624         ret = -EBUSY;
1625
1626 out_unlock:
1627         raw_spin_unlock_irqrestore(&desc->lock, flags);
1628
1629         if (!desc->action)
1630                 irq_release_resources(desc);
1631 out_bus_unlock:
1632         chip_bus_sync_unlock(desc);
1633         mutex_unlock(&desc->request_mutex);
1634
1635 out_thread:
1636         if (new->thread) {
1637                 struct task_struct *t = new->thread;
1638
1639                 new->thread = NULL;
1640                 kthread_stop(t);
1641                 put_task_struct(t);
1642         }
1643         if (new->secondary && new->secondary->thread) {
1644                 struct task_struct *t = new->secondary->thread;
1645
1646                 new->secondary->thread = NULL;
1647                 kthread_stop(t);
1648                 put_task_struct(t);
1649         }
1650 out_mput:
1651         module_put(desc->owner);
1652         return ret;
1653 }
1654
1655 /**
1656  *      setup_irq - setup an interrupt
1657  *      @irq: Interrupt line to setup
1658  *      @act: irqaction for the interrupt
1659  *
1660  * Used to statically setup interrupts in the early boot process.
1661  */
1662 int setup_irq(unsigned int irq, struct irqaction *act)
1663 {
1664         int retval;
1665         struct irq_desc *desc = irq_to_desc(irq);
1666
1667         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1668                 return -EINVAL;
1669
1670         retval = irq_chip_pm_get(&desc->irq_data);
1671         if (retval < 0)
1672                 return retval;
1673
1674         retval = __setup_irq(irq, desc, act);
1675
1676         if (retval)
1677                 irq_chip_pm_put(&desc->irq_data);
1678
1679         return retval;
1680 }
1681 EXPORT_SYMBOL_GPL(setup_irq);
1682
1683 /*
1684  * Internal function to unregister an irqaction - used to free
1685  * regular and special interrupts that are part of the architecture.
1686  */
1687 static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1688 {
1689         unsigned irq = desc->irq_data.irq;
1690         struct irqaction *action, **action_ptr;
1691         unsigned long flags;
1692
1693         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1694
1695         mutex_lock(&desc->request_mutex);
1696         chip_bus_lock(desc);
1697         raw_spin_lock_irqsave(&desc->lock, flags);
1698
1699         /*
1700          * There can be multiple actions per IRQ descriptor, find the right
1701          * one based on the dev_id:
1702          */
1703         action_ptr = &desc->action;
1704         for (;;) {
1705                 action = *action_ptr;
1706
1707                 if (!action) {
1708                         WARN(1, "Trying to free already-free IRQ %d\n", irq);
1709                         raw_spin_unlock_irqrestore(&desc->lock, flags);
1710                         chip_bus_sync_unlock(desc);
1711                         mutex_unlock(&desc->request_mutex);
1712                         return NULL;
1713                 }
1714
1715                 if (action->dev_id == dev_id)
1716                         break;
1717                 action_ptr = &action->next;
1718         }
1719
1720         /* Found it - now remove it from the list of entries: */
1721         *action_ptr = action->next;
1722
1723         irq_pm_remove_action(desc, action);
1724
1725         /* If this was the last handler, shut down the IRQ line: */
1726         if (!desc->action) {
1727                 irq_settings_clr_disable_unlazy(desc);
1728                 /* Only shutdown. Deactivate after synchronize_hardirq() */
1729                 irq_shutdown(desc);
1730         }
1731
1732 #ifdef CONFIG_SMP
1733         /* make sure affinity_hint is cleaned up */
1734         if (WARN_ON_ONCE(desc->affinity_hint))
1735                 desc->affinity_hint = NULL;
1736 #endif
1737
1738         raw_spin_unlock_irqrestore(&desc->lock, flags);
1739         /*
1740          * Drop bus_lock here so the changes which were done in the chip
1741          * callbacks above are synced out to the irq chips which hang
1742          * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1743          *
1744          * Aside of that the bus_lock can also be taken from the threaded
1745          * handler in irq_finalize_oneshot() which results in a deadlock
1746          * because kthread_stop() would wait forever for the thread to
1747          * complete, which is blocked on the bus lock.
1748          *
1749          * The still held desc->request_mutex() protects against a
1750          * concurrent request_irq() of this irq so the release of resources
1751          * and timing data is properly serialized.
1752          */
1753         chip_bus_sync_unlock(desc);
1754
1755         unregister_handler_proc(irq, action);
1756
1757         /*
1758          * Make sure it's not being used on another CPU and if the chip
1759          * supports it also make sure that there is no (not yet serviced)
1760          * interrupt in flight at the hardware level.
