Merge branch 'x86-entry-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.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/slab.h>
17 #include <linux/sched.h>
18 #include <linux/sched/rt.h>
19 #include <linux/sched/task.h>
20 #include <uapi/linux/sched/types.h>
21 #include <linux/task_work.h>
22
23 #include "internals.h"
24
25 #ifdef CONFIG_IRQ_FORCED_THREADING
26 __read_mostly bool force_irqthreads;
27 EXPORT_SYMBOL_GPL(force_irqthreads);
28
29 static int __init setup_forced_irqthreads(char *arg)
30 {
31         force_irqthreads = true;
32         return 0;
33 }
34 early_param("threadirqs", setup_forced_irqthreads);
35 #endif
36
37 static void __synchronize_hardirq(struct irq_desc *desc)
38 {
39         bool inprogress;
40
41         do {
42                 unsigned long flags;
43
44                 /*
45                  * Wait until we're out of the critical section.  This might
46                  * give the wrong answer due to the lack of memory barriers.
47                  */
48                 while (irqd_irq_inprogress(&desc->irq_data))
49                         cpu_relax();
50
51                 /* Ok, that indicated we're done: double-check carefully. */
52                 raw_spin_lock_irqsave(&desc->lock, flags);
53                 inprogress = irqd_irq_inprogress(&desc->irq_data);
54                 raw_spin_unlock_irqrestore(&desc->lock, flags);
55
56                 /* Oops, that failed? */
57         } while (inprogress);
58 }
59
60 /**
61  *      synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
62  *      @irq: interrupt number to wait for
63  *
64  *      This function waits for any pending hard IRQ handlers for this
65  *      interrupt to complete before returning. If you use this
66  *      function while holding a resource the IRQ handler may need you
67  *      will deadlock. It does not take associated threaded handlers
68  *      into account.
69  *
70  *      Do not use this for shutdown scenarios where you must be sure
71  *      that all parts (hardirq and threaded handler) have completed.
72  *
73  *      Returns: false if a threaded handler is active.
74  *
75  *      This function may be called - with care - from IRQ context.
76  */
77 bool synchronize_hardirq(unsigned int irq)
78 {
79         struct irq_desc *desc = irq_to_desc(irq);
80
81         if (desc) {
82                 __synchronize_hardirq(desc);
83                 return !atomic_read(&desc->threads_active);
84         }
85
86         return true;
87 }
88 EXPORT_SYMBOL(synchronize_hardirq);
89
90 /**
91  *      synchronize_irq - wait for pending IRQ handlers (on other CPUs)
92  *      @irq: interrupt number to wait for
93  *
94  *      This function waits for any pending IRQ handlers for this interrupt
95  *      to complete before returning. If you use this function while
96  *      holding a resource the IRQ handler may need you will deadlock.
97  *
98  *      This function may be called - with care - from IRQ context.
99  */
100 void synchronize_irq(unsigned int irq)
101 {
102         struct irq_desc *desc = irq_to_desc(irq);
103
104         if (desc) {
105                 __synchronize_hardirq(desc);
106                 /*
107                  * We made sure that no hardirq handler is
108                  * running. Now verify that no threaded handlers are
109                  * active.
110                  */
111                 wait_event(desc->wait_for_threads,
112                            !atomic_read(&desc->threads_active));
113         }
114 }
115 EXPORT_SYMBOL(synchronize_irq);
116
117 #ifdef CONFIG_SMP
118 cpumask_var_t irq_default_affinity;
119
120 static bool __irq_can_set_affinity(struct irq_desc *desc)
121 {
122         if (!desc || !irqd_can_balance(&desc->irq_data) ||
123             !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
124                 return false;
125         return true;
126 }
127
128 /**
129  *      irq_can_set_affinity - Check if the affinity of a given irq can be set
130  *      @irq:           Interrupt to check
131  *
132  */
133 int irq_can_set_affinity(unsigned int irq)
134 {
135         return __irq_can_set_affinity(irq_to_desc(irq));
136 }
137
138 /**
139  * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
140  * @irq:        Interrupt to check
141  *
142  * Like irq_can_set_affinity() above, but additionally checks for the
143  * AFFINITY_MANAGED flag.
144  */
145 bool irq_can_set_affinity_usr(unsigned int irq)
146 {
147         struct irq_desc *desc = irq_to_desc(irq);
148
149         return __irq_can_set_affinity(desc) &&
150                 !irqd_affinity_is_managed(&desc->irq_data);
151 }
152
153 /**
154  *      irq_set_thread_affinity - Notify irq threads to adjust affinity
155  *      @desc:          irq descriptor which has affitnity changed
156  *
157  *      We just set IRQTF_AFFINITY and delegate the affinity setting
158  *      to the interrupt thread itself. We can not call
159  *      set_cpus_allowed_ptr() here as we hold desc->lock and this
160  *      code can be called from hard interrupt context.
161  */
162 void irq_set_thread_affinity(struct irq_desc *desc)
163 {
164         struct irqaction *action;
165
166         for_each_action_of_desc(desc, action)
167                 if (action->thread)
168                         set_bit(IRQTF_AFFINITY, &action->thread_flags);
169 }
170
171 static void irq_validate_effective_affinity(struct irq_data *data)
172 {
173 #ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
174         const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
175         struct irq_chip *chip = irq_data_get_irq_chip(data);
176
177         if (!cpumask_empty(m))
178                 return;
179         pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
180                      chip->name, data->irq);
181 #endif
182 }
183
184 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
185                         bool force)
186 {
187         struct irq_desc *desc = irq_data_to_desc(data);
188         struct irq_chip *chip = irq_data_get_irq_chip(data);
189         int ret;
190
191         if (!chip || !chip->irq_set_affinity)
192                 return -EINVAL;
193
194         ret = chip->irq_set_affinity(data, mask, force);
195         switch (ret) {
196         case IRQ_SET_MASK_OK:
197         case IRQ_SET_MASK_OK_DONE:
198                 cpumask_copy(desc->irq_common_data.affinity, mask);
199                 /* fall through */
200         case IRQ_SET_MASK_OK_NOCOPY:
201                 irq_validate_effective_affinity(data);
202                 irq_set_thread_affinity(desc);
203                 ret = 0;
204         }
205
206         return ret;
207 }
208
209 #ifdef CONFIG_GENERIC_PENDING_IRQ
210 static inline int irq_set_affinity_pending(struct irq_data *data,
211                                            const struct cpumask *dest)
212 {
213         struct irq_desc *desc = irq_data_to_desc(data);
214
215         irqd_set_move_pending(data);
216         irq_copy_pending(desc, dest);
217         return 0;
218 }
219 #else
220 static inline int irq_set_affinity_pending(struct irq_data *data,
221                                            const struct cpumask *dest)
222 {
223         return -EBUSY;
224 }
225 #endif
226
227 static int irq_try_set_affinity(struct irq_data *data,
228                                 const struct cpumask *dest, bool force)
229 {
230         int ret = irq_do_set_affinity(data, dest, force);
231
232         /*
233          * In case that the underlying vector management is busy and the
234          * architecture supports the generic pending mechanism then utilize
235          * this to avoid returning an error to user space.
236          */
237         if (ret == -EBUSY && !force)
238                 ret = irq_set_affinity_pending(data, dest);
239         return ret;
240 }
241
242 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
243                             bool force)
244 {
245         struct irq_chip *chip = irq_data_get_irq_chip(data);
246         struct irq_desc *desc = irq_data_to_desc(data);
247         int ret = 0;
248
249         if (!chip || !chip->irq_set_affinity)
250                 return -EINVAL;
251
252         if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
253                 ret = irq_try_set_affinity(data, mask, force);
254         } else {
255                 irqd_set_move_pending(data);
256                 irq_copy_pending(desc, mask);
257         }
258
259         if (desc->affinity_notify) {
260                 kref_get(&desc->affinity_notify->kref);
261                 schedule_work(&desc->affinity_notify->work);
262         }
263         irqd_set(data, IRQD_AFFINITY_SET);
264
265         return ret;
266 }
267
268 int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
269 {
270         struct irq_desc *desc = irq_to_desc(irq);
271         unsigned long flags;
272         int ret;
273
274         if (!desc)
275                 return -EINVAL;
276
277         raw_spin_lock_irqsave(&desc->lock, flags);
278         ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
279         raw_spin_unlock_irqrestore(&desc->lock, flags);
280         return ret;
281 }
282
283 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
284 {
285         unsigned long flags;
286         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
287
288         if (!desc)
289                 return -EINVAL;
290         desc->affinity_hint = m;
291         irq_put_desc_unlock(desc, flags);
292         /* set the initial affinity to prevent every interrupt being on CPU0 */
293         if (m)
294                 __irq_set_affinity(irq, m, false);
295         return 0;
296 }
297 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
298
299 static void irq_affinity_notify(struct work_struct *work)
300 {
301         struct irq_affinity_notify *notify =
302                 container_of(work, struct irq_affinity_notify, work);
303         struct irq_desc *desc = irq_to_desc(notify->irq);
304         cpumask_var_t cpumask;
305         unsigned long flags;
306
307         if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
308                 goto out;
309
310         raw_spin_lock_irqsave(&desc->lock, flags);
311         if (irq_move_pending(&desc->irq_data))
312                 irq_get_pending(cpumask, desc);
313         else
314                 cpumask_copy(cpumask, desc->irq_common_data.affinity);
315         raw_spin_unlock_irqrestore(&desc->lock, flags);
316
317         notify->notify(notify, cpumask);
318
319         free_cpumask_var(cpumask);
320 out:
321         kref_put(&notify->kref, notify->release);
322 }
323
324 /**
325  *      irq_set_affinity_notifier - control notification of IRQ affinity changes
326  *      @irq:           Interrupt for which to enable/disable notification
327  *      @notify:        Context for notification, or %NULL to disable
328  *                      notification.  Function pointers must be initialised;
329  *                      the other fields will be initialised by this function.
330  *
331  *      Must be called in process context.  Notification may only be enabled
332  *      after the IRQ is allocated and must be disabled before the IRQ is
333  *      freed using free_irq().
334  */
335 int
336 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
337 {
338         struct irq_desc *desc = irq_to_desc(irq);
339         struct irq_affinity_notify *old_notify;
340         unsigned long flags;
341
342         /* The release function is promised process context */
343         might_sleep();
344
345         if (!desc || desc->istate & IRQS_NMI)
346                 return -EINVAL;
347
348         /* Complete initialisation of *notify */
349         if (notify) {
350                 notify->irq = irq;
351                 kref_init(&notify->kref);
352                 INIT_WORK(&notify->work, irq_affinity_notify);
353         }
354
355         raw_spin_lock_irqsave(&desc->lock, flags);
356         old_notify = desc->affinity_notify;
357         desc->affinity_notify = notify;
358         raw_spin_unlock_irqrestore(&desc->lock, flags);
359
360         if (old_notify) {
361                 cancel_work_sync(&old_notify->work);
362                 kref_put(&old_notify->kref, old_notify->release);
363         }
364
365         return 0;
366 }
367 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
368
369 #ifndef CONFIG_AUTO_IRQ_AFFINITY
370 /*
371  * Generic version of the affinity autoselector.
