Loading...
1/*
2 * linux/kernel/irq/manage.c
3 *
4 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
5 * Copyright (C) 2005-2006 Thomas Gleixner
6 *
7 * This file contains driver APIs to the irq subsystem.
8 */
9
10#define pr_fmt(fmt) "genirq: " fmt
11
12#include <linux/irq.h>
13#include <linux/kthread.h>
14#include <linux/module.h>
15#include <linux/random.h>
16#include <linux/interrupt.h>
17#include <linux/slab.h>
18#include <linux/sched.h>
19#include <linux/sched/rt.h>
20#include <linux/task_work.h>
21
22#include "internals.h"
23
24#ifdef CONFIG_IRQ_FORCED_THREADING
25__read_mostly bool force_irqthreads;
26
27static int __init setup_forced_irqthreads(char *arg)
28{
29 force_irqthreads = true;
30 return 0;
31}
32early_param("threadirqs", setup_forced_irqthreads);
33#endif
34
35static void __synchronize_hardirq(struct irq_desc *desc)
36{
37 bool inprogress;
38
39 do {
40 unsigned long flags;
41
42 /*
43 * Wait until we're out of the critical section. This might
44 * give the wrong answer due to the lack of memory barriers.
45 */
46 while (irqd_irq_inprogress(&desc->irq_data))
47 cpu_relax();
48
49 /* Ok, that indicated we're done: double-check carefully. */
50 raw_spin_lock_irqsave(&desc->lock, flags);
51 inprogress = irqd_irq_inprogress(&desc->irq_data);
52 raw_spin_unlock_irqrestore(&desc->lock, flags);
53
54 /* Oops, that failed? */
55 } while (inprogress);
56}
57
58/**
59 * synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
60 * @irq: interrupt number to wait for
61 *
62 * This function waits for any pending hard IRQ handlers for this
63 * interrupt to complete before returning. If you use this
64 * function while holding a resource the IRQ handler may need you
65 * will deadlock. It does not take associated threaded handlers
66 * into account.
67 *
68 * Do not use this for shutdown scenarios where you must be sure
69 * that all parts (hardirq and threaded handler) have completed.
70 *
71 * This function may be called - with care - from IRQ context.
72 */
73void synchronize_hardirq(unsigned int irq)
74{
75 struct irq_desc *desc = irq_to_desc(irq);
76
77 if (desc)
78 __synchronize_hardirq(desc);
79}
80EXPORT_SYMBOL(synchronize_hardirq);
81
82/**
83 * synchronize_irq - wait for pending IRQ handlers (on other CPUs)
84 * @irq: interrupt number to wait for
85 *
86 * This function waits for any pending IRQ handlers for this interrupt
87 * to complete before returning. If you use this function while
88 * holding a resource the IRQ handler may need you will deadlock.
89 *
90 * This function may be called - with care - from IRQ context.
91 */
92void synchronize_irq(unsigned int irq)
93{
94 struct irq_desc *desc = irq_to_desc(irq);
95
96 if (desc) {
97 __synchronize_hardirq(desc);
98 /*
99 * We made sure that no hardirq handler is
100 * running. Now verify that no threaded handlers are
101 * active.
102 */
103 wait_event(desc->wait_for_threads,
104 !atomic_read(&desc->threads_active));
105 }
106}
107EXPORT_SYMBOL(synchronize_irq);
108
109#ifdef CONFIG_SMP
110cpumask_var_t irq_default_affinity;
111
112/**
113 * irq_can_set_affinity - Check if the affinity of a given irq can be set
114 * @irq: Interrupt to check
115 *
116 */
117int irq_can_set_affinity(unsigned int irq)
118{
119 struct irq_desc *desc = irq_to_desc(irq);
120
121 if (!desc || !irqd_can_balance(&desc->irq_data) ||
122 !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
123 return 0;
124
125 return 1;
126}
127
128/**
129 * irq_set_thread_affinity - Notify irq threads to adjust affinity
130 * @desc: irq descriptor which has affitnity changed
131 *
132 * We just set IRQTF_AFFINITY and delegate the affinity setting
133 * to the interrupt thread itself. We can not call
134 * set_cpus_allowed_ptr() here as we hold desc->lock and this
135 * code can be called from hard interrupt context.
136 */
137void irq_set_thread_affinity(struct irq_desc *desc)
138{
139 struct irqaction *action = desc->action;
140
141 while (action) {
142 if (action->thread)
143 set_bit(IRQTF_AFFINITY, &action->thread_flags);
144 action = action->next;
145 }
146}
147
148#ifdef CONFIG_GENERIC_PENDING_IRQ
149static inline bool irq_can_move_pcntxt(struct irq_data *data)
150{
151 return irqd_can_move_in_process_context(data);
152}
153static inline bool irq_move_pending(struct irq_data *data)
154{
155 return irqd_is_setaffinity_pending(data);
156}
157static inline void
158irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
159{
160 cpumask_copy(desc->pending_mask, mask);
161}
162static inline void
163irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
164{
165 cpumask_copy(mask, desc->pending_mask);
166}
167#else
168static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
169static inline bool irq_move_pending(struct irq_data *data) { return false; }
170static inline void
171irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
172static inline void
173irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
174#endif
175
176int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
177 bool force)
178{
179 struct irq_desc *desc = irq_data_to_desc(data);
180 struct irq_chip *chip = irq_data_get_irq_chip(data);
181 int ret;
182
183 ret = chip->irq_set_affinity(data, mask, force);
184 switch (ret) {
185 case IRQ_SET_MASK_OK:
186 cpumask_copy(data->affinity, mask);
187 case IRQ_SET_MASK_OK_NOCOPY:
188 irq_set_thread_affinity(desc);
189 ret = 0;
190 }
191
192 return ret;
193}
194
195int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
196 bool force)
197{
198 struct irq_chip *chip = irq_data_get_irq_chip(data);
199 struct irq_desc *desc = irq_data_to_desc(data);
200 int ret = 0;
201
202 if (!chip || !chip->irq_set_affinity)
203 return -EINVAL;
204
205 if (irq_can_move_pcntxt(data)) {
206 ret = irq_do_set_affinity(data, mask, force);
207 } else {
208 irqd_set_move_pending(data);
209 irq_copy_pending(desc, mask);
210 }
211
212 if (desc->affinity_notify) {
213 kref_get(&desc->affinity_notify->kref);
214 schedule_work(&desc->affinity_notify->work);
215 }
216 irqd_set(data, IRQD_AFFINITY_SET);
217
218 return ret;
219}
220
221int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
222{
223 struct irq_desc *desc = irq_to_desc(irq);
224 unsigned long flags;
225 int ret;
226
227 if (!desc)
228 return -EINVAL;
229
230 raw_spin_lock_irqsave(&desc->lock, flags);
231 ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
232 raw_spin_unlock_irqrestore(&desc->lock, flags);
233 return ret;
234}
235
236int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
237{
238 unsigned long flags;
239 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
240
241 if (!desc)
242 return -EINVAL;
243 desc->affinity_hint = m;
244 irq_put_desc_unlock(desc, flags);
245 return 0;
246}
247EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
248
249static void irq_affinity_notify(struct work_struct *work)
250{
251 struct irq_affinity_notify *notify =
252 container_of(work, struct irq_affinity_notify, work);
253 struct irq_desc *desc = irq_to_desc(notify->irq);
254 cpumask_var_t cpumask;
255 unsigned long flags;
256
257 if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
258 goto out;
259
260 raw_spin_lock_irqsave(&desc->lock, flags);
261 if (irq_move_pending(&desc->irq_data))
262 irq_get_pending(cpumask, desc);
263 else
264 cpumask_copy(cpumask, desc->irq_data.affinity);
265 raw_spin_unlock_irqrestore(&desc->lock, flags);
266
267 notify->notify(notify, cpumask);
268
269 free_cpumask_var(cpumask);
270out:
271 kref_put(¬ify->kref, notify->release);
272}
273
274/**
275 * irq_set_affinity_notifier - control notification of IRQ affinity changes
276 * @irq: Interrupt for which to enable/disable notification
277 * @notify: Context for notification, or %NULL to disable
278 * notification. Function pointers must be initialised;
279 * the other fields will be initialised by this function.
280 *
281 * Must be called in process context. Notification may only be enabled
282 * after the IRQ is allocated and must be disabled before the IRQ is
283 * freed using free_irq().
284 */
285int
286irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
287{
288 struct irq_desc *desc = irq_to_desc(irq);
289 struct irq_affinity_notify *old_notify;
290 unsigned long flags;
291
292 /* The release function is promised process context */
293 might_sleep();
294
295 if (!desc)
296 return -EINVAL;
297
298 /* Complete initialisation of *notify */
299 if (notify) {
300 notify->irq = irq;
301 kref_init(¬ify->kref);
302 INIT_WORK(¬ify->work, irq_affinity_notify);
303 }
304
305 raw_spin_lock_irqsave(&desc->lock, flags);
306 old_notify = desc->affinity_notify;
307 desc->affinity_notify = notify;
308 raw_spin_unlock_irqrestore(&desc->lock, flags);
309
310 if (old_notify)
311 kref_put(&old_notify->kref, old_notify->release);
312
313 return 0;
314}
315EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
316
317#ifndef CONFIG_AUTO_IRQ_AFFINITY
318/*
319 * Generic version of the affinity autoselector.
320 */
321static int
322setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
323{
324 struct cpumask *set = irq_default_affinity;
325 int node = desc->irq_data.node;
326
327 /* Excludes PER_CPU and NO_BALANCE interrupts */
328 if (!irq_can_set_affinity(irq))
329 return 0;
330
331 /*
332 * Preserve an userspace affinity setup, but make sure that
333 * one of the targets is online.
334 */
335 if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
336 if (cpumask_intersects(desc->irq_data.affinity,
337 cpu_online_mask))
338 set = desc->irq_data.affinity;
339 else
340 irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
341 }
342
343 cpumask_and(mask, cpu_online_mask, set);
344 if (node != NUMA_NO_NODE) {
345 const struct cpumask *nodemask = cpumask_of_node(node);
346
347 /* make sure at least one of the cpus in nodemask is online */
348 if (cpumask_intersects(mask, nodemask))
349 cpumask_and(mask, mask, nodemask);
350 }
351 irq_do_set_affinity(&desc->irq_data, mask, false);
352 return 0;
353}
354#else
355static inline int
356setup_affinity(unsigned int irq, struct irq_desc *d, struct cpumask *mask)
357{
358 return irq_select_affinity(irq);
359}
360#endif
361
362/*
363 * Called when affinity is set via /proc/irq
364 */
365int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
366{
367 struct irq_desc *desc = irq_to_desc(irq);
368 unsigned long flags;
369 int ret;
370
371 raw_spin_lock_irqsave(&desc->lock, flags);
372 ret = setup_affinity(irq, desc, mask);
373 raw_spin_unlock_irqrestore(&desc->lock, flags);
374 return ret;
375}
376
377#else
378static inline int
379setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
380{
381 return 0;
382}
383#endif
384
385void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
386{
387 if (suspend) {
388 if (!desc->action || (desc->action->flags & IRQF_NO_SUSPEND))
389 return;
390 desc->istate |= IRQS_SUSPENDED;
391 }
392
393 if (!desc->depth++)
394 irq_disable(desc);
395}
396
397static int __disable_irq_nosync(unsigned int irq)
398{
399 unsigned long flags;
400 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
401
402 if (!desc)
403 return -EINVAL;
404 __disable_irq(desc, irq, false);
405 irq_put_desc_busunlock(desc, flags);
406 return 0;
407}
408
409/**
410 * disable_irq_nosync - disable an irq without waiting
411 * @irq: Interrupt to disable
412 *
413 * Disable the selected interrupt line. Disables and Enables are
414 * nested.
415 * Unlike disable_irq(), this function does not ensure existing
416 * instances of the IRQ handler have completed before returning.
417 *
418 * This function may be called from IRQ context.
419 */
420void disable_irq_nosync(unsigned int irq)
421{
422 __disable_irq_nosync(irq);
423}
424EXPORT_SYMBOL(disable_irq_nosync);
425
426/**
427 * disable_irq - disable an irq and wait for completion
428 * @irq: Interrupt to disable
429 *
430 * Disable the selected interrupt line. Enables and Disables are
431 * nested.
432 * This function waits for any pending IRQ handlers for this interrupt
433 * to complete before returning. If you use this function while
434 * holding a resource the IRQ handler may need you will deadlock.
435 *
436 * This function may be called - with care - from IRQ context.
437 */
438void disable_irq(unsigned int irq)
439{
440 if (!__disable_irq_nosync(irq))
441 synchronize_irq(irq);
442}
443EXPORT_SYMBOL(disable_irq);
444
445void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
446{
447 if (resume) {
448 if (!(desc->istate & IRQS_SUSPENDED)) {
449 if (!desc->action)
450 return;
451 if (!(desc->action->flags & IRQF_FORCE_RESUME))
452 return;
453 /* Pretend that it got disabled ! */
454 desc->depth++;
455 }
456 desc->istate &= ~IRQS_SUSPENDED;
457 }
458
459 switch (desc->depth) {
460 case 0:
461 err_out:
462 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
463 break;
464 case 1: {
465 if (desc->istate & IRQS_SUSPENDED)
466 goto err_out;
467 /* Prevent probing on this irq: */
468 irq_settings_set_noprobe(desc);
469 irq_enable(desc);
470 check_irq_resend(desc, irq);
471 /* fall-through */
472 }
473 default:
474 desc->depth--;
475 }
476}
477
478/**
479 * enable_irq - enable handling of an irq
480 * @irq: Interrupt to enable
481 *
482 * Undoes the effect of one call to disable_irq(). If this
483 * matches the last disable, processing of interrupts on this
484 * IRQ line is re-enabled.
485 *
486 * This function may be called from IRQ context only when
487 * desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
488 */
489void enable_irq(unsigned int irq)
490{
491 unsigned long flags;
492 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
493
494 if (!desc)
495 return;
496 if (WARN(!desc->irq_data.chip,
497 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
498 goto out;
499
500 __enable_irq(desc, irq, false);
501out:
502 irq_put_desc_busunlock(desc, flags);
503}
504EXPORT_SYMBOL(enable_irq);
505
506static int set_irq_wake_real(unsigned int irq, unsigned int on)
507{
508 struct irq_desc *desc = irq_to_desc(irq);
509 int ret = -ENXIO;
510
511 if (irq_desc_get_chip(desc)->flags & IRQCHIP_SKIP_SET_WAKE)
512 return 0;
513
514 if (desc->irq_data.chip->irq_set_wake)
515 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
516
517 return ret;
518}
519
520/**
521 * irq_set_irq_wake - control irq power management wakeup
522 * @irq: interrupt to control
523 * @on: enable/disable power management wakeup
524 *
525 * Enable/disable power management wakeup mode, which is
526 * disabled by default. Enables and disables must match,
527 * just as they match for non-wakeup mode support.
528 *
529 * Wakeup mode lets this IRQ wake the system from sleep
530 * states like "suspend to RAM".
531 */
532int irq_set_irq_wake(unsigned int irq, unsigned int on)
533{
534 unsigned long flags;
535 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
536 int ret = 0;
537
538 if (!desc)
539 return -EINVAL;
540
541 /* wakeup-capable irqs can be shared between drivers that
542 * don't need to have the same sleep mode behaviors.
543 */
544 if (on) {
545 if (desc->wake_depth++ == 0) {
546 ret = set_irq_wake_real(irq, on);
547 if (ret)
548 desc->wake_depth = 0;
549 else
550 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
551 }
552 } else {
553 if (desc->wake_depth == 0) {
554 WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
555 } else if (--desc->wake_depth == 0) {
556 ret = set_irq_wake_real(irq, on);
557 if (ret)
558 desc->wake_depth = 1;
559 else
560 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
561 }
562 }
563 irq_put_desc_busunlock(desc, flags);
564 return ret;
565}
566EXPORT_SYMBOL(irq_set_irq_wake);
567
568/*
569 * Internal function that tells the architecture code whether a
570 * particular irq has been exclusively allocated or is available
571 * for driver use.
572 */
573int can_request_irq(unsigned int irq, unsigned long irqflags)
574{
575 unsigned long flags;
576 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
577 int canrequest = 0;
578
579 if (!desc)
580 return 0;
581
582 if (irq_settings_can_request(desc)) {
583 if (!desc->action ||
584 irqflags & desc->action->flags & IRQF_SHARED)
585 canrequest = 1;
586 }
587 irq_put_desc_unlock(desc, flags);
588 return canrequest;
589}
590
591int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
592 unsigned long flags)
593{
594 struct irq_chip *chip = desc->irq_data.chip;
595 int ret, unmask = 0;
596
597 if (!chip || !chip->irq_set_type) {
598 /*
599 * IRQF_TRIGGER_* but the PIC does not support multiple
600 * flow-types?