1761          */
1762         __synchronize_hardirq(desc, true);
1763
1764 #ifdef CONFIG_DEBUG_SHIRQ
1765         /*
1766          * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1767          * event to happen even now it's being freed, so let's make sure that
1768          * is so by doing an extra call to the handler ....
1769          *
1770          * ( We do this after actually deregistering it, to make sure that a
1771          *   'real' IRQ doesn't run in parallel with our fake. )
1772          */
1773         if (action->flags & IRQF_SHARED) {
1774                 local_irq_save(flags);
1775                 action->handler(irq, dev_id);
1776                 local_irq_restore(flags);
1777         }
1778 #endif
1779
1780         /*
1781          * The action has already been removed above, but the thread writes
1782          * its oneshot mask bit when it completes. Though request_mutex is
1783          * held across this which prevents __setup_irq() from handing out
1784          * the same bit to a newly requested action.
1785          */
1786         if (action->thread) {
1787                 kthread_stop(action->thread);
1788                 put_task_struct(action->thread);
1789                 if (action->secondary && action->secondary->thread) {
1790                         kthread_stop(action->secondary->thread);
1791                         put_task_struct(action->secondary->thread);
1792                 }
1793         }
1794
1795         /* Last action releases resources */
1796         if (!desc->action) {
1797                 /*
1798                  * Reaquire bus lock as irq_release_resources() might
1799                  * require it to deallocate resources over the slow bus.
1800                  */
1801                 chip_bus_lock(desc);
1802                 /*
1803                  * There is no interrupt on the fly anymore. Deactivate it
1804                  * completely.
1805                  */
1806                 raw_spin_lock_irqsave(&desc->lock, flags);
1807                 irq_domain_deactivate_irq(&desc->irq_data);
1808                 raw_spin_unlock_irqrestore(&desc->lock, flags);
1809
1810                 irq_release_resources(desc);
1811                 chip_bus_sync_unlock(desc);
1812                 irq_remove_timings(desc);
1813         }
1814
1815         mutex_unlock(&desc->request_mutex);
1816
1817         irq_chip_pm_put(&desc->irq_data);
1818         module_put(desc->owner);
1819         kfree(action->secondary);
1820         return action;
1821 }
1822
1823 /**
1824  *      remove_irq - free an interrupt
1825  *      @irq: Interrupt line to free
1826  *      @act: irqaction for the interrupt
1827  *
1828  * Used to remove interrupts statically setup by the early boot process.
1829  */
1830 void remove_irq(unsigned int irq, struct irqaction *act)
1831 {
1832         struct irq_desc *desc = irq_to_desc(irq);
1833
1834         if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1835                 __free_irq(desc, act->dev_id);
1836 }
1837 EXPORT_SYMBOL_GPL(remove_irq);
1838
1839 /**
1840  *      free_irq - free an interrupt allocated with request_irq
1841  *      @irq: Interrupt line to free
1842  *      @dev_id: Device identity to free
1843  *
1844  *      Remove an interrupt handler. The handler is removed and if the
1845  *      interrupt line is no longer in use by any driver it is disabled.
1846  *      On a shared IRQ the caller must ensure the interrupt is disabled
1847  *      on the card it drives before calling this function. The function
1848  *      does not return until any executing interrupts for this IRQ
1849  *      have completed.
1850  *
1851  *      This function must not be called from interrupt context.
1852  *
1853  *      Returns the devname argument passed to request_irq.
