xref: /openbmc/linux/kernel/irq/manage.c (revision d7a3d85e)
1 /*
2  * linux/kernel/irq/manage.c
3  *
4  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
5  * Copyright (C) 2005-2006 Thomas Gleixner
6  *
7  * This file contains driver APIs to the irq subsystem.
8  */
9 
10 #define pr_fmt(fmt) "genirq: " fmt
11 
12 #include <linux/irq.h>
13 #include <linux/kthread.h>
14 #include <linux/module.h>
15 #include <linux/random.h>
16 #include <linux/interrupt.h>
17 #include <linux/slab.h>
18 #include <linux/sched.h>
19 #include <linux/sched/rt.h>
20 #include <linux/task_work.h>
21 
22 #include "internals.h"
23 
24 #ifdef CONFIG_IRQ_FORCED_THREADING
25 __read_mostly bool force_irqthreads;
26 
27 static int __init setup_forced_irqthreads(char *arg)
28 {
29 	force_irqthreads = true;
30 	return 0;
31 }
32 early_param("threadirqs", setup_forced_irqthreads);
33 #endif
34 
35 static void __synchronize_hardirq(struct irq_desc *desc)
36 {
37 	bool inprogress;
38 
39 	do {
40 		unsigned long flags;
41 
42 		/*
43 		 * Wait until we're out of the critical section.  This might
44 		 * give the wrong answer due to the lack of memory barriers.
45 		 */
46 		while (irqd_irq_inprogress(&desc->irq_data))
47 			cpu_relax();
48 
49 		/* Ok, that indicated we're done: double-check carefully. */
50 		raw_spin_lock_irqsave(&desc->lock, flags);
51 		inprogress = irqd_irq_inprogress(&desc->irq_data);
52 		raw_spin_unlock_irqrestore(&desc->lock, flags);
53 
54 		/* Oops, that failed? */
55 	} while (inprogress);
56 }
57 
58 /**
59  *	synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
60  *	@irq: interrupt number to wait for
61  *
62  *	This function waits for any pending hard IRQ handlers for this
63  *	interrupt to complete before returning. If you use this
64  *	function while holding a resource the IRQ handler may need you
65  *	will deadlock. It does not take associated threaded handlers
66  *	into account.
67  *
68  *	Do not use this for shutdown scenarios where you must be sure
69  *	that all parts (hardirq and threaded handler) have completed.
70  *
71  *	Returns: false if a threaded handler is active.
72  *
73  *	This function may be called - with care - from IRQ context.
74  */
75 bool synchronize_hardirq(unsigned int irq)
76 {
77 	struct irq_desc *desc = irq_to_desc(irq);
78 
79 	if (desc) {
80 		__synchronize_hardirq(desc);
81 		return !atomic_read(&desc->threads_active);
82 	}
83 
84 	return true;
85 }
86 EXPORT_SYMBOL(synchronize_hardirq);
87 
88 /**
89  *	synchronize_irq - wait for pending IRQ handlers (on other CPUs)
90  *	@irq: interrupt number to wait for
91  *
92  *	This function waits for any pending IRQ handlers for this interrupt
93  *	to complete before returning. If you use this function while
94  *	holding a resource the IRQ handler may need you will deadlock.
95  *
96  *	This function may be called - with care - from IRQ context.
97  */
98 void synchronize_irq(unsigned int irq)
99 {
100 	struct irq_desc *desc = irq_to_desc(irq);
101 
102 	if (desc) {
103 		__synchronize_hardirq(desc);
104 		/*
105 		 * We made sure that no hardirq handler is
106 		 * running. Now verify that no threaded handlers are
107 		 * active.
108 		 */
109 		wait_event(desc->wait_for_threads,
110 			   !atomic_read(&desc->threads_active));
111 	}
112 }
113 EXPORT_SYMBOL(synchronize_irq);
114 
115 #ifdef CONFIG_SMP
116 cpumask_var_t irq_default_affinity;
117 
118 /**
119  *	irq_can_set_affinity - Check if the affinity of a given irq can be set
120  *	@irq:		Interrupt to check
121  *
122  */
123 int irq_can_set_affinity(unsigned int irq)
124 {
125 	struct irq_desc *desc = irq_to_desc(irq);
126 
127 	if (!desc || !irqd_can_balance(&desc->irq_data) ||
128 	    !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
129 		return 0;
130 
131 	return 1;
132 }
133 
134 /**
135  *	irq_set_thread_affinity - Notify irq threads to adjust affinity
136  *	@desc:		irq descriptor which has affitnity changed
137  *
138  *	We just set IRQTF_AFFINITY and delegate the affinity setting
139  *	to the interrupt thread itself. We can not call
140  *	set_cpus_allowed_ptr() here as we hold desc->lock and this
141  *	code can be called from hard interrupt context.
142  */
143 void irq_set_thread_affinity(struct irq_desc *desc)
144 {
145 	struct irqaction *action = desc->action;
146 
147 	while (action) {
148 		if (action->thread)
149 			set_bit(IRQTF_AFFINITY, &action->thread_flags);
150 		action = action->next;
151 	}
152 }
153 
154 #ifdef CONFIG_GENERIC_PENDING_IRQ
155 static inline bool irq_can_move_pcntxt(struct irq_data *data)
156 {
157 	return irqd_can_move_in_process_context(data);
158 }
159 static inline bool irq_move_pending(struct irq_data *data)
160 {
161 	return irqd_is_setaffinity_pending(data);
162 }
163 static inline void
164 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
165 {
166 	cpumask_copy(desc->pending_mask, mask);
167 }
168 static inline void
169 irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
170 {
171 	cpumask_copy(mask, desc->pending_mask);
172 }
173 #else
174 static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
175 static inline bool irq_move_pending(struct irq_data *data) { return false; }
176 static inline void
177 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
178 static inline void
179 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
180 #endif
181 
182 int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
183 			bool force)
184 {
185 	struct irq_desc *desc = irq_data_to_desc(data);
186 	struct irq_chip *chip = irq_data_get_irq_chip(data);
187 	int ret;
188 
189 	ret = chip->irq_set_affinity(data, mask, force);
190 	switch (ret) {
191 	case IRQ_SET_MASK_OK:
192 	case IRQ_SET_MASK_OK_DONE:
193 		cpumask_copy(data->affinity, mask);
194 	case IRQ_SET_MASK_OK_NOCOPY:
195 		irq_set_thread_affinity(desc);
196 		ret = 0;
197 	}
198 
199 	return ret;
200 }
201 
202 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
203 			    bool force)
204 {
205 	struct irq_chip *chip = irq_data_get_irq_chip(data);
206 	struct irq_desc *desc = irq_data_to_desc(data);
207 	int ret = 0;
208 
209 	if (!chip || !chip->irq_set_affinity)
210 		return -EINVAL;
211 
212 	if (irq_can_move_pcntxt(data)) {
213 		ret = irq_do_set_affinity(data, mask, force);
214 	} else {
215 		irqd_set_move_pending(data);
216 		irq_copy_pending(desc, mask);
217 	}
218 
219 	if (desc->affinity_notify) {
220 		kref_get(&desc->affinity_notify->kref);
221 		schedule_work(&desc->affinity_notify->work);
222 	}
223 	irqd_set(data, IRQD_AFFINITY_SET);
224 
225 	return ret;
226 }
227 
228 int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
229 {
230 	struct irq_desc *desc = irq_to_desc(irq);
231 	unsigned long flags;
232 	int ret;
233 
234 	if (!desc)
235 		return -EINVAL;
236 
237 	raw_spin_lock_irqsave(&desc->lock, flags);
238 	ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
239 	raw_spin_unlock_irqrestore(&desc->lock, flags);
240 	return ret;
241 }
242 
243 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
244 {
245 	unsigned long flags;
246 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
247 
248 	if (!desc)
249 		return -EINVAL;
250 	desc->affinity_hint = m;
251 	irq_put_desc_unlock(desc, flags);
252 	/* set the initial affinity to prevent every interrupt being on CPU0 */
253 	if (m)
254 		__irq_set_affinity(irq, m, false);
255 	return 0;
256 }
257 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
258 
259 static void irq_affinity_notify(struct work_struct *work)
260 {
261 	struct irq_affinity_notify *notify =
262 		container_of(work, struct irq_affinity_notify, work);
263 	struct irq_desc *desc = irq_to_desc(notify->irq);
264 	cpumask_var_t cpumask;
265 	unsigned long flags;
266 
267 	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
268 		goto out;
269 
270 	raw_spin_lock_irqsave(&desc->lock, flags);
271 	if (irq_move_pending(&desc->irq_data))
272 		irq_get_pending(cpumask, desc);
273 	else
274 		cpumask_copy(cpumask, desc->irq_data.affinity);
275 	raw_spin_unlock_irqrestore(&desc->lock, flags);
276 
277 	notify->notify(notify, cpumask);
278 
279 	free_cpumask_var(cpumask);
280 out:
281 	kref_put(&notify->kref, notify->release);
282 }
283 
284 /**
285  *	irq_set_affinity_notifier - control notification of IRQ affinity changes
286  *	@irq:		Interrupt for which to enable/disable notification
287  *	@notify:	Context for notification, or %NULL to disable
288  *			notification.  Function pointers must be initialised;
289  *			the other fields will be initialised by this function.
