xref: /openbmc/linux/kernel/irq/manage.c (revision b34e08d5)
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  *	This function may be called - with care - from IRQ context.
72  */
73 void synchronize_hardirq(unsigned int irq)
74 {
75 	struct irq_desc *desc = irq_to_desc(irq);
76 
77 	if (desc)
78 		__synchronize_hardirq(desc);
79 }
80 EXPORT_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  */
92 void 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 }
107 EXPORT_SYMBOL(synchronize_irq);
108 
109 #ifdef CONFIG_SMP
110 cpumask_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  */
117 int 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  */
137 void 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
149 static inline bool irq_can_move_pcntxt(struct irq_data *data)
150 {
151 	return irqd_can_move_in_process_context(data);
152 }
153 static inline bool irq_move_pending(struct irq_data *data)
154 {
155 	return irqd_is_setaffinity_pending(data);
156 }
157 static inline void
158 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
159 {
160 	cpumask_copy(desc->pending_mask, mask);
161 }
162 static inline void
163 irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
164 {
165 	cpumask_copy(mask, desc->pending_mask);
166 }
167 #else
168 static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
169 static inline bool irq_move_pending(struct irq_data *data) { return false; }
170 static inline void
171 irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
172 static inline void
173 irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
174 #endif
175 
176 int 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, false);
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 
195 int __irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask)
196 {
197 	struct irq_chip *chip = irq_data_get_irq_chip(data);
198 	struct irq_desc *desc = irq_data_to_desc(data);
199 	int ret = 0;
200 
201 	if (!chip || !chip->irq_set_affinity)
202 		return -EINVAL;
203 
204 	if (irq_can_move_pcntxt(data)) {
205 		ret = irq_do_set_affinity(data, mask, false);
206 	} else {
207 		irqd_set_move_pending(data);
208 		irq_copy_pending(desc, mask);
209 	}
210 
211 	if (desc->affinity_notify) {
212 		kref_get(&desc->affinity_notify->kref);
213 		schedule_work(&desc->affinity_notify->work);
214 	}
215 	irqd_set(data, IRQD_AFFINITY_SET);
216 
217 	return ret;
218 }
219 
220 /**
221  *	irq_set_affinity - Set the irq affinity of a given irq
222  *	@irq:		Interrupt to set affinity
223  *	@mask:		cpumask
224  *
225  */
226 int irq_set_affinity(unsigned int irq, const struct cpumask *mask)
227 {
228 	struct irq_desc *desc = irq_to_desc(irq);
229 	unsigned long flags;
230 	int ret;
231 
232 	if (!desc)
233 		return -EINVAL;
234 
235 	raw_spin_lock_irqsave(&desc->lock, flags);
236 	ret =  __irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask);
237 	raw_spin_unlock_irqrestore(&desc->lock, flags);
238 	return ret;
239 }
240 
241 int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
242 {
243 	unsigned long flags;
244 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
245 
246 	if (!desc)
247 		return -EINVAL;
248 	desc->affinity_hint = m;
249 	irq_put_desc_unlock(desc, flags);
250 	return 0;
251 }
252 EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
253 
254 static void irq_affinity_notify(struct work_struct *work)
255 {
256 	struct irq_affinity_notify *notify =
257 		container_of(work, struct irq_affinity_notify, work);
258 	struct irq_desc *desc = irq_to_desc(notify->irq);
259 	cpumask_var_t cpumask;
260 	unsigned long flags;
261 
262 	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
263 		goto out;
264 
265 	raw_spin_lock_irqsave(&desc->lock, flags);
266 	if (irq_move_pending(&desc->irq_data))
267 		irq_get_pending(cpumask, desc);
268 	else
269 		cpumask_copy(cpumask, desc->irq_data.affinity);
270 	raw_spin_unlock_irqrestore(&desc->lock, flags);
271 
272 	notify->notify(notify, cpumask);
273 
274 	free_cpumask_var(cpumask);
275 out:
276 	kref_put(&notify->kref, notify->release);
277 }
278 
279 /**
280  *	irq_set_affinity_notifier - control notification of IRQ affinity changes
281  *	@irq:		Interrupt for which to enable/disable notification
282  *	@notify:	Context for notification, or %NULL to disable
283  *			notification.  Function pointers must be initialised;
284  *			the other fields will be initialised by this function.
285  *
286  *	Must be called in process context.  Notification may only be enabled
287  *	after the IRQ is allocated and must be disabled before the IRQ is
288  *	freed using free_irq().
289  */
290 int
291 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
292 {
293 	struct irq_desc *desc = irq_to_desc(irq);
294 	struct irq_affinity_notify *old_notify;
295 	unsigned long flags;
296 
297 	/* The release function is promised process context */
298 	might_sleep();
299 
300 	if (!desc)
301 		return -EINVAL;
302 
303 	/* Complete initialisation of *notify */
304 	if (notify) {
305 		notify->irq = irq;
306 		kref_init(&notify->kref);
307 		INIT_WORK(&notify->work, irq_affinity_notify);
308 	}
309 
310 	raw_spin_lock_irqsave(&desc->lock, flags);
311 	old_notify = desc->affinity_notify;
312 	desc->affinity_notify = notify;
313 	raw_spin_unlock_irqrestore(&desc->lock, flags);
314 
315 	if (old_notify)
316 		kref_put(&old_notify->kref, old_notify->release);
317 
318 	return 0;
319 }
320 EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
321 
322 #ifndef CONFIG_AUTO_IRQ_AFFINITY
323 /*
324  * Generic version of the affinity autoselector.