601 */
602 pr_debug("No set_type function for IRQ %d (%s)\n", irq,
603 chip ? (chip->name ? : "unknown") : "unknown");
604 return 0;
605 }
606
607 flags &= IRQ_TYPE_SENSE_MASK;
608
609 if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
610 if (!irqd_irq_masked(&desc->irq_data))
611 mask_irq(desc);
612 if (!irqd_irq_disabled(&desc->irq_data))
613 unmask = 1;
614 }
615
616 /* caller masked out all except trigger mode flags */
617 ret = chip->irq_set_type(&desc->irq_data, flags);
618
619 switch (ret) {
620 case IRQ_SET_MASK_OK:
621 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
622 irqd_set(&desc->irq_data, flags);
623
624 case IRQ_SET_MASK_OK_NOCOPY:
625 flags = irqd_get_trigger_type(&desc->irq_data);
626 irq_settings_set_trigger_mask(desc, flags);
627 irqd_clear(&desc->irq_data, IRQD_LEVEL);
628 irq_settings_clr_level(desc);
629 if (flags & IRQ_TYPE_LEVEL_MASK) {
630 irq_settings_set_level(desc);
631 irqd_set(&desc->irq_data, IRQD_LEVEL);
632 }
633
634 ret = 0;
635 break;
636 default:
637 pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
638 flags, irq, chip->irq_set_type);
639 }
640 if (unmask)
641 unmask_irq(desc);
642 return ret;
643}
644
645#ifdef CONFIG_HARDIRQS_SW_RESEND
646int irq_set_parent(int irq, int parent_irq)
647{
648 unsigned long flags;
649 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
650
651 if (!desc)
652 return -EINVAL;
653
654 desc->parent_irq = parent_irq;
655
656 irq_put_desc_unlock(desc, flags);
657 return 0;
658}
659#endif
660
661/*
662 * Default primary interrupt handler for threaded interrupts. Is
663 * assigned as primary handler when request_threaded_irq is called
664 * with handler == NULL. Useful for oneshot interrupts.
665 */
666static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
667{
668 return IRQ_WAKE_THREAD;
669}
670
671/*
672 * Primary handler for nested threaded interrupts. Should never be
673 * called.
674 */
675static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
676{
677 WARN(1, "Primary handler called for nested irq %d\n", irq);
678 return IRQ_NONE;
679}
680
681static int irq_wait_for_interrupt(struct irqaction *action)
682{
683 set_current_state(TASK_INTERRUPTIBLE);
684
685 while (!kthread_should_stop()) {
686
687 if (test_and_clear_bit(IRQTF_RUNTHREAD,
688 &action->thread_flags)) {
689 __set_current_state(TASK_RUNNING);
690 return 0;
691 }
692 schedule();
693 set_current_state(TASK_INTERRUPTIBLE);
694 }
695 __set_current_state(TASK_RUNNING);
696 return -1;
697}
698
699/*
700 * Oneshot interrupts keep the irq line masked until the threaded
701 * handler finished. unmask if the interrupt has not been disabled and
702 * is marked MASKED.
703 */
704static void irq_finalize_oneshot(struct irq_desc *desc,
705 struct irqaction *action)
706{
707 if (!(desc->istate & IRQS_ONESHOT))
708 return;
709again:
710 chip_bus_lock(desc);
711 raw_spin_lock_irq(&desc->lock);
712
713 /*
714 * Implausible though it may be we need to protect us against
715 * the following scenario:
716 *
717 * The thread is faster done than the hard interrupt handler
718 * on the other CPU. If we unmask the irq line then the
719 * interrupt can come in again and masks the line, leaves due
720 * to IRQS_INPROGRESS and the irq line is masked forever.
721 *
722 * This also serializes the state of shared oneshot handlers
723 * versus "desc->threads_onehsot |= action->thread_mask;" in
724 * irq_wake_thread(). See the comment there which explains the
725 * serialization.
726 */
727 if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
728 raw_spin_unlock_irq(&desc->lock);
729 chip_bus_sync_unlock(desc);
730 cpu_relax();
731 goto again;
732 }
733
734 /*
735 * Now check again, whether the thread should run. Otherwise
736 * we would clear the threads_oneshot bit of this thread which
737 * was just set.
738 */
739 if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
740 goto out_unlock;
741
742 desc->threads_oneshot &= ~action->thread_mask;
743
744 if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
745 irqd_irq_masked(&desc->irq_data))
746 unmask_threaded_irq(desc);
747
748out_unlock:
749 raw_spin_unlock_irq(&desc->lock);
750 chip_bus_sync_unlock(desc);
751}
752
753#ifdef CONFIG_SMP
754/*
755 * Check whether we need to change the affinity of the interrupt thread.
756 */
757static void
758irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
759{
760 cpumask_var_t mask;
761 bool valid = true;
762
763 if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
764 return;
765
766 /*
767 * In case we are out of memory we set IRQTF_AFFINITY again and
768 * try again next time
769 */
770 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
771 set_bit(IRQTF_AFFINITY, &action->thread_flags);
772 return;
773 }
774
775 raw_spin_lock_irq(&desc->lock);
776 /*
777 * This code is triggered unconditionally. Check the affinity
778 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
779 */
780 if (desc->irq_data.affinity)
781 cpumask_copy(mask, desc->irq_data.affinity);
782 else
783 valid = false;
784 raw_spin_unlock_irq(&desc->lock);
785
786 if (valid)
787 set_cpus_allowed_ptr(current, mask);
788 free_cpumask_var(mask);
789}
790#else
791static inline void
792irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
793#endif
794
795/*
796 * Interrupts which are not explicitely requested as threaded
797 * interrupts rely on the implicit bh/preempt disable of the hard irq
798 * context. So we need to disable bh here to avoid deadlocks and other
799 * side effects.
800 */
801static irqreturn_t
802irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
803{
804 irqreturn_t ret;
805
806 local_bh_disable();
807 ret = action->thread_fn(action->irq, action->dev_id);
808 irq_finalize_oneshot(desc, action);
809 local_bh_enable();
810 return ret;
811}
812
813/*
814 * Interrupts explicitly requested as threaded interrupts want to be
815 * preemtible - many of them need to sleep and wait for slow busses to
816 * complete.
817 */
818static irqreturn_t irq_thread_fn(struct irq_desc *desc,
819 struct irqaction *action)
820{
821 irqreturn_t ret;
822
823 ret = action->thread_fn(action->irq, action->dev_id);
824 irq_finalize_oneshot(desc, action);
825 return ret;
826}
827
828static void wake_threads_waitq(struct irq_desc *desc)
829{
830 if (atomic_dec_and_test(&desc->threads_active))
831 wake_up(&desc->wait_for_threads);
832}
833
834static void irq_thread_dtor(struct callback_head *unused)
835{
836 struct task_struct *tsk = current;
837 struct irq_desc *desc;
838 struct irqaction *action;
839
840 if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
841 return;
842
843 action = kthread_data(tsk);
844
845 pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
846 tsk->comm, tsk->pid, action->irq);
847
848
849 desc = irq_to_desc(action->irq);
850 /*
851 * If IRQTF_RUNTHREAD is set, we need to decrement
852 * desc->threads_active and wake possible waiters.
853 */
854 if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
855 wake_threads_waitq(desc);
856
857 /* Prevent a stale desc->threads_oneshot */
858 irq_finalize_oneshot(desc, action);
859}
860
861/*
862 * Interrupt handler thread
863 */
864static int irq_thread(void *data)
865{
866 struct callback_head on_exit_work;
867 struct irqaction *action = data;
868 struct irq_desc *desc = irq_to_desc(action->irq);
869 irqreturn_t (*handler_fn)(struct irq_desc *desc,
870 struct irqaction *action);
871
872 if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
873 &action->thread_flags))
874 handler_fn = irq_forced_thread_fn;
875 else
876 handler_fn = irq_thread_fn;
877
878 init_task_work(&on_exit_work, irq_thread_dtor);
879 task_work_add(current, &on_exit_work, false);
880
881 irq_thread_check_affinity(desc, action);
882
883 while (!irq_wait_for_interrupt(action)) {
884 irqreturn_t action_ret;
885
886 irq_thread_check_affinity(desc, action);
887
888 action_ret = handler_fn(desc, action);
889 if (!noirqdebug)
890 note_interrupt(action->irq, desc, action_ret);
891
892 wake_threads_waitq(desc);
893 }
894
895 /*
896 * This is the regular exit path. __free_irq() is stopping the
897 * thread via kthread_stop() after calling
898 * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
899 * oneshot mask bit can be set. We cannot verify that as we
900 * cannot touch the oneshot mask at this point anymore as
901 * __setup_irq() might have given out currents thread_mask
902 * again.
903 */
904 task_work_cancel(current, irq_thread_dtor);
905 return 0;
906}
907
908/**
909 * irq_wake_thread - wake the irq thread for the action identified by dev_id
910 * @irq: Interrupt line
911 * @dev_id: Device identity for which the thread should be woken
912 *
913 */
914void irq_wake_thread(unsigned int irq, void *dev_id)
915{
916 struct irq_desc *desc = irq_to_desc(irq);
917 struct irqaction *action;
918 unsigned long flags;
919
920 if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
921 return;
922
923 raw_spin_lock_irqsave(&desc->lock, flags);
924 for (action = desc->action; action; action = action->next) {
925 if (action->dev_id == dev_id) {
926 if (action->thread)
927 __irq_wake_thread(desc, action);
928 break;
929 }
930 }
931 raw_spin_unlock_irqrestore(&desc->lock, flags);
932}
933EXPORT_SYMBOL_GPL(irq_wake_thread);
934
935static void irq_setup_forced_threading(struct irqaction *new)
936{
937 if (!force_irqthreads)
938 return;
939 if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
940 return;
941
942 new->flags |= IRQF_ONESHOT;
943
944 if (!new->thread_fn) {
945 set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
946 new->thread_fn = new->handler;
947 new->handler = irq_default_primary_handler;
948 }
949}
950
951static int irq_request_resources(struct irq_desc *desc)
952{
953 struct irq_data *d = &desc->irq_data;
954 struct irq_chip *c = d->chip;
955
956 return c->irq_request_resources ? c->irq_request_resources(d) : 0;
957}
958
959static void irq_release_resources(struct irq_desc *desc)
960{
961 struct irq_data *d = &desc->irq_data;
962 struct irq_chip *c = d->chip;
963
964 if (c->irq_release_resources)
965 c->irq_release_resources(d);
966}
967
968/*
969 * Internal function to register an irqaction - typically used to
970 * allocate special interrupts that are part of the architecture.
971 */
972static int
973__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
974{
975 struct irqaction *old, **old_ptr;
976 unsigned long flags, thread_mask = 0;
977 int ret, nested, shared = 0;
978 cpumask_var_t mask;
979
980 if (!desc)
981 return -EINVAL;
982
983 if (desc->irq_data.chip == &no_irq_chip)
984 return -ENOSYS;
985 if (!try_module_get(desc->owner))
986 return -ENODEV;
987
988 /*
989 * Check whether the interrupt nests into another interrupt
990 * thread.
991 */
992 nested = irq_settings_is_nested_thread(desc);
993 if (nested) {
994 if (!new->thread_fn) {
995 ret = -EINVAL;
996 goto out_mput;
997 }
998 /*
999 * Replace the primary handler which was provided from
1000 * the driver for non nested interrupt handling by the
1001 * dummy function which warns when called.
1002 */
1003 new->handler = irq_nested_primary_handler;
1004 } else {
1005 if (irq_settings_can_thread(desc))
1006 irq_setup_forced_threading(new);
1007 }
1008
1009 /*
1010 * Create a handler thread when a thread function is supplied
1011 * and the interrupt does not nest into another interrupt
1012 * thread.
1013 */
1014 if (new->thread_fn && !nested) {
1015 struct task_struct *t;
1016 static const struct sched_param param = {
1017 .sched_priority = MAX_USER_RT_PRIO/2,
1018 };
1019
1020 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1021 new->name);
1022 if (IS_ERR(t)) {
1023 ret = PTR_ERR(t);
1024 goto out_mput;
1025 }
1026
1027 sched_setscheduler_nocheck(t, SCHED_FIFO, ¶m);
1028
1029 /*
1030 * We keep the reference to the task struct even if
1031 * the thread dies to avoid that the interrupt code
1032 * references an already freed task_struct.
1033 */
1034 get_task_struct(t);
1035 new->thread = t;
1036 /*
1037 * Tell the thread to set its affinity. This is
1038 * important for shared interrupt handlers as we do
1039 * not invoke setup_affinity() for the secondary
1040 * handlers as everything is already set up. Even for
1041 * interrupts marked with IRQF_NO_BALANCE this is
1042 * correct as we want the thread to move to the cpu(s)
1043 * on which the requesting code placed the interrupt.
1044 */
1045 set_bit(IRQTF_AFFINITY, &new->thread_flags);
1046 }
1047
1048 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1049 ret = -ENOMEM;
1050 goto out_thread;
1051 }
1052
1053 /*
1054 * Drivers are often written to work w/o knowledge about the
1055 * underlying irq chip implementation, so a request for a
1056 * threaded irq without a primary hard irq context handler
1057 * requires the ONESHOT flag to be set. Some irq chips like
1058 * MSI based interrupts are per se one shot safe. Check the
1059 * chip flags, so we can avoid the unmask dance at the end of
1060 * the threaded handler for those.
1061 */
1062 if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1063 new->flags &= ~IRQF_ONESHOT;
1064
1065 /*
1066 * The following block of code has to be executed atomically
1067 */
1068 raw_spin_lock_irqsave(&desc->lock, flags);
1069 old_ptr = &desc->action;
1070 old = *old_ptr;
1071 if (old) {
1072 /*
1073 * Can't share interrupts unless both agree to and are
1074 * the same type (level, edge, polarity). So both flag
1075 * fields must have IRQF_SHARED set and the bits which
1076 * set the trigger type must match. Also all must
1077 * agree on ONESHOT.
1078 */
1079 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1080 ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1081 ((old->flags ^ new->flags) & IRQF_ONESHOT))
1082 goto mismatch;
1083
1084 /* All handlers must agree on per-cpuness */
1085 if ((old->flags & IRQF_PERCPU) !=
1086 (new->flags & IRQF_PERCPU))
1087 goto mismatch;
1088
1089 /* add new interrupt at end of irq queue */
1090 do {
1091 /*
1092 * Or all existing action->thread_mask bits,
1093 * so we can find the next zero bit for this
1094 * new action.
1095 */
1096 thread_mask |= old->thread_mask;
1097 old_ptr = &old->next;
1098 old = *old_ptr;
1099 } while (old);
1100 shared = 1;
1101 }
1102
1103 /*
1104 * Setup the thread mask for this irqaction for ONESHOT. For
1105 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1106 * conditional in irq_wake_thread().
1107 */
1108 if (new->flags & IRQF_ONESHOT) {
1109 /*
1110 * Unlikely to have 32 resp 64 irqs sharing one line,
1111 * but who knows.
1112 */
1113 if (thread_mask == ~0UL) {
1114 ret = -EBUSY;
1115 goto out_mask;
1116 }
1117 /*
1118 * The thread_mask for the action is or'ed to
1119 * desc->thread_active to indicate that the
1120 * IRQF_ONESHOT thread handler has been woken, but not
1121 * yet finished. The bit is cleared when a thread
1122 * completes. When all threads of a shared interrupt
1123 * line have completed desc->threads_active becomes
1124 * zero and the interrupt line is unmasked. See
1125 * handle.c:irq_wake_thread() for further information.
1126 *
1127 * If no thread is woken by primary (hard irq context)
1128 * interrupt handlers, then desc->threads_active is
1129 * also checked for zero to unmask the irq line in the
1130 * affected hard irq flow handlers
1131 * (handle_[fasteoi|level]_irq).
1132 *
1133 * The new action gets the first zero bit of
1134 * thread_mask assigned. See the loop above which or's
1135 * all existing action->thread_mask bits.
1136 */
1137 new->thread_mask = 1 << ffz(thread_mask);
1138
1139 } else if (new->handler == irq_default_primary_handler &&
1140 !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1141 /*
1142 * The interrupt was requested with handler = NULL, so
1143 * we use the default primary handler for it. But it
1144 * does not have the oneshot flag set. In combination
1145 * with level interrupts this is deadly, because the
1146 * default primary handler just wakes the thread, then
1147 * the irq lines is reenabled, but the device still
1148 * has the level irq asserted. Rinse and repeat....
1149 *
1150 * While this works for edge type interrupts, we play
1151 * it safe and reject unconditionally because we can't
1152 * say for sure which type this interrupt really
1153 * has. The type flags are unreliable as the
1154 * underlying chip implementation can override them.
1155 */
1156 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1157 irq);
1158 ret = -EINVAL;
1159 goto out_mask;
1160 }
1161
1162 if (!shared) {
1163 ret = irq_request_resources(desc);
1164 if (ret) {
1165 pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1166 new->name, irq, desc->irq_data.chip->name);
1167 goto out_mask;
1168 }
1169
1170 init_waitqueue_head(&desc->wait_for_threads);
1171
1172 /* Setup the type (level, edge polarity) if configured: */
1173 if (new->flags & IRQF_TRIGGER_MASK) {
1174 ret = __irq_set_trigger(desc, irq,
1175 new->flags & IRQF_TRIGGER_MASK);
1176
1177 if (ret)
1178 goto out_mask;
1179 }
1180
1181 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1182 IRQS_ONESHOT | IRQS_WAITING);
1183 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1184
1185 if (new->flags & IRQF_PERCPU) {
1186 irqd_set(&desc->irq_data, IRQD_PER_CPU);
1187 irq_settings_set_per_cpu(desc);
1188 }
1189
1190 if (new->flags & IRQF_ONESHOT)
1191 desc->istate |= IRQS_ONESHOT;
1192
1193 if (irq_settings_can_autoenable(desc))
1194 irq_startup(desc, true);
1195 else
1196 /* Undo nested disables: */
1197 desc->depth = 1;
1198
1199 /* Exclude IRQ from balancing if requested */
1200 if (new->flags & IRQF_NOBALANCING) {
1201 irq_settings_set_no_balancing(desc);
1202 irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1203 }
1204
1205 /* Set default affinity mask once everything is setup */
1206 setup_affinity(irq, desc, mask);
1207
1208 } else if (new->flags & IRQF_TRIGGER_MASK) {
1209 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1210 unsigned int omsk = irq_settings_get_trigger_mask(desc);
1211
1212 if (nmsk != omsk)
1213 /* hope the handler works with current trigger mode */
1214 pr_warning("irq %d uses trigger mode %u; requested %u\n",
1215 irq, nmsk, omsk);
1216 }
1217
1218 new->irq = irq;
1219 *old_ptr = new;
1220
1221 /* Reset broken irq detection when installing new handler */
1222 desc->irq_count = 0;
1223 desc->irqs_unhandled = 0;
1224
1225 /*
1226 * Check whether we disabled the irq via the spurious handler
1227 * before. Reenable it and give it another chance.