290  *
291  *	Must be called in process context.  Notification may only be enabled
292  *	after the IRQ is allocated and must be disabled before the IRQ is
293  *	freed using free_irq().
294  */
295 int
296 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
297 {
298 	struct irq_desc *desc = irq_to_desc(irq);
299 	struct irq_affinity_notify *old_notify;
300 	unsigned long flags;
301 
302 	/* The release function is promised process context */
303 	might_sleep();
304 
305 	if (!desc)
306 		return -EINVAL;
307 
308 	/* Complete initialisation of *notify */
309 	if (notify) {
310 		notify->irq = irq;
311 		kref_init(&notify->kref);
312 		INIT_WORK(&notify->work, irq_affinity_notify);
313 	}
314 
315 	raw_spin_lock_irqsave(&desc->lock, flags);
316 	old_notify = desc->affinity_notify;
317 	desc->affinity_notify = notify;
318 	raw_spin_unlock_irqrestore(&desc->lock, flags);
319 
320 	if (old_notify)
321 		kref_put(&old_notify->kref, old_notify->release);
322 
323 	return 0;
324 }
325 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
326 
327 #ifndef CONFIG_AUTO_IRQ_AFFINITY
328 /*
329  * Generic version of the affinity autoselector.
330  */
331 static int
332 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
333 {
334 	struct cpumask *set = irq_default_affinity;
335 	int node = desc->irq_data.node;
336 
337 	/* Excludes PER_CPU and NO_BALANCE interrupts */
338 	if (!irq_can_set_affinity(irq))
339 		return 0;
340 
341 	/*
342 	 * Preserve an userspace affinity setup, but make sure that
343 	 * one of the targets is online.
344 	 */
345 	if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
346 		if (cpumask_intersects(desc->irq_data.affinity,
347 				       cpu_online_mask))
348 			set = desc->irq_data.affinity;
349 		else
350 			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
351 	}
352 
353 	cpumask_and(mask, cpu_online_mask, set);
354 	if (node != NUMA_NO_NODE) {
355 		const struct cpumask *nodemask = cpumask_of_node(node);
356 
357 		/* make sure at least one of the cpus in nodemask is online */
358 		if (cpumask_intersects(mask, nodemask))
359 			cpumask_and(mask, mask, nodemask);
360 	}
361 	irq_do_set_affinity(&desc->irq_data, mask, false);
362 	return 0;
363 }
364 #else
365 static inline int
366 setup_affinity(unsigned int irq, struct irq_desc *d, struct cpumask *mask)
367 {
368 	return irq_select_affinity(irq);
369 }
370 #endif
371 
372 /*
373  * Called when affinity is set via /proc/irq
374  */
375 int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
376 {
377 	struct irq_desc *desc = irq_to_desc(irq);
378 	unsigned long flags;
379 	int ret;
380 
381 	raw_spin_lock_irqsave(&desc->lock, flags);
382 	ret = setup_affinity(irq, desc, mask);
383 	raw_spin_unlock_irqrestore(&desc->lock, flags);
384 	return ret;
385 }
386 
387 #else
388 static inline int
389 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
390 {
391 	return 0;
392 }
393 #endif
394 
395 void __disable_irq(struct irq_desc *desc, unsigned int irq)
396 {
397 	if (!desc->depth++)
398 		irq_disable(desc);
399 }
400 
401 static int __disable_irq_nosync(unsigned int irq)
402 {
403 	unsigned long flags;
404 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
405 
406 	if (!desc)
407 		return -EINVAL;
408 	__disable_irq(desc, irq);
409 	irq_put_desc_busunlock(desc, flags);
410 	return 0;
411 }
412 
413 /**
414  *	disable_irq_nosync - disable an irq without waiting
415  *	@irq: Interrupt to disable
416  *
417  *	Disable the selected interrupt line.  Disables and Enables are
418  *	nested.
419  *	Unlike disable_irq(), this function does not ensure existing
420  *	instances of the IRQ handler have completed before returning.
421  *
422  *	This function may be called from IRQ context.
423  */
424 void disable_irq_nosync(unsigned int irq)
425 {
426 	__disable_irq_nosync(irq);
427 }
428 EXPORT_SYMBOL(disable_irq_nosync);
429 
430 /**
431  *	disable_irq - disable an irq and wait for completion
432  *	@irq: Interrupt to disable
433  *
434  *	Disable the selected interrupt line.  Enables and Disables are
435  *	nested.
436  *	This function waits for any pending IRQ handlers for this interrupt
437  *	to complete before returning. If you use this function while
438  *	holding a resource the IRQ handler may need you will deadlock.
439  *
440  *	This function may be called - with care - from IRQ context.
441  */
442 void disable_irq(unsigned int irq)
443 {
444 	if (!__disable_irq_nosync(irq))
445 		synchronize_irq(irq);
446 }
447 EXPORT_SYMBOL(disable_irq);
448 
449 /**
450  *	disable_hardirq - disables an irq and waits for hardirq completion
451  *	@irq: Interrupt to disable
452  *
453  *	Disable the selected interrupt line.  Enables and Disables are
454  *	nested.
455  *	This function waits for any pending hard IRQ handlers for this
456  *	interrupt to complete before returning. If you use this function while
457  *	holding a resource the hard IRQ handler may need you will deadlock.
458  *
459  *	When used to optimistically disable an interrupt from atomic context
460  *	the return value must be checked.
461  *
462  *	Returns: false if a threaded handler is active.
463  *
464  *	This function may be called - with care - from IRQ context.
465  */
466 bool disable_hardirq(unsigned int irq)
467 {
468 	if (!__disable_irq_nosync(irq))
469 		return synchronize_hardirq(irq);
470 
471 	return false;
472 }
473 EXPORT_SYMBOL_GPL(disable_hardirq);
474 
475 void __enable_irq(struct irq_desc *desc, unsigned int irq)
476 {
477 	switch (desc->depth) {
478 	case 0:
479  err_out:
480 		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
481 		break;
482 	case 1: {
483 		if (desc->istate & IRQS_SUSPENDED)
484 			goto err_out;
485 		/* Prevent probing on this irq: */
486 		irq_settings_set_noprobe(desc);
487 		irq_enable(desc);
488 		check_irq_resend(desc, irq);
489 		/* fall-through */
490 	}
491 	default:
492 		desc->depth--;
493 	}
494 }
495 
496 /**
497  *	enable_irq - enable handling of an irq
498  *	@irq: Interrupt to enable
499  *
500  *	Undoes the effect of one call to disable_irq().  If this
501  *	matches the last disable, processing of interrupts on this
502  *	IRQ line is re-enabled.