325  */
326 static int
327 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
328 {
329 	struct cpumask *set = irq_default_affinity;
330 	int node = desc->irq_data.node;
331 
332 	/* Excludes PER_CPU and NO_BALANCE interrupts */
333 	if (!irq_can_set_affinity(irq))
334 		return 0;
335 
336 	/*
337 	 * Preserve an userspace affinity setup, but make sure that
338 	 * one of the targets is online.
339 	 */
340 	if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
341 		if (cpumask_intersects(desc->irq_data.affinity,
342 				       cpu_online_mask))
343 			set = desc->irq_data.affinity;
344 		else
345 			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
346 	}
347 
348 	cpumask_and(mask, cpu_online_mask, set);
349 	if (node != NUMA_NO_NODE) {
350 		const struct cpumask *nodemask = cpumask_of_node(node);
351 
352 		/* make sure at least one of the cpus in nodemask is online */
353 		if (cpumask_intersects(mask, nodemask))
354 			cpumask_and(mask, mask, nodemask);
355 	}
356 	irq_do_set_affinity(&desc->irq_data, mask, false);
357 	return 0;
358 }
359 #else
360 static inline int
361 setup_affinity(unsigned int irq, struct irq_desc *d, struct cpumask *mask)
362 {
363 	return irq_select_affinity(irq);
364 }
365 #endif
366 
367 /*
368  * Called when affinity is set via /proc/irq
369  */
370 int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
371 {
372 	struct irq_desc *desc = irq_to_desc(irq);
373 	unsigned long flags;
374 	int ret;
375 
376 	raw_spin_lock_irqsave(&desc->lock, flags);
377 	ret = setup_affinity(irq, desc, mask);
378 	raw_spin_unlock_irqrestore(&desc->lock, flags);
379 	return ret;
380 }
381 
382 #else
383 static inline int
384 setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
385 {
386 	return 0;
387 }
388 #endif
389 
390 void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
391 {
392 	if (suspend) {
393 		if (!desc->action || (desc->action->flags & IRQF_NO_SUSPEND))
394 			return;
395 		desc->istate |= IRQS_SUSPENDED;
396 	}
397 
398 	if (!desc->depth++)
399 		irq_disable(desc);
400 }
401 
402 static int __disable_irq_nosync(unsigned int irq)
403 {
404 	unsigned long flags;
405 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
406 
407 	if (!desc)
408 		return -EINVAL;
409 	__disable_irq(desc, irq, false);
410 	irq_put_desc_busunlock(desc, flags);
411 	return 0;
412 }
413 
414 /**
415  *	disable_irq_nosync - disable an irq without waiting
416  *	@irq: Interrupt to disable
417  *
418  *	Disable the selected interrupt line.  Disables and Enables are
419  *	nested.
420  *	Unlike disable_irq(), this function does not ensure existing
421  *	instances of the IRQ handler have completed before returning.
422  *
423  *	This function may be called from IRQ context.
424  */
425 void disable_irq_nosync(unsigned int irq)
426 {
427 	__disable_irq_nosync(irq);
428 }
429 EXPORT_SYMBOL(disable_irq_nosync);
430 
431 /**
432  *	disable_irq - disable an irq and wait for completion
433  *	@irq: Interrupt to disable
434  *
435  *	Disable the selected interrupt line.  Enables and Disables are
436  *	nested.
437  *	This function waits for any pending IRQ handlers for this interrupt
438  *	to complete before returning. If you use this function while
439  *	holding a resource the IRQ handler may need you will deadlock.
440  *
441  *	This function may be called - with care - from IRQ context.
442  */
443 void disable_irq(unsigned int irq)
444 {
445 	if (!__disable_irq_nosync(irq))
446 		synchronize_irq(irq);
447 }
448 EXPORT_SYMBOL(disable_irq);
449 
450 void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
451 {
452 	if (resume) {
453 		if (!(desc->istate & IRQS_SUSPENDED)) {
454 			if (!desc->action)
455 				return;
456 			if (!(desc->action->flags & IRQF_FORCE_RESUME))
457 				return;
458 			/* Pretend that it got disabled ! */
459 			desc->depth++;
460 		}
461 		desc->istate &= ~IRQS_SUSPENDED;
462 	}
463 
464 	switch (desc->depth) {
465 	case 0:
466  err_out:
467 		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
468 		break;
469 	case 1: {
470 		if (desc->istate & IRQS_SUSPENDED)
471 			goto err_out;
472 		/* Prevent probing on this irq: */
473 		irq_settings_set_noprobe(desc);
474 		irq_enable(desc);
475 		check_irq_resend(desc, irq);
476 		/* fall-through */
477 	}
478 	default:
479 		desc->depth--;
480 	}
481 }
482 
483 /**
484  *	enable_irq - enable handling of an irq
485  *	@irq: Interrupt to enable
486  *
487  *	Undoes the effect of one call to disable_irq().  If this
488  *	matches the last disable, processing of interrupts on this
489  *	IRQ line is re-enabled.