1228 */
1229 if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1230 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1231 __enable_irq(desc, irq, false);
1232 }
1233
1234 raw_spin_unlock_irqrestore(&desc->lock, flags);
1235
1236 /*
1237 * Strictly no need to wake it up, but hung_task complains
1238 * when no hard interrupt wakes the thread up.
1239 */
1240 if (new->thread)
1241 wake_up_process(new->thread);
1242
1243 register_irq_proc(irq, desc);
1244 new->dir = NULL;
1245 register_handler_proc(irq, new);
1246 free_cpumask_var(mask);
1247
1248 return 0;
1249
1250mismatch:
1251 if (!(new->flags & IRQF_PROBE_SHARED)) {
1252 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1253 irq, new->flags, new->name, old->flags, old->name);
1254#ifdef CONFIG_DEBUG_SHIRQ
1255 dump_stack();
1256#endif
1257 }
1258 ret = -EBUSY;
1259
1260out_mask:
1261 raw_spin_unlock_irqrestore(&desc->lock, flags);
1262 free_cpumask_var(mask);
1263
1264out_thread:
1265 if (new->thread) {
1266 struct task_struct *t = new->thread;
1267
1268 new->thread = NULL;
1269 kthread_stop(t);
1270 put_task_struct(t);
1271 }
1272out_mput:
1273 module_put(desc->owner);
1274 return ret;
1275}
1276
1277/**
1278 * setup_irq - setup an interrupt
1279 * @irq: Interrupt line to setup
1280 * @act: irqaction for the interrupt
1281 *
1282 * Used to statically setup interrupts in the early boot process.
1283 */
1284int setup_irq(unsigned int irq, struct irqaction *act)
1285{
1286 int retval;
1287 struct irq_desc *desc = irq_to_desc(irq);
1288
1289 if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1290 return -EINVAL;
1291 chip_bus_lock(desc);
1292 retval = __setup_irq(irq, desc, act);
1293 chip_bus_sync_unlock(desc);
1294
1295 return retval;
1296}
1297EXPORT_SYMBOL_GPL(setup_irq);
1298
1299/*
1300 * Internal function to unregister an irqaction - used to free
1301 * regular and special interrupts that are part of the architecture.
1302 */
1303static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1304{
1305 struct irq_desc *desc = irq_to_desc(irq);
1306 struct irqaction *action, **action_ptr;
1307 unsigned long flags;
1308
1309 WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1310
1311 if (!desc)
1312 return NULL;
1313
1314 raw_spin_lock_irqsave(&desc->lock, flags);
1315
1316 /*
1317 * There can be multiple actions per IRQ descriptor, find the right
1318 * one based on the dev_id:
1319 */
1320 action_ptr = &desc->action;
1321 for (;;) {
1322 action = *action_ptr;
1323
1324 if (!action) {
1325 WARN(1, "Trying to free already-free IRQ %d\n", irq);
1326 raw_spin_unlock_irqrestore(&desc->lock, flags);
1327
1328 return NULL;
1329 }
1330
1331 if (action->dev_id == dev_id)
1332 break;
1333 action_ptr = &action->next;
1334 }
1335
1336 /* Found it - now remove it from the list of entries: */
1337 *action_ptr = action->next;
1338
1339 /* If this was the last handler, shut down the IRQ line: */
1340 if (!desc->action) {
1341 irq_shutdown(desc);
1342 irq_release_resources(desc);
1343 }
1344
1345#ifdef CONFIG_SMP
1346 /* make sure affinity_hint is cleaned up */
1347 if (WARN_ON_ONCE(desc->affinity_hint))
1348 desc->affinity_hint = NULL;
1349#endif
1350
1351 raw_spin_unlock_irqrestore(&desc->lock, flags);
1352
1353 unregister_handler_proc(irq, action);
1354
1355 /* Make sure it's not being used on another CPU: */
1356 synchronize_irq(irq);
1357
1358#ifdef CONFIG_DEBUG_SHIRQ
1359 /*
1360 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1361 * event to happen even now it's being freed, so let's make sure that
1362 * is so by doing an extra call to the handler ....
1363 *
1364 * ( We do this after actually deregistering it, to make sure that a
1365 * 'real' IRQ doesn't run in * parallel with our fake. )
1366 */
1367 if (action->flags & IRQF_SHARED) {
1368 local_irq_save(flags);
1369 action->handler(irq, dev_id);
1370 local_irq_restore(flags);
1371 }
1372#endif
1373
1374 if (action->thread) {
1375 kthread_stop(action->thread);
1376 put_task_struct(action->thread);
1377 }
1378
1379 module_put(desc->owner);
1380 return action;
1381}
1382
1383/**
1384 * remove_irq - free an interrupt
1385 * @irq: Interrupt line to free
1386 * @act: irqaction for the interrupt
1387 *
1388 * Used to remove interrupts statically setup by the early boot process.
1389 */
1390void remove_irq(unsigned int irq, struct irqaction *act)
1391{
1392 struct irq_desc *desc = irq_to_desc(irq);
1393
1394 if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1395 __free_irq(irq, act->dev_id);
1396}
1397EXPORT_SYMBOL_GPL(remove_irq);
1398
1399/**
1400 * free_irq - free an interrupt allocated with request_irq
1401 * @irq: Interrupt line to free
1402 * @dev_id: Device identity to free
1403 *
1404 * Remove an interrupt handler. The handler is removed and if the
1405 * interrupt line is no longer in use by any driver it is disabled.
1406 * On a shared IRQ the caller must ensure the interrupt is disabled
1407 * on the card it drives before calling this function. The function
1408 * does not return until any executing interrupts for this IRQ
1409 * have completed.
1410 *
1411 * This function must not be called from interrupt context.
1412 */
1413void free_irq(unsigned int irq, void *dev_id)
1414{
1415 struct irq_desc *desc = irq_to_desc(irq);
1416
1417 if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1418 return;
1419
1420#ifdef CONFIG_SMP
1421 if (WARN_ON(desc->affinity_notify))
1422 desc->affinity_notify = NULL;
1423#endif
1424
1425 chip_bus_lock(desc);
1426 kfree(__free_irq(irq, dev_id));
1427 chip_bus_sync_unlock(desc);
1428}
1429EXPORT_SYMBOL(free_irq);
1430
1431/**
1432 * request_threaded_irq - allocate an interrupt line
1433 * @irq: Interrupt line to allocate
1434 * @handler: Function to be called when the IRQ occurs.
1435 * Primary handler for threaded interrupts
1436 * If NULL and thread_fn != NULL the default
1437 * primary handler is installed
1438 * @thread_fn: Function called from the irq handler thread
1439 * If NULL, no irq thread is created
1440 * @irqflags: Interrupt type flags
1441 * @devname: An ascii name for the claiming device
1442 * @dev_id: A cookie passed back to the handler function
1443 *
1444 * This call allocates interrupt resources and enables the
1445 * interrupt line and IRQ handling. From the point this
1446 * call is made your handler function may be invoked. Since
1447 * your handler function must clear any interrupt the board
1448 * raises, you must take care both to initialise your hardware
1449 * and to set up the interrupt handler in the right order.
1450 *
1451 * If you want to set up a threaded irq handler for your device
1452 * then you need to supply @handler and @thread_fn. @handler is
1453 * still called in hard interrupt context and has to check
1454 * whether the interrupt originates from the device. If yes it
1455 * needs to disable the interrupt on the device and return
1456 * IRQ_WAKE_THREAD which will wake up the handler thread and run
1457 * @thread_fn. This split handler design is necessary to support
1458 * shared interrupts.
1459 *
1460 * Dev_id must be globally unique. Normally the address of the
1461 * device data structure is used as the cookie. Since the handler
1462 * receives this value it makes sense to use it.
1463 *
1464 * If your interrupt is shared you must pass a non NULL dev_id
1465 * as this is required when freeing the interrupt.
1466 *
1467 * Flags:
1468 *
1469 * IRQF_SHARED Interrupt is shared
1470 * IRQF_TRIGGER_* Specify active edge(s) or level
1471 *
1472 */
1473int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1474 irq_handler_t thread_fn, unsigned long irqflags,
1475 const char *devname, void *dev_id)
1476{
1477 struct irqaction *action;
1478 struct irq_desc *desc;
1479 int retval;
1480
1481 /*
1482 * Sanity-check: shared interrupts must pass in a real dev-ID,
1483 * otherwise we'll have trouble later trying to figure out
1484 * which interrupt is which (messes up the interrupt freeing
1485 * logic etc).
1486 */
1487 if ((irqflags & IRQF_SHARED) && !dev_id)
1488 return -EINVAL;
1489
1490 desc = irq_to_desc(irq);
1491 if (!desc)
1492 return -EINVAL;
1493
1494 if (!irq_settings_can_request(desc) ||
1495 WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1496 return -EINVAL;
1497
1498 if (!handler) {
1499 if (!thread_fn)
1500 return -EINVAL;
1501 handler = irq_default_primary_handler;
1502 }
1503
1504 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1505 if (!action)
1506 return -ENOMEM;
1507
1508 action->handler = handler;
1509 action->thread_fn = thread_fn;
1510 action->flags = irqflags;
1511 action->name = devname;
1512 action->dev_id = dev_id;
1513
1514 chip_bus_lock(desc);
1515 retval = __setup_irq(irq, desc, action);
1516 chip_bus_sync_unlock(desc);
1517
1518 if (retval)
1519 kfree(action);
1520
1521#ifdef CONFIG_DEBUG_SHIRQ_FIXME
1522 if (!retval && (irqflags & IRQF_SHARED)) {
1523 /*
1524 * It's a shared IRQ -- the driver ought to be prepared for it
1525 * to happen immediately, so let's make sure....
1526 * We disable the irq to make sure that a 'real' IRQ doesn't
1527 * run in parallel with our fake.
1528 */
1529 unsigned long flags;
1530
1531 disable_irq(irq);
1532 local_irq_save(flags);
1533
1534 handler(irq, dev_id);
1535
1536 local_irq_restore(flags);
1537 enable_irq(irq);
1538 }
1539#endif
1540 return retval;
1541}
1542EXPORT_SYMBOL(request_threaded_irq);
1543
1544/**
1545 * request_any_context_irq - allocate an interrupt line
1546 * @irq: Interrupt line to allocate
1547 * @handler: Function to be called when the IRQ occurs.
1548 * Threaded handler for threaded interrupts.
1549 * @flags: Interrupt type flags
1550 * @name: An ascii name for the claiming device
1551 * @dev_id: A cookie passed back to the handler function
1552 *
1553 * This call allocates interrupt resources and enables the
1554 * interrupt line and IRQ handling. It selects either a
1555 * hardirq or threaded handling method depending on the
1556 * context.
1557 *
1558 * On failure, it returns a negative value. On success,
1559 * it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1560 */
1561int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1562 unsigned long flags, const char *name, void *dev_id)
1563{
1564 struct irq_desc *desc = irq_to_desc(irq);
1565 int ret;
1566
1567 if (!desc)
1568 return -EINVAL;
1569
1570 if (irq_settings_is_nested_thread(desc)) {
1571 ret = request_threaded_irq(irq, NULL, handler,
1572 flags, name, dev_id);
1573 return !ret ? IRQC_IS_NESTED : ret;
1574 }
1575
1576 ret = request_irq(irq, handler, flags, name, dev_id);
1577 return !ret ? IRQC_IS_HARDIRQ : ret;
1578}
1579EXPORT_SYMBOL_GPL(request_any_context_irq);
1580
1581void enable_percpu_irq(unsigned int irq, unsigned int type)
1582{
1583 unsigned int cpu = smp_processor_id();
1584 unsigned long flags;
1585 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1586
1587 if (!desc)
1588 return;
1589
1590 type &= IRQ_TYPE_SENSE_MASK;
1591 if (type != IRQ_TYPE_NONE) {
1592 int ret;
1593
1594 ret = __irq_set_trigger(desc, irq, type);
1595
1596 if (ret) {
1597 WARN(1, "failed to set type for IRQ%d\n", irq);
1598 goto out;
1599 }
1600 }
1601
1602 irq_percpu_enable(desc, cpu);
1603out:
1604 irq_put_desc_unlock(desc, flags);
1605}
1606EXPORT_SYMBOL_GPL(enable_percpu_irq);
1607
1608void disable_percpu_irq(unsigned int irq)
1609{
1610 unsigned int cpu = smp_processor_id();
1611 unsigned long flags;
1612 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1613
1614 if (!desc)
1615 return;
1616
1617 irq_percpu_disable(desc, cpu);
1618 irq_put_desc_unlock(desc, flags);
1619}
1620EXPORT_SYMBOL_GPL(disable_percpu_irq);
1621
1622/*
1623 * Internal function to unregister a percpu irqaction.
1624 */
1625static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1626{
1627 struct irq_desc *desc = irq_to_desc(irq);
1628 struct irqaction *action;
1629 unsigned long flags;
1630
1631 WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1632
1633 if (!desc)
1634 return NULL;
1635
1636 raw_spin_lock_irqsave(&desc->lock, flags);
1637
1638 action = desc->action;
1639 if (!action || action->percpu_dev_id != dev_id) {
1640 WARN(1, "Trying to free already-free IRQ %d\n", irq);
1641 goto bad;
1642 }
1643
1644 if (!cpumask_empty(desc->percpu_enabled)) {
1645 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1646 irq, cpumask_first(desc->percpu_enabled));
1647 goto bad;
1648 }
1649
1650 /* Found it - now remove it from the list of entries: */
1651 desc->action = NULL;
1652
1653 raw_spin_unlock_irqrestore(&desc->lock, flags);
1654
1655 unregister_handler_proc(irq, action);
1656
1657 module_put(desc->owner);
1658 return action;
1659
1660bad:
1661 raw_spin_unlock_irqrestore(&desc->lock, flags);
1662 return NULL;
1663}
1664
1665/**
1666 * remove_percpu_irq - free a per-cpu interrupt
1667 * @irq: Interrupt line to free
1668 * @act: irqaction for the interrupt
1669 *
1670 * Used to remove interrupts statically setup by the early boot process.
1671 */
1672void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1673{
1674 struct irq_desc *desc = irq_to_desc(irq);
1675
1676 if (desc && irq_settings_is_per_cpu_devid(desc))
1677 __free_percpu_irq(irq, act->percpu_dev_id);
1678}
1679
1680/**
1681 * free_percpu_irq - free an interrupt allocated with request_percpu_irq
1682 * @irq: Interrupt line to free
1683 * @dev_id: Device identity to free
1684 *
1685 * Remove a percpu interrupt handler. The handler is removed, but
1686 * the interrupt line is not disabled. This must be done on each
1687 * CPU before calling this function. The function does not return
1688 * until any executing interrupts for this IRQ have completed.
1689 *
1690 * This function must not be called from interrupt context.
1691 */
1692void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1693{
1694 struct irq_desc *desc = irq_to_desc(irq);
1695
1696 if (!desc || !irq_settings_is_per_cpu_devid(desc))
1697 return;
1698
1699 chip_bus_lock(desc);
1700 kfree(__free_percpu_irq(irq, dev_id));
1701 chip_bus_sync_unlock(desc);
1702}
1703
1704/**
1705 * setup_percpu_irq - setup a per-cpu interrupt
1706 * @irq: Interrupt line to setup
1707 * @act: irqaction for the interrupt
1708 *
1709 * Used to statically setup per-cpu interrupts in the early boot process.
1710 */
1711int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1712{
1713 struct irq_desc *desc = irq_to_desc(irq);
1714 int retval;
1715
1716 if (!desc || !irq_settings_is_per_cpu_devid(desc))
1717 return -EINVAL;
1718 chip_bus_lock(desc);
1719 retval = __setup_irq(irq, desc, act);
1720 chip_bus_sync_unlock(desc);
1721
1722 return retval;
1723}
1724
1725/**
1726 * request_percpu_irq - allocate a percpu interrupt line
1727 * @irq: Interrupt line to allocate
1728 * @handler: Function to be called when the IRQ occurs.
1729 * @devname: An ascii name for the claiming device
1730 * @dev_id: A percpu cookie passed back to the handler function
1731 *
1732 * This call allocates interrupt resources, but doesn't
1733 * automatically enable the interrupt. It has to be done on each
1734 * CPU using enable_percpu_irq().
1735 *
1736 * Dev_id must be globally unique. It is a per-cpu variable, and
1737 * the handler gets called with the interrupted CPU's instance of
1738 * that variable.