503  *
504  *	This function may be called from IRQ context only when
505  *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
506  */
507 void enable_irq(unsigned int irq)
508 {
509 	unsigned long flags;
510 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
511 
512 	if (!desc)
513 		return;
514 	if (WARN(!desc->irq_data.chip,
515 		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
516 		goto out;
517 
518 	__enable_irq(desc, irq);
519 out:
520 	irq_put_desc_busunlock(desc, flags);
521 }
522 EXPORT_SYMBOL(enable_irq);
523 
524 static int set_irq_wake_real(unsigned int irq, unsigned int on)
525 {
526 	struct irq_desc *desc = irq_to_desc(irq);
527 	int ret = -ENXIO;
528 
529 	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
530 		return 0;
531 
532 	if (desc->irq_data.chip->irq_set_wake)
533 		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
534 
535 	return ret;
536 }
537 
538 /**
539  *	irq_set_irq_wake - control irq power management wakeup
540  *	@irq:	interrupt to control
541  *	@on:	enable/disable power management wakeup
542  *
543  *	Enable/disable power management wakeup mode, which is
544  *	disabled by default.  Enables and disables must match,
545  *	just as they match for non-wakeup mode support.
546  *
547  *	Wakeup mode lets this IRQ wake the system from sleep
548  *	states like "suspend to RAM".
549  */
550 int irq_set_irq_wake(unsigned int irq, unsigned int on)
551 {
552 	unsigned long flags;
553 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
554 	int ret = 0;
555 
556 	if (!desc)
557 		return -EINVAL;
558 
559 	/* wakeup-capable irqs can be shared between drivers that
560 	 * don't need to have the same sleep mode behaviors.
561 	 */
562 	if (on) {
563 		if (desc->wake_depth++ == 0) {
564 			ret = set_irq_wake_real(irq, on);
565 			if (ret)
566 				desc->wake_depth = 0;
567 			else
568 				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
569 		}
570 	} else {
571 		if (desc->wake_depth == 0) {
572 			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
573 		} else if (--desc->wake_depth == 0) {
574 			ret = set_irq_wake_real(irq, on);
575 			if (ret)
576 				desc->wake_depth = 1;
577 			else
578 				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
579 		}
580 	}
581 	irq_put_desc_busunlock(desc, flags);
582 	return ret;
583 }
584 EXPORT_SYMBOL(irq_set_irq_wake);
585 
586 /*
587  * Internal function that tells the architecture code whether a
588  * particular irq has been exclusively allocated or is available
589  * for driver use.
590  */
591 int can_request_irq(unsigned int irq, unsigned long irqflags)
592 {
593 	unsigned long flags;
594 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
595 	int canrequest = 0;
596 
597 	if (!desc)
598 		return 0;
599 
600 	if (irq_settings_can_request(desc)) {
601 		if (!desc->action ||
602 		    irqflags & desc->action->flags & IRQF_SHARED)
603 			canrequest = 1;
604 	}
605 	irq_put_desc_unlock(desc, flags);
606 	return canrequest;
607 }
608 
609 int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
610 		      unsigned long flags)
611 {
612 	struct irq_chip *chip = desc->irq_data.chip;
613 	int ret, unmask = 0;
614 
615 	if (!chip || !chip->irq_set_type) {
616 		/*
617 		 * IRQF_TRIGGER_* but the PIC does not support multiple
618 		 * flow-types?
619 		 */
620 		pr_debug("No set_type function for IRQ %d (%s)\n", irq,
621 			 chip ? (chip->name ? : "unknown") : "unknown");
622 		return 0;
623 	}
624 
625 	flags &= IRQ_TYPE_SENSE_MASK;
626 
627 	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
628 		if (!irqd_irq_masked(&desc->irq_data))
629 			mask_irq(desc);
630 		if (!irqd_irq_disabled(&desc->irq_data))
631 			unmask = 1;
632 	}
633 
634 	/* caller masked out all except trigger mode flags */
635 	ret = chip->irq_set_type(&desc->irq_data, flags);
636 
637 	switch (ret) {
638 	case IRQ_SET_MASK_OK:
639 	case IRQ_SET_MASK_OK_DONE:
640 		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
641 		irqd_set(&desc->irq_data, flags);
642 
643 	case IRQ_SET_MASK_OK_NOCOPY:
644 		flags = irqd_get_trigger_type(&desc->irq_data);
645 		irq_settings_set_trigger_mask(desc, flags);
646 		irqd_clear(&desc->irq_data, IRQD_LEVEL);
647 		irq_settings_clr_level(desc);
648 		if (flags & IRQ_TYPE_LEVEL_MASK) {
649 			irq_settings_set_level(desc);
650 			irqd_set(&desc->irq_data, IRQD_LEVEL);
651 		}
652 
653 		ret = 0;
654 		break;
655 	default:
656 		pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
657 		       flags, irq, chip->irq_set_type);
658 	}
659 	if (unmask)
660 		unmask_irq(desc);
661 	return ret;
662 }
663 
664 #ifdef CONFIG_HARDIRQS_SW_RESEND
665 int irq_set_parent(int irq, int parent_irq)
666 {
667 	unsigned long flags;
668 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
669 
670 	if (!desc)
671 		return -EINVAL;
672 
673 	desc->parent_irq = parent_irq;
674 
675 	irq_put_desc_unlock(desc, flags);
676 	return 0;
677 }
678 #endif
679 
680 /*
681  * Default primary interrupt handler for threaded interrupts. Is
682  * assigned as primary handler when request_threaded_irq is called
683  * with handler == NULL. Useful for oneshot interrupts.
684  */
685 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
686 {
687 	return IRQ_WAKE_THREAD;
688 }
689 
690 /*
691  * Primary handler for nested threaded interrupts. Should never be
692  * called.
693  */
694 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
695 {
696 	WARN(1, "Primary handler called for nested irq %d\n", irq);
697 	return IRQ_NONE;
698 }
699 
700 static int irq_wait_for_interrupt(struct irqaction *action)
701 {
702 	set_current_state(TASK_INTERRUPTIBLE);
703 
704 	while (!kthread_should_stop()) {
705 
706 		if (test_and_clear_bit(IRQTF_RUNTHREAD,
707 				       &action->thread_flags)) {
708 			__set_current_state(TASK_RUNNING);
709 			return 0;
710 		}
711 		schedule();
712 		set_current_state(TASK_INTERRUPTIBLE);
713 	}
714 	__set_current_state(TASK_RUNNING);
715 	return -1;
716 }
717 
718 /*
719  * Oneshot interrupts keep the irq line masked until the threaded
720  * handler finished. unmask if the interrupt has not been disabled and
721  * is marked MASKED.
722  */
723 static void irq_finalize_oneshot(struct irq_desc *desc,
724 				 struct irqaction *action)
725 {
726 	if (!(desc->istate & IRQS_ONESHOT))
727 		return;
728 again:
729 	chip_bus_lock(desc);
730 	raw_spin_lock_irq(&desc->lock);
731 
732 	/*
733 	 * Implausible though it may be we need to protect us against
734 	 * the following scenario:
735 	 *
736 	 * The thread is faster done than the hard interrupt handler
737 	 * on the other CPU. If we unmask the irq line then the
738 	 * interrupt can come in again and masks the line, leaves due
739 	 * to IRQS_INPROGRESS and the irq line is masked forever.