1854  */
1855 const void *free_irq(unsigned int irq, void *dev_id)
1856 {
1857         struct irq_desc *desc = irq_to_desc(irq);
1858         struct irqaction *action;
1859         const char *devname;
1860
1861         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1862                 return NULL;
1863
1864 #ifdef CONFIG_SMP
1865         if (WARN_ON(desc->affinity_notify))
1866                 desc->affinity_notify = NULL;
1867 #endif
1868
1869         action = __free_irq(desc, dev_id);
1870
1871         if (!action)
1872                 return NULL;
1873
1874         devname = action->name;
1875         kfree(action);
1876         return devname;
1877 }
1878 EXPORT_SYMBOL(free_irq);
1879
1880 /* This function must be called with desc->lock held */
1881 static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
1882 {
1883         const char *devname = NULL;
1884
1885         desc->istate &= ~IRQS_NMI;
1886
1887         if (!WARN_ON(desc->action == NULL)) {
1888                 irq_pm_remove_action(desc, desc->action);
1889                 devname = desc->action->name;
1890                 unregister_handler_proc(irq, desc->action);
1891
1892                 kfree(desc->action);
1893                 desc->action = NULL;
1894         }
1895
1896         irq_settings_clr_disable_unlazy(desc);
1897         irq_shutdown_and_deactivate(desc);
1898
1899         irq_release_resources(desc);
1900
1901         irq_chip_pm_put(&desc->irq_data);
1902         module_put(desc->owner);
1903
1904         return devname;
1905 }
1906
1907 const void *free_nmi(unsigned int irq, void *dev_id)
1908 {
1909         struct irq_desc *desc = irq_to_desc(irq);
1910         unsigned long flags;
1911         const void *devname;
1912
1913         if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
1914                 return NULL;
1915
1916         if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1917                 return NULL;
1918
1919         /* NMI still enabled */
1920         if (WARN_ON(desc->depth == 0))
1921                 disable_nmi_nosync(irq);
1922
1923         raw_spin_lock_irqsave(&desc->lock, flags);
1924
1925         irq_nmi_teardown(desc);
1926         devname = __cleanup_nmi(irq, desc);
1927
1928         raw_spin_unlock_irqrestore(&desc->lock, flags);
1929
1930         return devname;
1931 }
1932
1933 /**
1934  *      request_threaded_irq - allocate an interrupt line
1935  *      @irq: Interrupt line to allocate
1936  *      @handler: Function to be called when the IRQ occurs.
1937  *                Primary handler for threaded interrupts
1938  *                If NULL and thread_fn != NULL the default
1939  *                primary handler is installed
1940  *      @thread_fn: Function called from the irq handler thread
1941  *                  If NULL, no irq thread is created
1942  *      @irqflags: Interrupt type flags
1943  *      @devname: An ascii name for the claiming device
1944  *      @dev_id: A cookie passed back to the handler function
1945  *
1946  *      This call allocates interrupt resources and enables the
1947  *      interrupt line and IRQ handling. From the point this
1948  *      call is made your handler function may be invoked. Since
1949  *      your handler function must clear any interrupt the board
1950  *      raises, you must take care both to initialise your hardware
1951  *      and to set up the interrupt handler in the right order.
1952  *
1953  *      If you want to set up a threaded irq handler for your device
1954  *      then you need to supply @handler and @thread_fn. @handler is
1955  *      still called in hard interrupt context and has to check
1956  *      whether the interrupt originates from the device. If yes it
1957  *      needs to disable the interrupt on the device and return
1958  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1959  *      @thread_fn. This split handler design is necessary to support
1960  *      shared interrupts.
1961  *
1962  *      Dev_id must be globally unique. Normally the address of the
1963  *      device data structure is used as the cookie. Since the handler
1964  *      receives this value it makes sense to use it.
1965  *
1966  *      If your interrupt is shared you must pass a non NULL dev_id
1967  *      as this is required when freeing the interrupt.
1968  *
1969  *      Flags:
1970  *
1971  *      IRQF_SHARED             Interrupt is shared
1972  *      IRQF_TRIGGER_*          Specify active edge(s) or level
1973  *
1974  */
1975 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1976                          irq_handler_t thread_fn, unsigned long irqflags,
1977                          const char *devname, void *dev_id)
1978 {
1979         struct irqaction *action;
1980         struct irq_desc *desc;
1981         int retval;
1982
1983         if (irq == IRQ_NOTCONNECTED)
1984                 return -ENOTCONN;
1985
1986         /*
1987          * Sanity-check: shared interrupts must pass in a real dev-ID,
1988          * otherwise we'll have trouble later trying to figure out
1989          * which interrupt is which (messes up the interrupt freeing
1990          * logic etc).
1991          *
1992          * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1993          * it cannot be set along with IRQF_NO_SUSPEND.
1994          */
1995         if (((irqflags & IRQF_SHARED) && !dev_id) ||
1996             (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1997             ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1998                 return -EINVAL;
1999
2000         desc = irq_to_desc(irq);
2001         if (!desc)
2002                 return -EINVAL;
2003
2004         if (!irq_settings_can_request(desc) ||
2005             WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2006                 return -EINVAL;
2007
2008         if (!handler) {
2009                 if (!thread_fn)
2010                         return -EINVAL;
2011                 handler = irq_default_primary_handler;
2012         }
2013
2014         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2015         if (!action)
2016                 return -ENOMEM;
2017
2018         action->handler = handler;
2019         action->thread_fn = thread_fn;
2020         action->flags = irqflags;
2021         action->name = devname;
2022         action->dev_id = dev_id;
2023
2024         retval = irq_chip_pm_get(&desc->irq_data);
2025         if (retval < 0) {
2026                 kfree(action);
2027                 return retval;
2028         }
2029
2030         retval = __setup_irq(irq, desc, action);
2031
2032         if (retval) {
2033                 irq_chip_pm_put(&desc->irq_data);
2034                 kfree(action->secondary);
2035                 kfree(action);
2036         }
2037
2038 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
2039         if (!retval && (irqflags & IRQF_SHARED)) {
2040                 /*
2041                  * It's a shared IRQ -- the driver ought to be prepared for it
2042                  * to happen immediately, so let's make sure....