490  *
491  *	This function may be called from IRQ context only when
492  *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
493  */
494 void enable_irq(unsigned int irq)
495 {
496 	unsigned long flags;
497 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
498 
499 	if (!desc)
500 		return;
501 	if (WARN(!desc->irq_data.chip,
502 		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
503 		goto out;
504 
505 	__enable_irq(desc, irq, false);
506 out:
507 	irq_put_desc_busunlock(desc, flags);
508 }
509 EXPORT_SYMBOL(enable_irq);
510 
511 static int set_irq_wake_real(unsigned int irq, unsigned int on)
512 {
513 	struct irq_desc *desc = irq_to_desc(irq);
514 	int ret = -ENXIO;
515 
516 	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
517 		return 0;
518 
519 	if (desc->irq_data.chip->irq_set_wake)
520 		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
521 
522 	return ret;
523 }
524 
525 /**
526  *	irq_set_irq_wake - control irq power management wakeup
527  *	@irq:	interrupt to control
528  *	@on:	enable/disable power management wakeup
529  *
530  *	Enable/disable power management wakeup mode, which is
531  *	disabled by default.  Enables and disables must match,
532  *	just as they match for non-wakeup mode support.
533  *
534  *	Wakeup mode lets this IRQ wake the system from sleep
535  *	states like "suspend to RAM".
536  */
537 int irq_set_irq_wake(unsigned int irq, unsigned int on)
538 {
539 	unsigned long flags;
540 	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
541 	int ret = 0;
542 
543 	if (!desc)
544 		return -EINVAL;
545 
546 	/* wakeup-capable irqs can be shared between drivers that
547 	 * don't need to have the same sleep mode behaviors.
548 	 */
549 	if (on) {
550 		if (desc->wake_depth++ == 0) {
551 			ret = set_irq_wake_real(irq, on);
552 			if (ret)
553 				desc->wake_depth = 0;
554 			else
555 				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
556 		}
557 	} else {
558 		if (desc->wake_depth == 0) {
559 			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
560 		} else if (--desc->wake_depth == 0) {
561 			ret = set_irq_wake_real(irq, on);
562 			if (ret)
563 				desc->wake_depth = 1;
564 			else
565 				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
566 		}
567 	}
568 	irq_put_desc_busunlock(desc, flags);
569 	return ret;
570 }
571 EXPORT_SYMBOL(irq_set_irq_wake);
572 
573 /*
574  * Internal function that tells the architecture code whether a
575  * particular irq has been exclusively allocated or is available
576  * for driver use.
577  */
578 int can_request_irq(unsigned int irq, unsigned long irqflags)
579 {
580 	unsigned long flags;
581 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
582 	int canrequest = 0;
583 
584 	if (!desc)
585 		return 0;
586 
587 	if (irq_settings_can_request(desc)) {
588 		if (!desc->action ||
589 		    irqflags & desc->action->flags & IRQF_SHARED)
590 			canrequest = 1;
591 	}
592 	irq_put_desc_unlock(desc, flags);
593 	return canrequest;
594 }
595 
596 int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
597 		      unsigned long flags)
598 {
599 	struct irq_chip *chip = desc->irq_data.chip;
600 	int ret, unmask = 0;
601 
602 	if (!chip || !chip->irq_set_type) {
603 		/*
604 		 * IRQF_TRIGGER_* but the PIC does not support multiple
605 		 * flow-types?
606 		 */
607 		pr_debug("No set_type function for IRQ %d (%s)\n", irq,
608 			 chip ? (chip->name ? : "unknown") : "unknown");
609 		return 0;
610 	}
611 
612 	flags &= IRQ_TYPE_SENSE_MASK;
613 
614 	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
615 		if (!irqd_irq_masked(&desc->irq_data))
616 			mask_irq(desc);
617 		if (!irqd_irq_disabled(&desc->irq_data))
618 			unmask = 1;
619 	}
620 
621 	/* caller masked out all except trigger mode flags */
622 	ret = chip->irq_set_type(&desc->irq_data, flags);
623 
624 	switch (ret) {
625 	case IRQ_SET_MASK_OK:
626 		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
627 		irqd_set(&desc->irq_data, flags);
628 
629 	case IRQ_SET_MASK_OK_NOCOPY:
630 		flags = irqd_get_trigger_type(&desc->irq_data);
631 		irq_settings_set_trigger_mask(desc, flags);
632 		irqd_clear(&desc->irq_data, IRQD_LEVEL);
633 		irq_settings_clr_level(desc);
634 		if (flags & IRQ_TYPE_LEVEL_MASK) {
635 			irq_settings_set_level(desc);
636 			irqd_set(&desc->irq_data, IRQD_LEVEL);
637 		}
638 
639 		ret = 0;
640 		break;
641 	default:
642 		pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
643 		       flags, irq, chip->irq_set_type);
644 	}
645 	if (unmask)
646 		unmask_irq(desc);
647 	return ret;
648 }
649 
650 #ifdef CONFIG_HARDIRQS_SW_RESEND
651 int irq_set_parent(int irq, int parent_irq)
652 {
653 	unsigned long flags;
654 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
655 
656 	if (!desc)
657 		return -EINVAL;
658 
659 	desc->parent_irq = parent_irq;
660 
661 	irq_put_desc_unlock(desc, flags);
662 	return 0;
663 }
664 #endif
665 
666 /*
667  * Default primary interrupt handler for threaded interrupts. Is
668  * assigned as primary handler when request_threaded_irq is called
669  * with handler == NULL. Useful for oneshot interrupts.