740 	 *
741 	 * This also serializes the state of shared oneshot handlers
742 	 * versus "desc->threads_onehsot |= action->thread_mask;" in
743 	 * irq_wake_thread(). See the comment there which explains the
744 	 * serialization.
745 	 */
746 	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
747 		raw_spin_unlock_irq(&desc->lock);
748 		chip_bus_sync_unlock(desc);
749 		cpu_relax();
750 		goto again;
751 	}
752 
753 	/*
754 	 * Now check again, whether the thread should run. Otherwise
755 	 * we would clear the threads_oneshot bit of this thread which
756 	 * was just set.
757 	 */
758 	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
759 		goto out_unlock;
760 
761 	desc->threads_oneshot &= ~action->thread_mask;
762 
763 	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
764 	    irqd_irq_masked(&desc->irq_data))
765 		unmask_threaded_irq(desc);
766 
767 out_unlock:
768 	raw_spin_unlock_irq(&desc->lock);
769 	chip_bus_sync_unlock(desc);
770 }
771 
772 #ifdef CONFIG_SMP
773 /*
774  * Check whether we need to change the affinity of the interrupt thread.
775  */
776 static void
777 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
778 {
779 	cpumask_var_t mask;
780 	bool valid = true;
781 
782 	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
783 		return;
784 
785 	/*
786 	 * In case we are out of memory we set IRQTF_AFFINITY again and
787 	 * try again next time
788 	 */
789 	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
790 		set_bit(IRQTF_AFFINITY, &action->thread_flags);
791 		return;
792 	}
793 
794 	raw_spin_lock_irq(&desc->lock);
795 	/*
796 	 * This code is triggered unconditionally. Check the affinity
797 	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
798 	 */
799 	if (desc->irq_data.affinity)
800 		cpumask_copy(mask, desc->irq_data.affinity);
801 	else
802 		valid = false;
803 	raw_spin_unlock_irq(&desc->lock);
804 
805 	if (valid)
806 		set_cpus_allowed_ptr(current, mask);
807 	free_cpumask_var(mask);
808 }
809 #else
810 static inline void
811 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
812 #endif
813 
814 /*
815  * Interrupts which are not explicitely requested as threaded
816  * interrupts rely on the implicit bh/preempt disable of the hard irq
817  * context. So we need to disable bh here to avoid deadlocks and other
818  * side effects.
819  */
820 static irqreturn_t
821 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
822 {
823 	irqreturn_t ret;
824 
825 	local_bh_disable();
826 	ret = action->thread_fn(action->irq, action->dev_id);
827 	irq_finalize_oneshot(desc, action);
828 	local_bh_enable();
829 	return ret;
830 }
831 
832 /*
833  * Interrupts explicitly requested as threaded interrupts want to be
834  * preemtible - many of them need to sleep and wait for slow busses to
835  * complete.
836  */
837 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
838 		struct irqaction *action)
839 {
840 	irqreturn_t ret;
841 
842 	ret = action->thread_fn(action->irq, action->dev_id);
843 	irq_finalize_oneshot(desc, action);
844 	return ret;
845 }
846 
847 static void wake_threads_waitq(struct irq_desc *desc)
848 {
849 	if (atomic_dec_and_test(&desc->threads_active))
850 		wake_up(&desc->wait_for_threads);
851 }
852 
853 static void irq_thread_dtor(struct callback_head *unused)
854 {
855 	struct task_struct *tsk = current;
856 	struct irq_desc *desc;
857 	struct irqaction *action;
858 
859 	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
860 		return;
861 
862 	action = kthread_data(tsk);
863 
864 	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
865 	       tsk->comm, tsk->pid, action->irq);
866 
867 
868 	desc = irq_to_desc(action->irq);
869 	/*
870 	 * If IRQTF_RUNTHREAD is set, we need to decrement
871 	 * desc->threads_active and wake possible waiters.
872 	 */
873 	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
874 		wake_threads_waitq(desc);
875 
876 	/* Prevent a stale desc->threads_oneshot */
877 	irq_finalize_oneshot(desc, action);
878 }
879 
880 /*
881  * Interrupt handler thread
882  */
883 static int irq_thread(void *data)
884 {
885 	struct callback_head on_exit_work;
886 	struct irqaction *action = data;
887 	struct irq_desc *desc = irq_to_desc(action->irq);
888 	irqreturn_t (*handler_fn)(struct irq_desc *desc,
889 			struct irqaction *action);
890 
891 	if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
892 					&action->thread_flags))
893 		handler_fn = irq_forced_thread_fn;
894 	else
895 		handler_fn = irq_thread_fn;
896 
897 	init_task_work(&on_exit_work, irq_thread_dtor);
898 	task_work_add(current, &on_exit_work, false);
899 
900 	irq_thread_check_affinity(desc, action);
901 
902 	while (!irq_wait_for_interrupt(action)) {
903 		irqreturn_t action_ret;
904 
905 		irq_thread_check_affinity(desc, action);
906 
907 		action_ret = handler_fn(desc, action);
908 		if (action_ret == IRQ_HANDLED)
909 			atomic_inc(&desc->threads_handled);
910 
911 		wake_threads_waitq(desc);
912 	}
913 
914 	/*
915 	 * This is the regular exit path. __free_irq() is stopping the
916 	 * thread via kthread_stop() after calling
917 	 * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
918 	 * oneshot mask bit can be set. We cannot verify that as we
919 	 * cannot touch the oneshot mask at this point anymore as
920 	 * __setup_irq() might have given out currents thread_mask
921 	 * again.
922 	 */
923 	task_work_cancel(current, irq_thread_dtor);
924 	return 0;
925 }
926 
927 /**
928  *	irq_wake_thread - wake the irq thread for the action identified by dev_id
929  *	@irq:		Interrupt line
930  *	@dev_id:	Device identity for which the thread should be woken
931  *
932  */
933 void irq_wake_thread(unsigned int irq, void *dev_id)
934 {
935 	struct irq_desc *desc = irq_to_desc(irq);
936 	struct irqaction *action;
937 	unsigned long flags;
938 
939 	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
940 		return;
941 
942 	raw_spin_lock_irqsave(&desc->lock, flags);
943 	for (action = desc->action; action; action = action->next) {
944 		if (action->dev_id == dev_id) {
945 			if (action->thread)
946 				__irq_wake_thread(desc, action);
947 			break;
948 		}
949 	}
950 	raw_spin_unlock_irqrestore(&desc->lock, flags);
951 }
952 EXPORT_SYMBOL_GPL(irq_wake_thread);
953 
954 static void irq_setup_forced_threading(struct irqaction *new)
955 {
956 	if (!force_irqthreads)
957 		return;
958 	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
959 		return;
960 
961 	new->flags |= IRQF_ONESHOT;
962 
963 	if (!new->thread_fn) {
964 		set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
965 		new->thread_fn = new->handler;
966 		new->handler = irq_default_primary_handler;
967 	}
968 }
969 
970 static int irq_request_resources(struct irq_desc *desc)
971 {
972 	struct irq_data *d = &desc->irq_data;
973 	struct irq_chip *c = d->chip;
974 
975 	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
976 }
977 
978 static void irq_release_resources(struct irq_desc *desc)
979 {
980 	struct irq_data *d = &desc->irq_data;
981 	struct irq_chip *c = d->chip;
982 
983 	if (c->irq_release_resources)
984 		c->irq_release_resources(d);
985 }
986 
987 /*
988  * Internal function to register an irqaction - typically used to
989  * allocate special interrupts that are part of the architecture.