2043                  * We disable the irq to make sure that a 'real' IRQ doesn't
2044                  * run in parallel with our fake.
2045                  */
2046                 unsigned long flags;
2047
2048                 disable_irq(irq);
2049                 local_irq_save(flags);
2050
2051                 handler(irq, dev_id);
2052
2053                 local_irq_restore(flags);
2054                 enable_irq(irq);
2055         }
2056 #endif
2057         return retval;
2058 }
2059 EXPORT_SYMBOL(request_threaded_irq);
2060
2061 /**
2062  *      request_any_context_irq - allocate an interrupt line
2063  *      @irq: Interrupt line to allocate
2064  *      @handler: Function to be called when the IRQ occurs.
2065  *                Threaded handler for threaded interrupts.
2066  *      @flags: Interrupt type flags
2067  *      @name: An ascii name for the claiming device
2068  *      @dev_id: A cookie passed back to the handler function
2069  *
2070  *      This call allocates interrupt resources and enables the
2071  *      interrupt line and IRQ handling. It selects either a
2072  *      hardirq or threaded handling method depending on the
2073  *      context.
2074  *
2075  *      On failure, it returns a negative value. On success,
2076  *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2077  */
2078 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2079                             unsigned long flags, const char *name, void *dev_id)
2080 {
2081         struct irq_desc *desc;
2082         int ret;
2083
2084         if (irq == IRQ_NOTCONNECTED)
2085                 return -ENOTCONN;
2086
2087         desc = irq_to_desc(irq);
2088         if (!desc)
2089                 return -EINVAL;
2090
2091         if (irq_settings_is_nested_thread(desc)) {
2092                 ret = request_threaded_irq(irq, NULL, handler,
2093                                            flags, name, dev_id);
2094                 return !ret ? IRQC_IS_NESTED : ret;
2095         }
2096
2097         ret = request_irq(irq, handler, flags, name, dev_id);
2098         return !ret ? IRQC_IS_HARDIRQ : ret;
2099 }
2100 EXPORT_SYMBOL_GPL(request_any_context_irq);
2101
2102 /**
2103  *      request_nmi - allocate an interrupt line for NMI delivery
2104  *      @irq: Interrupt line to allocate
2105  *      @handler: Function to be called when the IRQ occurs.
2106  *                Threaded handler for threaded interrupts.
2107  *      @irqflags: Interrupt type flags
2108  *      @name: An ascii name for the claiming device
2109  *      @dev_id: A cookie passed back to the handler function
2110  *
2111  *      This call allocates interrupt resources and enables the
2112  *      interrupt line and IRQ handling. It sets up the IRQ line
2113  *      to be handled as an NMI.
2114  *
2115  *      An interrupt line delivering NMIs cannot be shared and IRQ handling
2116  *      cannot be threaded.
2117  *
2118  *      Interrupt lines requested for NMI delivering must produce per cpu
2119  *      interrupts and have auto enabling setting disabled.
2120  *
2121  *      Dev_id must be globally unique. Normally the address of the
2122  *      device data structure is used as the cookie. Since the handler
2123  *      receives this value it makes sense to use it.
2124  *
2125  *      If the interrupt line cannot be used to deliver NMIs, function
2126  *      will fail and return a negative value.