372  */
373 int irq_setup_affinity(struct irq_desc *desc)
374 {
375         struct cpumask *set = irq_default_affinity;
376         int ret, node = irq_desc_get_node(desc);
377         static DEFINE_RAW_SPINLOCK(mask_lock);
378         static struct cpumask mask;
379
380         /* Excludes PER_CPU and NO_BALANCE interrupts */
381         if (!__irq_can_set_affinity(desc))
382                 return 0;
383
384         raw_spin_lock(&mask_lock);
385         /*
386          * Preserve the managed affinity setting and a userspace affinity
387          * setup, but make sure that one of the targets is online.
388          */
389         if (irqd_affinity_is_managed(&desc->irq_data) ||
390             irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
391                 if (cpumask_intersects(desc->irq_common_data.affinity,
392                                        cpu_online_mask))
393                         set = desc->irq_common_data.affinity;
394                 else
395                         irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
396         }
397
398         cpumask_and(&mask, cpu_online_mask, set);
399         if (cpumask_empty(&mask))
400                 cpumask_copy(&mask, cpu_online_mask);
401
402         if (node != NUMA_NO_NODE) {
403                 const struct cpumask *nodemask = cpumask_of_node(node);
404
405                 /* make sure at least one of the cpus in nodemask is online */
406                 if (cpumask_intersects(&mask, nodemask))
407                         cpumask_and(&mask, &mask, nodemask);
408         }
409         ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
410         raw_spin_unlock(&mask_lock);
411         return ret;
412 }
413 #else
414 /* Wrapper for ALPHA specific affinity selector magic */
415 int irq_setup_affinity(struct irq_desc *desc)
416 {
417         return irq_select_affinity(irq_desc_get_irq(desc));
418 }
419 #endif
420
421 /*
422  * Called when a bogus affinity is set via /proc/irq
423  */
424 int irq_select_affinity_usr(unsigned int irq)
425 {
426         struct irq_desc *desc = irq_to_desc(irq);
427         unsigned long flags;
428         int ret;
429
430         raw_spin_lock_irqsave(&desc->lock, flags);
431         ret = irq_setup_affinity(desc);
432         raw_spin_unlock_irqrestore(&desc->lock, flags);
433         return ret;
434 }
435 #endif
436
437 /**
438  *      irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
439  *      @irq: interrupt number to set affinity
440  *      @vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
441  *                  specific data for percpu_devid interrupts
442  *
443  *      This function uses the vCPU specific data to set the vCPU
444  *      affinity for an irq. The vCPU specific data is passed from
445  *      outside, such as KVM. One example code path is as below:
446  *      KVM -> IOMMU -> irq_set_vcpu_affinity().
447  */
448 int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
449 {
450         unsigned long flags;
451         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
452         struct irq_data *data;
453         struct irq_chip *chip;
454         int ret = -ENOSYS;
455
456         if (!desc)
457                 return -EINVAL;
458
459         data = irq_desc_get_irq_data(desc);
460         do {
461                 chip = irq_data_get_irq_chip(data);
462                 if (chip && chip->irq_set_vcpu_affinity)
463                         break;
464 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
465                 data = data->parent_data;
466 #else
467                 data = NULL;
468 #endif
469         } while (data);
470
471         if (data)
472                 ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
473         irq_put_desc_unlock(desc, flags);
474
475         return ret;
476 }
477 EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
478
479 void __disable_irq(struct irq_desc *desc)
480 {
481         if (!desc->depth++)
482                 irq_disable(desc);
483 }
484
485 static int __disable_irq_nosync(unsigned int irq)
486 {
487         unsigned long flags;
488         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
489
490         if (!desc)
491                 return -EINVAL;
492         __disable_irq(desc);
493         irq_put_desc_busunlock(desc, flags);
494         return 0;
495 }
496
497 /**
498  *      disable_irq_nosync - disable an irq without waiting
499  *      @irq: Interrupt to disable
500  *
501  *      Disable the selected interrupt line.  Disables and Enables are
502  *      nested.
503  *      Unlike disable_irq(), this function does not ensure existing
504  *      instances of the IRQ handler have completed before returning.
505  *
506  *      This function may be called from IRQ context.
507  */
508 void disable_irq_nosync(unsigned int irq)
509 {
510         __disable_irq_nosync(irq);
511 }
512 EXPORT_SYMBOL(disable_irq_nosync);
513
514 /**
515  *      disable_irq - disable an irq and wait for completion
516  *      @irq: Interrupt to disable
517  *
518  *      Disable the selected interrupt line.  Enables and Disables are
519  *      nested.
520  *      This function waits for any pending IRQ handlers for this interrupt
521  *      to complete before returning. If you use this function while
522  *      holding a resource the IRQ handler may need you will deadlock.
523  *
524  *      This function may be called - with care - from IRQ context.
525  */
526 void disable_irq(unsigned int irq)
527 {
528         if (!__disable_irq_nosync(irq))
529                 synchronize_irq(irq);
530 }
531 EXPORT_SYMBOL(disable_irq);
532
533 /**
534  *      disable_hardirq - disables an irq and waits for hardirq completion
535  *      @irq: Interrupt to disable
536  *
537  *      Disable the selected interrupt line.  Enables and Disables are
538  *      nested.
539  *      This function waits for any pending hard IRQ handlers for this
540  *      interrupt to complete before returning. If you use this function while
541  *      holding a resource the hard IRQ handler may need you will deadlock.
542  *
543  *      When used to optimistically disable an interrupt from atomic context
544  *      the return value must be checked.
545  *
546  *      Returns: false if a threaded handler is active.
547  *
548  *      This function may be called - with care - from IRQ context.
549  */
550 bool disable_hardirq(unsigned int irq)
551 {
552         if (!__disable_irq_nosync(irq))
553                 return synchronize_hardirq(irq);
554
555         return false;
556 }
557 EXPORT_SYMBOL_GPL(disable_hardirq);
558
559 /**
560  *      disable_nmi_nosync - disable an nmi without waiting
561  *      @irq: Interrupt to disable
562  *
563  *      Disable the selected interrupt line. Disables and enables are
564  *      nested.
565  *      The interrupt to disable must have been requested through request_nmi.
566  *      Unlike disable_nmi(), this function does not ensure existing
567  *      instances of the IRQ handler have completed before returning.
568  */
569 void disable_nmi_nosync(unsigned int irq)
570 {
571         disable_irq_nosync(irq);
572 }
573
574 void __enable_irq(struct irq_desc *desc)
575 {
576         switch (desc->depth) {
577         case 0:
578  err_out:
579                 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
580                      irq_desc_get_irq(desc));
581                 break;
582         case 1: {
583                 if (desc->istate & IRQS_SUSPENDED)
584                         goto err_out;
585                 /* Prevent probing on this irq: */
586                 irq_settings_set_noprobe(desc);
587                 /*
588                  * Call irq_startup() not irq_enable() here because the
589                  * interrupt might be marked NOAUTOEN. So irq_startup()
590                  * needs to be invoked when it gets enabled the first
591                  * time. If it was already started up, then irq_startup()
592                  * will invoke irq_enable() under the hood.
593                  */
594                 irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
595                 break;
596         }
597         default:
598                 desc->depth--;
599         }
600 }
601
602 /**
603  *      enable_irq - enable handling of an irq
604  *      @irq: Interrupt to enable
605  *
606  *      Undoes the effect of one call to disable_irq().  If this
607  *      matches the last disable, processing of interrupts on this
608  *      IRQ line is re-enabled.
609  *
610  *      This function may be called from IRQ context only when
611  *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
612  */
613 void enable_irq(unsigned int irq)
614 {
615         unsigned long flags;
616         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
617
618         if (!desc)
619                 return;
620         if (WARN(!desc->irq_data.chip,
621                  KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
622                 goto out;
623
624         __enable_irq(desc);
625 out:
626         irq_put_desc_busunlock(desc, flags);
627 }
628 EXPORT_SYMBOL(enable_irq);
629
630 /**
631  *      enable_nmi - enable handling of an nmi
632  *      @irq: Interrupt to enable
633  *
634  *      The interrupt to enable must have been requested through request_nmi.
635  *      Undoes the effect of one call to disable_nmi(). If this
636  *      matches the last disable, processing of interrupts on this
637  *      IRQ line is re-enabled.
638  */
639 void enable_nmi(unsigned int irq)
640 {
641         enable_irq(irq);
642 }
643
644 static int set_irq_wake_real(unsigned int irq, unsigned int on)
645 {
646         struct irq_desc *desc = irq_to_desc(irq);
647         int ret = -ENXIO;
648
649         if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
650                 return 0;
651
652         if (desc->irq_data.chip->irq_set_wake)
653                 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
654
655         return ret;
656 }
657
658 /**
659  *      irq_set_irq_wake - control irq power management wakeup
660  *      @irq:   interrupt to control
661  *      @on:    enable/disable power management wakeup
662  *
663  *      Enable/disable power management wakeup mode, which is
664  *      disabled by default.  Enables and disables must match,
665  *      just as they match for non-wakeup mode support.
666  *
667  *      Wakeup mode lets this IRQ wake the system from sleep
668  *      states like "suspend to RAM".
669  */
670 int irq_set_irq_wake(unsigned int irq, unsigned int on)
671 {
672         unsigned long flags;
673         struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
674         int ret = 0;
675
676         if (!desc)
677                 return -EINVAL;
678
679         /* Don't use NMIs as wake up interrupts please */
680         if (desc->istate & IRQS_NMI) {
681                 ret = -EINVAL;
682                 goto out_unlock;
683         }
684
685         /* wakeup-capable irqs can be shared between drivers that
686          * don't need to have the same sleep mode behaviors.
687          */
688         if (on) {
689                 if (desc->wake_depth++ == 0) {
690                         ret = set_irq_wake_real(irq, on);
691                         if (ret)
692                                 desc->wake_depth = 0;
693                         else
694                                 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
695                 }
696         } else {
697                 if (desc->wake_depth == 0) {
698                         WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
699                 } else if (--desc->wake_depth == 0) {
700                         ret = set_irq_wake_real(irq, on);
701                         if (ret)
702                                 desc->wake_depth = 1;
703                         else
704                                 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
705                 }
706         }
707
708 out_unlock:
709         irq_put_desc_busunlock(desc, flags);
710         return ret;
711 }
712 EXPORT_SYMBOL(irq_set_irq_wake);
713
714 /*
715  * Internal function that tells the architecture code whether a
716  * particular irq has been exclusively allocated or is available
717  * for driver use.