990  */
991 static int
992 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
993 {
994 	struct irqaction *old, **old_ptr;
995 	unsigned long flags, thread_mask = 0;
996 	int ret, nested, shared = 0;
997 	cpumask_var_t mask;
998 
999 	if (!desc)
1000 		return -EINVAL;
1001 
1002 	if (desc->irq_data.chip == &no_irq_chip)
1003 		return -ENOSYS;
1004 	if (!try_module_get(desc->owner))
1005 		return -ENODEV;
1006 
1007 	/*
1008 	 * Check whether the interrupt nests into another interrupt
1009 	 * thread.
1010 	 */
1011 	nested = irq_settings_is_nested_thread(desc);
1012 	if (nested) {
1013 		if (!new->thread_fn) {
1014 			ret = -EINVAL;
1015 			goto out_mput;
1016 		}
1017 		/*
1018 		 * Replace the primary handler which was provided from
1019 		 * the driver for non nested interrupt handling by the
1020 		 * dummy function which warns when called.
1021 		 */
1022 		new->handler = irq_nested_primary_handler;
1023 	} else {
1024 		if (irq_settings_can_thread(desc))
1025 			irq_setup_forced_threading(new);
1026 	}
1027 
1028 	/*
1029 	 * Create a handler thread when a thread function is supplied
1030 	 * and the interrupt does not nest into another interrupt
1031 	 * thread.
1032 	 */
1033 	if (new->thread_fn && !nested) {
1034 		struct task_struct *t;
1035 		static const struct sched_param param = {
1036 			.sched_priority = MAX_USER_RT_PRIO/2,
1037 		};
1038 
1039 		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1040 				   new->name);
1041 		if (IS_ERR(t)) {
1042 			ret = PTR_ERR(t);
1043 			goto out_mput;
1044 		}
1045 
1046 		sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1047 
1048 		/*
1049 		 * We keep the reference to the task struct even if
1050 		 * the thread dies to avoid that the interrupt code
1051 		 * references an already freed task_struct.
1052 		 */
1053 		get_task_struct(t);
1054 		new->thread = t;
1055 		/*
1056 		 * Tell the thread to set its affinity. This is
1057 		 * important for shared interrupt handlers as we do
1058 		 * not invoke setup_affinity() for the secondary
1059 		 * handlers as everything is already set up. Even for
1060 		 * interrupts marked with IRQF_NO_BALANCE this is
1061 		 * correct as we want the thread to move to the cpu(s)
1062 		 * on which the requesting code placed the interrupt.
1063 		 */
1064 		set_bit(IRQTF_AFFINITY, &new->thread_flags);
1065 	}
1066 
1067 	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1068 		ret = -ENOMEM;
1069 		goto out_thread;
1070 	}
1071 
1072 	/*
1073 	 * Drivers are often written to work w/o knowledge about the
1074 	 * underlying irq chip implementation, so a request for a
1075 	 * threaded irq without a primary hard irq context handler
1076 	 * requires the ONESHOT flag to be set. Some irq chips like
1077 	 * MSI based interrupts are per se one shot safe. Check the
1078 	 * chip flags, so we can avoid the unmask dance at the end of
1079 	 * the threaded handler for those.
1080 	 */
1081 	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1082 		new->flags &= ~IRQF_ONESHOT;
1083 
1084 	/*
1085 	 * The following block of code has to be executed atomically
1086 	 */
1087 	raw_spin_lock_irqsave(&desc->lock, flags);
1088 	old_ptr = &desc->action;
1089 	old = *old_ptr;
1090 	if (old) {
1091 		/*
1092 		 * Can't share interrupts unless both agree to and are
1093 		 * the same type (level, edge, polarity). So both flag
1094 		 * fields must have IRQF_SHARED set and the bits which
1095 		 * set the trigger type must match. Also all must
1096 		 * agree on ONESHOT.
1097 		 */
1098 		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1099 		    ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1100 		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1101 			goto mismatch;
1102 
1103 		/* All handlers must agree on per-cpuness */
1104 		if ((old->flags & IRQF_PERCPU) !=
1105 		    (new->flags & IRQF_PERCPU))
1106 			goto mismatch;
1107 
1108 		/* add new interrupt at end of irq queue */
1109 		do {
1110 			/*
1111 			 * Or all existing action->thread_mask bits,
1112 			 * so we can find the next zero bit for this
1113 			 * new action.
1114 			 */
1115 			thread_mask |= old->thread_mask;
1116 			old_ptr = &old->next;
1117 			old = *old_ptr;
1118 		} while (old);
1119 		shared = 1;
1120 	}
1121 
1122 	/*
1123 	 * Setup the thread mask for this irqaction for ONESHOT. For
1124 	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1125 	 * conditional in irq_wake_thread().
1126 	 */
1127 	if (new->flags & IRQF_ONESHOT) {
1128 		/*
1129 		 * Unlikely to have 32 resp 64 irqs sharing one line,
1130 		 * but who knows.
1131 		 */
1132 		if (thread_mask == ~0UL) {
1133 			ret = -EBUSY;
1134 			goto out_mask;
1135 		}
1136 		/*
1137 		 * The thread_mask for the action is or'ed to
1138 		 * desc->thread_active to indicate that the
1139 		 * IRQF_ONESHOT thread handler has been woken, but not
1140 		 * yet finished. The bit is cleared when a thread
1141 		 * completes. When all threads of a shared interrupt
1142 		 * line have completed desc->threads_active becomes
1143 		 * zero and the interrupt line is unmasked. See
1144 		 * handle.c:irq_wake_thread() for further information.
1145 		 *
1146 		 * If no thread is woken by primary (hard irq context)
1147 		 * interrupt handlers, then desc->threads_active is
1148 		 * also checked for zero to unmask the irq line in the
1149 		 * affected hard irq flow handlers
1150 		 * (handle_[fasteoi|level]_irq).
1151 		 *
1152 		 * The new action gets the first zero bit of
1153 		 * thread_mask assigned. See the loop above which or's
1154 		 * all existing action->thread_mask bits.
1155 		 */
1156 		new->thread_mask = 1 << ffz(thread_mask);
1157 
1158 	} else if (new->handler == irq_default_primary_handler &&
1159 		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1160 		/*
1161 		 * The interrupt was requested with handler = NULL, so
1162 		 * we use the default primary handler for it. But it
1163 		 * does not have the oneshot flag set. In combination
1164 		 * with level interrupts this is deadly, because the
1165 		 * default primary handler just wakes the thread, then
1166 		 * the irq lines is reenabled, but the device still
1167 		 * has the level irq asserted. Rinse and repeat....
1168 		 *
1169 		 * While this works for edge type interrupts, we play
1170 		 * it safe and reject unconditionally because we can't
1171 		 * say for sure which type this interrupt really
1172 		 * has. The type flags are unreliable as the
1173 		 * underlying chip implementation can override them.