2127  */
2128 int request_nmi(unsigned int irq, irq_handler_t handler,
2129                 unsigned long irqflags, const char *name, void *dev_id)
2130 {
2131         struct irqaction *action;
2132         struct irq_desc *desc;
2133         unsigned long flags;
2134         int retval;
2135
2136         if (irq == IRQ_NOTCONNECTED)
2137                 return -ENOTCONN;
2138
2139         /* NMI cannot be shared, used for Polling */
2140         if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2141                 return -EINVAL;
2142
2143         if (!(irqflags & IRQF_PERCPU))
2144                 return -EINVAL;
2145
2146         if (!handler)
2147                 return -EINVAL;
2148
2149         desc = irq_to_desc(irq);
2150
2151         if (!desc || irq_settings_can_autoenable(desc) ||
2152             !irq_settings_can_request(desc) ||
2153             WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2154             !irq_supports_nmi(desc))
2155                 return -EINVAL;
2156
2157         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2158         if (!action)
2159                 return -ENOMEM;
2160
2161         action->handler = handler;
2162         action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2163         action->name = name;
2164         action->dev_id = dev_id;
2165
2166         retval = irq_chip_pm_get(&desc->irq_data);
2167         if (retval < 0)
2168                 goto err_out;
2169
2170         retval = __setup_irq(irq, desc, action);
2171         if (retval)
2172                 goto err_irq_setup;
2173
2174         raw_spin_lock_irqsave(&desc->lock, flags);
2175
2176         /* Setup NMI state */
2177         desc->istate |= IRQS_NMI;
2178         retval = irq_nmi_setup(desc);
2179         if (retval) {
2180                 __cleanup_nmi(irq, desc);
2181                 raw_spin_unlock_irqrestore(&desc->lock, flags);
2182                 return -EINVAL;
2183         }
2184
2185         raw_spin_unlock_irqrestore(&desc->lock, flags);
2186
2187         return 0;
2188
2189 err_irq_setup:
2190         irq_chip_pm_put(&desc->irq_data);
2191 err_out:
2192         kfree(action);
2193
2194         return retval;
2195 }
2196
2197 void enable_percpu_irq(unsigned int irq, unsigned int type)
2198 {
2199         unsigned int cpu = smp_processor_id();
2200         unsigned long flags;
2201         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2202
2203         if (!desc)
2204                 return;
2205
2206         /*
2207          * If the trigger type is not specified by the caller, then
2208          * use the default for this interrupt.
2209          */
2210         type &= IRQ_TYPE_SENSE_MASK;
2211         if (type == IRQ_TYPE_NONE)
2212                 type = irqd_get_trigger_type(&desc->irq_data);
2213
2214         if (type != IRQ_TYPE_NONE) {
2215                 int ret;
2216
2217                 ret = __irq_set_trigger(desc, type);
2218
2219                 if (ret) {
2220                         WARN(1, "failed to set type for IRQ%d\n", irq);
2221                         goto out;
2222                 }
2223         }
2224
2225         irq_percpu_enable(desc, cpu);
2226 out:
2227         irq_put_desc_unlock(desc, flags);
2228 }
2229 EXPORT_SYMBOL_GPL(enable_percpu_irq);
2230
2231 void enable_percpu_nmi(unsigned int irq, unsigned int type)
2232 {
2233         enable_percpu_irq(irq, type);
2234 }
2235
2236 /**
2237  * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2238  * @irq:        Linux irq number to check for
2239  *
2240  * Must be called from a non migratable context. Returns the enable
2241  * state of a per cpu interrupt on the current cpu.
2242  */
2243 bool irq_percpu_is_enabled(unsigned int irq)
2244 {
2245         unsigned int cpu = smp_processor_id();
2246         struct irq_desc *desc;
2247         unsigned long flags;
2248         bool is_enabled;
2249
2250         desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2251         if (!desc)
2252                 return false;
2253
2254         is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2255         irq_put_desc_unlock(desc, flags);
2256
2257         return is_enabled;
2258 }
2259 EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2260
2261 void disable_percpu_irq(unsigned int irq)
2262 {
2263         unsigned int cpu = smp_processor_id();
2264         unsigned long flags;
2265         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2266
2267         if (!desc)
2268                 return;
2269
2270         irq_percpu_disable(desc, cpu);
2271         irq_put_desc_unlock(desc, flags);
2272 }
2273 EXPORT_SYMBOL_GPL(disable_percpu_irq);
2274
2275 void disable_percpu_nmi(unsigned int irq)
2276 {
2277         disable_percpu_irq(irq);
2278 }
2279
2280 /*
2281  * Internal function to unregister a percpu irqaction.