1739 */
1740int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1741 const char *devname, void __percpu *dev_id)
1742{
1743 struct irqaction *action;
1744 struct irq_desc *desc;
1745 int retval;
1746
1747 if (!dev_id)
1748 return -EINVAL;
1749
1750 desc = irq_to_desc(irq);
1751 if (!desc || !irq_settings_can_request(desc) ||
1752 !irq_settings_is_per_cpu_devid(desc))
1753 return -EINVAL;
1754
1755 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1756 if (!action)
1757 return -ENOMEM;
1758
1759 action->handler = handler;
1760 action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1761 action->name = devname;
1762 action->percpu_dev_id = dev_id;
1763
1764 chip_bus_lock(desc);
1765 retval = __setup_irq(irq, desc, action);
1766 chip_bus_sync_unlock(desc);
1767
1768 if (retval)
1769 kfree(action);
1770
1771 return retval;
1772}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
4 * Copyright (C) 2005-2006 Thomas Gleixner
5 *
6 * This file contains driver APIs to the irq subsystem.
7 */
8
9#define pr_fmt(fmt) "genirq: " fmt
10
11#include <linux/irq.h>
12#include <linux/kthread.h>
13#include <linux/module.h>
14#include <linux/random.h>
15#include <linux/interrupt.h>
16#include <linux/irqdomain.h>
17#include <linux/slab.h>
18#include <linux/sched.h>
19#include <linux/sched/rt.h>
20#include <linux/sched/task.h>
21#include <linux/sched/isolation.h>
22#include <uapi/linux/sched/types.h>
23#include <linux/task_work.h>
24
25#include "internals.h"
26
27#if defined(CONFIG_IRQ_FORCED_THREADING) && !defined(CONFIG_PREEMPT_RT)
28__read_mostly bool force_irqthreads;
29EXPORT_SYMBOL_GPL(force_irqthreads);
30
31static int __init setup_forced_irqthreads(char *arg)
32{
33 force_irqthreads = true;
34 return 0;
35}
36early_param("threadirqs", setup_forced_irqthreads);
37#endif
38
39static void __synchronize_hardirq(struct irq_desc *desc, bool sync_chip)
40{
41 struct irq_data *irqd = irq_desc_get_irq_data(desc);
42 bool inprogress;
43
44 do {
45 unsigned long flags;
46
47 /*
48 * Wait until we're out of the critical section. This might
49 * give the wrong answer due to the lack of memory barriers.
50 */
51 while (irqd_irq_inprogress(&desc->irq_data))
52 cpu_relax();
53
54 /* Ok, that indicated we're done: double-check carefully. */
55 raw_spin_lock_irqsave(&desc->lock, flags);
56 inprogress = irqd_irq_inprogress(&desc->irq_data);
57
58 /*
59 * If requested and supported, check at the chip whether it
60 * is in flight at the hardware level, i.e. already pending
61 * in a CPU and waiting for service and acknowledge.
62 */
63 if (!inprogress && sync_chip) {
64 /*
65 * Ignore the return code. inprogress is only updated
66 * when the chip supports it.
67 */
68 __irq_get_irqchip_state(irqd, IRQCHIP_STATE_ACTIVE,
69 &inprogress);
70 }
71 raw_spin_unlock_irqrestore(&desc->lock, flags);
72
73 /* Oops, that failed? */
74 } while (inprogress);
75}
76
77/**
78 * synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
79 * @irq: interrupt number to wait for
80 *
81 * This function waits for any pending hard IRQ handlers for this
82 * interrupt to complete before returning. If you use this
83 * function while holding a resource the IRQ handler may need you
84 * will deadlock. It does not take associated threaded handlers
85 * into account.
86 *
87 * Do not use this for shutdown scenarios where you must be sure
88 * that all parts (hardirq and threaded handler) have completed.
89 *
90 * Returns: false if a threaded handler is active.
91 *
92 * This function may be called - with care - from IRQ context.
93 *
94 * It does not check whether there is an interrupt in flight at the
95 * hardware level, but not serviced yet, as this might deadlock when
96 * called with interrupts disabled and the target CPU of the interrupt
97 * is the current CPU.
98 */
99bool synchronize_hardirq(unsigned int irq)
100{
101 struct irq_desc *desc = irq_to_desc(irq);
102
103 if (desc) {
104 __synchronize_hardirq(desc, false);
105 return !atomic_read(&desc->threads_active);
106 }
107
108 return true;
109}
110EXPORT_SYMBOL(synchronize_hardirq);
111
112/**
113 * synchronize_irq - wait for pending IRQ handlers (on other CPUs)
114 * @irq: interrupt number to wait for
115 *
116 * This function waits for any pending IRQ handlers for this interrupt
117 * to complete before returning. If you use this function while
118 * holding a resource the IRQ handler may need you will deadlock.
119 *
120 * Can only be called from preemptible code as it might sleep when
121 * an interrupt thread is associated to @irq.
122 *
123 * It optionally makes sure (when the irq chip supports that method)
124 * that the interrupt is not pending in any CPU and waiting for
125 * service.
126 */
127void synchronize_irq(unsigned int irq)
128{
129 struct irq_desc *desc = irq_to_desc(irq);
130
131 if (desc) {
132 __synchronize_hardirq(desc, true);
133 /*
134 * We made sure that no hardirq handler is
135 * running. Now verify that no threaded handlers are
136 * active.
137 */
138 wait_event(desc->wait_for_threads,
139 !atomic_read(&desc->threads_active));
140 }
141}
142EXPORT_SYMBOL(synchronize_irq);
143
144#ifdef CONFIG_SMP
145cpumask_var_t irq_default_affinity;
146
147static bool __irq_can_set_affinity(struct irq_desc *desc)
148{
149 if (!desc || !irqd_can_balance(&desc->irq_data) ||
150 !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
151 return false;
152 return true;
153}
154
155/**
156 * irq_can_set_affinity - Check if the affinity of a given irq can be set
157 * @irq: Interrupt to check
158 *
159 */
160int irq_can_set_affinity(unsigned int irq)
161{
162 return __irq_can_set_affinity(irq_to_desc(irq));
163}
164
165/**
166 * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
167 * @irq: Interrupt to check
168 *
169 * Like irq_can_set_affinity() above, but additionally checks for the
170 * AFFINITY_MANAGED flag.
171 */
172bool irq_can_set_affinity_usr(unsigned int irq)
173{
174 struct irq_desc *desc = irq_to_desc(irq);
175
176 return __irq_can_set_affinity(desc) &&
177 !irqd_affinity_is_managed(&desc->irq_data);
178}
179
180/**
181 * irq_set_thread_affinity - Notify irq threads to adjust affinity
182 * @desc: irq descriptor which has affitnity changed
183 *
184 * We just set IRQTF_AFFINITY and delegate the affinity setting
185 * to the interrupt thread itself. We can not call
186 * set_cpus_allowed_ptr() here as we hold desc->lock and this
187 * code can be called from hard interrupt context.
188 */
189void irq_set_thread_affinity(struct irq_desc *desc)
190{
191 struct irqaction *action;
192
193 for_each_action_of_desc(desc, action)
194 if (action->thread)
195 set_bit(IRQTF_AFFINITY, &action->thread_flags);
196}
197
198#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
199static void irq_validate_effective_affinity(struct irq_data *data)
200{
201 const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
202 struct irq_chip *chip = irq_data_get_irq_chip(data);
203
204 if (!cpumask_empty(m))
205 return;
206 pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
207 chip->name, data->irq);
208}
209
210static inline void irq_init_effective_affinity(struct irq_data *data,
211 const struct cpumask *mask)
212{
213 cpumask_copy(irq_data_get_effective_affinity_mask(data), mask);
214}
215#else
216static inline void irq_validate_effective_affinity(struct irq_data *data) { }
217static inline void irq_init_effective_affinity(struct irq_data *data,
218 const struct cpumask *mask) { }
219#endif
220
221int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
222 bool force)
223{
224 struct irq_desc *desc = irq_data_to_desc(data);
225 struct irq_chip *chip = irq_data_get_irq_chip(data);
226 int ret;
227
228 if (!chip || !chip->irq_set_affinity)
229 return -EINVAL;
230
231 /*
232 * If this is a managed interrupt and housekeeping is enabled on
233 * it check whether the requested affinity mask intersects with
234 * a housekeeping CPU. If so, then remove the isolated CPUs from
235 * the mask and just keep the housekeeping CPU(s). This prevents
236 * the affinity setter from routing the interrupt to an isolated
237 * CPU to avoid that I/O submitted from a housekeeping CPU causes
238 * interrupts on an isolated one.
239 *
240 * If the masks do not intersect or include online CPU(s) then
241 * keep the requested mask. The isolated target CPUs are only
242 * receiving interrupts when the I/O operation was submitted
243 * directly from them.
244 *
245 * If all housekeeping CPUs in the affinity mask are offline, the
246 * interrupt will be migrated by the CPU hotplug code once a
247 * housekeeping CPU which belongs to the affinity mask comes
248 * online.
249 */
250 if (irqd_affinity_is_managed(data) &&
251 housekeeping_enabled(HK_FLAG_MANAGED_IRQ)) {
252 const struct cpumask *hk_mask, *prog_mask;
253
254 static DEFINE_RAW_SPINLOCK(tmp_mask_lock);
255 static struct cpumask tmp_mask;
256
257 hk_mask = housekeeping_cpumask(HK_FLAG_MANAGED_IRQ);
258
259 raw_spin_lock(&tmp_mask_lock);
260 cpumask_and(&tmp_mask, mask, hk_mask);
261 if (!cpumask_intersects(&tmp_mask, cpu_online_mask))
262 prog_mask = mask;
263 else
264 prog_mask = &tmp_mask;
265 ret = chip->irq_set_affinity(data, prog_mask, force);
266 raw_spin_unlock(&tmp_mask_lock);
267 } else {
268 ret = chip->irq_set_affinity(data, mask, force);
269 }
270 switch (ret) {
271 case IRQ_SET_MASK_OK:
272 case IRQ_SET_MASK_OK_DONE:
273 cpumask_copy(desc->irq_common_data.affinity, mask);
274 fallthrough;
275 case IRQ_SET_MASK_OK_NOCOPY:
276 irq_validate_effective_affinity(data);
277 irq_set_thread_affinity(desc);
278 ret = 0;
279 }
280
281 return ret;
282}
283
284#ifdef CONFIG_GENERIC_PENDING_IRQ
285static inline int irq_set_affinity_pending(struct irq_data *data,
286 const struct cpumask *dest)
287{
288 struct irq_desc *desc = irq_data_to_desc(data);
289
290 irqd_set_move_pending(data);
291 irq_copy_pending(desc, dest);
292 return 0;
293}
294#else
295static inline int irq_set_affinity_pending(struct irq_data *data,
296 const struct cpumask *dest)
297{
298 return -EBUSY;
299}
300#endif
301
302static int irq_try_set_affinity(struct irq_data *data,
303 const struct cpumask *dest, bool force)
304{
305 int ret = irq_do_set_affinity(data, dest, force);
306
307 /*
308 * In case that the underlying vector management is busy and the
309 * architecture supports the generic pending mechanism then utilize
310 * this to avoid returning an error to user space.
311 */
312 if (ret == -EBUSY && !force)
313 ret = irq_set_affinity_pending(data, dest);
314 return ret;
315}
316
317static bool irq_set_affinity_deactivated(struct irq_data *data,
318 const struct cpumask *mask, bool force)
319{
320 struct irq_desc *desc = irq_data_to_desc(data);
321
322 /*
323 * Handle irq chips which can handle affinity only in activated
324 * state correctly
325 *
326 * If the interrupt is not yet activated, just store the affinity
327 * mask and do not call the chip driver at all. On activation the
328 * driver has to make sure anyway that the interrupt is in a
329 * useable state so startup works.
330 */
331 if (!IS_ENABLED(CONFIG_IRQ_DOMAIN_HIERARCHY) ||
332 irqd_is_activated(data) || !irqd_affinity_on_activate(data))
333 return false;
334
335 cpumask_copy(desc->irq_common_data.affinity, mask);
336 irq_init_effective_affinity(data, mask);
337 irqd_set(data, IRQD_AFFINITY_SET);
338 return true;
339}
340
341int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
342 bool force)
343{
344 struct irq_chip *chip = irq_data_get_irq_chip(data);
345 struct irq_desc *desc = irq_data_to_desc(data);
346 int ret = 0;
347
348 if (!chip || !chip->irq_set_affinity)
349 return -EINVAL;
350
351 if (irq_set_affinity_deactivated(data, mask, force))
352 return 0;
353
354 if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
355 ret = irq_try_set_affinity(data, mask, force);
356 } else {
357 irqd_set_move_pending(data);
358 irq_copy_pending(desc, mask);
359 }
360
361 if (desc->affinity_notify) {
362 kref_get(&desc->affinity_notify->kref);
363 if (!schedule_work(&desc->affinity_notify->work)) {
364 /* Work was already scheduled, drop our extra ref */
365 kref_put(&desc->affinity_notify->kref,
366 desc->affinity_notify->release);
367 }
368 }
369 irqd_set(data, IRQD_AFFINITY_SET);
370
371 return ret;
372}
373
374int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
375{
376 struct irq_desc *desc = irq_to_desc(irq);
377 unsigned long flags;
378 int ret;
379
380 if (!desc)
381 return -EINVAL;
382
383 raw_spin_lock_irqsave(&desc->lock, flags);
384 ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
385 raw_spin_unlock_irqrestore(&desc->lock, flags);
386 return ret;
387}
388
389int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
390{
391 unsigned long flags;
392 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
393
394 if (!desc)
395 return -EINVAL;
396 desc->affinity_hint = m;
397 irq_put_desc_unlock(desc, flags);
398 /* set the initial affinity to prevent every interrupt being on CPU0 */
399 if (m)
400 __irq_set_affinity(irq, m, false);
401 return 0;
402}
403EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
404
405static void irq_affinity_notify(struct work_struct *work)
406{
407 struct irq_affinity_notify *notify =
408 container_of(work, struct irq_affinity_notify, work);
409 struct irq_desc *desc = irq_to_desc(notify->irq);
410 cpumask_var_t cpumask;
411 unsigned long flags;
412
413 if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
414 goto out;
415
416 raw_spin_lock_irqsave(&desc->lock, flags);
417 if (irq_move_pending(&desc->irq_data))
418 irq_get_pending(cpumask, desc);
419 else
420 cpumask_copy(cpumask, desc->irq_common_data.affinity);
421 raw_spin_unlock_irqrestore(&desc->lock, flags);
422
423 notify->notify(notify, cpumask);
424
425 free_cpumask_var(cpumask);
426out:
427 kref_put(¬ify->kref, notify->release);
428}
429
430/**
431 * irq_set_affinity_notifier - control notification of IRQ affinity changes
432 * @irq: Interrupt for which to enable/disable notification
433 * @notify: Context for notification, or %NULL to disable
434 * notification. Function pointers must be initialised;
435 * the other fields will be initialised by this function.
436 *
437 * Must be called in process context. Notification may only be enabled
438 * after the IRQ is allocated and must be disabled before the IRQ is
439 * freed using free_irq().
440 */
441int
442irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
443{
444 struct irq_desc *desc = irq_to_desc(irq);
445 struct irq_affinity_notify *old_notify;
446 unsigned long flags;
447
448 /* The release function is promised process context */
449 might_sleep();
450
451 if (!desc || desc->istate & IRQS_NMI)
452 return -EINVAL;
453
454 /* Complete initialisation of *notify */
455 if (notify) {
456 notify->irq = irq;
457 kref_init(¬ify->kref);
458 INIT_WORK(¬ify->work, irq_affinity_notify);
459 }
460
461 raw_spin_lock_irqsave(&desc->lock, flags);
462 old_notify = desc->affinity_notify;
463 desc->affinity_notify = notify;
464 raw_spin_unlock_irqrestore(&desc->lock, flags);
465
466 if (old_notify) {
467 if (cancel_work_sync(&old_notify->work)) {
468 /* Pending work had a ref, put that one too */
469 kref_put(&old_notify->kref, old_notify->release);
470 }
471 kref_put(&old_notify->kref, old_notify->release);
472 }
473
474 return 0;
475}
476EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
477
478#ifndef CONFIG_AUTO_IRQ_AFFINITY
479/*
480 * Generic version of the affinity autoselector.
481 */
482int irq_setup_affinity(struct irq_desc *desc)
483{
484 struct cpumask *set = irq_default_affinity;
485 int ret, node = irq_desc_get_node(desc);
486 static DEFINE_RAW_SPINLOCK(mask_lock);
487 static struct cpumask mask;
488
489 /* Excludes PER_CPU and NO_BALANCE interrupts */
490 if (!__irq_can_set_affinity(desc))
491 return 0;
492
493 raw_spin_lock(&mask_lock);
494 /*
495 * Preserve the managed affinity setting and a userspace affinity
496 * setup, but make sure that one of the targets is online.