670  */
671 static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
672 {
673 	return IRQ_WAKE_THREAD;
674 }
675 
676 /*
677  * Primary handler for nested threaded interrupts. Should never be
678  * called.
679  */
680 static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
681 {
682 	WARN(1, "Primary handler called for nested irq %d\n", irq);
683 	return IRQ_NONE;
684 }
685 
686 static int irq_wait_for_interrupt(struct irqaction *action)
687 {
688 	set_current_state(TASK_INTERRUPTIBLE);
689 
690 	while (!kthread_should_stop()) {
691 
692 		if (test_and_clear_bit(IRQTF_RUNTHREAD,
693 				       &action->thread_flags)) {
694 			__set_current_state(TASK_RUNNING);
695 			return 0;
696 		}
697 		schedule();
698 		set_current_state(TASK_INTERRUPTIBLE);
699 	}
700 	__set_current_state(TASK_RUNNING);
701 	return -1;
702 }
703 
704 /*
705  * Oneshot interrupts keep the irq line masked until the threaded
706  * handler finished. unmask if the interrupt has not been disabled and
707  * is marked MASKED.
708  */
709 static void irq_finalize_oneshot(struct irq_desc *desc,
710 				 struct irqaction *action)
711 {
712 	if (!(desc->istate & IRQS_ONESHOT))
713 		return;
714 again:
715 	chip_bus_lock(desc);
716 	raw_spin_lock_irq(&desc->lock);
717 
718 	/*
719 	 * Implausible though it may be we need to protect us against
720 	 * the following scenario:
721 	 *
722 	 * The thread is faster done than the hard interrupt handler
723 	 * on the other CPU. If we unmask the irq line then the
724 	 * interrupt can come in again and masks the line, leaves due
725 	 * to IRQS_INPROGRESS and the irq line is masked forever.
726 	 *
727 	 * This also serializes the state of shared oneshot handlers
728 	 * versus "desc->threads_onehsot |= action->thread_mask;" in
729 	 * irq_wake_thread(). See the comment there which explains the
730 	 * serialization.
731 	 */
732 	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
733 		raw_spin_unlock_irq(&desc->lock);
734 		chip_bus_sync_unlock(desc);
735 		cpu_relax();
736 		goto again;
737 	}
738 
739 	/*
740 	 * Now check again, whether the thread should run. Otherwise
741 	 * we would clear the threads_oneshot bit of this thread which
742 	 * was just set.
743 	 */
744 	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
745 		goto out_unlock;
746 
747 	desc->threads_oneshot &= ~action->thread_mask;
748 
749 	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
750 	    irqd_irq_masked(&desc->irq_data))
751 		unmask_threaded_irq(desc);
752 
753 out_unlock:
754 	raw_spin_unlock_irq(&desc->lock);
755 	chip_bus_sync_unlock(desc);
756 }
757 
758 #ifdef CONFIG_SMP
759 /*
760  * Check whether we need to change the affinity of the interrupt thread.
761  */
762 static void
763 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
764 {
765 	cpumask_var_t mask;
766 	bool valid = true;
767 
768 	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
769 		return;
770 
771 	/*
772 	 * In case we are out of memory we set IRQTF_AFFINITY again and
773 	 * try again next time
774 	 */
775 	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
776 		set_bit(IRQTF_AFFINITY, &action->thread_flags);
777 		return;
778 	}
779 
780 	raw_spin_lock_irq(&desc->lock);
781 	/*
782 	 * This code is triggered unconditionally. Check the affinity
783 	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
784 	 */
785 	if (desc->irq_data.affinity)
786 		cpumask_copy(mask, desc->irq_data.affinity);
787 	else
788 		valid = false;
789 	raw_spin_unlock_irq(&desc->lock);
790 
791 	if (valid)
792 		set_cpus_allowed_ptr(current, mask);
793 	free_cpumask_var(mask);
794 }
795 #else
796 static inline void
797 irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
798 #endif
799 
800 /*
801  * Interrupts which are not explicitely requested as threaded
802  * interrupts rely on the implicit bh/preempt disable of the hard irq
803  * context. So we need to disable bh here to avoid deadlocks and other
804  * side effects.
805  */
806 static irqreturn_t
807 irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
808 {
809 	irqreturn_t ret;
810 
811 	local_bh_disable();
812 	ret = action->thread_fn(action->irq, action->dev_id);
813 	irq_finalize_oneshot(desc, action);
814 	local_bh_enable();
815 	return ret;
816 }
817 
818 /*
819  * Interrupts explicitly requested as threaded interrupts want to be
820  * preemtible - many of them need to sleep and wait for slow busses to
821  * complete.