1174 		 */
1175 		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1176 		       irq);
1177 		ret = -EINVAL;
1178 		goto out_mask;
1179 	}
1180 
1181 	if (!shared) {
1182 		ret = irq_request_resources(desc);
1183 		if (ret) {
1184 			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1185 			       new->name, irq, desc->irq_data.chip->name);
1186 			goto out_mask;
1187 		}
1188 
1189 		init_waitqueue_head(&desc->wait_for_threads);
1190 
1191 		/* Setup the type (level, edge polarity) if configured: */
1192 		if (new->flags & IRQF_TRIGGER_MASK) {
1193 			ret = __irq_set_trigger(desc, irq,
1194 					new->flags & IRQF_TRIGGER_MASK);
1195 
1196 			if (ret)
1197 				goto out_mask;
1198 		}
1199 
1200 		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1201 				  IRQS_ONESHOT | IRQS_WAITING);
1202 		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1203 
1204 		if (new->flags & IRQF_PERCPU) {
1205 			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1206 			irq_settings_set_per_cpu(desc);
1207 		}
1208 
1209 		if (new->flags & IRQF_ONESHOT)
1210 			desc->istate |= IRQS_ONESHOT;
1211 
1212 		if (irq_settings_can_autoenable(desc))
1213 			irq_startup(desc, true);
1214 		else
1215 			/* Undo nested disables: */
1216 			desc->depth = 1;
1217 
1218 		/* Exclude IRQ from balancing if requested */
1219 		if (new->flags & IRQF_NOBALANCING) {
1220 			irq_settings_set_no_balancing(desc);
1221 			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1222 		}
1223 
1224 		/* Set default affinity mask once everything is setup */
1225 		setup_affinity(irq, desc, mask);
1226 
1227 	} else if (new->flags & IRQF_TRIGGER_MASK) {
1228 		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1229 		unsigned int omsk = irq_settings_get_trigger_mask(desc);
1230 
1231 		if (nmsk != omsk)
1232 			/* hope the handler works with current  trigger mode */
1233 			pr_warning("irq %d uses trigger mode %u; requested %u\n",
1234 				   irq, nmsk, omsk);
1235 	}
1236 
1237 	new->irq = irq;
1238 	*old_ptr = new;
1239 
1240 	irq_pm_install_action(desc, new);
1241 
1242 	/* Reset broken irq detection when installing new handler */
1243 	desc->irq_count = 0;
1244 	desc->irqs_unhandled = 0;
1245 
1246 	/*
1247 	 * Check whether we disabled the irq via the spurious handler
1248 	 * before. Reenable it and give it another chance.
1249 	 */
1250 	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1251 		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1252 		__enable_irq(desc, irq);
1253 	}
1254 
1255 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1256 
1257 	/*
1258 	 * Strictly no need to wake it up, but hung_task complains
1259 	 * when no hard interrupt wakes the thread up.
1260 	 */
1261 	if (new->thread)
1262 		wake_up_process(new->thread);
1263 
1264 	register_irq_proc(irq, desc);
1265 	new->dir = NULL;
1266 	register_handler_proc(irq, new);
1267 	free_cpumask_var(mask);
1268 
1269 	return 0;
1270 
1271 mismatch:
1272 	if (!(new->flags & IRQF_PROBE_SHARED)) {
1273 		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1274 		       irq, new->flags, new->name, old->flags, old->name);
1275 #ifdef CONFIG_DEBUG_SHIRQ
1276 		dump_stack();
1277 #endif
1278 	}
1279 	ret = -EBUSY;
1280 
1281 out_mask:
1282 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1283 	free_cpumask_var(mask);
1284 
1285 out_thread:
1286 	if (new->thread) {
1287 		struct task_struct *t = new->thread;
1288 
1289 		new->thread = NULL;
1290 		kthread_stop(t);
1291 		put_task_struct(t);
1292 	}
1293 out_mput:
1294 	module_put(desc->owner);
1295 	return ret;
1296 }
1297 
1298 /**
1299  *	setup_irq - setup an interrupt
1300  *	@irq: Interrupt line to setup
1301  *	@act: irqaction for the interrupt
1302  *
1303  * Used to statically setup interrupts in the early boot process.
1304  */
1305 int setup_irq(unsigned int irq, struct irqaction *act)
1306 {
1307 	int retval;
1308 	struct irq_desc *desc = irq_to_desc(irq);
1309 
1310 	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1311 		return -EINVAL;
1312 	chip_bus_lock(desc);
1313 	retval = __setup_irq(irq, desc, act);
1314 	chip_bus_sync_unlock(desc);
1315 
1316 	return retval;
1317 }
1318 EXPORT_SYMBOL_GPL(setup_irq);
1319 
1320 /*
1321  * Internal function to unregister an irqaction - used to free
1322  * regular and special interrupts that are part of the architecture.
1323  */
1324 static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1325 {
1326 	struct irq_desc *desc = irq_to_desc(irq);
1327 	struct irqaction *action, **action_ptr;
1328 	unsigned long flags;
1329 
1330 	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1331 
1332 	if (!desc)
1333 		return NULL;
1334 
1335 	raw_spin_lock_irqsave(&desc->lock, flags);
1336 
1337 	/*
1338 	 * There can be multiple actions per IRQ descriptor, find the right
1339 	 * one based on the dev_id:
1340 	 */
1341 	action_ptr = &desc->action;
1342 	for (;;) {
1343 		action = *action_ptr;
1344 
1345 		if (!action) {
1346 			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1347 			raw_spin_unlock_irqrestore(&desc->lock, flags);
1348 
1349 			return NULL;
1350 		}
1351 
1352 		if (action->dev_id == dev_id)
1353 			break;
1354 		action_ptr = &action->next;
1355 	}
1356 
1357 	/* Found it - now remove it from the list of entries: */
1358 	*action_ptr = action->next;
1359 
1360 	irq_pm_remove_action(desc, action);
1361 
1362 	/* If this was the last handler, shut down the IRQ line: */
1363 	if (!desc->action) {
1364 		irq_shutdown(desc);
1365 		irq_release_resources(desc);
1366 	}
1367 
1368 #ifdef CONFIG_SMP
1369 	/* make sure affinity_hint is cleaned up */
1370 	if (WARN_ON_ONCE(desc->affinity_hint))
1371 		desc->affinity_hint = NULL;
1372 #endif
1373 
1374 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1375 
1376 	unregister_handler_proc(irq, action);
1377 
1378 	/* Make sure it's not being used on another CPU: */
1379 	synchronize_irq(irq);
1380 
1381 #ifdef CONFIG_DEBUG_SHIRQ
1382 	/*
1383 	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1384 	 * event to happen even now it's being freed, so let's make sure that
1385 	 * is so by doing an extra call to the handler ....
1386 	 *
1387 	 * ( We do this after actually deregistering it, to make sure that a
1388 	 *   'real' IRQ doesn't run in * parallel with our fake. )
1389 	 */
1390 	if (action->flags & IRQF_SHARED) {
1391 		local_irq_save(flags);
1392 		action->handler(irq, dev_id);
1393 		local_irq_restore(flags);
1394 	}
1395 #endif
1396 
1397 	if (action->thread) {
1398 		kthread_stop(action->thread);
1399 		put_task_struct(action->thread);
1400 	}
1401 
1402 	module_put(desc->owner);
1403 	return action;
1404 }
1405 
1406 /**
1407  *	remove_irq - free an interrupt
1408  *	@irq: Interrupt line to free
1409  *	@act: irqaction for the interrupt
1410  *
1411  * Used to remove interrupts statically setup by the early boot process.
1412  */
1413 void remove_irq(unsigned int irq, struct irqaction *act)
1414 {
1415 	struct irq_desc *desc = irq_to_desc(irq);
1416 
1417 	if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1418 	    __free_irq(irq, act->dev_id);
1419 }
1420 EXPORT_SYMBOL_GPL(remove_irq);
1421 
1422 /**
1423  *	free_irq - free an interrupt allocated with request_irq
1424  *	@irq: Interrupt line to free
1425  *	@dev_id: Device identity to free
1426  *
1427  *	Remove an interrupt handler. The handler is removed and if the
1428  *	interrupt line is no longer in use by any driver it is disabled.
1429  *	On a shared IRQ the caller must ensure the interrupt is disabled
1430  *	on the card it drives before calling this function. The function
1431  *	does not return until any executing interrupts for this IRQ
1432  *	have completed.
1433  *
1434  *	This function must not be called from interrupt context.