718  */
719 int can_request_irq(unsigned int irq, unsigned long irqflags)
720 {
721         unsigned long flags;
722         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
723         int canrequest = 0;
724
725         if (!desc)
726                 return 0;
727
728         if (irq_settings_can_request(desc)) {
729                 if (!desc->action ||
730                     irqflags & desc->action->flags & IRQF_SHARED)
731                         canrequest = 1;
732         }
733         irq_put_desc_unlock(desc, flags);
734         return canrequest;
735 }
736
737 int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
738 {
739         struct irq_chip *chip = desc->irq_data.chip;
740         int ret, unmask = 0;
741
742         if (!chip || !chip->irq_set_type) {
743                 /*
744                  * IRQF_TRIGGER_* but the PIC does not support multiple
745                  * flow-types?
746                  */
747                 pr_debug("No set_type function for IRQ %d (%s)\n",
748                          irq_desc_get_irq(desc),
749                          chip ? (chip->name ? : "unknown") : "unknown");
750                 return 0;
751         }
752
753         if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
754                 if (!irqd_irq_masked(&desc->irq_data))
755                         mask_irq(desc);
756                 if (!irqd_irq_disabled(&desc->irq_data))
757                         unmask = 1;
758         }
759
760         /* Mask all flags except trigger mode */
761         flags &= IRQ_TYPE_SENSE_MASK;
762         ret = chip->irq_set_type(&desc->irq_data, flags);
763
764         switch (ret) {
765         case IRQ_SET_MASK_OK:
766         case IRQ_SET_MASK_OK_DONE:
767                 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
768                 irqd_set(&desc->irq_data, flags);
769                 /* fall through */
770
771         case IRQ_SET_MASK_OK_NOCOPY:
772                 flags = irqd_get_trigger_type(&desc->irq_data);
773                 irq_settings_set_trigger_mask(desc, flags);
774                 irqd_clear(&desc->irq_data, IRQD_LEVEL);
775                 irq_settings_clr_level(desc);
776                 if (flags & IRQ_TYPE_LEVEL_MASK) {
777                         irq_settings_set_level(desc);
778                         irqd_set(&desc->irq_data, IRQD_LEVEL);
779                 }
780
781                 ret = 0;
782                 break;
783         default:
784                 pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
785                        flags, irq_desc_get_irq(desc), chip->irq_set_type);
786         }
787         if (unmask)
788                 unmask_irq(desc);
789         return ret;
790 }
791
792 #ifdef CONFIG_HARDIRQS_SW_RESEND
793 int irq_set_parent(int irq, int parent_irq)
794 {
795         unsigned long flags;
796         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
797
798         if (!desc)
799                 return -EINVAL;
800
801         desc->parent_irq = parent_irq;
802
803         irq_put_desc_unlock(desc, flags);
804         return 0;
805 }
806 EXPORT_SYMBOL_GPL(irq_set_parent);
807 #endif
808
809 /*
810  * Default primary interrupt handler for threaded interrupts. Is
811  * assigned as primary handler when request_threaded_irq is called
812  * with handler == NULL. Useful for oneshot interrupts.
813  */
814 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
815 {
816         return IRQ_WAKE_THREAD;
817 }
818
819 /*
820  * Primary handler for nested threaded interrupts. Should never be
821  * called.
822  */
823 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
824 {
825         WARN(1, "Primary handler called for nested irq %d\n", irq);
826         return IRQ_NONE;
827 }
828
829 static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
830 {
831         WARN(1, "Secondary action handler called for irq %d\n", irq);
832         return IRQ_NONE;
833 }
834
835 static int irq_wait_for_interrupt(struct irqaction *action)
836 {
837         for (;;) {
838                 set_current_state(TASK_INTERRUPTIBLE);
839
840                 if (kthread_should_stop()) {
841                         /* may need to run one last time */
842                         if (test_and_clear_bit(IRQTF_RUNTHREAD,
843                                                &action->thread_flags)) {
844                                 __set_current_state(TASK_RUNNING);
845                                 return 0;
846                         }
847                         __set_current_state(TASK_RUNNING);
848                         return -1;
849                 }
850
851                 if (test_and_clear_bit(IRQTF_RUNTHREAD,
852                                        &action->thread_flags)) {
853                         __set_current_state(TASK_RUNNING);
854                         return 0;
855                 }
856                 schedule();
857         }
858 }
859
860 /*
861  * Oneshot interrupts keep the irq line masked until the threaded
862  * handler finished. unmask if the interrupt has not been disabled and
863  * is marked MASKED.
864  */
865 static void irq_finalize_oneshot(struct irq_desc *desc,
866                                  struct irqaction *action)
867 {
868         if (!(desc->istate & IRQS_ONESHOT) ||
869             action->handler == irq_forced_secondary_handler)
870                 return;
871 again:
872         chip_bus_lock(desc);
873         raw_spin_lock_irq(&desc->lock);
874
875         /*
876          * Implausible though it may be we need to protect us against
877          * the following scenario:
878          *
879          * The thread is faster done than the hard interrupt handler
880          * on the other CPU. If we unmask the irq line then the
881          * interrupt can come in again and masks the line, leaves due
882          * to IRQS_INPROGRESS and the irq line is masked forever.
883          *
884          * This also serializes the state of shared oneshot handlers
885          * versus "desc->threads_onehsot |= action->thread_mask;" in
886          * irq_wake_thread(). See the comment there which explains the
887          * serialization.
888          */
889         if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
890                 raw_spin_unlock_irq(&desc->lock);
891                 chip_bus_sync_unlock(desc);
892                 cpu_relax();
893                 goto again;
894         }
895
896         /*
897          * Now check again, whether the thread should run. Otherwise
898          * we would clear the threads_oneshot bit of this thread which
899          * was just set.
900          */
901         if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
902                 goto out_unlock;
903
904         desc->threads_oneshot &= ~action->thread_mask;
905
906         if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
907             irqd_irq_masked(&desc->irq_data))
908                 unmask_threaded_irq(desc);
909
910 out_unlock:
911         raw_spin_unlock_irq(&desc->lock);
912         chip_bus_sync_unlock(desc);
913 }
914
915 #ifdef CONFIG_SMP
916 /*
917  * Check whether we need to change the affinity of the interrupt thread.
918  */
919 static void
920 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
921 {
922         cpumask_var_t mask;
923         bool valid = true;
924
925         if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
926                 return;
927
928         /*
929          * In case we are out of memory we set IRQTF_AFFINITY again and
930          * try again next time
931          */
932         if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
933                 set_bit(IRQTF_AFFINITY, &action->thread_flags);
934                 return;
935         }
936
937         raw_spin_lock_irq(&desc->lock);
938         /*
939          * This code is triggered unconditionally. Check the affinity
940          * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
941          */
942         if (cpumask_available(desc->irq_common_data.affinity)) {
943                 const struct cpumask *m;
944
945                 m = irq_data_get_effective_affinity_mask(&desc->irq_data);
946                 cpumask_copy(mask, m);
947         } else {
948                 valid = false;
949         }
950         raw_spin_unlock_irq(&desc->lock);
951
952         if (valid)
953                 set_cpus_allowed_ptr(current, mask);
954         free_cpumask_var(mask);
955 }
956 #else
957 static inline void
958 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
959 #endif
960
961 /*
962  * Interrupts which are not explicitly requested as threaded
963  * interrupts rely on the implicit bh/preempt disable of the hard irq
964  * context. So we need to disable bh here to avoid deadlocks and other
965  * side effects.
966  */
967 static irqreturn_t
968 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
969 {
970         irqreturn_t ret;
971
972         local_bh_disable();
973         ret = action->thread_fn(action->irq, action->dev_id);
974         if (ret == IRQ_HANDLED)
975                 atomic_inc(&desc->threads_handled);
976
977         irq_finalize_oneshot(desc, action);
978         local_bh_enable();
979         return ret;
980 }
981
982 /*
983  * Interrupts explicitly requested as threaded interrupts want to be
984  * preemtible - many of them need to sleep and wait for slow busses to
985  * complete.
986  */
987 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
988                 struct irqaction *action)
989 {
990         irqreturn_t ret;
991
992         ret = action->thread_fn(action->irq, action->dev_id);
993         if (ret == IRQ_HANDLED)
994                 atomic_inc(&desc->threads_handled);
995
996         irq_finalize_oneshot(desc, action);
997         return ret;
998 }
999
1000 static void wake_threads_waitq(struct irq_desc *desc)
1001 {
1002         if (atomic_dec_and_test(&desc->threads_active))
1003                 wake_up(&desc->wait_for_threads);
1004 }
1005
1006 static void irq_thread_dtor(struct callback_head *unused)
1007 {
1008         struct task_struct *tsk = current;
1009         struct irq_desc *desc;
1010         struct irqaction *action;
1011
1012         if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1013                 return;
1014
1015         action = kthread_data(tsk);
1016
1017         pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1018                tsk->comm, tsk->pid, action->irq);
1019
1020
1021         desc = irq_to_desc(action->irq);
1022         /*
1023          * If IRQTF_RUNTHREAD is set, we need to decrement
1024          * desc->threads_active and wake possible waiters.
1025          */
1026         if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1027                 wake_threads_waitq(desc);
1028
1029         /* Prevent a stale desc->threads_oneshot */
1030         irq_finalize_oneshot(desc, action);
1031 }
1032
1033 static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1034 {
1035         struct irqaction *secondary = action->secondary;
1036
1037         if (WARN_ON_ONCE(!secondary))
1038                 return;
1039
1040         raw_spin_lock_irq(&desc->lock);
1041         __irq_wake_thread(desc, secondary);
1042         raw_spin_unlock_irq(&desc->lock);
1043 }
1044
1045 /*
1046  * Interrupt handler thread
1047  */
1048 static int irq_thread(void *data)
1049 {
1050         struct callback_head on_exit_work;
1051         struct irqaction *action = data;
1052         struct irq_desc *desc = irq_to_desc(action->irq);
1053         irqreturn_t (*handler_fn)(struct irq_desc *desc,
1054                         struct irqaction *action);
1055
1056         if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
1057                                         &action->thread_flags))
1058                 handler_fn = irq_forced_thread_fn;
1059         else
1060                 handler_fn = irq_thread_fn;
1061
1062         init_task_work(&on_exit_work, irq_thread_dtor);
1063         task_work_add(current, &on_exit_work, false);
1064
1065         irq_thread_check_affinity(desc, action);
1066
1067         while (!irq_wait_for_interrupt(action)) {
1068                 irqreturn_t action_ret;
1069
1070                 irq_thread_check_affinity(desc, action);
1071
1072                 action_ret = handler_fn(desc, action);
1073                 if (action_ret == IRQ_WAKE_THREAD)
1074                         irq_wake_secondary(desc, action);
1075
1076                 wake_threads_waitq(desc);
1077         }
1078
1079         /*
1080          * This is the regular exit path. __free_irq() is stopping the
1081          * thread via kthread_stop() after calling
1082          * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1083          * oneshot mask bit can be set.