822  */
823 static irqreturn_t irq_thread_fn(struct irq_desc *desc,
824 		struct irqaction *action)
825 {
826 	irqreturn_t ret;
827 
828 	ret = action->thread_fn(action->irq, action->dev_id);
829 	irq_finalize_oneshot(desc, action);
830 	return ret;
831 }
832 
833 static void wake_threads_waitq(struct irq_desc *desc)
834 {
835 	if (atomic_dec_and_test(&desc->threads_active))
836 		wake_up(&desc->wait_for_threads);
837 }
838 
839 static void irq_thread_dtor(struct callback_head *unused)
840 {
841 	struct task_struct *tsk = current;
842 	struct irq_desc *desc;
843 	struct irqaction *action;
844 
845 	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
846 		return;
847 
848 	action = kthread_data(tsk);
849 
850 	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
851 	       tsk->comm, tsk->pid, action->irq);
852 
853 
854 	desc = irq_to_desc(action->irq);
855 	/*
856 	 * If IRQTF_RUNTHREAD is set, we need to decrement
857 	 * desc->threads_active and wake possible waiters.
858 	 */
859 	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
860 		wake_threads_waitq(desc);
861 
862 	/* Prevent a stale desc->threads_oneshot */
863 	irq_finalize_oneshot(desc, action);
864 }
865 
866 /*
867  * Interrupt handler thread
868  */
869 static int irq_thread(void *data)
870 {
871 	struct callback_head on_exit_work;
872 	struct irqaction *action = data;
873 	struct irq_desc *desc = irq_to_desc(action->irq);
874 	irqreturn_t (*handler_fn)(struct irq_desc *desc,
875 			struct irqaction *action);
876 
877 	if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
878 					&action->thread_flags))
879 		handler_fn = irq_forced_thread_fn;
880 	else
881 		handler_fn = irq_thread_fn;
882 
883 	init_task_work(&on_exit_work, irq_thread_dtor);
884 	task_work_add(current, &on_exit_work, false);
885 
886 	irq_thread_check_affinity(desc, action);
887 
888 	while (!irq_wait_for_interrupt(action)) {
889 		irqreturn_t action_ret;
890 
891 		irq_thread_check_affinity(desc, action);
892 
893 		action_ret = handler_fn(desc, action);
894 		if (!noirqdebug)
895 			note_interrupt(action->irq, desc, action_ret);
896 
897 		wake_threads_waitq(desc);
898 	}
899 
900 	/*
901 	 * This is the regular exit path. __free_irq() is stopping the
902 	 * thread via kthread_stop() after calling
903 	 * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
904 	 * oneshot mask bit can be set. We cannot verify that as we
905 	 * cannot touch the oneshot mask at this point anymore as
906 	 * __setup_irq() might have given out currents thread_mask
907 	 * again.
908 	 */
909 	task_work_cancel(current, irq_thread_dtor);
910 	return 0;
911 }
912 
913 /**
914  *	irq_wake_thread - wake the irq thread for the action identified by dev_id
915  *	@irq:		Interrupt line
916  *	@dev_id:	Device identity for which the thread should be woken
917  *
918  */
919 void irq_wake_thread(unsigned int irq, void *dev_id)
920 {
921 	struct irq_desc *desc = irq_to_desc(irq);
922 	struct irqaction *action;
923 	unsigned long flags;
924 
925 	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
926 		return;
927 
928 	raw_spin_lock_irqsave(&desc->lock, flags);
929 	for (action = desc->action; action; action = action->next) {
930 		if (action->dev_id == dev_id) {
931 			if (action->thread)
932 				__irq_wake_thread(desc, action);
933 			break;
934 		}
935 	}
936 	raw_spin_unlock_irqrestore(&desc->lock, flags);
937 }
938 EXPORT_SYMBOL_GPL(irq_wake_thread);
939 
940 static void irq_setup_forced_threading(struct irqaction *new)
941 {
942 	if (!force_irqthreads)
943 		return;
944 	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
945 		return;
946 
947 	new->flags |= IRQF_ONESHOT;
948 
949 	if (!new->thread_fn) {
950 		set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
951 		new->thread_fn = new->handler;
952 		new->handler = irq_default_primary_handler;
953 	}
954 }
955 
956 static int irq_request_resources(struct irq_desc *desc)
957 {
958 	struct irq_data *d = &desc->irq_data;
959 	struct irq_chip *c = d->chip;
960 
961 	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
962 }
963 
964 static void irq_release_resources(struct irq_desc *desc)
965 {
966 	struct irq_data *d = &desc->irq_data;
967 	struct irq_chip *c = d->chip;
968 
969 	if (c->irq_release_resources)
970 		c->irq_release_resources(d);
971 }
972 
973 /*
974  * Internal function to register an irqaction - typically used to
975  * allocate special interrupts that are part of the architecture.
976  */
977 static int
978 __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
979 {
980 	struct irqaction *old, **old_ptr;
981 	unsigned long flags, thread_mask = 0;
982 	int ret, nested, shared = 0;
983 	cpumask_var_t mask;
984 
985 	if (!desc)
986 		return -EINVAL;
987 
988 	if (desc->irq_data.chip == &no_irq_chip)
989 		return -ENOSYS;
990 	if (!try_module_get(desc->owner))
991 		return -ENODEV;
992 
993 	/*
994 	 * Check whether the interrupt nests into another interrupt
995 	 * thread.