497 */
498 if (irqd_affinity_is_managed(&desc->irq_data) ||
499 irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
500 if (cpumask_intersects(desc->irq_common_data.affinity,
501 cpu_online_mask))
502 set = desc->irq_common_data.affinity;
503 else
504 irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
505 }
506
507 cpumask_and(&mask, cpu_online_mask, set);
508 if (cpumask_empty(&mask))
509 cpumask_copy(&mask, cpu_online_mask);
510
511 if (node != NUMA_NO_NODE) {
512 const struct cpumask *nodemask = cpumask_of_node(node);
513
514 /* make sure at least one of the cpus in nodemask is online */
515 if (cpumask_intersects(&mask, nodemask))
516 cpumask_and(&mask, &mask, nodemask);
517 }
518 ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
519 raw_spin_unlock(&mask_lock);
520 return ret;
521}
522#else
523/* Wrapper for ALPHA specific affinity selector magic */
524int irq_setup_affinity(struct irq_desc *desc)
525{
526 return irq_select_affinity(irq_desc_get_irq(desc));
527}
528#endif /* CONFIG_AUTO_IRQ_AFFINITY */
529#endif /* CONFIG_SMP */
530
531
532/**
533 * irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
534 * @irq: interrupt number to set affinity
535 * @vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
536 * specific data for percpu_devid interrupts
537 *
538 * This function uses the vCPU specific data to set the vCPU
539 * affinity for an irq. The vCPU specific data is passed from
540 * outside, such as KVM. One example code path is as below:
541 * KVM -> IOMMU -> irq_set_vcpu_affinity().
542 */
543int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
544{
545 unsigned long flags;
546 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
547 struct irq_data *data;
548 struct irq_chip *chip;
549 int ret = -ENOSYS;
550
551 if (!desc)
552 return -EINVAL;
553
554 data = irq_desc_get_irq_data(desc);
555 do {
556 chip = irq_data_get_irq_chip(data);
557 if (chip && chip->irq_set_vcpu_affinity)
558 break;
559#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
560 data = data->parent_data;
561#else
562 data = NULL;
563#endif
564 } while (data);
565
566 if (data)
567 ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
568 irq_put_desc_unlock(desc, flags);
569
570 return ret;
571}
572EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
573
574void __disable_irq(struct irq_desc *desc)
575{
576 if (!desc->depth++)
577 irq_disable(desc);
578}
579
580static int __disable_irq_nosync(unsigned int irq)
581{
582 unsigned long flags;
583 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
584
585 if (!desc)
586 return -EINVAL;
587 __disable_irq(desc);
588 irq_put_desc_busunlock(desc, flags);
589 return 0;
590}
591
592/**
593 * disable_irq_nosync - disable an irq without waiting
594 * @irq: Interrupt to disable
595 *
596 * Disable the selected interrupt line. Disables and Enables are
597 * nested.
598 * Unlike disable_irq(), this function does not ensure existing
599 * instances of the IRQ handler have completed before returning.
600 *
601 * This function may be called from IRQ context.
602 */
603void disable_irq_nosync(unsigned int irq)
604{
605 __disable_irq_nosync(irq);
606}
607EXPORT_SYMBOL(disable_irq_nosync);
608
609/**
610 * disable_irq - disable an irq and wait for completion
611 * @irq: Interrupt to disable
612 *
613 * Disable the selected interrupt line. Enables and Disables are
614 * nested.
615 * This function waits for any pending IRQ handlers for this interrupt
616 * to complete before returning. If you use this function while
617 * holding a resource the IRQ handler may need you will deadlock.
618 *
619 * This function may be called - with care - from IRQ context.
620 */
621void disable_irq(unsigned int irq)
622{
623 if (!__disable_irq_nosync(irq))
624 synchronize_irq(irq);
625}
626EXPORT_SYMBOL(disable_irq);
627
628/**
629 * disable_hardirq - disables an irq and waits for hardirq completion
630 * @irq: Interrupt to disable
631 *
632 * Disable the selected interrupt line. Enables and Disables are
633 * nested.
634 * This function waits for any pending hard IRQ handlers for this
635 * interrupt to complete before returning. If you use this function while
636 * holding a resource the hard IRQ handler may need you will deadlock.
637 *
638 * When used to optimistically disable an interrupt from atomic context
639 * the return value must be checked.
640 *
641 * Returns: false if a threaded handler is active.
642 *
643 * This function may be called - with care - from IRQ context.
644 */
645bool disable_hardirq(unsigned int irq)
646{
647 if (!__disable_irq_nosync(irq))
648 return synchronize_hardirq(irq);
649
650 return false;
651}
652EXPORT_SYMBOL_GPL(disable_hardirq);
653
654/**
655 * disable_nmi_nosync - disable an nmi without waiting
656 * @irq: Interrupt to disable
657 *
658 * Disable the selected interrupt line. Disables and enables are
659 * nested.
660 * The interrupt to disable must have been requested through request_nmi.
661 * Unlike disable_nmi(), this function does not ensure existing
662 * instances of the IRQ handler have completed before returning.
663 */
664void disable_nmi_nosync(unsigned int irq)
665{
666 disable_irq_nosync(irq);
667}
668
669void __enable_irq(struct irq_desc *desc)
670{
671 switch (desc->depth) {
672 case 0:
673 err_out:
674 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
675 irq_desc_get_irq(desc));
676 break;
677 case 1: {
678 if (desc->istate & IRQS_SUSPENDED)
679 goto err_out;
680 /* Prevent probing on this irq: */
681 irq_settings_set_noprobe(desc);
682 /*
683 * Call irq_startup() not irq_enable() here because the
684 * interrupt might be marked NOAUTOEN. So irq_startup()
685 * needs to be invoked when it gets enabled the first
686 * time. If it was already started up, then irq_startup()
687 * will invoke irq_enable() under the hood.
688 */
689 irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
690 break;
691 }
692 default:
693 desc->depth--;
694 }
695}
696
697/**
698 * enable_irq - enable handling of an irq
699 * @irq: Interrupt to enable
700 *
701 * Undoes the effect of one call to disable_irq(). If this
702 * matches the last disable, processing of interrupts on this
703 * IRQ line is re-enabled.
704 *
705 * This function may be called from IRQ context only when
706 * desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
707 */
708void enable_irq(unsigned int irq)
709{
710 unsigned long flags;
711 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
712
713 if (!desc)
714 return;
715 if (WARN(!desc->irq_data.chip,
716 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
717 goto out;
718
719 __enable_irq(desc);
720out:
721 irq_put_desc_busunlock(desc, flags);
722}
723EXPORT_SYMBOL(enable_irq);
724
725/**
726 * enable_nmi - enable handling of an nmi
727 * @irq: Interrupt to enable
728 *
729 * The interrupt to enable must have been requested through request_nmi.
730 * Undoes the effect of one call to disable_nmi(). If this
731 * matches the last disable, processing of interrupts on this
732 * IRQ line is re-enabled.
733 */
734void enable_nmi(unsigned int irq)
735{
736 enable_irq(irq);
737}
738
739static int set_irq_wake_real(unsigned int irq, unsigned int on)
740{
741 struct irq_desc *desc = irq_to_desc(irq);
742 int ret = -ENXIO;
743
744 if (irq_desc_get_chip(desc)->flags & IRQCHIP_SKIP_SET_WAKE)
745 return 0;
746
747 if (desc->irq_data.chip->irq_set_wake)
748 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
749
750 return ret;
751}
752
753/**
754 * irq_set_irq_wake - control irq power management wakeup
755 * @irq: interrupt to control
756 * @on: enable/disable power management wakeup
757 *
758 * Enable/disable power management wakeup mode, which is
759 * disabled by default. Enables and disables must match,
760 * just as they match for non-wakeup mode support.
761 *
762 * Wakeup mode lets this IRQ wake the system from sleep
763 * states like "suspend to RAM".
764 *
765 * Note: irq enable/disable state is completely orthogonal
766 * to the enable/disable state of irq wake. An irq can be
767 * disabled with disable_irq() and still wake the system as
768 * long as the irq has wake enabled. If this does not hold,
769 * then the underlying irq chip and the related driver need
770 * to be investigated.
771 */
772int irq_set_irq_wake(unsigned int irq, unsigned int on)
773{
774 unsigned long flags;
775 struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
776 int ret = 0;
777
778 if (!desc)
779 return -EINVAL;
780
781 /* Don't use NMIs as wake up interrupts please */
782 if (desc->istate & IRQS_NMI) {
783 ret = -EINVAL;
784 goto out_unlock;
785 }
786
787 /* wakeup-capable irqs can be shared between drivers that
788 * don't need to have the same sleep mode behaviors.
789 */
790 if (on) {
791 if (desc->wake_depth++ == 0) {
792 ret = set_irq_wake_real(irq, on);
793 if (ret)
794 desc->wake_depth = 0;
795 else
796 irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
797 }
798 } else {
799 if (desc->wake_depth == 0) {
800 WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
801 } else if (--desc->wake_depth == 0) {
802 ret = set_irq_wake_real(irq, on);
803 if (ret)
804 desc->wake_depth = 1;
805 else
806 irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
807 }
808 }
809
810out_unlock:
811 irq_put_desc_busunlock(desc, flags);
812 return ret;
813}
814EXPORT_SYMBOL(irq_set_irq_wake);
815
816/*
817 * Internal function that tells the architecture code whether a
818 * particular irq has been exclusively allocated or is available
819 * for driver use.
820 */
821int can_request_irq(unsigned int irq, unsigned long irqflags)
822{
823 unsigned long flags;
824 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
825 int canrequest = 0;
826
827 if (!desc)
828 return 0;
829
830 if (irq_settings_can_request(desc)) {
831 if (!desc->action ||
832 irqflags & desc->action->flags & IRQF_SHARED)
833 canrequest = 1;
834 }
835 irq_put_desc_unlock(desc, flags);
836 return canrequest;
837}
838
839int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
840{
841 struct irq_chip *chip = desc->irq_data.chip;
842 int ret, unmask = 0;
843
844 if (!chip || !chip->irq_set_type) {
845 /*
846 * IRQF_TRIGGER_* but the PIC does not support multiple
847 * flow-types?
848 */
849 pr_debug("No set_type function for IRQ %d (%s)\n",
850 irq_desc_get_irq(desc),
851 chip ? (chip->name ? : "unknown") : "unknown");
852 return 0;
853 }
854
855 if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
856 if (!irqd_irq_masked(&desc->irq_data))
857 mask_irq(desc);
858 if (!irqd_irq_disabled(&desc->irq_data))
859 unmask = 1;
860 }
861
862 /* Mask all flags except trigger mode */
863 flags &= IRQ_TYPE_SENSE_MASK;
864 ret = chip->irq_set_type(&desc->irq_data, flags);
865
866 switch (ret) {
867 case IRQ_SET_MASK_OK:
868 case IRQ_SET_MASK_OK_DONE:
869 irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
870 irqd_set(&desc->irq_data, flags);
871 fallthrough;
872
873 case IRQ_SET_MASK_OK_NOCOPY:
874 flags = irqd_get_trigger_type(&desc->irq_data);
875 irq_settings_set_trigger_mask(desc, flags);
876 irqd_clear(&desc->irq_data, IRQD_LEVEL);
877 irq_settings_clr_level(desc);
878 if (flags & IRQ_TYPE_LEVEL_MASK) {
879 irq_settings_set_level(desc);
880 irqd_set(&desc->irq_data, IRQD_LEVEL);
881 }
882
883 ret = 0;
884 break;
885 default:
886 pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
887 flags, irq_desc_get_irq(desc), chip->irq_set_type);
888 }
889 if (unmask)
890 unmask_irq(desc);
891 return ret;
892}
893
894#ifdef CONFIG_HARDIRQS_SW_RESEND
895int irq_set_parent(int irq, int parent_irq)
896{
897 unsigned long flags;
898 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
899
900 if (!desc)
901 return -EINVAL;
902
903 desc->parent_irq = parent_irq;
904
905 irq_put_desc_unlock(desc, flags);
906 return 0;
907}
908EXPORT_SYMBOL_GPL(irq_set_parent);
909#endif
910
911/*
912 * Default primary interrupt handler for threaded interrupts. Is
913 * assigned as primary handler when request_threaded_irq is called
914 * with handler == NULL. Useful for oneshot interrupts.
915 */
916static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
917{
918 return IRQ_WAKE_THREAD;
919}
920
921/*
922 * Primary handler for nested threaded interrupts. Should never be
923 * called.
924 */
925static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
926{
927 WARN(1, "Primary handler called for nested irq %d\n", irq);
928 return IRQ_NONE;
929}
930
931static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
932{
933 WARN(1, "Secondary action handler called for irq %d\n", irq);
934 return IRQ_NONE;
935}
936
937static int irq_wait_for_interrupt(struct irqaction *action)
938{
939 for (;;) {
940 set_current_state(TASK_INTERRUPTIBLE);
941
942 if (kthread_should_stop()) {
943 /* may need to run one last time */
944 if (test_and_clear_bit(IRQTF_RUNTHREAD,
945 &action->thread_flags)) {
946 __set_current_state(TASK_RUNNING);
947 return 0;
948 }
949 __set_current_state(TASK_RUNNING);
950 return -1;
951 }
952
953 if (test_and_clear_bit(IRQTF_RUNTHREAD,
954 &action->thread_flags)) {
955 __set_current_state(TASK_RUNNING);
956 return 0;
957 }
958 schedule();
959 }
960}
961
962/*
963 * Oneshot interrupts keep the irq line masked until the threaded
964 * handler finished. unmask if the interrupt has not been disabled and
965 * is marked MASKED.
966 */
967static void irq_finalize_oneshot(struct irq_desc *desc,
968 struct irqaction *action)
969{
970 if (!(desc->istate & IRQS_ONESHOT) ||
971 action->handler == irq_forced_secondary_handler)
972 return;
973again:
974 chip_bus_lock(desc);
975 raw_spin_lock_irq(&desc->lock);
976
977 /*
978 * Implausible though it may be we need to protect us against
979 * the following scenario:
980 *
981 * The thread is faster done than the hard interrupt handler
982 * on the other CPU. If we unmask the irq line then the
983 * interrupt can come in again and masks the line, leaves due
984 * to IRQS_INPROGRESS and the irq line is masked forever.
985 *
986 * This also serializes the state of shared oneshot handlers
987 * versus "desc->threads_onehsot |= action->thread_mask;" in
988 * irq_wake_thread(). See the comment there which explains the
989 * serialization.
990 */
991 if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
992 raw_spin_unlock_irq(&desc->lock);
993 chip_bus_sync_unlock(desc);
994 cpu_relax();
995 goto again;
996 }
997
998 /*
999 * Now check again, whether the thread should run. Otherwise
1000 * we would clear the threads_oneshot bit of this thread which
1001 * was just set.
1002 */
1003 if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1004 goto out_unlock;
1005
1006 desc->threads_oneshot &= ~action->thread_mask;
1007
1008 if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
1009 irqd_irq_masked(&desc->irq_data))
1010 unmask_threaded_irq(desc);
1011
1012out_unlock:
1013 raw_spin_unlock_irq(&desc->lock);
1014 chip_bus_sync_unlock(desc);
1015}
1016
1017#ifdef CONFIG_SMP
1018/*
1019 * Check whether we need to change the affinity of the interrupt thread.
1020 */
1021static void
1022irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
1023{
1024 cpumask_var_t mask;
1025 bool valid = true;
1026
1027 if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
1028 return;
1029
1030 /*
1031 * In case we are out of memory we set IRQTF_AFFINITY again and
1032 * try again next time
1033 */
1034 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1035 set_bit(IRQTF_AFFINITY, &action->thread_flags);
1036 return;
1037 }
1038
1039 raw_spin_lock_irq(&desc->lock);
1040 /*
1041 * This code is triggered unconditionally. Check the affinity
1042 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
1043 */
1044 if (cpumask_available(desc->irq_common_data.affinity)) {
1045 const struct cpumask *m;
1046
1047 m = irq_data_get_effective_affinity_mask(&desc->irq_data);
1048 cpumask_copy(mask, m);
1049 } else {
1050 valid = false;
1051 }
1052 raw_spin_unlock_irq(&desc->lock);
1053
1054 if (valid)
1055 set_cpus_allowed_ptr(current, mask);
1056 free_cpumask_var(mask);
1057}
1058#else
1059static inline void
1060irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
1061#endif
1062
1063/*
1064 * Interrupts which are not explicitly requested as threaded
1065 * interrupts rely on the implicit bh/preempt disable of the hard irq
1066 * context. So we need to disable bh here to avoid deadlocks and other
1067 * side effects.
1068 */
1069static irqreturn_t
1070irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
1071{
1072 irqreturn_t ret;
1073
1074 local_bh_disable();
1075 ret = action->thread_fn(action->irq, action->dev_id);
1076 if (ret == IRQ_HANDLED)
1077 atomic_inc(&desc->threads_handled);
1078
1079 irq_finalize_oneshot(desc, action);
1080 local_bh_enable();
1081 return ret;
1082}
1083
1084/*
1085 * Interrupts explicitly requested as threaded interrupts want to be
1086 * preemtible - many of them need to sleep and wait for slow busses to
1087 * complete.
1088 */
1089static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1090 struct irqaction *action)
1091{
1092 irqreturn_t ret;
1093
1094 ret = action->thread_fn(action->irq, action->dev_id);
1095 if (ret == IRQ_HANDLED)
1096 atomic_inc(&desc->threads_handled);
1097
1098 irq_finalize_oneshot(desc, action);
1099 return ret;
1100}
1101
1102static void wake_threads_waitq(struct irq_desc *desc)
1103{
1104 if (atomic_dec_and_test(&desc->threads_active))
1105 wake_up(&desc->wait_for_threads);
1106}
1107
1108static void irq_thread_dtor(struct callback_head *unused)
1109{
1110 struct task_struct *tsk = current;
1111 struct irq_desc *desc;
1112 struct irqaction *action;
1113
1114 if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1115 return;
1116
1117 action = kthread_data(tsk);
1118
1119 pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1120 tsk->comm, tsk->pid, action->irq);
1121
1122
1123 desc = irq_to_desc(action->irq);
1124 /*
1125 * If IRQTF_RUNTHREAD is set, we need to decrement
1126 * desc->threads_active and wake possible waiters.