1084          */
1085         task_work_cancel(current, irq_thread_dtor);
1086         return 0;
1087 }
1088
1089 /**
1090  *      irq_wake_thread - wake the irq thread for the action identified by dev_id
1091  *      @irq:           Interrupt line
1092  *      @dev_id:        Device identity for which the thread should be woken
1093  *
1094  */
1095 void irq_wake_thread(unsigned int irq, void *dev_id)
1096 {
1097         struct irq_desc *desc = irq_to_desc(irq);
1098         struct irqaction *action;
1099         unsigned long flags;
1100
1101         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1102                 return;
1103
1104         raw_spin_lock_irqsave(&desc->lock, flags);
1105         for_each_action_of_desc(desc, action) {
1106                 if (action->dev_id == dev_id) {
1107                         if (action->thread)
1108                                 __irq_wake_thread(desc, action);
1109                         break;
1110                 }
1111         }
1112         raw_spin_unlock_irqrestore(&desc->lock, flags);
1113 }
1114 EXPORT_SYMBOL_GPL(irq_wake_thread);
1115
1116 static int irq_setup_forced_threading(struct irqaction *new)
1117 {
1118         if (!force_irqthreads)
1119                 return 0;
1120         if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1121                 return 0;
1122
1123         /*
1124          * No further action required for interrupts which are requested as
1125          * threaded interrupts already
1126          */
1127         if (new->handler == irq_default_primary_handler)
1128                 return 0;
1129
1130         new->flags |= IRQF_ONESHOT;
1131
1132         /*
1133          * Handle the case where we have a real primary handler and a
1134          * thread handler. We force thread them as well by creating a
1135          * secondary action.
1136          */
1137         if (new->handler && new->thread_fn) {
1138                 /* Allocate the secondary action */
1139                 new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1140                 if (!new->secondary)
1141                         return -ENOMEM;
1142                 new->secondary->handler = irq_forced_secondary_handler;
1143                 new->secondary->thread_fn = new->thread_fn;
1144                 new->secondary->dev_id = new->dev_id;
1145                 new->secondary->irq = new->irq;
1146                 new->secondary->name = new->name;
1147         }
1148         /* Deal with the primary handler */
1149         set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1150         new->thread_fn = new->handler;
1151         new->handler = irq_default_primary_handler;
1152         return 0;
1153 }
1154
1155 static int irq_request_resources(struct irq_desc *desc)
1156 {
1157         struct irq_data *d = &desc->irq_data;
1158         struct irq_chip *c = d->chip;
1159
1160         return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1161 }
1162
1163 static void irq_release_resources(struct irq_desc *desc)
1164 {
1165         struct irq_data *d = &desc->irq_data;
1166         struct irq_chip *c = d->chip;
1167
1168         if (c->irq_release_resources)
1169                 c->irq_release_resources(d);
1170 }
1171
1172 static bool irq_supports_nmi(struct irq_desc *desc)
1173 {
1174         struct irq_data *d = irq_desc_get_irq_data(desc);
1175
1176 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1177         /* Only IRQs directly managed by the root irqchip can be set as NMI */
1178         if (d->parent_data)
1179                 return false;
1180 #endif
1181         /* Don't support NMIs for chips behind a slow bus */
1182         if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1183                 return false;
1184
1185         return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1186 }
1187
1188 static int irq_nmi_setup(struct irq_desc *desc)
1189 {
1190         struct irq_data *d = irq_desc_get_irq_data(desc);
1191         struct irq_chip *c = d->chip;
1192
1193         return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1194 }
1195
1196 static void irq_nmi_teardown(struct irq_desc *desc)
1197 {
1198         struct irq_data *d = irq_desc_get_irq_data(desc);
1199         struct irq_chip *c = d->chip;
1200
1201         if (c->irq_nmi_teardown)
1202                 c->irq_nmi_teardown(d);
1203 }
1204
1205 static int
1206 setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1207 {
1208         struct task_struct *t;
1209         struct sched_param param = {
1210                 .sched_priority = MAX_USER_RT_PRIO/2,
1211         };
1212
1213         if (!secondary) {
1214                 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1215                                    new->name);
1216         } else {
1217                 t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1218                                    new->name);
1219                 param.sched_priority -= 1;
1220         }
1221
1222         if (IS_ERR(t))
1223                 return PTR_ERR(t);
1224
1225         sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1226
1227         /*
1228          * We keep the reference to the task struct even if
1229          * the thread dies to avoid that the interrupt code
1230          * references an already freed task_struct.
1231          */
1232         get_task_struct(t);
1233         new->thread = t;
1234         /*
1235          * Tell the thread to set its affinity. This is
1236          * important for shared interrupt handlers as we do
1237          * not invoke setup_affinity() for the secondary
1238          * handlers as everything is already set up. Even for
1239          * interrupts marked with IRQF_NO_BALANCE this is
1240          * correct as we want the thread to move to the cpu(s)
1241          * on which the requesting code placed the interrupt.
1242          */
1243         set_bit(IRQTF_AFFINITY, &new->thread_flags);
1244         return 0;
1245 }
1246
1247 /*
1248  * Internal function to register an irqaction - typically used to
1249  * allocate special interrupts that are part of the architecture.
1250  *
1251  * Locking rules:
1252  *
1253  * desc->request_mutex  Provides serialization against a concurrent free_irq()
1254  *   chip_bus_lock      Provides serialization for slow bus operations
1255  *     desc->lock       Provides serialization against hard interrupts
1256  *
1257  * chip_bus_lock and desc->lock are sufficient for all other management and
1258  * interrupt related functions. desc->request_mutex solely serializes
1259  * request/free_irq().
1260  */
1261 static int
1262 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1263 {
1264         struct irqaction *old, **old_ptr;
1265         unsigned long flags, thread_mask = 0;
1266         int ret, nested, shared = 0;
1267
1268         if (!desc)
1269                 return -EINVAL;
1270
1271         if (desc->irq_data.chip == &no_irq_chip)
1272                 return -ENOSYS;
1273         if (!try_module_get(desc->owner))
1274                 return -ENODEV;
1275
1276         new->irq = irq;
1277
1278         /*
1279          * If the trigger type is not specified by the caller,
1280          * then use the default for this interrupt.
1281          */
1282         if (!(new->flags & IRQF_TRIGGER_MASK))
1283                 new->flags |= irqd_get_trigger_type(&desc->irq_data);
1284
1285         /*
1286          * Check whether the interrupt nests into another interrupt
1287          * thread.
1288          */
1289         nested = irq_settings_is_nested_thread(desc);
1290         if (nested) {
1291                 if (!new->thread_fn) {
1292                         ret = -EINVAL;
1293                         goto out_mput;
1294                 }
1295                 /*
1296                  * Replace the primary handler which was provided from
1297                  * the driver for non nested interrupt handling by the
1298                  * dummy function which warns when called.
1299                  */
1300                 new->handler = irq_nested_primary_handler;
1301         } else {
1302                 if (irq_settings_can_thread(desc)) {
1303                         ret = irq_setup_forced_threading(new);
1304                         if (ret)
1305                                 goto out_mput;
1306                 }
1307         }
1308
1309         /*
1310          * Create a handler thread when a thread function is supplied
1311          * and the interrupt does not nest into another interrupt
1312          * thread.
1313          */
1314         if (new->thread_fn && !nested) {
1315                 ret = setup_irq_thread(new, irq, false);
1316                 if (ret)
1317                         goto out_mput;
1318                 if (new->secondary) {
1319                         ret = setup_irq_thread(new->secondary, irq, true);
1320                         if (ret)
1321                                 goto out_thread;
1322                 }
1323         }
1324
1325         /*
1326          * Drivers are often written to work w/o knowledge about the
1327          * underlying irq chip implementation, so a request for a
1328          * threaded irq without a primary hard irq context handler
1329          * requires the ONESHOT flag to be set. Some irq chips like
1330          * MSI based interrupts are per se one shot safe. Check the
1331          * chip flags, so we can avoid the unmask dance at the end of
1332          * the threaded handler for those.
1333          */
1334         if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1335                 new->flags &= ~IRQF_ONESHOT;
1336
1337         /*
1338          * Protects against a concurrent __free_irq() call which might wait
1339          * for synchronize_hardirq() to complete without holding the optional
1340          * chip bus lock and desc->lock. Also protects against handing out
1341          * a recycled oneshot thread_mask bit while it's still in use by
1342          * its previous owner.
1343          */
1344         mutex_lock(&desc->request_mutex);
1345
1346         /*
1347          * Acquire bus lock as the irq_request_resources() callback below
1348          * might rely on the serialization or the magic power management
1349          * functions which are abusing the irq_bus_lock() callback,
1350          */
1351         chip_bus_lock(desc);
1352
1353         /* First installed action requests resources. */
1354         if (!desc->action) {
1355                 ret = irq_request_resources(desc);
1356                 if (ret) {
1357                         pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1358                                new->name, irq, desc->irq_data.chip->name);
1359                         goto out_bus_unlock;
1360                 }
1361         }
1362
1363         /*
1364          * The following block of code has to be executed atomically
1365          * protected against a concurrent interrupt and any of the other
1366          * management calls which are not serialized via
1367          * desc->request_mutex or the optional bus lock.
1368          */
1369         raw_spin_lock_irqsave(&desc->lock, flags);
1370         old_ptr = &desc->action;
1371         old = *old_ptr;
1372         if (old) {
1373                 /*
1374                  * Can't share interrupts unless both agree to and are
1375                  * the same type (level, edge, polarity). So both flag
1376                  * fields must have IRQF_SHARED set and the bits which
1377                  * set the trigger type must match. Also all must
1378                  * agree on ONESHOT.
1379                  * Interrupt lines used for NMIs cannot be shared.
1380                  */
1381                 unsigned int oldtype;
1382
1383                 if (desc->istate & IRQS_NMI) {
1384                         pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1385                                 new->name, irq, desc->irq_data.chip->name);
1386                         ret = -EINVAL;
1387                         goto out_unlock;
1388                 }
1389
1390                 /*
1391                  * If nobody did set the configuration before, inherit
1392                  * the one provided by the requester.