996 	 */
997 	nested = irq_settings_is_nested_thread(desc);
998 	if (nested) {
999 		if (!new->thread_fn) {
1000 			ret = -EINVAL;
1001 			goto out_mput;
1002 		}
1003 		/*
1004 		 * Replace the primary handler which was provided from
1005 		 * the driver for non nested interrupt handling by the
1006 		 * dummy function which warns when called.
1007 		 */
1008 		new->handler = irq_nested_primary_handler;
1009 	} else {
1010 		if (irq_settings_can_thread(desc))
1011 			irq_setup_forced_threading(new);
1012 	}
1013 
1014 	/*
1015 	 * Create a handler thread when a thread function is supplied
1016 	 * and the interrupt does not nest into another interrupt
1017 	 * thread.
1018 	 */
1019 	if (new->thread_fn && !nested) {
1020 		struct task_struct *t;
1021 		static const struct sched_param param = {
1022 			.sched_priority = MAX_USER_RT_PRIO/2,
1023 		};
1024 
1025 		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1026 				   new->name);
1027 		if (IS_ERR(t)) {
1028 			ret = PTR_ERR(t);
1029 			goto out_mput;
1030 		}
1031 
1032 		sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1033 
1034 		/*
1035 		 * We keep the reference to the task struct even if
1036 		 * the thread dies to avoid that the interrupt code
1037 		 * references an already freed task_struct.
1038 		 */
1039 		get_task_struct(t);
1040 		new->thread = t;
1041 		/*
1042 		 * Tell the thread to set its affinity. This is
1043 		 * important for shared interrupt handlers as we do
1044 		 * not invoke setup_affinity() for the secondary
1045 		 * handlers as everything is already set up. Even for
1046 		 * interrupts marked with IRQF_NO_BALANCE this is
1047 		 * correct as we want the thread to move to the cpu(s)
1048 		 * on which the requesting code placed the interrupt.
1049 		 */
1050 		set_bit(IRQTF_AFFINITY, &new->thread_flags);
1051 	}
1052 
1053 	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1054 		ret = -ENOMEM;
1055 		goto out_thread;
1056 	}
1057 
1058 	/*
1059 	 * Drivers are often written to work w/o knowledge about the
1060 	 * underlying irq chip implementation, so a request for a
1061 	 * threaded irq without a primary hard irq context handler
1062 	 * requires the ONESHOT flag to be set. Some irq chips like
1063 	 * MSI based interrupts are per se one shot safe. Check the
1064 	 * chip flags, so we can avoid the unmask dance at the end of
1065 	 * the threaded handler for those.
1066 	 */
1067 	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1068 		new->flags &= ~IRQF_ONESHOT;
1069 
1070 	/*
1071 	 * The following block of code has to be executed atomically
1072 	 */
1073 	raw_spin_lock_irqsave(&desc->lock, flags);
1074 	old_ptr = &desc->action;
1075 	old = *old_ptr;
1076 	if (old) {
1077 		/*
1078 		 * Can't share interrupts unless both agree to and are
1079 		 * the same type (level, edge, polarity). So both flag
1080 		 * fields must have IRQF_SHARED set and the bits which
1081 		 * set the trigger type must match. Also all must
1082 		 * agree on ONESHOT.
1083 		 */
1084 		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1085 		    ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1086 		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1087 			goto mismatch;
1088 
1089 		/* All handlers must agree on per-cpuness */
1090 		if ((old->flags & IRQF_PERCPU) !=
1091 		    (new->flags & IRQF_PERCPU))
1092 			goto mismatch;
1093 
1094 		/* add new interrupt at end of irq queue */
1095 		do {
1096 			/*
1097 			 * Or all existing action->thread_mask bits,
1098 			 * so we can find the next zero bit for this
1099 			 * new action.
1100 			 */
1101 			thread_mask |= old->thread_mask;
1102 			old_ptr = &old->next;
1103 			old = *old_ptr;
1104 		} while (old);
1105 		shared = 1;
1106 	}
1107 
1108 	/*
1109 	 * Setup the thread mask for this irqaction for ONESHOT. For
1110 	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1111 	 * conditional in irq_wake_thread().
1112 	 */
1113 	if (new->flags & IRQF_ONESHOT) {
1114 		/*
1115 		 * Unlikely to have 32 resp 64 irqs sharing one line,
1116 		 * but who knows.
1117 		 */
1118 		if (thread_mask == ~0UL) {
1119 			ret = -EBUSY;
1120 			goto out_mask;
1121 		}
1122 		/*
1123 		 * The thread_mask for the action is or'ed to
1124 		 * desc->thread_active to indicate that the
1125 		 * IRQF_ONESHOT thread handler has been woken, but not
1126 		 * yet finished. The bit is cleared when a thread
1127 		 * completes. When all threads of a shared interrupt
1128 		 * line have completed desc->threads_active becomes
1129 		 * zero and the interrupt line is unmasked. See
1130 		 * handle.c:irq_wake_thread() for further information.