2282  */
2283 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2284 {
2285         struct irq_desc *desc = irq_to_desc(irq);
2286         struct irqaction *action;
2287         unsigned long flags;
2288
2289         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2290
2291         if (!desc)
2292                 return NULL;
2293
2294         raw_spin_lock_irqsave(&desc->lock, flags);
2295
2296         action = desc->action;
2297         if (!action || action->percpu_dev_id != dev_id) {
2298                 WARN(1, "Trying to free already-free IRQ %d\n", irq);
2299                 goto bad;
2300         }
2301
2302         if (!cpumask_empty(desc->percpu_enabled)) {
2303                 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2304                      irq, cpumask_first(desc->percpu_enabled));
2305                 goto bad;
2306         }
2307
2308         /* Found it - now remove it from the list of entries: */
2309         desc->action = NULL;
2310
2311         desc->istate &= ~IRQS_NMI;
2312
2313         raw_spin_unlock_irqrestore(&desc->lock, flags);
2314
2315         unregister_handler_proc(irq, action);
2316
2317         irq_chip_pm_put(&desc->irq_data);
2318         module_put(desc->owner);
2319         return action;
2320
2321 bad:
2322         raw_spin_unlock_irqrestore(&desc->lock, flags);
2323         return NULL;
2324 }
2325
2326 /**
2327  *      remove_percpu_irq - free a per-cpu interrupt
2328  *      @irq: Interrupt line to free
2329  *      @act: irqaction for the interrupt
2330  *
2331  * Used to remove interrupts statically setup by the early boot process.
2332  */
2333 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2334 {
2335         struct irq_desc *desc = irq_to_desc(irq);
2336
2337         if (desc && irq_settings_is_per_cpu_devid(desc))
2338             __free_percpu_irq(irq, act->percpu_dev_id);
2339 }
2340
2341 /**
2342  *      free_percpu_irq - free an interrupt allocated with request_percpu_irq
2343  *      @irq: Interrupt line to free
2344  *      @dev_id: Device identity to free
2345  *
2346  *      Remove a percpu interrupt handler. The handler is removed, but
2347  *      the interrupt line is not disabled. This must be done on each
2348  *      CPU before calling this function. The function does not return
2349  *      until any executing interrupts for this IRQ have completed.
2350  *
2351  *      This function must not be called from interrupt context.
2352  */
2353 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2354 {
2355         struct irq_desc *desc = irq_to_desc(irq);
2356
2357         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2358                 return;
2359
2360         chip_bus_lock(desc);
2361         kfree(__free_percpu_irq(irq, dev_id));
2362         chip_bus_sync_unlock(desc);
2363 }
2364 EXPORT_SYMBOL_GPL(free_percpu_irq);
2365
2366 void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2367 {
2368         struct irq_desc *desc = irq_to_desc(irq);
2369
2370         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2371                 return;
2372
2373         if (WARN_ON(!(desc->istate & IRQS_NMI)))
2374                 return;
2375
2376         kfree(__free_percpu_irq(irq, dev_id));
2377 }
2378
2379 /**
2380  *      setup_percpu_irq - setup a per-cpu interrupt
2381  *      @irq: Interrupt line to setup
2382  *      @act: irqaction for the interrupt
2383  *
2384  * Used to statically setup per-cpu interrupts in the early boot process.
2385  */
2386 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2387 {
2388         struct irq_desc *desc = irq_to_desc(irq);
2389         int retval;
2390
2391         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2392                 return -EINVAL;
2393
2394         retval = irq_chip_pm_get(&desc->irq_data);
2395         if (retval < 0)
2396                 return retval;
2397
2398         retval = __setup_irq(irq, desc, act);
2399
2400         if (retval)
2401                 irq_chip_pm_put(&desc->irq_data);
2402
2403         return retval;
2404 }
2405
2406 /**
2407  *      __request_percpu_irq - allocate a percpu interrupt line
2408  *      @irq: Interrupt line to allocate
2409  *      @handler: Function to be called when the IRQ occurs.
2410  *      @flags: Interrupt type flags (IRQF_TIMER only)
2411  *      @devname: An ascii name for the claiming device
2412  *      @dev_id: A percpu cookie passed back to the handler function
2413  *
2414  *      This call allocates interrupt resources and enables the
2415  *      interrupt on the local CPU. If the interrupt is supposed to be
2416  *      enabled on other CPUs, it has to be done on each CPU using
2417  *      enable_percpu_irq().
2418  *
2419  *      Dev_id must be globally unique. It is a per-cpu variable, and
2420  *      the handler gets called with the interrupted CPU's instance of
2421  *      that variable.