1393                  */
1394                 if (irqd_trigger_type_was_set(&desc->irq_data)) {
1395                         oldtype = irqd_get_trigger_type(&desc->irq_data);
1396                 } else {
1397                         oldtype = new->flags & IRQF_TRIGGER_MASK;
1398                         irqd_set_trigger_type(&desc->irq_data, oldtype);
1399                 }
1400
1401                 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1402                     (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1403                     ((old->flags ^ new->flags) & IRQF_ONESHOT))
1404                         goto mismatch;
1405
1406                 /* All handlers must agree on per-cpuness */
1407                 if ((old->flags & IRQF_PERCPU) !=
1408                     (new->flags & IRQF_PERCPU))
1409                         goto mismatch;
1410
1411                 /* add new interrupt at end of irq queue */
1412                 do {
1413                         /*
1414                          * Or all existing action->thread_mask bits,
1415                          * so we can find the next zero bit for this
1416                          * new action.
1417                          */
1418                         thread_mask |= old->thread_mask;
1419                         old_ptr = &old->next;
1420                         old = *old_ptr;
1421                 } while (old);
1422                 shared = 1;
1423         }
1424
1425         /*
1426          * Setup the thread mask for this irqaction for ONESHOT. For
1427          * !ONESHOT irqs the thread mask is 0 so we can avoid a
1428          * conditional in irq_wake_thread().
1429          */
1430         if (new->flags & IRQF_ONESHOT) {
1431                 /*
1432                  * Unlikely to have 32 resp 64 irqs sharing one line,
1433                  * but who knows.
1434                  */
1435                 if (thread_mask == ~0UL) {
1436                         ret = -EBUSY;
1437                         goto out_unlock;
1438                 }
1439                 /*
1440                  * The thread_mask for the action is or'ed to
1441                  * desc->thread_active to indicate that the
1442                  * IRQF_ONESHOT thread handler has been woken, but not
1443                  * yet finished. The bit is cleared when a thread
1444                  * completes. When all threads of a shared interrupt
1445                  * line have completed desc->threads_active becomes
1446                  * zero and the interrupt line is unmasked. See
1447                  * handle.c:irq_wake_thread() for further information.
1448                  *
1449                  * If no thread is woken by primary (hard irq context)
1450                  * interrupt handlers, then desc->threads_active is
1451                  * also checked for zero to unmask the irq line in the
1452                  * affected hard irq flow handlers
1453                  * (handle_[fasteoi|level]_irq).
1454                  *
1455                  * The new action gets the first zero bit of
1456                  * thread_mask assigned. See the loop above which or's
1457                  * all existing action->thread_mask bits.
1458                  */
1459                 new->thread_mask = 1UL << ffz(thread_mask);
1460
1461         } else if (new->handler == irq_default_primary_handler &&
1462                    !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1463                 /*
1464                  * The interrupt was requested with handler = NULL, so
1465                  * we use the default primary handler for it. But it
1466                  * does not have the oneshot flag set. In combination
1467                  * with level interrupts this is deadly, because the
1468                  * default primary handler just wakes the thread, then
1469                  * the irq lines is reenabled, but the device still
1470                  * has the level irq asserted. Rinse and repeat....
1471                  *
1472                  * While this works for edge type interrupts, we play
1473                  * it safe and reject unconditionally because we can't
1474                  * say for sure which type this interrupt really
1475                  * has. The type flags are unreliable as the
1476                  * underlying chip implementation can override them.
1477                  */
1478                 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1479                        irq);
1480                 ret = -EINVAL;
1481                 goto out_unlock;
1482         }
1483
1484         if (!shared) {
1485                 init_waitqueue_head(&desc->wait_for_threads);
1486
1487                 /* Setup the type (level, edge polarity) if configured: */
1488                 if (new->flags & IRQF_TRIGGER_MASK) {
1489                         ret = __irq_set_trigger(desc,
1490                                                 new->flags & IRQF_TRIGGER_MASK);
1491
1492                         if (ret)
1493                                 goto out_unlock;
1494                 }
1495
1496                 /*
1497                  * Activate the interrupt. That activation must happen
1498                  * independently of IRQ_NOAUTOEN. request_irq() can fail
1499                  * and the callers are supposed to handle
1500                  * that. enable_irq() of an interrupt requested with
1501                  * IRQ_NOAUTOEN is not supposed to fail. The activation
1502                  * keeps it in shutdown mode, it merily associates
1503                  * resources if necessary and if that's not possible it
1504                  * fails. Interrupts which are in managed shutdown mode
1505                  * will simply ignore that activation request.
1506                  */
1507                 ret = irq_activate(desc);
1508                 if (ret)
1509                         goto out_unlock;
1510
1511                 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1512                                   IRQS_ONESHOT | IRQS_WAITING);
1513                 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1514
1515                 if (new->flags & IRQF_PERCPU) {
1516                         irqd_set(&desc->irq_data, IRQD_PER_CPU);
1517                         irq_settings_set_per_cpu(desc);
1518                 }
1519
1520                 if (new->flags & IRQF_ONESHOT)
1521                         desc->istate |= IRQS_ONESHOT;
1522
1523                 /* Exclude IRQ from balancing if requested */
1524                 if (new->flags & IRQF_NOBALANCING) {
1525                         irq_settings_set_no_balancing(desc);
1526                         irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1527                 }
1528
1529                 if (irq_settings_can_autoenable(desc)) {
1530                         irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1531                 } else {
1532                         /*
1533                          * Shared interrupts do not go well with disabling
1534                          * auto enable. The sharing interrupt might request
1535                          * it while it's still disabled and then wait for
1536                          * interrupts forever.
1537                          */
1538                         WARN_ON_ONCE(new->flags & IRQF_SHARED);
1539                         /* Undo nested disables: */
1540                         desc->depth = 1;
1541                 }
1542
1543         } else if (new->flags & IRQF_TRIGGER_MASK) {
1544                 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1545                 unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1546
1547                 if (nmsk != omsk)
1548                         /* hope the handler works with current  trigger mode */
1549                         pr_warn("irq %d uses trigger mode %u; requested %u\n",
1550                                 irq, omsk, nmsk);
1551         }
1552
1553         *old_ptr = new;
1554
1555         irq_pm_install_action(desc, new);
1556
1557         /* Reset broken irq detection when installing new handler */
1558         desc->irq_count = 0;
1559         desc->irqs_unhandled = 0;
1560
1561         /*
1562          * Check whether we disabled the irq via the spurious handler
1563          * before. Reenable it and give it another chance.
1564          */
1565         if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1566                 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1567                 __enable_irq(desc);
1568         }
1569
1570         raw_spin_unlock_irqrestore(&desc->lock, flags);
1571         chip_bus_sync_unlock(desc);
1572         mutex_unlock(&desc->request_mutex);
1573
1574         irq_setup_timings(desc, new);
1575
1576         /*
1577          * Strictly no need to wake it up, but hung_task complains
1578          * when no hard interrupt wakes the thread up.
1579          */
1580         if (new->thread)
1581                 wake_up_process(new->thread);
1582         if (new->secondary)
1583                 wake_up_process(new->secondary->thread);
1584
1585         register_irq_proc(irq, desc);
1586         new->dir = NULL;
1587         register_handler_proc(irq, new);
1588         return 0;
1589
1590 mismatch:
1591         if (!(new->flags & IRQF_PROBE_SHARED)) {
1592                 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1593                        irq, new->flags, new->name, old->flags, old->name);
1594 #ifdef CONFIG_DEBUG_SHIRQ
1595                 dump_stack();
1596 #endif
1597         }
1598         ret = -EBUSY;
1599
1600 out_unlock:
1601         raw_spin_unlock_irqrestore(&desc->lock, flags);
1602
1603         if (!desc->action)
1604                 irq_release_resources(desc);
1605 out_bus_unlock:
1606         chip_bus_sync_unlock(desc);
1607         mutex_unlock(&desc->request_mutex);
1608
1609 out_thread:
1610         if (new->thread) {
1611                 struct task_struct *t = new->thread;
1612
1613                 new->thread = NULL;
1614                 kthread_stop(t);
1615                 put_task_struct(t);
1616         }
1617         if (new->secondary && new->secondary->thread) {
1618                 struct task_struct *t = new->secondary->thread;
1619
1620                 new->secondary->thread = NULL;
1621                 kthread_stop(t);
1622                 put_task_struct(t);
1623         }
1624 out_mput:
1625         module_put(desc->owner);
1626         return ret;
1627 }
1628
1629 /**
1630  *      setup_irq - setup an interrupt
1631  *      @irq: Interrupt line to setup
1632  *      @act: irqaction for the interrupt
1633  *
1634  * Used to statically setup interrupts in the early boot process.
1635  */
1636 int setup_irq(unsigned int irq, struct irqaction *act)
1637 {
1638         int retval;
1639         struct irq_desc *desc = irq_to_desc(irq);
1640
1641         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1642                 return -EINVAL;
1643
1644         retval = irq_chip_pm_get(&desc->irq_data);
1645         if (retval < 0)
1646                 return retval;
1647
1648         retval = __setup_irq(irq, desc, act);
1649
1650         if (retval)
1651                 irq_chip_pm_put(&desc->irq_data);
1652
1653         return retval;
1654 }
1655 EXPORT_SYMBOL_GPL(setup_irq);
1656
1657 /*
1658  * Internal function to unregister an irqaction - used to free
1659  * regular and special interrupts that are part of the architecture.
1660  */
1661 static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1662 {
1663         unsigned irq = desc->irq_data.irq;
1664         struct irqaction *action, **action_ptr;
1665         unsigned long flags;
1666
1667         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1668
1669         mutex_lock(&desc->request_mutex);
1670         chip_bus_lock(desc);
1671         raw_spin_lock_irqsave(&desc->lock, flags);
1672
1673         /*
1674          * There can be multiple actions per IRQ descriptor, find the right
1675          * one based on the dev_id:
1676          */
1677         action_ptr = &desc->action;
1678         for (;;) {
1679                 action = *action_ptr;
1680
1681                 if (!action) {
1682                         WARN(1, "Trying to free already-free IRQ %d\n", irq);
1683                         raw_spin_unlock_irqrestore(&desc->lock, flags);
1684                         chip_bus_sync_unlock(desc);
1685                         mutex_unlock(&desc->request_mutex);
1686                         return NULL;
1687                 }
1688
1689                 if (action->dev_id == dev_id)
1690                         break;
1691                 action_ptr = &action->next;
1692         }
1693
1694         /* Found it - now remove it from the list of entries: */
1695         *action_ptr = action->next;
1696
1697         irq_pm_remove_action(desc, action);
1698
1699         /* If this was the last handler, shut down the IRQ line: */
1700         if (!desc->action) {
1701                 irq_settings_clr_disable_unlazy(desc);
1702                 irq_shutdown(desc);
1703         }
1704
1705 #ifdef CONFIG_SMP
1706         /* make sure affinity_hint is cleaned up */
1707         if (WARN_ON_ONCE(desc->affinity_hint))
1708                 desc->affinity_hint = NULL;
1709 #endif
1710
1711         raw_spin_unlock_irqrestore(&desc->lock, flags);
1712         /*
1713          * Drop bus_lock here so the changes which were done in the chip
1714          * callbacks above are synced out to the irq chips which hang
1715          * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1716          *
1717          * Aside of that the bus_lock can also be taken from the threaded
1718          * handler in irq_finalize_oneshot() which results in a deadlock
1719          * because kthread_stop() would wait forever for the thread to
1720          * complete, which is blocked on the bus lock.