1131 		 *
1132 		 * If no thread is woken by primary (hard irq context)
1133 		 * interrupt handlers, then desc->threads_active is
1134 		 * also checked for zero to unmask the irq line in the
1135 		 * affected hard irq flow handlers
1136 		 * (handle_[fasteoi|level]_irq).
1137 		 *
1138 		 * The new action gets the first zero bit of
1139 		 * thread_mask assigned. See the loop above which or's
1140 		 * all existing action->thread_mask bits.
1141 		 */
1142 		new->thread_mask = 1 << ffz(thread_mask);
1143 
1144 	} else if (new->handler == irq_default_primary_handler &&
1145 		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1146 		/*
1147 		 * The interrupt was requested with handler = NULL, so
1148 		 * we use the default primary handler for it. But it
1149 		 * does not have the oneshot flag set. In combination
1150 		 * with level interrupts this is deadly, because the
1151 		 * default primary handler just wakes the thread, then
1152 		 * the irq lines is reenabled, but the device still
1153 		 * has the level irq asserted. Rinse and repeat....
1154 		 *
1155 		 * While this works for edge type interrupts, we play
1156 		 * it safe and reject unconditionally because we can't
1157 		 * say for sure which type this interrupt really
1158 		 * has. The type flags are unreliable as the
1159 		 * underlying chip implementation can override them.
1160 		 */
1161 		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1162 		       irq);
1163 		ret = -EINVAL;
1164 		goto out_mask;
1165 	}
1166 
1167 	if (!shared) {
1168 		ret = irq_request_resources(desc);
1169 		if (ret) {
1170 			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1171 			       new->name, irq, desc->irq_data.chip->name);
1172 			goto out_mask;
1173 		}
1174 
1175 		init_waitqueue_head(&desc->wait_for_threads);
1176 
1177 		/* Setup the type (level, edge polarity) if configured: */
1178 		if (new->flags & IRQF_TRIGGER_MASK) {
1179 			ret = __irq_set_trigger(desc, irq,
1180 					new->flags & IRQF_TRIGGER_MASK);
1181 
1182 			if (ret)
1183 				goto out_mask;
1184 		}
1185 
1186 		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1187 				  IRQS_ONESHOT | IRQS_WAITING);
1188 		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1189 
1190 		if (new->flags & IRQF_PERCPU) {
1191 			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1192 			irq_settings_set_per_cpu(desc);
1193 		}
1194 
1195 		if (new->flags & IRQF_ONESHOT)
1196 			desc->istate |= IRQS_ONESHOT;
1197 
1198 		if (irq_settings_can_autoenable(desc))
1199 			irq_startup(desc, true);
1200 		else
1201 			/* Undo nested disables: */
1202 			desc->depth = 1;
1203 
1204 		/* Exclude IRQ from balancing if requested */
1205 		if (new->flags & IRQF_NOBALANCING) {
1206 			irq_settings_set_no_balancing(desc);
1207 			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1208 		}
1209 
1210 		/* Set default affinity mask once everything is setup */
1211 		setup_affinity(irq, desc, mask);
1212 
1213 	} else if (new->flags & IRQF_TRIGGER_MASK) {
1214 		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1215 		unsigned int omsk = irq_settings_get_trigger_mask(desc);
1216 
1217 		if (nmsk != omsk)
1218 			/* hope the handler works with current  trigger mode */
1219 			pr_warning("irq %d uses trigger mode %u; requested %u\n",
1220 				   irq, nmsk, omsk);
1221 	}
1222 
1223 	new->irq = irq;
1224 	*old_ptr = new;
1225 
1226 	/* Reset broken irq detection when installing new handler */
1227 	desc->irq_count = 0;
1228 	desc->irqs_unhandled = 0;
1229 
1230 	/*
1231 	 * Check whether we disabled the irq via the spurious handler
1232 	 * before. Reenable it and give it another chance.
1233 	 */
1234 	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1235 		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1236 		__enable_irq(desc, irq, false);
1237 	}
1238 
1239 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1240 
1241 	/*
1242 	 * Strictly no need to wake it up, but hung_task complains
1243 	 * when no hard interrupt wakes the thread up.
1244 	 */
1245 	if (new->thread)
1246 		wake_up_process(new->thread);
1247 
1248 	register_irq_proc(irq, desc);
1249 	new->dir = NULL;
1250 	register_handler_proc(irq, new);
1251 	free_cpumask_var(mask);
1252 
1253 	return 0;
1254 
1255 mismatch:
1256 	if (!(new->flags & IRQF_PROBE_SHARED)) {
1257 		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1258 		       irq, new->flags, new->name, old->flags, old->name);
1259 #ifdef CONFIG_DEBUG_SHIRQ
1260 		dump_stack();
1261 #endif
1262 	}
1263 	ret = -EBUSY;
1264 
1265 out_mask:
1266 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1267 	free_cpumask_var(mask);
1268 
1269 out_thread:
1270 	if (new->thread) {
1271 		struct task_struct *t = new->thread;
1272 
1273 		new->thread = NULL;
1274 		kthread_stop(t);
1275 		put_task_struct(t);
1276 	}
1277 out_mput:
1278 	module_put(desc->owner);
1279 	return ret;
1280 }
1281 
1282 /**
1283  *	setup_irq - setup an interrupt
1284  *	@irq: Interrupt line to setup
1285  *	@act: irqaction for the interrupt
1286  *
1287  * Used to statically setup interrupts in the early boot process.