1127 */
1128 if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1129 wake_threads_waitq(desc);
1130
1131 /* Prevent a stale desc->threads_oneshot */
1132 irq_finalize_oneshot(desc, action);
1133}
1134
1135static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1136{
1137 struct irqaction *secondary = action->secondary;
1138
1139 if (WARN_ON_ONCE(!secondary))
1140 return;
1141
1142 raw_spin_lock_irq(&desc->lock);
1143 __irq_wake_thread(desc, secondary);
1144 raw_spin_unlock_irq(&desc->lock);
1145}
1146
1147/*
1148 * Interrupt handler thread
1149 */
1150static int irq_thread(void *data)
1151{
1152 struct callback_head on_exit_work;
1153 struct irqaction *action = data;
1154 struct irq_desc *desc = irq_to_desc(action->irq);
1155 irqreturn_t (*handler_fn)(struct irq_desc *desc,
1156 struct irqaction *action);
1157
1158 if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
1159 &action->thread_flags))
1160 handler_fn = irq_forced_thread_fn;
1161 else
1162 handler_fn = irq_thread_fn;
1163
1164 init_task_work(&on_exit_work, irq_thread_dtor);
1165 task_work_add(current, &on_exit_work, false);
1166
1167 irq_thread_check_affinity(desc, action);
1168
1169 while (!irq_wait_for_interrupt(action)) {
1170 irqreturn_t action_ret;
1171
1172 irq_thread_check_affinity(desc, action);
1173
1174 action_ret = handler_fn(desc, action);
1175 if (action_ret == IRQ_WAKE_THREAD)
1176 irq_wake_secondary(desc, action);
1177
1178 wake_threads_waitq(desc);
1179 }
1180
1181 /*
1182 * This is the regular exit path. __free_irq() is stopping the
1183 * thread via kthread_stop() after calling
1184 * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1185 * oneshot mask bit can be set.
1186 */
1187 task_work_cancel(current, irq_thread_dtor);
1188 return 0;
1189}
1190
1191/**
1192 * irq_wake_thread - wake the irq thread for the action identified by dev_id
1193 * @irq: Interrupt line
1194 * @dev_id: Device identity for which the thread should be woken
1195 *
1196 */
1197void irq_wake_thread(unsigned int irq, void *dev_id)
1198{
1199 struct irq_desc *desc = irq_to_desc(irq);
1200 struct irqaction *action;
1201 unsigned long flags;
1202
1203 if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1204 return;
1205
1206 raw_spin_lock_irqsave(&desc->lock, flags);
1207 for_each_action_of_desc(desc, action) {
1208 if (action->dev_id == dev_id) {
1209 if (action->thread)
1210 __irq_wake_thread(desc, action);
1211 break;
1212 }
1213 }
1214 raw_spin_unlock_irqrestore(&desc->lock, flags);
1215}
1216EXPORT_SYMBOL_GPL(irq_wake_thread);
1217
1218static int irq_setup_forced_threading(struct irqaction *new)
1219{
1220 if (!force_irqthreads)
1221 return 0;
1222 if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1223 return 0;
1224
1225 /*
1226 * No further action required for interrupts which are requested as
1227 * threaded interrupts already
1228 */
1229 if (new->handler == irq_default_primary_handler)
1230 return 0;
1231
1232 new->flags |= IRQF_ONESHOT;
1233
1234 /*
1235 * Handle the case where we have a real primary handler and a
1236 * thread handler. We force thread them as well by creating a
1237 * secondary action.
1238 */
1239 if (new->handler && new->thread_fn) {
1240 /* Allocate the secondary action */
1241 new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1242 if (!new->secondary)
1243 return -ENOMEM;
1244 new->secondary->handler = irq_forced_secondary_handler;
1245 new->secondary->thread_fn = new->thread_fn;
1246 new->secondary->dev_id = new->dev_id;
1247 new->secondary->irq = new->irq;
1248 new->secondary->name = new->name;
1249 }
1250 /* Deal with the primary handler */
1251 set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1252 new->thread_fn = new->handler;
1253 new->handler = irq_default_primary_handler;
1254 return 0;
1255}
1256
1257static int irq_request_resources(struct irq_desc *desc)
1258{
1259 struct irq_data *d = &desc->irq_data;
1260 struct irq_chip *c = d->chip;
1261
1262 return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1263}
1264
1265static void irq_release_resources(struct irq_desc *desc)
1266{
1267 struct irq_data *d = &desc->irq_data;
1268 struct irq_chip *c = d->chip;
1269
1270 if (c->irq_release_resources)
1271 c->irq_release_resources(d);
1272}
1273
1274static bool irq_supports_nmi(struct irq_desc *desc)
1275{
1276 struct irq_data *d = irq_desc_get_irq_data(desc);
1277
1278#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1279 /* Only IRQs directly managed by the root irqchip can be set as NMI */
1280 if (d->parent_data)
1281 return false;
1282#endif
1283 /* Don't support NMIs for chips behind a slow bus */
1284 if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1285 return false;
1286
1287 return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1288}
1289
1290static int irq_nmi_setup(struct irq_desc *desc)
1291{
1292 struct irq_data *d = irq_desc_get_irq_data(desc);
1293 struct irq_chip *c = d->chip;
1294
1295 return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1296}
1297
1298static void irq_nmi_teardown(struct irq_desc *desc)
1299{
1300 struct irq_data *d = irq_desc_get_irq_data(desc);
1301 struct irq_chip *c = d->chip;
1302
1303 if (c->irq_nmi_teardown)
1304 c->irq_nmi_teardown(d);
1305}
1306
1307static int
1308setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1309{
1310 struct task_struct *t;
1311
1312 if (!secondary) {
1313 t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1314 new->name);
1315 } else {
1316 t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1317 new->name);
1318 }
1319
1320 if (IS_ERR(t))
1321 return PTR_ERR(t);
1322
1323 sched_set_fifo(t);
1324
1325 /*
1326 * We keep the reference to the task struct even if
1327 * the thread dies to avoid that the interrupt code
1328 * references an already freed task_struct.
1329 */
1330 new->thread = get_task_struct(t);
1331 /*
1332 * Tell the thread to set its affinity. This is
1333 * important for shared interrupt handlers as we do
1334 * not invoke setup_affinity() for the secondary
1335 * handlers as everything is already set up. Even for
1336 * interrupts marked with IRQF_NO_BALANCE this is
1337 * correct as we want the thread to move to the cpu(s)
1338 * on which the requesting code placed the interrupt.
1339 */
1340 set_bit(IRQTF_AFFINITY, &new->thread_flags);
1341 return 0;
1342}
1343
1344/*
1345 * Internal function to register an irqaction - typically used to
1346 * allocate special interrupts that are part of the architecture.
1347 *
1348 * Locking rules:
1349 *
1350 * desc->request_mutex Provides serialization against a concurrent free_irq()
1351 * chip_bus_lock Provides serialization for slow bus operations
1352 * desc->lock Provides serialization against hard interrupts
1353 *
1354 * chip_bus_lock and desc->lock are sufficient for all other management and
1355 * interrupt related functions. desc->request_mutex solely serializes
1356 * request/free_irq().
1357 */
1358static int
1359__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1360{
1361 struct irqaction *old, **old_ptr;
1362 unsigned long flags, thread_mask = 0;
1363 int ret, nested, shared = 0;
1364
1365 if (!desc)
1366 return -EINVAL;
1367
1368 if (desc->irq_data.chip == &no_irq_chip)
1369 return -ENOSYS;
1370 if (!try_module_get(desc->owner))
1371 return -ENODEV;
1372
1373 new->irq = irq;
1374
1375 /*
1376 * If the trigger type is not specified by the caller,
1377 * then use the default for this interrupt.
1378 */
1379 if (!(new->flags & IRQF_TRIGGER_MASK))
1380 new->flags |= irqd_get_trigger_type(&desc->irq_data);
1381
1382 /*
1383 * Check whether the interrupt nests into another interrupt
1384 * thread.
1385 */
1386 nested = irq_settings_is_nested_thread(desc);
1387 if (nested) {
1388 if (!new->thread_fn) {
1389 ret = -EINVAL;
1390 goto out_mput;
1391 }
1392 /*
1393 * Replace the primary handler which was provided from
1394 * the driver for non nested interrupt handling by the
1395 * dummy function which warns when called.
1396 */
1397 new->handler = irq_nested_primary_handler;
1398 } else {
1399 if (irq_settings_can_thread(desc)) {
1400 ret = irq_setup_forced_threading(new);
1401 if (ret)
1402 goto out_mput;
1403 }
1404 }
1405
1406 /*
1407 * Create a handler thread when a thread function is supplied
1408 * and the interrupt does not nest into another interrupt
1409 * thread.
1410 */
1411 if (new->thread_fn && !nested) {
1412 ret = setup_irq_thread(new, irq, false);
1413 if (ret)
1414 goto out_mput;
1415 if (new->secondary) {
1416 ret = setup_irq_thread(new->secondary, irq, true);
1417 if (ret)
1418 goto out_thread;
1419 }
1420 }
1421
1422 /*
1423 * Drivers are often written to work w/o knowledge about the
1424 * underlying irq chip implementation, so a request for a
1425 * threaded irq without a primary hard irq context handler
1426 * requires the ONESHOT flag to be set. Some irq chips like
1427 * MSI based interrupts are per se one shot safe. Check the
1428 * chip flags, so we can avoid the unmask dance at the end of
1429 * the threaded handler for those.
1430 */
1431 if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1432 new->flags &= ~IRQF_ONESHOT;
1433
1434 /*
1435 * Protects against a concurrent __free_irq() call which might wait
1436 * for synchronize_hardirq() to complete without holding the optional
1437 * chip bus lock and desc->lock. Also protects against handing out
1438 * a recycled oneshot thread_mask bit while it's still in use by
1439 * its previous owner.
1440 */
1441 mutex_lock(&desc->request_mutex);
1442
1443 /*
1444 * Acquire bus lock as the irq_request_resources() callback below
1445 * might rely on the serialization or the magic power management
1446 * functions which are abusing the irq_bus_lock() callback,
1447 */
1448 chip_bus_lock(desc);
1449
1450 /* First installed action requests resources. */
1451 if (!desc->action) {
1452 ret = irq_request_resources(desc);
1453 if (ret) {
1454 pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1455 new->name, irq, desc->irq_data.chip->name);
1456 goto out_bus_unlock;
1457 }
1458 }
1459
1460 /*
1461 * The following block of code has to be executed atomically
1462 * protected against a concurrent interrupt and any of the other
1463 * management calls which are not serialized via
1464 * desc->request_mutex or the optional bus lock.
1465 */
1466 raw_spin_lock_irqsave(&desc->lock, flags);
1467 old_ptr = &desc->action;
1468 old = *old_ptr;
1469 if (old) {
1470 /*
1471 * Can't share interrupts unless both agree to and are
1472 * the same type (level, edge, polarity). So both flag
1473 * fields must have IRQF_SHARED set and the bits which
1474 * set the trigger type must match. Also all must
1475 * agree on ONESHOT.
1476 * Interrupt lines used for NMIs cannot be shared.
1477 */
1478 unsigned int oldtype;
1479
1480 if (desc->istate & IRQS_NMI) {
1481 pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1482 new->name, irq, desc->irq_data.chip->name);
1483 ret = -EINVAL;
1484 goto out_unlock;
1485 }
1486
1487 /*
1488 * If nobody did set the configuration before, inherit
1489 * the one provided by the requester.
1490 */
1491 if (irqd_trigger_type_was_set(&desc->irq_data)) {
1492 oldtype = irqd_get_trigger_type(&desc->irq_data);
1493 } else {
1494 oldtype = new->flags & IRQF_TRIGGER_MASK;
1495 irqd_set_trigger_type(&desc->irq_data, oldtype);
1496 }
1497
1498 if (!((old->flags & new->flags) & IRQF_SHARED) ||
1499 (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1500 ((old->flags ^ new->flags) & IRQF_ONESHOT))
1501 goto mismatch;
1502
1503 /* All handlers must agree on per-cpuness */
1504 if ((old->flags & IRQF_PERCPU) !=
1505 (new->flags & IRQF_PERCPU))
1506 goto mismatch;
1507
1508 /* add new interrupt at end of irq queue */
1509 do {
1510 /*
1511 * Or all existing action->thread_mask bits,
1512 * so we can find the next zero bit for this
1513 * new action.
1514 */
1515 thread_mask |= old->thread_mask;
1516 old_ptr = &old->next;
1517 old = *old_ptr;
1518 } while (old);
1519 shared = 1;
1520 }
1521
1522 /*
1523 * Setup the thread mask for this irqaction for ONESHOT. For
1524 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1525 * conditional in irq_wake_thread().
1526 */
1527 if (new->flags & IRQF_ONESHOT) {
1528 /*
1529 * Unlikely to have 32 resp 64 irqs sharing one line,
1530 * but who knows.
1531 */
1532 if (thread_mask == ~0UL) {
1533 ret = -EBUSY;
1534 goto out_unlock;
1535 }
1536 /*
1537 * The thread_mask for the action is or'ed to
1538 * desc->thread_active to indicate that the
1539 * IRQF_ONESHOT thread handler has been woken, but not
1540 * yet finished. The bit is cleared when a thread
1541 * completes. When all threads of a shared interrupt
1542 * line have completed desc->threads_active becomes
1543 * zero and the interrupt line is unmasked. See
1544 * handle.c:irq_wake_thread() for further information.
1545 *
1546 * If no thread is woken by primary (hard irq context)
1547 * interrupt handlers, then desc->threads_active is
1548 * also checked for zero to unmask the irq line in the
1549 * affected hard irq flow handlers
1550 * (handle_[fasteoi|level]_irq).
1551 *
1552 * The new action gets the first zero bit of
1553 * thread_mask assigned. See the loop above which or's
1554 * all existing action->thread_mask bits.
1555 */
1556 new->thread_mask = 1UL << ffz(thread_mask);
1557
1558 } else if (new->handler == irq_default_primary_handler &&
1559 !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1560 /*
1561 * The interrupt was requested with handler = NULL, so
1562 * we use the default primary handler for it. But it
1563 * does not have the oneshot flag set. In combination
1564 * with level interrupts this is deadly, because the
1565 * default primary handler just wakes the thread, then
1566 * the irq lines is reenabled, but the device still
1567 * has the level irq asserted. Rinse and repeat....
1568 *
1569 * While this works for edge type interrupts, we play
1570 * it safe and reject unconditionally because we can't
1571 * say for sure which type this interrupt really
1572 * has. The type flags are unreliable as the
1573 * underlying chip implementation can override them.
1574 */
1575 pr_err("Threaded irq requested with handler=NULL and !ONESHOT for %s (irq %d)\n",
1576 new->name, irq);
1577 ret = -EINVAL;
1578 goto out_unlock;
1579 }
1580
1581 if (!shared) {
1582 init_waitqueue_head(&desc->wait_for_threads);
1583
1584 /* Setup the type (level, edge polarity) if configured: */
1585 if (new->flags & IRQF_TRIGGER_MASK) {
1586 ret = __irq_set_trigger(desc,
1587 new->flags & IRQF_TRIGGER_MASK);
1588
1589 if (ret)
1590 goto out_unlock;
1591 }
1592
1593 /*
1594 * Activate the interrupt. That activation must happen
1595 * independently of IRQ_NOAUTOEN. request_irq() can fail
1596 * and the callers are supposed to handle
1597 * that. enable_irq() of an interrupt requested with
1598 * IRQ_NOAUTOEN is not supposed to fail. The activation
1599 * keeps it in shutdown mode, it merily associates
1600 * resources if necessary and if that's not possible it
1601 * fails. Interrupts which are in managed shutdown mode
1602 * will simply ignore that activation request.
1603 */
1604 ret = irq_activate(desc);
1605 if (ret)
1606 goto out_unlock;
1607
1608 desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1609 IRQS_ONESHOT | IRQS_WAITING);
1610 irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1611
1612 if (new->flags & IRQF_PERCPU) {
1613 irqd_set(&desc->irq_data, IRQD_PER_CPU);
1614 irq_settings_set_per_cpu(desc);
1615 }
1616
1617 if (new->flags & IRQF_ONESHOT)
1618 desc->istate |= IRQS_ONESHOT;
1619
1620 /* Exclude IRQ from balancing if requested */
1621 if (new->flags & IRQF_NOBALANCING) {
1622 irq_settings_set_no_balancing(desc);
1623 irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1624 }
1625
1626 if (irq_settings_can_autoenable(desc)) {
1627 irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1628 } else {
1629 /*
1630 * Shared interrupts do not go well with disabling
1631 * auto enable. The sharing interrupt might request
1632 * it while it's still disabled and then wait for
1633 * interrupts forever.