1721          *
1722          * The still held desc->request_mutex() protects against a
1723          * concurrent request_irq() of this irq so the release of resources
1724          * and timing data is properly serialized.
1725          */
1726         chip_bus_sync_unlock(desc);
1727
1728         unregister_handler_proc(irq, action);
1729
1730         /* Make sure it's not being used on another CPU: */
1731         synchronize_hardirq(irq);
1732
1733 #ifdef CONFIG_DEBUG_SHIRQ
1734         /*
1735          * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1736          * event to happen even now it's being freed, so let's make sure that
1737          * is so by doing an extra call to the handler ....
1738          *
1739          * ( We do this after actually deregistering it, to make sure that a
1740          *   'real' IRQ doesn't run in parallel with our fake. )
1741          */
1742         if (action->flags & IRQF_SHARED) {
1743                 local_irq_save(flags);
1744                 action->handler(irq, dev_id);
1745                 local_irq_restore(flags);
1746         }
1747 #endif
1748
1749         /*
1750          * The action has already been removed above, but the thread writes
1751          * its oneshot mask bit when it completes. Though request_mutex is
1752          * held across this which prevents __setup_irq() from handing out
1753          * the same bit to a newly requested action.
1754          */
1755         if (action->thread) {
1756                 kthread_stop(action->thread);
1757                 put_task_struct(action->thread);
1758                 if (action->secondary && action->secondary->thread) {
1759                         kthread_stop(action->secondary->thread);
1760                         put_task_struct(action->secondary->thread);
1761                 }
1762         }
1763
1764         /* Last action releases resources */
1765         if (!desc->action) {
1766                 /*
1767                  * Reaquire bus lock as irq_release_resources() might
1768                  * require it to deallocate resources over the slow bus.
1769                  */
1770                 chip_bus_lock(desc);
1771                 irq_release_resources(desc);
1772                 chip_bus_sync_unlock(desc);
1773                 irq_remove_timings(desc);
1774         }
1775
1776         mutex_unlock(&desc->request_mutex);
1777
1778         irq_chip_pm_put(&desc->irq_data);
1779         module_put(desc->owner);
1780         kfree(action->secondary);
1781         return action;
1782 }
1783
1784 /**
1785  *      remove_irq - free an interrupt
1786  *      @irq: Interrupt line to free
1787  *      @act: irqaction for the interrupt
1788  *
1789  * Used to remove interrupts statically setup by the early boot process.
1790  */
1791 void remove_irq(unsigned int irq, struct irqaction *act)
1792 {
1793         struct irq_desc *desc = irq_to_desc(irq);
1794
1795         if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1796                 __free_irq(desc, act->dev_id);
1797 }
1798 EXPORT_SYMBOL_GPL(remove_irq);
1799
1800 /**
1801  *      free_irq - free an interrupt allocated with request_irq
1802  *      @irq: Interrupt line to free
1803  *      @dev_id: Device identity to free
1804  *
1805  *      Remove an interrupt handler. The handler is removed and if the
1806  *      interrupt line is no longer in use by any driver it is disabled.
1807  *      On a shared IRQ the caller must ensure the interrupt is disabled
1808  *      on the card it drives before calling this function. The function
1809  *      does not return until any executing interrupts for this IRQ
1810  *      have completed.
1811  *
1812  *      This function must not be called from interrupt context.
1813  *
1814  *      Returns the devname argument passed to request_irq.
1815  */
1816 const void *free_irq(unsigned int irq, void *dev_id)
1817 {
1818         struct irq_desc *desc = irq_to_desc(irq);
1819         struct irqaction *action;
1820         const char *devname;
1821
1822         if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1823                 return NULL;
1824
1825 #ifdef CONFIG_SMP
1826         if (WARN_ON(desc->affinity_notify))
1827                 desc->affinity_notify = NULL;
1828 #endif
1829
1830         action = __free_irq(desc, dev_id);
1831
1832         if (!action)
1833                 return NULL;
1834
1835         devname = action->name;
1836         kfree(action);
1837         return devname;
1838 }
1839 EXPORT_SYMBOL(free_irq);
1840
1841 /* This function must be called with desc->lock held */
1842 static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
1843 {
1844         const char *devname = NULL;
1845
1846         desc->istate &= ~IRQS_NMI;
1847
1848         if (!WARN_ON(desc->action == NULL)) {
1849                 irq_pm_remove_action(desc, desc->action);
1850                 devname = desc->action->name;
1851                 unregister_handler_proc(irq, desc->action);
1852
1853                 kfree(desc->action);
1854                 desc->action = NULL;
1855         }
1856
1857         irq_settings_clr_disable_unlazy(desc);
1858         irq_shutdown(desc);
1859
1860         irq_release_resources(desc);
1861
1862         irq_chip_pm_put(&desc->irq_data);
1863         module_put(desc->owner);
1864
1865         return devname;
1866 }
1867
1868 const void *free_nmi(unsigned int irq, void *dev_id)
1869 {
1870         struct irq_desc *desc = irq_to_desc(irq);
1871         unsigned long flags;
1872         const void *devname;
1873
1874         if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
1875                 return NULL;
1876
1877         if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1878                 return NULL;
1879
1880         /* NMI still enabled */
1881         if (WARN_ON(desc->depth == 0))
1882                 disable_nmi_nosync(irq);
1883
1884         raw_spin_lock_irqsave(&desc->lock, flags);
1885
1886         irq_nmi_teardown(desc);
1887         devname = __cleanup_nmi(irq, desc);
1888
1889         raw_spin_unlock_irqrestore(&desc->lock, flags);
1890
1891         return devname;
1892 }
1893
1894 /**
1895  *      request_threaded_irq - allocate an interrupt line
1896  *      @irq: Interrupt line to allocate
1897  *      @handler: Function to be called when the IRQ occurs.
1898  *                Primary handler for threaded interrupts
1899  *                If NULL and thread_fn != NULL the default
1900  *                primary handler is installed
1901  *      @thread_fn: Function called from the irq handler thread
1902  *                  If NULL, no irq thread is created
1903  *      @irqflags: Interrupt type flags
1904  *      @devname: An ascii name for the claiming device
1905  *      @dev_id: A cookie passed back to the handler function
1906  *
1907  *      This call allocates interrupt resources and enables the
1908  *      interrupt line and IRQ handling. From the point this
1909  *      call is made your handler function may be invoked. Since
1910  *      your handler function must clear any interrupt the board
1911  *      raises, you must take care both to initialise your hardware
1912  *      and to set up the interrupt handler in the right order.
1913  *
1914  *      If you want to set up a threaded irq handler for your device
1915  *      then you need to supply @handler and @thread_fn. @handler is
1916  *      still called in hard interrupt context and has to check
1917  *      whether the interrupt originates from the device. If yes it
1918  *      needs to disable the interrupt on the device and return
1919  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1920  *      @thread_fn. This split handler design is necessary to support
1921  *      shared interrupts.
1922  *
1923  *      Dev_id must be globally unique. Normally the address of the
1924  *      device data structure is used as the cookie. Since the handler
1925  *      receives this value it makes sense to use it.
1926  *
1927  *      If your interrupt is shared you must pass a non NULL dev_id
1928  *      as this is required when freeing the interrupt.
1929  *
1930  *      Flags:
1931  *
1932  *      IRQF_SHARED             Interrupt is shared
1933  *      IRQF_TRIGGER_*          Specify active edge(s) or level
1934  *
1935  */
1936 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1937                          irq_handler_t thread_fn, unsigned long irqflags,
1938                          const char *devname, void *dev_id)
1939 {
1940         struct irqaction *action;
1941         struct irq_desc *desc;
1942         int retval;
1943
1944         if (irq == IRQ_NOTCONNECTED)
1945                 return -ENOTCONN;
1946
1947         /*
1948          * Sanity-check: shared interrupts must pass in a real dev-ID,
1949          * otherwise we'll have trouble later trying to figure out
1950          * which interrupt is which (messes up the interrupt freeing
1951          * logic etc).
1952          *
1953          * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1954          * it cannot be set along with IRQF_NO_SUSPEND.
1955          */
1956         if (((irqflags & IRQF_SHARED) && !dev_id) ||
1957             (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1958             ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1959                 return -EINVAL;
1960
1961         desc = irq_to_desc(irq);
1962         if (!desc)
1963                 return -EINVAL;
1964
1965         if (!irq_settings_can_request(desc) ||
1966             WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1967                 return -EINVAL;
1968
1969         if (!handler) {
1970                 if (!thread_fn)
1971                         return -EINVAL;
1972                 handler = irq_default_primary_handler;
1973         }
1974
1975         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1976         if (!action)
1977                 return -ENOMEM;
1978
1979         action->handler = handler;
1980         action->thread_fn = thread_fn;
1981         action->flags = irqflags;
1982         action->name = devname;
1983         action->dev_id = dev_id;
1984
1985         retval = irq_chip_pm_get(&desc->irq_data);
1986         if (retval < 0) {
1987                 kfree(action);
1988                 return retval;
1989         }
1990
1991         retval = __setup_irq(irq, desc, action);
1992
1993         if (retval) {
1994                 irq_chip_pm_put(&desc->irq_data);
1995                 kfree(action->secondary);
1996                 kfree(action);
1997         }
1998
1999 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
2000         if (!retval && (irqflags & IRQF_SHARED)) {
2001                 /*
2002                  * It's a shared IRQ -- the driver ought to be prepared for it
2003                  * to happen immediately, so let's make sure....
2004                  * We disable the irq to make sure that a 'real' IRQ doesn't
2005                  * run in parallel with our fake.