2422  */
2423 int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2424                          unsigned long flags, const char *devname,
2425                          void __percpu *dev_id)
2426 {
2427         struct irqaction *action;
2428         struct irq_desc *desc;
2429         int retval;
2430
2431         if (!dev_id)
2432                 return -EINVAL;
2433
2434         desc = irq_to_desc(irq);
2435         if (!desc || !irq_settings_can_request(desc) ||
2436             !irq_settings_is_per_cpu_devid(desc))
2437                 return -EINVAL;
2438
2439         if (flags && flags != IRQF_TIMER)
2440                 return -EINVAL;
2441
2442         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2443         if (!action)
2444                 return -ENOMEM;
2445
2446         action->handler = handler;
2447         action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2448         action->name = devname;
2449         action->percpu_dev_id = dev_id;
2450
2451         retval = irq_chip_pm_get(&desc->irq_data);
2452         if (retval < 0) {
2453                 kfree(action);
2454                 return retval;
2455         }
2456
2457         retval = __setup_irq(irq, desc, action);
2458
2459         if (retval) {
2460                 irq_chip_pm_put(&desc->irq_data);
2461                 kfree(action);
2462         }
2463
2464         return retval;
2465 }
2466 EXPORT_SYMBOL_GPL(__request_percpu_irq);
2467
2468 /**
2469  *      request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2470  *      @irq: Interrupt line to allocate
2471  *      @handler: Function to be called when the IRQ occurs.
2472  *      @name: An ascii name for the claiming device
2473  *      @dev_id: A percpu cookie passed back to the handler function
2474  *
2475  *      This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2476  *      have to be setup on each CPU by calling prepare_percpu_nmi() before
2477  *      being enabled on the same CPU by using enable_percpu_nmi().
2478  *
2479  *      Dev_id must be globally unique. It is a per-cpu variable, and
2480  *      the handler gets called with the interrupted CPU's instance of
2481  *      that variable.
2482  *
2483  *      Interrupt lines requested for NMI delivering should have auto enabling
2484  *      setting disabled.
2485  *
2486  *      If the interrupt line cannot be used to deliver NMIs, function
2487  *      will fail returning a negative value.
2488  */
2489 int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2490                        const char *name, void __percpu *dev_id)
2491 {
2492         struct irqaction *action;
2493         struct irq_desc *desc;
2494         unsigned long flags;
2495         int retval;
2496
2497         if (!handler)
2498                 return -EINVAL;
2499
2500         desc = irq_to_desc(irq);
2501
2502         if (!desc || !irq_settings_can_request(desc) ||
2503             !irq_settings_is_per_cpu_devid(desc) ||
2504             irq_settings_can_autoenable(desc) ||
2505             !irq_supports_nmi(desc))
2506                 return -EINVAL;
2507
2508         /* The line cannot already be NMI */
2509         if (desc->istate & IRQS_NMI)
2510                 return -EINVAL;
2511
2512         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2513         if (!action)
2514                 return -ENOMEM;
2515
2516         action->handler = handler;
2517         action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2518                 | IRQF_NOBALANCING;
2519         action->name = name;
2520         action->percpu_dev_id = dev_id;
2521
2522         retval = irq_chip_pm_get(&desc->irq_data);
2523         if (retval < 0)
2524                 goto err_out;
2525
2526         retval = __setup_irq(irq, desc, action);
2527         if (retval)
2528                 goto err_irq_setup;
2529
2530         raw_spin_lock_irqsave(&desc->lock, flags);
2531         desc->istate |= IRQS_NMI;
2532         raw_spin_unlock_irqrestore(&desc->lock, flags);
2533
2534         return 0;
2535
2536 err_irq_setup:
2537         irq_chip_pm_put(&desc->irq_data);
2538 err_out:
2539         kfree(action);
2540
2541         return retval;
2542 }
2543
2544 /**
2545  *      prepare_percpu_nmi - performs CPU local setup for NMI delivery
2546  *      @irq: Interrupt line to prepare for NMI delivery
2547  *
2548  *      This call prepares an interrupt line to deliver NMI on the current CPU,
2549  *      before that interrupt line gets enabled with enable_percpu_nmi().
2550  *
2551  *      As a CPU local operation, this should be called from non-preemptible
2552  *      context.
2553  *
2554  *      If the interrupt line cannot be used to deliver NMIs, function
2555  *      will fail returning a negative value.
2556  */
2557 int prepare_percpu_nmi(unsigned int irq)
2558 {
2559         unsigned long flags;
2560         struct irq_desc *desc;
2561         int ret = 0;
2562
2563         WARN_ON(preemptible());
2564
2565         desc = irq_get_desc_lock(irq, &flags,
2566                                  IRQ_GET_DESC_CHECK_PERCPU);
2567         if (!desc)
2568                 return -EINVAL;
2569
2570         if (WARN(!(desc->istate & IRQS_NMI),
2571                  KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2572                  irq)) {
2573                 ret = -EINVAL;
2574                 goto out;
2575         }
2576
2577         ret = irq_nmi_setup(desc);
2578         if (ret) {
2579                 pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2580                 goto out;
2581         }
2582
2583 out:
2584         irq_put_desc_unlock(desc, flags);
2585         return ret;
2586 }
2587
2588 /**
2589  *      teardown_percpu_nmi - undoes NMI setup of IRQ line
2590  *      @irq: Interrupt line from which CPU local NMI configuration should be
2591  *            removed
2592  *
2593  *      This call undoes the setup done by prepare_percpu_nmi().