1634 */
1635 WARN_ON_ONCE(new->flags & IRQF_SHARED);
1636 /* Undo nested disables: */
1637 desc->depth = 1;
1638 }
1639
1640 } else if (new->flags & IRQF_TRIGGER_MASK) {
1641 unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1642 unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1643
1644 if (nmsk != omsk)
1645 /* hope the handler works with current trigger mode */
1646 pr_warn("irq %d uses trigger mode %u; requested %u\n",
1647 irq, omsk, nmsk);
1648 }
1649
1650 *old_ptr = new;
1651
1652 irq_pm_install_action(desc, new);
1653
1654 /* Reset broken irq detection when installing new handler */
1655 desc->irq_count = 0;
1656 desc->irqs_unhandled = 0;
1657
1658 /*
1659 * Check whether we disabled the irq via the spurious handler
1660 * before. Reenable it and give it another chance.
1661 */
1662 if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1663 desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1664 __enable_irq(desc);
1665 }
1666
1667 raw_spin_unlock_irqrestore(&desc->lock, flags);
1668 chip_bus_sync_unlock(desc);
1669 mutex_unlock(&desc->request_mutex);
1670
1671 irq_setup_timings(desc, new);
1672
1673 /*
1674 * Strictly no need to wake it up, but hung_task complains
1675 * when no hard interrupt wakes the thread up.
1676 */
1677 if (new->thread)
1678 wake_up_process(new->thread);
1679 if (new->secondary)
1680 wake_up_process(new->secondary->thread);
1681
1682 register_irq_proc(irq, desc);
1683 new->dir = NULL;
1684 register_handler_proc(irq, new);
1685 return 0;
1686
1687mismatch:
1688 if (!(new->flags & IRQF_PROBE_SHARED)) {
1689 pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1690 irq, new->flags, new->name, old->flags, old->name);
1691#ifdef CONFIG_DEBUG_SHIRQ
1692 dump_stack();
1693#endif
1694 }
1695 ret = -EBUSY;
1696
1697out_unlock:
1698 raw_spin_unlock_irqrestore(&desc->lock, flags);
1699
1700 if (!desc->action)
1701 irq_release_resources(desc);
1702out_bus_unlock:
1703 chip_bus_sync_unlock(desc);
1704 mutex_unlock(&desc->request_mutex);
1705
1706out_thread:
1707 if (new->thread) {
1708 struct task_struct *t = new->thread;
1709
1710 new->thread = NULL;
1711 kthread_stop(t);
1712 put_task_struct(t);
1713 }
1714 if (new->secondary && new->secondary->thread) {
1715 struct task_struct *t = new->secondary->thread;
1716
1717 new->secondary->thread = NULL;
1718 kthread_stop(t);
1719 put_task_struct(t);
1720 }
1721out_mput:
1722 module_put(desc->owner);
1723 return ret;
1724}
1725
1726/*
1727 * Internal function to unregister an irqaction - used to free
1728 * regular and special interrupts that are part of the architecture.
1729 */
1730static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1731{
1732 unsigned irq = desc->irq_data.irq;
1733 struct irqaction *action, **action_ptr;
1734 unsigned long flags;
1735
1736 WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1737
1738 mutex_lock(&desc->request_mutex);
1739 chip_bus_lock(desc);
1740 raw_spin_lock_irqsave(&desc->lock, flags);
1741
1742 /*
1743 * There can be multiple actions per IRQ descriptor, find the right
1744 * one based on the dev_id:
1745 */
1746 action_ptr = &desc->action;
1747 for (;;) {
1748 action = *action_ptr;
1749
1750 if (!action) {
1751 WARN(1, "Trying to free already-free IRQ %d\n", irq);
1752 raw_spin_unlock_irqrestore(&desc->lock, flags);
1753 chip_bus_sync_unlock(desc);
1754 mutex_unlock(&desc->request_mutex);
1755 return NULL;
1756 }
1757
1758 if (action->dev_id == dev_id)
1759 break;
1760 action_ptr = &action->next;
1761 }
1762
1763 /* Found it - now remove it from the list of entries: */
1764 *action_ptr = action->next;
1765
1766 irq_pm_remove_action(desc, action);
1767
1768 /* If this was the last handler, shut down the IRQ line: */
1769 if (!desc->action) {
1770 irq_settings_clr_disable_unlazy(desc);
1771 /* Only shutdown. Deactivate after synchronize_hardirq() */
1772 irq_shutdown(desc);
1773 }
1774
1775#ifdef CONFIG_SMP
1776 /* make sure affinity_hint is cleaned up */
1777 if (WARN_ON_ONCE(desc->affinity_hint))
1778 desc->affinity_hint = NULL;
1779#endif
1780
1781 raw_spin_unlock_irqrestore(&desc->lock, flags);
1782 /*
1783 * Drop bus_lock here so the changes which were done in the chip
1784 * callbacks above are synced out to the irq chips which hang
1785 * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1786 *
1787 * Aside of that the bus_lock can also be taken from the threaded
1788 * handler in irq_finalize_oneshot() which results in a deadlock
1789 * because kthread_stop() would wait forever for the thread to
1790 * complete, which is blocked on the bus lock.
1791 *
1792 * The still held desc->request_mutex() protects against a
1793 * concurrent request_irq() of this irq so the release of resources
1794 * and timing data is properly serialized.
1795 */
1796 chip_bus_sync_unlock(desc);
1797
1798 unregister_handler_proc(irq, action);
1799
1800 /*
1801 * Make sure it's not being used on another CPU and if the chip
1802 * supports it also make sure that there is no (not yet serviced)
1803 * interrupt in flight at the hardware level.
1804 */
1805 __synchronize_hardirq(desc, true);
1806
1807#ifdef CONFIG_DEBUG_SHIRQ
1808 /*
1809 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1810 * event to happen even now it's being freed, so let's make sure that
1811 * is so by doing an extra call to the handler ....
1812 *
1813 * ( We do this after actually deregistering it, to make sure that a
1814 * 'real' IRQ doesn't run in parallel with our fake. )
1815 */
1816 if (action->flags & IRQF_SHARED) {
1817 local_irq_save(flags);
1818 action->handler(irq, dev_id);
1819 local_irq_restore(flags);
1820 }
1821#endif
1822
1823 /*
1824 * The action has already been removed above, but the thread writes
1825 * its oneshot mask bit when it completes. Though request_mutex is
1826 * held across this which prevents __setup_irq() from handing out
1827 * the same bit to a newly requested action.
1828 */
1829 if (action->thread) {
1830 kthread_stop(action->thread);
1831 put_task_struct(action->thread);
1832 if (action->secondary && action->secondary->thread) {
1833 kthread_stop(action->secondary->thread);
1834 put_task_struct(action->secondary->thread);
1835 }
1836 }
1837
1838 /* Last action releases resources */
1839 if (!desc->action) {
1840 /*
1841 * Reaquire bus lock as irq_release_resources() might
1842 * require it to deallocate resources over the slow bus.
1843 */
1844 chip_bus_lock(desc);
1845 /*
1846 * There is no interrupt on the fly anymore. Deactivate it
1847 * completely.
1848 */
1849 raw_spin_lock_irqsave(&desc->lock, flags);
1850 irq_domain_deactivate_irq(&desc->irq_data);
1851 raw_spin_unlock_irqrestore(&desc->lock, flags);
1852
1853 irq_release_resources(desc);
1854 chip_bus_sync_unlock(desc);
1855 irq_remove_timings(desc);
1856 }
1857
1858 mutex_unlock(&desc->request_mutex);
1859
1860 irq_chip_pm_put(&desc->irq_data);
1861 module_put(desc->owner);
1862 kfree(action->secondary);
1863 return action;
1864}
1865
1866/**
1867 * free_irq - free an interrupt allocated with request_irq
1868 * @irq: Interrupt line to free
1869 * @dev_id: Device identity to free
1870 *
1871 * Remove an interrupt handler. The handler is removed and if the
1872 * interrupt line is no longer in use by any driver it is disabled.
1873 * On a shared IRQ the caller must ensure the interrupt is disabled
1874 * on the card it drives before calling this function. The function
1875 * does not return until any executing interrupts for this IRQ
1876 * have completed.
1877 *
1878 * This function must not be called from interrupt context.
1879 *
1880 * Returns the devname argument passed to request_irq.
1881 */
1882const void *free_irq(unsigned int irq, void *dev_id)
1883{
1884 struct irq_desc *desc = irq_to_desc(irq);
1885 struct irqaction *action;
1886 const char *devname;
1887
1888 if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1889 return NULL;
1890
1891#ifdef CONFIG_SMP
1892 if (WARN_ON(desc->affinity_notify))
1893 desc->affinity_notify = NULL;
1894#endif
1895
1896 action = __free_irq(desc, dev_id);
1897
1898 if (!action)
1899 return NULL;
1900
1901 devname = action->name;
1902 kfree(action);
1903 return devname;
1904}
1905EXPORT_SYMBOL(free_irq);
1906
1907/* This function must be called with desc->lock held */
1908static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
1909{
1910 const char *devname = NULL;
1911
1912 desc->istate &= ~IRQS_NMI;
1913
1914 if (!WARN_ON(desc->action == NULL)) {
1915 irq_pm_remove_action(desc, desc->action);
1916 devname = desc->action->name;
1917 unregister_handler_proc(irq, desc->action);
1918
1919 kfree(desc->action);
1920 desc->action = NULL;
1921 }
1922
1923 irq_settings_clr_disable_unlazy(desc);
1924 irq_shutdown_and_deactivate(desc);
1925
1926 irq_release_resources(desc);
1927
1928 irq_chip_pm_put(&desc->irq_data);
1929 module_put(desc->owner);
1930
1931 return devname;
1932}
1933
1934const void *free_nmi(unsigned int irq, void *dev_id)
1935{
1936 struct irq_desc *desc = irq_to_desc(irq);
1937 unsigned long flags;
1938 const void *devname;
1939
1940 if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
1941 return NULL;
1942
1943 if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1944 return NULL;
1945
1946 /* NMI still enabled */
1947 if (WARN_ON(desc->depth == 0))
1948 disable_nmi_nosync(irq);
1949
1950 raw_spin_lock_irqsave(&desc->lock, flags);
1951
1952 irq_nmi_teardown(desc);
1953 devname = __cleanup_nmi(irq, desc);
1954
1955 raw_spin_unlock_irqrestore(&desc->lock, flags);
1956
1957 return devname;
1958}
1959
1960/**
1961 * request_threaded_irq - allocate an interrupt line
1962 * @irq: Interrupt line to allocate
1963 * @handler: Function to be called when the IRQ occurs.
1964 * Primary handler for threaded interrupts
1965 * If NULL and thread_fn != NULL the default
1966 * primary handler is installed
1967 * @thread_fn: Function called from the irq handler thread
1968 * If NULL, no irq thread is created
1969 * @irqflags: Interrupt type flags
1970 * @devname: An ascii name for the claiming device
1971 * @dev_id: A cookie passed back to the handler function
1972 *
1973 * This call allocates interrupt resources and enables the
1974 * interrupt line and IRQ handling. From the point this
1975 * call is made your handler function may be invoked. Since
1976 * your handler function must clear any interrupt the board
1977 * raises, you must take care both to initialise your hardware
1978 * and to set up the interrupt handler in the right order.
1979 *
1980 * If you want to set up a threaded irq handler for your device
1981 * then you need to supply @handler and @thread_fn. @handler is
1982 * still called in hard interrupt context and has to check
1983 * whether the interrupt originates from the device. If yes it
1984 * needs to disable the interrupt on the device and return
1985 * IRQ_WAKE_THREAD which will wake up the handler thread and run
1986 * @thread_fn. This split handler design is necessary to support
1987 * shared interrupts.
1988 *
1989 * Dev_id must be globally unique. Normally the address of the
1990 * device data structure is used as the cookie. Since the handler
1991 * receives this value it makes sense to use it.
1992 *
1993 * If your interrupt is shared you must pass a non NULL dev_id
1994 * as this is required when freeing the interrupt.
1995 *
1996 * Flags:
1997 *
1998 * IRQF_SHARED Interrupt is shared
1999 * IRQF_TRIGGER_* Specify active edge(s) or level
2000 *
2001 */
2002int request_threaded_irq(unsigned int irq, irq_handler_t handler,
2003 irq_handler_t thread_fn, unsigned long irqflags,
2004 const char *devname, void *dev_id)
2005{
2006 struct irqaction *action;
2007 struct irq_desc *desc;
2008 int retval;
2009
2010 if (irq == IRQ_NOTCONNECTED)
2011 return -ENOTCONN;
2012
2013 /*
2014 * Sanity-check: shared interrupts must pass in a real dev-ID,
2015 * otherwise we'll have trouble later trying to figure out
2016 * which interrupt is which (messes up the interrupt freeing
2017 * logic etc).
2018 *
2019 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
2020 * it cannot be set along with IRQF_NO_SUSPEND.
2021 */
2022 if (((irqflags & IRQF_SHARED) && !dev_id) ||
2023 (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
2024 ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
2025 return -EINVAL;
2026
2027 desc = irq_to_desc(irq);
2028 if (!desc)
2029 return -EINVAL;
2030
2031 if (!irq_settings_can_request(desc) ||
2032 WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2033 return -EINVAL;
2034
2035 if (!handler) {
2036 if (!thread_fn)
2037 return -EINVAL;
2038 handler = irq_default_primary_handler;
2039 }
2040
2041 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2042 if (!action)
2043 return -ENOMEM;
2044
2045 action->handler = handler;
2046 action->thread_fn = thread_fn;
2047 action->flags = irqflags;
2048 action->name = devname;
2049 action->dev_id = dev_id;
2050
2051 retval = irq_chip_pm_get(&desc->irq_data);
2052 if (retval < 0) {
2053 kfree(action);
2054 return retval;
2055 }
2056
2057 retval = __setup_irq(irq, desc, action);
2058
2059 if (retval) {
2060 irq_chip_pm_put(&desc->irq_data);
2061 kfree(action->secondary);
2062 kfree(action);
2063 }
2064
2065#ifdef CONFIG_DEBUG_SHIRQ_FIXME
2066 if (!retval && (irqflags & IRQF_SHARED)) {
2067 /*
2068 * It's a shared IRQ -- the driver ought to be prepared for it
2069 * to happen immediately, so let's make sure....
2070 * We disable the irq to make sure that a 'real' IRQ doesn't
2071 * run in parallel with our fake.
2072 */
2073 unsigned long flags;
2074
2075 disable_irq(irq);
2076 local_irq_save(flags);
2077
2078 handler(irq, dev_id);
2079
2080 local_irq_restore(flags);
2081 enable_irq(irq);
2082 }
2083#endif
2084 return retval;
2085}
2086EXPORT_SYMBOL(request_threaded_irq);
2087
2088/**
2089 * request_any_context_irq - allocate an interrupt line
2090 * @irq: Interrupt line to allocate
2091 * @handler: Function to be called when the IRQ occurs.
2092 * Threaded handler for threaded interrupts.
2093 * @flags: Interrupt type flags
2094 * @name: An ascii name for the claiming device
2095 * @dev_id: A cookie passed back to the handler function
2096 *
2097 * This call allocates interrupt resources and enables the
2098 * interrupt line and IRQ handling. It selects either a
2099 * hardirq or threaded handling method depending on the
2100 * context.
2101 *
2102 * On failure, it returns a negative value. On success,
2103 * it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2104 */
2105int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2106 unsigned long flags, const char *name, void *dev_id)
2107{
2108 struct irq_desc *desc;
2109 int ret;
2110
2111 if (irq == IRQ_NOTCONNECTED)
2112 return -ENOTCONN;
2113
2114 desc = irq_to_desc(irq);
2115 if (!desc)
2116 return -EINVAL;
2117
2118 if (irq_settings_is_nested_thread(desc)) {
2119 ret = request_threaded_irq(irq, NULL, handler,
2120 flags, name, dev_id);
2121 return !ret ? IRQC_IS_NESTED : ret;
2122 }
2123
2124 ret = request_irq(irq, handler, flags, name, dev_id);
2125 return !ret ? IRQC_IS_HARDIRQ : ret;
2126}
2127EXPORT_SYMBOL_GPL(request_any_context_irq);
2128
2129/**
2130 * request_nmi - allocate an interrupt line for NMI delivery
2131 * @irq: Interrupt line to allocate
2132 * @handler: Function to be called when the IRQ occurs.
2133 * Threaded handler for threaded interrupts.
2134 * @irqflags: Interrupt type flags
2135 * @name: An ascii name for the claiming device
2136 * @dev_id: A cookie passed back to the handler function
2137 *
2138 * This call allocates interrupt resources and enables the
2139 * interrupt line and IRQ handling. It sets up the IRQ line
2140 * to be handled as an NMI.
2141 *
2142 * An interrupt line delivering NMIs cannot be shared and IRQ handling
2143 * cannot be threaded.
2144 *
2145 * Interrupt lines requested for NMI delivering must produce per cpu
2146 * interrupts and have auto enabling setting disabled.
2147 *
2148 * Dev_id must be globally unique. Normally the address of the
2149 * device data structure is used as the cookie. Since the handler
2150 * receives this value it makes sense to use it.
2151 *
2152 * If the interrupt line cannot be used to deliver NMIs, function
2153 * will fail and return a negative value.