2006                  */
2007                 unsigned long flags;
2008
2009                 disable_irq(irq);
2010                 local_irq_save(flags);
2011
2012                 handler(irq, dev_id);
2013
2014                 local_irq_restore(flags);
2015                 enable_irq(irq);
2016         }
2017 #endif
2018         return retval;
2019 }
2020 EXPORT_SYMBOL(request_threaded_irq);
2021
2022 /**
2023  *      request_any_context_irq - allocate an interrupt line
2024  *      @irq: Interrupt line to allocate
2025  *      @handler: Function to be called when the IRQ occurs.
2026  *                Threaded handler for threaded interrupts.
2027  *      @flags: Interrupt type flags
2028  *      @name: An ascii name for the claiming device
2029  *      @dev_id: A cookie passed back to the handler function
2030  *
2031  *      This call allocates interrupt resources and enables the
2032  *      interrupt line and IRQ handling. It selects either a
2033  *      hardirq or threaded handling method depending on the
2034  *      context.
2035  *
2036  *      On failure, it returns a negative value. On success,
2037  *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2038  */
2039 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2040                             unsigned long flags, const char *name, void *dev_id)
2041 {
2042         struct irq_desc *desc;
2043         int ret;
2044
2045         if (irq == IRQ_NOTCONNECTED)
2046                 return -ENOTCONN;
2047
2048         desc = irq_to_desc(irq);
2049         if (!desc)
2050                 return -EINVAL;
2051
2052         if (irq_settings_is_nested_thread(desc)) {
2053                 ret = request_threaded_irq(irq, NULL, handler,
2054                                            flags, name, dev_id);
2055                 return !ret ? IRQC_IS_NESTED : ret;
2056         }
2057
2058         ret = request_irq(irq, handler, flags, name, dev_id);
2059         return !ret ? IRQC_IS_HARDIRQ : ret;
2060 }
2061 EXPORT_SYMBOL_GPL(request_any_context_irq);
2062
2063 /**
2064  *      request_nmi - allocate an interrupt line for NMI delivery
2065  *      @irq: Interrupt line to allocate
2066  *      @handler: Function to be called when the IRQ occurs.
2067  *                Threaded handler for threaded interrupts.
2068  *      @irqflags: Interrupt type flags
2069  *      @name: An ascii name for the claiming device
2070  *      @dev_id: A cookie passed back to the handler function
2071  *
2072  *      This call allocates interrupt resources and enables the
2073  *      interrupt line and IRQ handling. It sets up the IRQ line
2074  *      to be handled as an NMI.
2075  *
2076  *      An interrupt line delivering NMIs cannot be shared and IRQ handling
2077  *      cannot be threaded.
2078  *
2079  *      Interrupt lines requested for NMI delivering must produce per cpu
2080  *      interrupts and have auto enabling setting disabled.
2081  *
2082  *      Dev_id must be globally unique. Normally the address of the
2083  *      device data structure is used as the cookie. Since the handler
2084  *      receives this value it makes sense to use it.
2085  *
2086  *      If the interrupt line cannot be used to deliver NMIs, function
2087  *      will fail and return a negative value.
2088  */
2089 int request_nmi(unsigned int irq, irq_handler_t handler,
2090                 unsigned long irqflags, const char *name, void *dev_id)
2091 {
2092         struct irqaction *action;
2093         struct irq_desc *desc;
2094         unsigned long flags;
2095         int retval;
2096
2097         if (irq == IRQ_NOTCONNECTED)
2098                 return -ENOTCONN;
2099
2100         /* NMI cannot be shared, used for Polling */
2101         if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2102                 return -EINVAL;
2103
2104         if (!(irqflags & IRQF_PERCPU))
2105                 return -EINVAL;
2106
2107         if (!handler)
2108                 return -EINVAL;
2109
2110         desc = irq_to_desc(irq);
2111
2112         if (!desc || irq_settings_can_autoenable(desc) ||
2113             !irq_settings_can_request(desc) ||
2114             WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2115             !irq_supports_nmi(desc))
2116                 return -EINVAL;
2117
2118         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2119         if (!action)
2120                 return -ENOMEM;
2121
2122         action->handler = handler;
2123         action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2124         action->name = name;
2125         action->dev_id = dev_id;
2126
2127         retval = irq_chip_pm_get(&desc->irq_data);
2128         if (retval < 0)
2129                 goto err_out;
2130
2131         retval = __setup_irq(irq, desc, action);
2132         if (retval)
2133                 goto err_irq_setup;
2134
2135         raw_spin_lock_irqsave(&desc->lock, flags);
2136
2137         /* Setup NMI state */
2138         desc->istate |= IRQS_NMI;
2139         retval = irq_nmi_setup(desc);
2140         if (retval) {
2141                 __cleanup_nmi(irq, desc);
2142                 raw_spin_unlock_irqrestore(&desc->lock, flags);
2143                 return -EINVAL;
2144         }
2145
2146         raw_spin_unlock_irqrestore(&desc->lock, flags);
2147
2148         return 0;
2149
2150 err_irq_setup:
2151         irq_chip_pm_put(&desc->irq_data);
2152 err_out:
2153         kfree(action);
2154
2155         return retval;
2156 }
2157
2158 void enable_percpu_irq(unsigned int irq, unsigned int type)
2159 {
2160         unsigned int cpu = smp_processor_id();
2161         unsigned long flags;
2162         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2163
2164         if (!desc)
2165                 return;
2166
2167         /*
2168          * If the trigger type is not specified by the caller, then
2169          * use the default for this interrupt.
2170          */
2171         type &= IRQ_TYPE_SENSE_MASK;
2172         if (type == IRQ_TYPE_NONE)
2173                 type = irqd_get_trigger_type(&desc->irq_data);
2174
2175         if (type != IRQ_TYPE_NONE) {
2176                 int ret;
2177
2178                 ret = __irq_set_trigger(desc, type);
2179
2180                 if (ret) {
2181                         WARN(1, "failed to set type for IRQ%d\n", irq);
2182                         goto out;
2183                 }
2184         }
2185
2186         irq_percpu_enable(desc, cpu);
2187 out:
2188         irq_put_desc_unlock(desc, flags);
2189 }
2190 EXPORT_SYMBOL_GPL(enable_percpu_irq);
2191
2192 void enable_percpu_nmi(unsigned int irq, unsigned int type)
2193 {
2194         enable_percpu_irq(irq, type);
2195 }
2196
2197 /**
2198  * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2199  * @irq:        Linux irq number to check for
2200  *
2201  * Must be called from a non migratable context. Returns the enable
2202  * state of a per cpu interrupt on the current cpu.
2203  */
2204 bool irq_percpu_is_enabled(unsigned int irq)
2205 {
2206         unsigned int cpu = smp_processor_id();
2207         struct irq_desc *desc;
2208         unsigned long flags;
2209         bool is_enabled;
2210
2211         desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2212         if (!desc)
2213                 return false;
2214
2215         is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2216         irq_put_desc_unlock(desc, flags);
2217
2218         return is_enabled;
2219 }
2220 EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2221
2222 void disable_percpu_irq(unsigned int irq)
2223 {
2224         unsigned int cpu = smp_processor_id();
2225         unsigned long flags;
2226         struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2227
2228         if (!desc)
2229                 return;
2230
2231         irq_percpu_disable(desc, cpu);
2232         irq_put_desc_unlock(desc, flags);
2233 }
2234 EXPORT_SYMBOL_GPL(disable_percpu_irq);
2235
2236 void disable_percpu_nmi(unsigned int irq)
2237 {
2238         disable_percpu_irq(irq);
2239 }
2240
2241 /*
2242  * Internal function to unregister a percpu irqaction.
2243  */
2244 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2245 {
2246         struct irq_desc *desc = irq_to_desc(irq);
2247         struct irqaction *action;
2248         unsigned long flags;
2249
2250         WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2251
2252         if (!desc)
2253                 return NULL;
2254
2255         raw_spin_lock_irqsave(&desc->lock, flags);
2256
2257         action = desc->action;
2258         if (!action || action->percpu_dev_id != dev_id) {
2259                 WARN(1, "Trying to free already-free IRQ %d\n", irq);
2260                 goto bad;
2261         }
2262
2263         if (!cpumask_empty(desc->percpu_enabled)) {
2264                 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2265                      irq, cpumask_first(desc->percpu_enabled));
2266                 goto bad;
2267         }
2268
2269         /* Found it - now remove it from the list of entries: */
2270         desc->action = NULL;
2271
2272         desc->istate &= ~IRQS_NMI;
2273
2274         raw_spin_unlock_irqrestore(&desc->lock, flags);
2275
2276         unregister_handler_proc(irq, action);
2277
2278         irq_chip_pm_put(&desc->irq_data);
2279         module_put(desc->owner);
2280         return action;
2281
2282 bad:
2283         raw_spin_unlock_irqrestore(&desc->lock, flags);
2284         return NULL;
2285 }
2286
2287 /**
2288  *      remove_percpu_irq - free a per-cpu interrupt
2289  *      @irq: Interrupt line to free
2290  *      @act: irqaction for the interrupt
2291  *
2292  * Used to remove interrupts statically setup by the early boot process.
2293  */
2294 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2295 {
2296         struct irq_desc *desc = irq_to_desc(irq);
2297
2298         if (desc && irq_settings_is_per_cpu_devid(desc))
2299             __free_percpu_irq(irq, act->percpu_dev_id);
2300 }
2301
2302 /**
2303  *      free_percpu_irq - free an interrupt allocated with request_percpu_irq
2304  *      @irq: Interrupt line to free
2305  *      @dev_id: Device identity to free
2306  *
2307  *      Remove a percpu interrupt handler. The handler is removed, but
2308  *      the interrupt line is not disabled. This must be done on each
2309  *      CPU before calling this function. The function does not return
2310  *      until any executing interrupts for this IRQ have completed.
2311  *
2312  *      This function must not be called from interrupt context.
2313  */
2314 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2315 {
2316         struct irq_desc *desc = irq_to_desc(irq);
2317
2318         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2319                 return;
2320
2321         chip_bus_lock(desc);
2322         kfree(__free_percpu_irq(irq, dev_id));
2323         chip_bus_sync_unlock(desc);
2324 }
2325 EXPORT_SYMBOL_GPL(free_percpu_irq);
2326
2327 void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2328 {
2329         struct irq_desc *desc = irq_to_desc(irq);
2330
2331         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2332                 return;
2333
2334         if (WARN_ON(!(desc->istate & IRQS_NMI)))
2335                 return;
2336
2337         kfree(__free_percpu_irq(irq, dev_id));
2338 }
2339
2340 /**
2341  *      setup_percpu_irq - setup a per-cpu interrupt
2342  *      @irq: Interrupt line to setup
2343  *      @act: irqaction for the interrupt
2344  *
2345  * Used to statically setup per-cpu interrupts in the early boot process.