1435  */
1436 void free_irq(unsigned int irq, void *dev_id)
1437 {
1438 	struct irq_desc *desc = irq_to_desc(irq);
1439 
1440 	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1441 		return;
1442 
1443 #ifdef CONFIG_SMP
1444 	if (WARN_ON(desc->affinity_notify))
1445 		desc->affinity_notify = NULL;
1446 #endif
1447 
1448 	chip_bus_lock(desc);
1449 	kfree(__free_irq(irq, dev_id));
1450 	chip_bus_sync_unlock(desc);
1451 }
1452 EXPORT_SYMBOL(free_irq);
1453 
1454 /**
1455  *	request_threaded_irq - allocate an interrupt line
1456  *	@irq: Interrupt line to allocate
1457  *	@handler: Function to be called when the IRQ occurs.
1458  *		  Primary handler for threaded interrupts
1459  *		  If NULL and thread_fn != NULL the default
1460  *		  primary handler is installed
1461  *	@thread_fn: Function called from the irq handler thread
1462  *		    If NULL, no irq thread is created
1463  *	@irqflags: Interrupt type flags
1464  *	@devname: An ascii name for the claiming device
1465  *	@dev_id: A cookie passed back to the handler function
1466  *
1467  *	This call allocates interrupt resources and enables the
1468  *	interrupt line and IRQ handling. From the point this
1469  *	call is made your handler function may be invoked. Since
1470  *	your handler function must clear any interrupt the board
1471  *	raises, you must take care both to initialise your hardware
1472  *	and to set up the interrupt handler in the right order.
1473  *
1474  *	If you want to set up a threaded irq handler for your device
1475  *	then you need to supply @handler and @thread_fn. @handler is
1476  *	still called in hard interrupt context and has to check
1477  *	whether the interrupt originates from the device. If yes it
1478  *	needs to disable the interrupt on the device and return
1479  *	IRQ_WAKE_THREAD which will wake up the handler thread and run
1480  *	@thread_fn. This split handler design is necessary to support
1481  *	shared interrupts.
1482  *
1483  *	Dev_id must be globally unique. Normally the address of the
1484  *	device data structure is used as the cookie. Since the handler
1485  *	receives this value it makes sense to use it.
1486  *
1487  *	If your interrupt is shared you must pass a non NULL dev_id
1488  *	as this is required when freeing the interrupt.
1489  *
1490  *	Flags:
1491  *
1492  *	IRQF_SHARED		Interrupt is shared
1493  *	IRQF_TRIGGER_*		Specify active edge(s) or level
1494  *
1495  */
1496 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1497 			 irq_handler_t thread_fn, unsigned long irqflags,
1498 			 const char *devname, void *dev_id)
1499 {
1500 	struct irqaction *action;
1501 	struct irq_desc *desc;
1502 	int retval;
1503 
1504 	/*
1505 	 * Sanity-check: shared interrupts must pass in a real dev-ID,
1506 	 * otherwise we'll have trouble later trying to figure out
1507 	 * which interrupt is which (messes up the interrupt freeing
1508 	 * logic etc).
1509 	 *
1510 	 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1511 	 * it cannot be set along with IRQF_NO_SUSPEND.
1512 	 */
1513 	if (((irqflags & IRQF_SHARED) && !dev_id) ||
1514 	    (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1515 	    ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1516 		return -EINVAL;
1517 
1518 	desc = irq_to_desc(irq);
1519 	if (!desc)
1520 		return -EINVAL;
1521 
1522 	if (!irq_settings_can_request(desc) ||
1523 	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1524 		return -EINVAL;
1525 
1526 	if (!handler) {
1527 		if (!thread_fn)
1528 			return -EINVAL;
1529 		handler = irq_default_primary_handler;
1530 	}
1531 
1532 	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1533 	if (!action)
1534 		return -ENOMEM;
1535 
1536 	action->handler = handler;
1537 	action->thread_fn = thread_fn;
1538 	action->flags = irqflags;
1539 	action->name = devname;
1540 	action->dev_id = dev_id;
1541 
1542 	chip_bus_lock(desc);
1543 	retval = __setup_irq(irq, desc, action);
1544 	chip_bus_sync_unlock(desc);
1545 
1546 	if (retval)
1547 		kfree(action);
1548 
1549 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
1550 	if (!retval && (irqflags & IRQF_SHARED)) {
1551 		/*
1552 		 * It's a shared IRQ -- the driver ought to be prepared for it
1553 		 * to happen immediately, so let's make sure....
1554 		 * We disable the irq to make sure that a 'real' IRQ doesn't
1555 		 * run in parallel with our fake.
1556 		 */
1557 		unsigned long flags;
1558 
1559 		disable_irq(irq);
1560 		local_irq_save(flags);
1561 
1562 		handler(irq, dev_id);
1563 
1564 		local_irq_restore(flags);
1565 		enable_irq(irq);
1566 	}
1567 #endif
1568 	return retval;
1569 }
1570 EXPORT_SYMBOL(request_threaded_irq);
1571 
1572 /**
1573  *	request_any_context_irq - allocate an interrupt line
1574  *	@irq: Interrupt line to allocate
1575  *	@handler: Function to be called when the IRQ occurs.
1576  *		  Threaded handler for threaded interrupts.
1577  *	@flags: Interrupt type flags
1578  *	@name: An ascii name for the claiming device
1579  *	@dev_id: A cookie passed back to the handler function
1580  *
1581  *	This call allocates interrupt resources and enables the
1582  *	interrupt line and IRQ handling. It selects either a
1583  *	hardirq or threaded handling method depending on the
1584  *	context.
1585  *
1586  *	On failure, it returns a negative value. On success,
1587  *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1588  */
1589 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1590 			    unsigned long flags, const char *name, void *dev_id)
1591 {
1592 	struct irq_desc *desc = irq_to_desc(irq);
1593 	int ret;
1594 
1595 	if (!desc)
1596 		return -EINVAL;
1597 
1598 	if (irq_settings_is_nested_thread(desc)) {
1599 		ret = request_threaded_irq(irq, NULL, handler,
1600 					   flags, name, dev_id);
1601 		return !ret ? IRQC_IS_NESTED : ret;
1602 	}
1603 
1604 	ret = request_irq(irq, handler, flags, name, dev_id);
1605 	return !ret ? IRQC_IS_HARDIRQ : ret;
1606 }
1607 EXPORT_SYMBOL_GPL(request_any_context_irq);
1608 
1609 void enable_percpu_irq(unsigned int irq, unsigned int type)
1610 {
1611 	unsigned int cpu = smp_processor_id();
1612 	unsigned long flags;
1613 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1614 
1615 	if (!desc)
1616 		return;
1617 
1618 	type &= IRQ_TYPE_SENSE_MASK;
1619 	if (type != IRQ_TYPE_NONE) {
1620 		int ret;
1621 
1622 		ret = __irq_set_trigger(desc, irq, type);
1623 
1624 		if (ret) {
1625 			WARN(1, "failed to set type for IRQ%d\n", irq);
1626 			goto out;
1627 		}
1628 	}
1629 
1630 	irq_percpu_enable(desc, cpu);
1631 out:
1632 	irq_put_desc_unlock(desc, flags);
1633 }
1634 EXPORT_SYMBOL_GPL(enable_percpu_irq);
1635 
1636 void disable_percpu_irq(unsigned int irq)
1637 {
1638 	unsigned int cpu = smp_processor_id();
1639 	unsigned long flags;
1640 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1641 
1642 	if (!desc)
1643 		return;
1644 
1645 	irq_percpu_disable(desc, cpu);
1646 	irq_put_desc_unlock(desc, flags);
1647 }
1648 EXPORT_SYMBOL_GPL(disable_percpu_irq);
1649 
1650 /*
1651  * Internal function to unregister a percpu irqaction.