2154 */
2155int request_nmi(unsigned int irq, irq_handler_t handler,
2156 unsigned long irqflags, const char *name, void *dev_id)
2157{
2158 struct irqaction *action;
2159 struct irq_desc *desc;
2160 unsigned long flags;
2161 int retval;
2162
2163 if (irq == IRQ_NOTCONNECTED)
2164 return -ENOTCONN;
2165
2166 /* NMI cannot be shared, used for Polling */
2167 if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2168 return -EINVAL;
2169
2170 if (!(irqflags & IRQF_PERCPU))
2171 return -EINVAL;
2172
2173 if (!handler)
2174 return -EINVAL;
2175
2176 desc = irq_to_desc(irq);
2177
2178 if (!desc || irq_settings_can_autoenable(desc) ||
2179 !irq_settings_can_request(desc) ||
2180 WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2181 !irq_supports_nmi(desc))
2182 return -EINVAL;
2183
2184 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2185 if (!action)
2186 return -ENOMEM;
2187
2188 action->handler = handler;
2189 action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2190 action->name = name;
2191 action->dev_id = dev_id;
2192
2193 retval = irq_chip_pm_get(&desc->irq_data);
2194 if (retval < 0)
2195 goto err_out;
2196
2197 retval = __setup_irq(irq, desc, action);
2198 if (retval)
2199 goto err_irq_setup;
2200
2201 raw_spin_lock_irqsave(&desc->lock, flags);
2202
2203 /* Setup NMI state */
2204 desc->istate |= IRQS_NMI;
2205 retval = irq_nmi_setup(desc);
2206 if (retval) {
2207 __cleanup_nmi(irq, desc);
2208 raw_spin_unlock_irqrestore(&desc->lock, flags);
2209 return -EINVAL;
2210 }
2211
2212 raw_spin_unlock_irqrestore(&desc->lock, flags);
2213
2214 return 0;
2215
2216err_irq_setup:
2217 irq_chip_pm_put(&desc->irq_data);
2218err_out:
2219 kfree(action);
2220
2221 return retval;
2222}
2223
2224void enable_percpu_irq(unsigned int irq, unsigned int type)
2225{
2226 unsigned int cpu = smp_processor_id();
2227 unsigned long flags;
2228 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2229
2230 if (!desc)
2231 return;
2232
2233 /*
2234 * If the trigger type is not specified by the caller, then
2235 * use the default for this interrupt.
2236 */
2237 type &= IRQ_TYPE_SENSE_MASK;
2238 if (type == IRQ_TYPE_NONE)
2239 type = irqd_get_trigger_type(&desc->irq_data);
2240
2241 if (type != IRQ_TYPE_NONE) {
2242 int ret;
2243
2244 ret = __irq_set_trigger(desc, type);
2245
2246 if (ret) {
2247 WARN(1, "failed to set type for IRQ%d\n", irq);
2248 goto out;
2249 }
2250 }
2251
2252 irq_percpu_enable(desc, cpu);
2253out:
2254 irq_put_desc_unlock(desc, flags);
2255}
2256EXPORT_SYMBOL_GPL(enable_percpu_irq);
2257
2258void enable_percpu_nmi(unsigned int irq, unsigned int type)
2259{
2260 enable_percpu_irq(irq, type);
2261}
2262
2263/**
2264 * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2265 * @irq: Linux irq number to check for
2266 *
2267 * Must be called from a non migratable context. Returns the enable
2268 * state of a per cpu interrupt on the current cpu.
2269 */
2270bool irq_percpu_is_enabled(unsigned int irq)
2271{
2272 unsigned int cpu = smp_processor_id();
2273 struct irq_desc *desc;
2274 unsigned long flags;
2275 bool is_enabled;
2276
2277 desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2278 if (!desc)
2279 return false;
2280
2281 is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2282 irq_put_desc_unlock(desc, flags);
2283
2284 return is_enabled;
2285}
2286EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2287
2288void disable_percpu_irq(unsigned int irq)
2289{
2290 unsigned int cpu = smp_processor_id();
2291 unsigned long flags;
2292 struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2293
2294 if (!desc)
2295 return;
2296
2297 irq_percpu_disable(desc, cpu);
2298 irq_put_desc_unlock(desc, flags);
2299}
2300EXPORT_SYMBOL_GPL(disable_percpu_irq);
2301
2302void disable_percpu_nmi(unsigned int irq)
2303{
2304 disable_percpu_irq(irq);
2305}
2306
2307/*
2308 * Internal function to unregister a percpu irqaction.
2309 */
2310static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2311{
2312 struct irq_desc *desc = irq_to_desc(irq);
2313 struct irqaction *action;
2314 unsigned long flags;
2315
2316 WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2317
2318 if (!desc)
2319 return NULL;
2320
2321 raw_spin_lock_irqsave(&desc->lock, flags);
2322
2323 action = desc->action;
2324 if (!action || action->percpu_dev_id != dev_id) {
2325 WARN(1, "Trying to free already-free IRQ %d\n", irq);
2326 goto bad;
2327 }
2328
2329 if (!cpumask_empty(desc->percpu_enabled)) {
2330 WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2331 irq, cpumask_first(desc->percpu_enabled));
2332 goto bad;
2333 }
2334
2335 /* Found it - now remove it from the list of entries: */
2336 desc->action = NULL;
2337
2338 desc->istate &= ~IRQS_NMI;
2339
2340 raw_spin_unlock_irqrestore(&desc->lock, flags);
2341
2342 unregister_handler_proc(irq, action);
2343
2344 irq_chip_pm_put(&desc->irq_data);
2345 module_put(desc->owner);
2346 return action;
2347
2348bad:
2349 raw_spin_unlock_irqrestore(&desc->lock, flags);
2350 return NULL;
2351}
2352
2353/**
2354 * remove_percpu_irq - free a per-cpu interrupt
2355 * @irq: Interrupt line to free
2356 * @act: irqaction for the interrupt
2357 *
2358 * Used to remove interrupts statically setup by the early boot process.
2359 */
2360void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2361{
2362 struct irq_desc *desc = irq_to_desc(irq);
2363
2364 if (desc && irq_settings_is_per_cpu_devid(desc))
2365 __free_percpu_irq(irq, act->percpu_dev_id);
2366}
2367
2368/**
2369 * free_percpu_irq - free an interrupt allocated with request_percpu_irq
2370 * @irq: Interrupt line to free
2371 * @dev_id: Device identity to free
2372 *
2373 * Remove a percpu interrupt handler. The handler is removed, but
2374 * the interrupt line is not disabled. This must be done on each
2375 * CPU before calling this function. The function does not return
2376 * until any executing interrupts for this IRQ have completed.
2377 *
2378 * This function must not be called from interrupt context.
2379 */
2380void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2381{
2382 struct irq_desc *desc = irq_to_desc(irq);
2383
2384 if (!desc || !irq_settings_is_per_cpu_devid(desc))
2385 return;
2386
2387 chip_bus_lock(desc);
2388 kfree(__free_percpu_irq(irq, dev_id));
2389 chip_bus_sync_unlock(desc);
2390}
2391EXPORT_SYMBOL_GPL(free_percpu_irq);
2392
2393void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2394{
2395 struct irq_desc *desc = irq_to_desc(irq);
2396
2397 if (!desc || !irq_settings_is_per_cpu_devid(desc))
2398 return;
2399
2400 if (WARN_ON(!(desc->istate & IRQS_NMI)))
2401 return;
2402
2403 kfree(__free_percpu_irq(irq, dev_id));
2404}
2405
2406/**
2407 * setup_percpu_irq - setup a per-cpu interrupt
2408 * @irq: Interrupt line to setup
2409 * @act: irqaction for the interrupt
2410 *
2411 * Used to statically setup per-cpu interrupts in the early boot process.
2412 */
2413int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2414{
2415 struct irq_desc *desc = irq_to_desc(irq);
2416 int retval;
2417
2418 if (!desc || !irq_settings_is_per_cpu_devid(desc))
2419 return -EINVAL;
2420
2421 retval = irq_chip_pm_get(&desc->irq_data);
2422 if (retval < 0)
2423 return retval;
2424
2425 retval = __setup_irq(irq, desc, act);
2426
2427 if (retval)
2428 irq_chip_pm_put(&desc->irq_data);
2429
2430 return retval;
2431}
2432
2433/**
2434 * __request_percpu_irq - allocate a percpu interrupt line
2435 * @irq: Interrupt line to allocate
2436 * @handler: Function to be called when the IRQ occurs.
2437 * @flags: Interrupt type flags (IRQF_TIMER only)
2438 * @devname: An ascii name for the claiming device
2439 * @dev_id: A percpu cookie passed back to the handler function
2440 *
2441 * This call allocates interrupt resources and enables the
2442 * interrupt on the local CPU. If the interrupt is supposed to be
2443 * enabled on other CPUs, it has to be done on each CPU using
2444 * enable_percpu_irq().
2445 *
2446 * Dev_id must be globally unique. It is a per-cpu variable, and
2447 * the handler gets called with the interrupted CPU's instance of
2448 * that variable.
2449 */
2450int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2451 unsigned long flags, const char *devname,
2452 void __percpu *dev_id)
2453{
2454 struct irqaction *action;
2455 struct irq_desc *desc;
2456 int retval;
2457
2458 if (!dev_id)
2459 return -EINVAL;
2460
2461 desc = irq_to_desc(irq);
2462 if (!desc || !irq_settings_can_request(desc) ||
2463 !irq_settings_is_per_cpu_devid(desc))
2464 return -EINVAL;
2465
2466 if (flags && flags != IRQF_TIMER)
2467 return -EINVAL;
2468
2469 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2470 if (!action)
2471 return -ENOMEM;
2472
2473 action->handler = handler;
2474 action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2475 action->name = devname;
2476 action->percpu_dev_id = dev_id;
2477
2478 retval = irq_chip_pm_get(&desc->irq_data);
2479 if (retval < 0) {
2480 kfree(action);
2481 return retval;
2482 }
2483
2484 retval = __setup_irq(irq, desc, action);
2485
2486 if (retval) {
2487 irq_chip_pm_put(&desc->irq_data);
2488 kfree(action);
2489 }
2490
2491 return retval;
2492}
2493EXPORT_SYMBOL_GPL(__request_percpu_irq);
2494
2495/**
2496 * request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2497 * @irq: Interrupt line to allocate
2498 * @handler: Function to be called when the IRQ occurs.
2499 * @name: An ascii name for the claiming device
2500 * @dev_id: A percpu cookie passed back to the handler function
2501 *
2502 * This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2503 * have to be setup on each CPU by calling prepare_percpu_nmi() before
2504 * being enabled on the same CPU by using enable_percpu_nmi().
2505 *
2506 * Dev_id must be globally unique. It is a per-cpu variable, and
2507 * the handler gets called with the interrupted CPU's instance of
2508 * that variable.
2509 *
2510 * Interrupt lines requested for NMI delivering should have auto enabling
2511 * setting disabled.
2512 *
2513 * If the interrupt line cannot be used to deliver NMIs, function
2514 * will fail returning a negative value.
2515 */
2516int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2517 const char *name, void __percpu *dev_id)
2518{
2519 struct irqaction *action;
2520 struct irq_desc *desc;
2521 unsigned long flags;
2522 int retval;
2523
2524 if (!handler)
2525 return -EINVAL;
2526
2527 desc = irq_to_desc(irq);
2528
2529 if (!desc || !irq_settings_can_request(desc) ||
2530 !irq_settings_is_per_cpu_devid(desc) ||
2531 irq_settings_can_autoenable(desc) ||
2532 !irq_supports_nmi(desc))
2533 return -EINVAL;
2534
2535 /* The line cannot already be NMI */
2536 if (desc->istate & IRQS_NMI)
2537 return -EINVAL;
2538
2539 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2540 if (!action)
2541 return -ENOMEM;
2542
2543 action->handler = handler;
2544 action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2545 | IRQF_NOBALANCING;
2546 action->name = name;
2547 action->percpu_dev_id = dev_id;
2548
2549 retval = irq_chip_pm_get(&desc->irq_data);
2550 if (retval < 0)
2551 goto err_out;
2552
2553 retval = __setup_irq(irq, desc, action);
2554 if (retval)
2555 goto err_irq_setup;
2556
2557 raw_spin_lock_irqsave(&desc->lock, flags);
2558 desc->istate |= IRQS_NMI;
2559 raw_spin_unlock_irqrestore(&desc->lock, flags);
2560
2561 return 0;
2562
2563err_irq_setup:
2564 irq_chip_pm_put(&desc->irq_data);
2565err_out:
2566 kfree(action);
2567
2568 return retval;
2569}
2570
2571/**
2572 * prepare_percpu_nmi - performs CPU local setup for NMI delivery
2573 * @irq: Interrupt line to prepare for NMI delivery
2574 *
2575 * This call prepares an interrupt line to deliver NMI on the current CPU,
2576 * before that interrupt line gets enabled with enable_percpu_nmi().
2577 *
2578 * As a CPU local operation, this should be called from non-preemptible
2579 * context.
2580 *
2581 * If the interrupt line cannot be used to deliver NMIs, function
2582 * will fail returning a negative value.
2583 */
2584int prepare_percpu_nmi(unsigned int irq)
2585{
2586 unsigned long flags;
2587 struct irq_desc *desc;
2588 int ret = 0;
2589
2590 WARN_ON(preemptible());
2591
2592 desc = irq_get_desc_lock(irq, &flags,
2593 IRQ_GET_DESC_CHECK_PERCPU);
2594 if (!desc)
2595 return -EINVAL;
2596
2597 if (WARN(!(desc->istate & IRQS_NMI),
2598 KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2599 irq)) {
2600 ret = -EINVAL;
2601 goto out;
2602 }
2603
2604 ret = irq_nmi_setup(desc);
2605 if (ret) {
2606 pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2607 goto out;
2608 }
2609
2610out:
2611 irq_put_desc_unlock(desc, flags);
2612 return ret;
2613}
2614
2615/**
2616 * teardown_percpu_nmi - undoes NMI setup of IRQ line
2617 * @irq: Interrupt line from which CPU local NMI configuration should be
2618 * removed
2619 *
2620 * This call undoes the setup done by prepare_percpu_nmi().
2621 *
2622 * IRQ line should not be enabled for the current CPU.
2623 *
2624 * As a CPU local operation, this should be called from non-preemptible
2625 * context.
2626 */
2627void teardown_percpu_nmi(unsigned int irq)
2628{
2629 unsigned long flags;
2630 struct irq_desc *desc;
2631
2632 WARN_ON(preemptible());
2633
2634 desc = irq_get_desc_lock(irq, &flags,
2635 IRQ_GET_DESC_CHECK_PERCPU);
2636 if (!desc)
2637 return;
2638
2639 if (WARN_ON(!(desc->istate & IRQS_NMI)))
2640 goto out;
2641
2642 irq_nmi_teardown(desc);
2643out:
2644 irq_put_desc_unlock(desc, flags);
2645}
2646
2647int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2648 bool *state)
2649{
2650 struct irq_chip *chip;
2651 int err = -EINVAL;
2652
2653 do {
2654 chip = irq_data_get_irq_chip(data);
2655 if (WARN_ON_ONCE(!chip))
2656 return -ENODEV;
2657 if (chip->irq_get_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_get_irqchip_state(data, which, state);
2668 return err;
2669}
2670
2671/**
2672 * irq_get_irqchip_state - returns the irqchip state of a interrupt.
2673 * @irq: Interrupt line that is forwarded to a VM
2674 * @which: One of IRQCHIP_STATE_* the caller wants to know about
2675 * @state: a pointer to a boolean where the state is to be storeed
2676 *
2677 * This call snapshots the internal irqchip state of an
2678 * interrupt, returning into @state the bit corresponding to
2679 * stage @which
2680 *
2681 * This function should be called with preemption disabled if the
2682 * interrupt controller has per-cpu registers.
2683 */
2684int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2685 bool *state)
2686{
2687 struct irq_desc *desc;
2688 struct irq_data *data;
2689 unsigned long flags;
2690 int err = -EINVAL;
2691
2692 desc = irq_get_desc_buslock(irq, &flags, 0);
2693 if (!desc)
2694 return err;
2695
2696 data = irq_desc_get_irq_data(desc);
2697
2698 err = __irq_get_irqchip_state(data, which, state);
2699
2700 irq_put_desc_busunlock(desc, flags);
2701 return err;
2702}
2703EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2704
2705/**
2706 * irq_set_irqchip_state - set the state of a forwarded interrupt.
2707 * @irq: Interrupt line that is forwarded to a VM
2708 * @which: State to be restored (one of IRQCHIP_STATE_*)
2709 * @val: Value corresponding to @which
2710 *
2711 * This call sets the internal irqchip state of an interrupt,
2712 * depending on the value of @which.
2713 *
2714 * This function should be called with preemption disabled if the
2715 * interrupt controller has per-cpu registers.
2716 */
2717int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2718 bool val)
2719{
2720 struct irq_desc *desc;
2721 struct irq_data *data;
2722 struct irq_chip *chip;
2723 unsigned long flags;
2724 int err = -EINVAL;
2725
2726 desc = irq_get_desc_buslock(irq, &flags, 0);
2727 if (!desc)
2728 return err;
2729
2730 data = irq_desc_get_irq_data(desc);
2731
2732 do {
2733 chip = irq_data_get_irq_chip(data);
2734 if (WARN_ON_ONCE(!chip)) {
2735 err = -ENODEV;
2736 goto out_unlock;
2737 }
2738 if (chip->irq_set_irqchip_state)
2739 break;
2740#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2741 data = data->parent_data;
2742#else
2743 data = NULL;
2744#endif
2745 } while (data);
2746
2747 if (data)
2748 err = chip->irq_set_irqchip_state(data, which, val);
2749
2750out_unlock:
2751 irq_put_desc_busunlock(desc, flags);
2752 return err;
2753}
2754EXPORT_SYMBOL_GPL(irq_set_irqchip_state);