2346  */
2347 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2348 {
2349         struct irq_desc *desc = irq_to_desc(irq);
2350         int retval;
2351
2352         if (!desc || !irq_settings_is_per_cpu_devid(desc))
2353                 return -EINVAL;
2354
2355         retval = irq_chip_pm_get(&desc->irq_data);
2356         if (retval < 0)
2357                 return retval;
2358
2359         retval = __setup_irq(irq, desc, act);
2360
2361         if (retval)
2362                 irq_chip_pm_put(&desc->irq_data);
2363
2364         return retval;
2365 }
2366
2367 /**
2368  *      __request_percpu_irq - allocate a percpu interrupt line
2369  *      @irq: Interrupt line to allocate
2370  *      @handler: Function to be called when the IRQ occurs.
2371  *      @flags: Interrupt type flags (IRQF_TIMER only)
2372  *      @devname: An ascii name for the claiming device
2373  *      @dev_id: A percpu cookie passed back to the handler function
2374  *
2375  *      This call allocates interrupt resources and enables the
2376  *      interrupt on the local CPU. If the interrupt is supposed to be
2377  *      enabled on other CPUs, it has to be done on each CPU using
2378  *      enable_percpu_irq().
2379  *
2380  *      Dev_id must be globally unique. It is a per-cpu variable, and
2381  *      the handler gets called with the interrupted CPU's instance of
2382  *      that variable.
2383  */
2384 int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2385                          unsigned long flags, const char *devname,
2386                          void __percpu *dev_id)
2387 {
2388         struct irqaction *action;
2389         struct irq_desc *desc;
2390         int retval;
2391
2392         if (!dev_id)
2393                 return -EINVAL;
2394
2395         desc = irq_to_desc(irq);
2396         if (!desc || !irq_settings_can_request(desc) ||
2397             !irq_settings_is_per_cpu_devid(desc))
2398                 return -EINVAL;
2399
2400         if (flags && flags != IRQF_TIMER)
2401                 return -EINVAL;
2402
2403         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2404         if (!action)
2405                 return -ENOMEM;
2406
2407         action->handler = handler;
2408         action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2409         action->name = devname;
2410         action->percpu_dev_id = dev_id;
2411
2412         retval = irq_chip_pm_get(&desc->irq_data);
2413         if (retval < 0) {
2414                 kfree(action);
2415                 return retval;
2416         }
2417
2418         retval = __setup_irq(irq, desc, action);
2419
2420         if (retval) {
2421                 irq_chip_pm_put(&desc->irq_data);
2422                 kfree(action);
2423         }
2424
2425         return retval;
2426 }
2427 EXPORT_SYMBOL_GPL(__request_percpu_irq);
2428
2429 /**
2430  *      request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2431  *      @irq: Interrupt line to allocate
2432  *      @handler: Function to be called when the IRQ occurs.
2433  *      @name: An ascii name for the claiming device
2434  *      @dev_id: A percpu cookie passed back to the handler function
2435  *
2436  *      This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2437  *      have to be setup on each CPU by calling prepare_percpu_nmi() before
2438  *      being enabled on the same CPU by using enable_percpu_nmi().
2439  *
2440  *      Dev_id must be globally unique. It is a per-cpu variable, and
2441  *      the handler gets called with the interrupted CPU's instance of
2442  *      that variable.
2443  *
2444  *      Interrupt lines requested for NMI delivering should have auto enabling
2445  *      setting disabled.
2446  *
2447  *      If the interrupt line cannot be used to deliver NMIs, function
2448  *      will fail returning a negative value.
2449  */
2450 int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2451                        const char *name, void __percpu *dev_id)
2452 {
2453         struct irqaction *action;
2454         struct irq_desc *desc;
2455         unsigned long flags;
2456         int retval;
2457
2458         if (!handler)
2459                 return -EINVAL;
2460
2461         desc = irq_to_desc(irq);
2462
2463         if (!desc || !irq_settings_can_request(desc) ||
2464             !irq_settings_is_per_cpu_devid(desc) ||
2465             irq_settings_can_autoenable(desc) ||
2466             !irq_supports_nmi(desc))
2467                 return -EINVAL;
2468
2469         /* The line cannot already be NMI */
2470         if (desc->istate & IRQS_NMI)
2471                 return -EINVAL;
2472
2473         action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2474         if (!action)
2475                 return -ENOMEM;
2476
2477         action->handler = handler;
2478         action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2479                 | IRQF_NOBALANCING;
2480         action->name = name;
2481         action->percpu_dev_id = dev_id;
2482
2483         retval = irq_chip_pm_get(&desc->irq_data);
2484         if (retval < 0)
2485                 goto err_out;
2486
2487         retval = __setup_irq(irq, desc, action);
2488         if (retval)
2489                 goto err_irq_setup;
2490
2491         raw_spin_lock_irqsave(&desc->lock, flags);
2492         desc->istate |= IRQS_NMI;
2493         raw_spin_unlock_irqrestore(&desc->lock, flags);
2494
2495         return 0;
2496
2497 err_irq_setup:
2498         irq_chip_pm_put(&desc->irq_data);
2499 err_out:
2500         kfree(action);
2501
2502         return retval;
2503 }
2504
2505 /**
2506  *      prepare_percpu_nmi - performs CPU local setup for NMI delivery
2507  *      @irq: Interrupt line to prepare for NMI delivery
2508  *
2509  *      This call prepares an interrupt line to deliver NMI on the current CPU,
2510  *      before that interrupt line gets enabled with enable_percpu_nmi().
2511  *
2512  *      As a CPU local operation, this should be called from non-preemptible
2513  *      context.
2514  *
2515  *      If the interrupt line cannot be used to deliver NMIs, function
2516  *      will fail returning a negative value.
2517  */
2518 int prepare_percpu_nmi(unsigned int irq)
2519 {
2520         unsigned long flags;
2521         struct irq_desc *desc;
2522         int ret = 0;
2523
2524         WARN_ON(preemptible());
2525
2526         desc = irq_get_desc_lock(irq, &flags,
2527                                  IRQ_GET_DESC_CHECK_PERCPU);
2528         if (!desc)
2529                 return -EINVAL;
2530
2531         if (WARN(!(desc->istate & IRQS_NMI),
2532                  KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2533                  irq)) {
2534                 ret = -EINVAL;
2535                 goto out;
2536         }
2537
2538         ret = irq_nmi_setup(desc);
2539         if (ret) {
2540                 pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2541                 goto out;
2542         }
2543
2544 out:
2545         irq_put_desc_unlock(desc, flags);
2546         return ret;
2547 }
2548
2549 /**
2550  *      teardown_percpu_nmi - undoes NMI setup of IRQ line
2551  *      @irq: Interrupt line from which CPU local NMI configuration should be
2552  *            removed
2553  *
2554  *      This call undoes the setup done by prepare_percpu_nmi().
2555  *
2556  *      IRQ line should not be enabled for the current CPU.
2557  *
2558  *      As a CPU local operation, this should be called from non-preemptible
2559  *      context.
2560  */
2561 void teardown_percpu_nmi(unsigned int irq)
2562 {
2563         unsigned long flags;
2564         struct irq_desc *desc;
2565
2566         WARN_ON(preemptible());
2567
2568         desc = irq_get_desc_lock(irq, &flags,
2569                                  IRQ_GET_DESC_CHECK_PERCPU);
2570         if (!desc)
2571                 return;
2572
2573         if (WARN_ON(!(desc->istate & IRQS_NMI)))
2574                 goto out;
2575
2576         irq_nmi_teardown(desc);
2577 out:
2578         irq_put_desc_unlock(desc, flags);
2579 }
2580
2581 /**
2582  *      irq_get_irqchip_state - returns the irqchip state of a interrupt.
2583  *      @irq: Interrupt line that is forwarded to a VM
2584  *      @which: One of IRQCHIP_STATE_* the caller wants to know about
2585  *      @state: a pointer to a boolean where the state is to be storeed
2586  *
2587  *      This call snapshots the internal irqchip state of an
2588  *      interrupt, returning into @state the bit corresponding to
2589  *      stage @which
2590  *
2591  *      This function should be called with preemption disabled if the
2592  *      interrupt controller has per-cpu registers.
2593  */
2594 int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2595                           bool *state)
2596 {
2597         struct irq_desc *desc;
2598         struct irq_data *data;
2599         struct irq_chip *chip;
2600         unsigned long flags;
2601         int err = -EINVAL;
2602
2603         desc = irq_get_desc_buslock(irq, &flags, 0);
2604         if (!desc)
2605                 return err;
2606
2607         data = irq_desc_get_irq_data(desc);
2608
2609         do {
2610                 chip = irq_data_get_irq_chip(data);
2611                 if (chip->irq_get_irqchip_state)
2612                         break;
2613 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2614                 data = data->parent_data;
2615 #else
2616                 data = NULL;
2617 #endif
2618         } while (data);
2619
2620         if (data)
2621                 err = chip->irq_get_irqchip_state(data, which, state);
2622
2623         irq_put_desc_busunlock(desc, flags);
2624         return err;
2625 }
2626 EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2627
2628 /**
2629  *      irq_set_irqchip_state - set the state of a forwarded interrupt.
2630  *      @irq: Interrupt line that is forwarded to a VM
2631  *      @which: State to be restored (one of IRQCHIP_STATE_*)
2632  *      @val: Value corresponding to @which
2633  *
2634  *      This call sets the internal irqchip state of an interrupt,
2635  *      depending on the value of @which.
2636  *
2637  *      This function should be called with preemption disabled if the
2638  *      interrupt controller has per-cpu registers.
2639  */
2640 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2641                           bool val)
2642 {
2643         struct irq_desc *desc;
2644         struct irq_data *data;
2645         struct irq_chip *chip;
2646         unsigned long flags;
2647         int err = -EINVAL;
2648
2649         desc = irq_get_desc_buslock(irq, &flags, 0);
2650         if (!desc)
2651                 return err;
2652
2653         data = irq_desc_get_irq_data(desc);
2654
2655         do {
2656                 chip = irq_data_get_irq_chip(data);
2657                 if (chip->irq_set_irqchip_state)
2658                         break;
2659 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2660                 data = data->parent_data;
2661 #else
2662                 data = NULL;
2663 #endif
2664         } while (data);
2665
2666         if (data)
2667                 err = chip->irq_set_irqchip_state(data, which, val);
2668
2669         irq_put_desc_busunlock(desc, flags);
2670         return err;
2671 }
2672 EXPORT_SYMBOL_GPL(irq_set_irqchip_state);