2594  *
2595  *      IRQ line should not be enabled for the current CPU.
2596  *
2597  *      As a CPU local operation, this should be called from non-preemptible
2598  *      context.
2599  */
2600 void teardown_percpu_nmi(unsigned int irq)
2601 {
2602         unsigned long flags;
2603         struct irq_desc *desc;
2604
2605         WARN_ON(preemptible());
2606
2607         desc = irq_get_desc_lock(irq, &flags,
2608                                  IRQ_GET_DESC_CHECK_PERCPU);
2609         if (!desc)
2610                 return;
2611
2612         if (WARN_ON(!(desc->istate & IRQS_NMI)))
2613                 goto out;
2614
2615         irq_nmi_teardown(desc);
2616 out:
2617         irq_put_desc_unlock(desc, flags);
2618 }
2619
2620 int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2621                             bool *state)
2622 {
2623         struct irq_chip *chip;
2624         int err = -EINVAL;
2625
2626         do {
2627                 chip = irq_data_get_irq_chip(data);
2628                 if (chip->irq_get_irqchip_state)
2629                         break;
2630 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2631                 data = data->parent_data;
2632 #else
2633                 data = NULL;
2634 #endif
2635         } while (data);
2636
2637         if (data)
2638                 err = chip->irq_get_irqchip_state(data, which, state);
2639         return err;
2640 }
2641
2642 /**
2643  *      irq_get_irqchip_state - returns the irqchip state of a interrupt.
2644  *      @irq: Interrupt line that is forwarded to a VM
2645  *      @which: One of IRQCHIP_STATE_* the caller wants to know about
2646  *      @state: a pointer to a boolean where the state is to be storeed
2647  *
2648  *      This call snapshots the internal irqchip state of an
2649  *      interrupt, returning into @state the bit corresponding to
2650  *      stage @which
2651  *
2652  *      This function should be called with preemption disabled if the
2653  *      interrupt controller has per-cpu registers.
2654  */
2655 int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2656                           bool *state)
2657 {
2658         struct irq_desc *desc;
2659         struct irq_data *data;
2660         unsigned long flags;
2661         int err = -EINVAL;
2662
2663         desc = irq_get_desc_buslock(irq, &flags, 0);
2664         if (!desc)
2665                 return err;
2666
2667         data = irq_desc_get_irq_data(desc);
2668
2669         err = __irq_get_irqchip_state(data, which, state);
2670
2671         irq_put_desc_busunlock(desc, flags);
2672         return err;
2673 }
2674 EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2675
2676 /**
2677  *      irq_set_irqchip_state - set the state of a forwarded interrupt.
2678  *      @irq: Interrupt line that is forwarded to a VM
2679  *      @which: State to be restored (one of IRQCHIP_STATE_*)
2680  *      @val: Value corresponding to @which
2681  *
2682  *      This call sets the internal irqchip state of an interrupt,
2683  *      depending on the value of @which.
2684  *
2685  *      This function should be called with preemption disabled if the
2686  *      interrupt controller has per-cpu registers.
2687  */
2688 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2689                           bool val)
2690 {
2691         struct irq_desc *desc;
2692         struct irq_data *data;
2693         struct irq_chip *chip;
2694         unsigned long flags;
2695         int err = -EINVAL;
2696
2697         desc = irq_get_desc_buslock(irq, &flags, 0);
2698         if (!desc)
2699                 return err;
2700
2701         data = irq_desc_get_irq_data(desc);
2702
2703         do {
2704                 chip = irq_data_get_irq_chip(data);
2705                 if (chip->irq_set_irqchip_state)
2706                         break;
2707 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2708                 data = data->parent_data;
2709 #else
2710                 data = NULL;
2711 #endif
2712         } while (data);
2713
2714         if (data)
2715                 err = chip->irq_set_irqchip_state(data, which, val);
2716
2717         irq_put_desc_busunlock(desc, flags);
2718         return err;
2719 }
2720 EXPORT_SYMBOL_GPL(irq_set_irqchip_state);