1652  */
1653 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1654 {
1655 	struct irq_desc *desc = irq_to_desc(irq);
1656 	struct irqaction *action;
1657 	unsigned long flags;
1658 
1659 	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1660 
1661 	if (!desc)
1662 		return NULL;
1663 
1664 	raw_spin_lock_irqsave(&desc->lock, flags);
1665 
1666 	action = desc->action;
1667 	if (!action || action->percpu_dev_id != dev_id) {
1668 		WARN(1, "Trying to free already-free IRQ %d\n", irq);
1669 		goto bad;
1670 	}
1671 
1672 	if (!cpumask_empty(desc->percpu_enabled)) {
1673 		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1674 		     irq, cpumask_first(desc->percpu_enabled));
1675 		goto bad;
1676 	}
1677 
1678 	/* Found it - now remove it from the list of entries: */
1679 	desc->action = NULL;
1680 
1681 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1682 
1683 	unregister_handler_proc(irq, action);
1684 
1685 	module_put(desc->owner);
1686 	return action;
1687 
1688 bad:
1689 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1690 	return NULL;
1691 }
1692 
1693 /**
1694  *	remove_percpu_irq - free a per-cpu interrupt
1695  *	@irq: Interrupt line to free
1696  *	@act: irqaction for the interrupt
1697  *
1698  * Used to remove interrupts statically setup by the early boot process.
1699  */
1700 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1701 {
1702 	struct irq_desc *desc = irq_to_desc(irq);
1703 
1704 	if (desc && irq_settings_is_per_cpu_devid(desc))
1705 	    __free_percpu_irq(irq, act->percpu_dev_id);
1706 }
1707 
1708 /**
1709  *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
1710  *	@irq: Interrupt line to free
1711  *	@dev_id: Device identity to free
1712  *
1713  *	Remove a percpu interrupt handler. The handler is removed, but
1714  *	the interrupt line is not disabled. This must be done on each
1715  *	CPU before calling this function. The function does not return
1716  *	until any executing interrupts for this IRQ have completed.
1717  *
1718  *	This function must not be called from interrupt context.
1719  */
1720 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1721 {
1722 	struct irq_desc *desc = irq_to_desc(irq);
1723 
1724 	if (!desc || !irq_settings_is_per_cpu_devid(desc))
1725 		return;
1726 
1727 	chip_bus_lock(desc);
1728 	kfree(__free_percpu_irq(irq, dev_id));
1729 	chip_bus_sync_unlock(desc);
1730 }
1731 
1732 /**
1733  *	setup_percpu_irq - setup a per-cpu interrupt
1734  *	@irq: Interrupt line to setup
1735  *	@act: irqaction for the interrupt
1736  *
1737  * Used to statically setup per-cpu interrupts in the early boot process.
1738  */
1739 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1740 {
1741 	struct irq_desc *desc = irq_to_desc(irq);
1742 	int retval;
1743 
1744 	if (!desc || !irq_settings_is_per_cpu_devid(desc))
1745 		return -EINVAL;
1746 	chip_bus_lock(desc);
1747 	retval = __setup_irq(irq, desc, act);
1748 	chip_bus_sync_unlock(desc);
1749 
1750 	return retval;
1751 }
1752 
1753 /**
1754  *	request_percpu_irq - allocate a percpu interrupt line
1755  *	@irq: Interrupt line to allocate
1756  *	@handler: Function to be called when the IRQ occurs.
1757  *	@devname: An ascii name for the claiming device
1758  *	@dev_id: A percpu cookie passed back to the handler function
1759  *
1760  *	This call allocates interrupt resources, but doesn't
1761  *	automatically enable the interrupt. It has to be done on each
1762  *	CPU using enable_percpu_irq().
1763  *
1764  *	Dev_id must be globally unique. It is a per-cpu variable, and
1765  *	the handler gets called with the interrupted CPU's instance of
1766  *	that variable.
1767  */
1768 int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1769 		       const char *devname, void __percpu *dev_id)
1770 {
1771 	struct irqaction *action;
1772 	struct irq_desc *desc;
1773 	int retval;
1774 
1775 	if (!dev_id)
1776 		return -EINVAL;
1777 
1778 	desc = irq_to_desc(irq);
1779 	if (!desc || !irq_settings_can_request(desc) ||
1780 	    !irq_settings_is_per_cpu_devid(desc))
1781 		return -EINVAL;
1782 
1783 	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1784 	if (!action)
1785 		return -ENOMEM;
1786 
1787 	action->handler = handler;
1788 	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1789 	action->name = devname;
1790 	action->percpu_dev_id = dev_id;
1791 
1792 	chip_bus_lock(desc);
1793 	retval = __setup_irq(irq, desc, action);
1794 	chip_bus_sync_unlock(desc);
1795 
1796 	if (retval)
1797 		kfree(action);
1798 
1799 	return retval;
1800 }
1801 
1802 /**
1803  *	irq_get_irqchip_state - returns the irqchip state of a interrupt.
1804  *	@irq: Interrupt line that is forwarded to a VM
1805  *	@which: One of IRQCHIP_STATE_* the caller wants to know about
1806  *	@state: a pointer to a boolean where the state is to be storeed
1807  *
1808  *	This call snapshots the internal irqchip state of an
1809  *	interrupt, returning into @state the bit corresponding to
1810  *	stage @which
1811  *
1812  *	This function should be called with preemption disabled if the
1813  *	interrupt controller has per-cpu registers.
1814  */
1815 int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1816 			  bool *state)
1817 {
1818 	struct irq_desc *desc;
1819 	struct irq_data *data;
1820 	struct irq_chip *chip;
1821 	unsigned long flags;
1822 	int err = -EINVAL;
1823 
1824 	desc = irq_get_desc_buslock(irq, &flags, 0);
1825 	if (!desc)
1826 		return err;
1827 
1828 	data = irq_desc_get_irq_data(desc);
1829 
1830 	do {
1831 		chip = irq_data_get_irq_chip(data);
1832 		if (chip->irq_get_irqchip_state)
1833 			break;
1834 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1835 		data = data->parent_data;
1836 #else
1837 		data = NULL;
1838 #endif
1839 	} while (data);
1840 
1841 	if (data)
1842 		err = chip->irq_get_irqchip_state(data, which, state);
1843 
1844 	irq_put_desc_busunlock(desc, flags);
1845 	return err;
1846 }
1847 
1848 /**
1849  *	irq_set_irqchip_state - set the state of a forwarded interrupt.
1850  *	@irq: Interrupt line that is forwarded to a VM
1851  *	@which: State to be restored (one of IRQCHIP_STATE_*)
1852  *	@val: Value corresponding to @which
1853  *
1854  *	This call sets the internal irqchip state of an interrupt,
1855  *	depending on the value of @which.
1856  *
1857  *	This function should be called with preemption disabled if the
1858  *	interrupt controller has per-cpu registers.
1859  */
1860 int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1861 			  bool val)
1862 {
1863 	struct irq_desc *desc;
1864 	struct irq_data *data;
1865 	struct irq_chip *chip;
1866 	unsigned long flags;
1867 	int err = -EINVAL;
1868 
1869 	desc = irq_get_desc_buslock(irq, &flags, 0);
1870 	if (!desc)
1871 		return err;
1872 
1873 	data = irq_desc_get_irq_data(desc);
1874 
1875 	do {
1876 		chip = irq_data_get_irq_chip(data);
1877 		if (chip->irq_set_irqchip_state)
1878 			break;
1879 #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1880 		data = data->parent_data;
1881 #else
1882 		data = NULL;
1883 #endif
1884 	} while (data);
1885 
1886 	if (data)
1887 		err = chip->irq_set_irqchip_state(data, which, val);
1888 
1889 	irq_put_desc_busunlock(desc, flags);
1890 	return err;
1891 }
1892