1288  */
1289 int setup_irq(unsigned int irq, struct irqaction *act)
1290 {
1291 	int retval;
1292 	struct irq_desc *desc = irq_to_desc(irq);
1293 
1294 	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1295 		return -EINVAL;
1296 	chip_bus_lock(desc);
1297 	retval = __setup_irq(irq, desc, act);
1298 	chip_bus_sync_unlock(desc);
1299 
1300 	return retval;
1301 }
1302 EXPORT_SYMBOL_GPL(setup_irq);
1303 
1304 /*
1305  * Internal function to unregister an irqaction - used to free
1306  * regular and special interrupts that are part of the architecture.
1307  */
1308 static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1309 {
1310 	struct irq_desc *desc = irq_to_desc(irq);
1311 	struct irqaction *action, **action_ptr;
1312 	unsigned long flags;
1313 
1314 	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1315 
1316 	if (!desc)
1317 		return NULL;
1318 
1319 	raw_spin_lock_irqsave(&desc->lock, flags);
1320 
1321 	/*
1322 	 * There can be multiple actions per IRQ descriptor, find the right
1323 	 * one based on the dev_id:
1324 	 */
1325 	action_ptr = &desc->action;
1326 	for (;;) {
1327 		action = *action_ptr;
1328 
1329 		if (!action) {
1330 			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1331 			raw_spin_unlock_irqrestore(&desc->lock, flags);
1332 
1333 			return NULL;
1334 		}
1335 
1336 		if (action->dev_id == dev_id)
1337 			break;
1338 		action_ptr = &action->next;
1339 	}
1340 
1341 	/* Found it - now remove it from the list of entries: */
1342 	*action_ptr = action->next;
1343 
1344 	/* If this was the last handler, shut down the IRQ line: */
1345 	if (!desc->action) {
1346 		irq_shutdown(desc);
1347 		irq_release_resources(desc);
1348 	}
1349 
1350 #ifdef CONFIG_SMP
1351 	/* make sure affinity_hint is cleaned up */
1352 	if (WARN_ON_ONCE(desc->affinity_hint))
1353 		desc->affinity_hint = NULL;
1354 #endif
1355 
1356 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1357 
1358 	unregister_handler_proc(irq, action);
1359 
1360 	/* Make sure it's not being used on another CPU: */
1361 	synchronize_irq(irq);
1362 
1363 #ifdef CONFIG_DEBUG_SHIRQ
1364 	/*
1365 	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1366 	 * event to happen even now it's being freed, so let's make sure that
1367 	 * is so by doing an extra call to the handler ....
1368 	 *
1369 	 * ( We do this after actually deregistering it, to make sure that a
1370 	 *   'real' IRQ doesn't run in * parallel with our fake. )
1371 	 */
1372 	if (action->flags & IRQF_SHARED) {
1373 		local_irq_save(flags);
1374 		action->handler(irq, dev_id);
1375 		local_irq_restore(flags);
1376 	}
1377 #endif
1378 
1379 	if (action->thread) {
1380 		kthread_stop(action->thread);
1381 		put_task_struct(action->thread);
1382 	}
1383 
1384 	module_put(desc->owner);
1385 	return action;
1386 }
1387 
1388 /**
1389  *	remove_irq - free an interrupt
1390  *	@irq: Interrupt line to free
1391  *	@act: irqaction for the interrupt
1392  *
1393  * Used to remove interrupts statically setup by the early boot process.
1394  */
1395 void remove_irq(unsigned int irq, struct irqaction *act)
1396 {
1397 	struct irq_desc *desc = irq_to_desc(irq);
1398 
1399 	if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1400 	    __free_irq(irq, act->dev_id);
1401 }
1402 EXPORT_SYMBOL_GPL(remove_irq);
1403 
1404 /**
1405  *	free_irq - free an interrupt allocated with request_irq
1406  *	@irq: Interrupt line to free
1407  *	@dev_id: Device identity to free
1408  *
1409  *	Remove an interrupt handler. The handler is removed and if the
1410  *	interrupt line is no longer in use by any driver it is disabled.
1411  *	On a shared IRQ the caller must ensure the interrupt is disabled
1412  *	on the card it drives before calling this function. The function
1413  *	does not return until any executing interrupts for this IRQ
1414  *	have completed.
1415  *
1416  *	This function must not be called from interrupt context.
1417  */
1418 void free_irq(unsigned int irq, void *dev_id)
1419 {
1420 	struct irq_desc *desc = irq_to_desc(irq);
1421 
1422 	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1423 		return;
1424 
1425 #ifdef CONFIG_SMP
1426 	if (WARN_ON(desc->affinity_notify))
1427 		desc->affinity_notify = NULL;
1428 #endif
1429 
1430 	chip_bus_lock(desc);
1431 	kfree(__free_irq(irq, dev_id));
1432 	chip_bus_sync_unlock(desc);
1433 }
1434 EXPORT_SYMBOL(free_irq);
1435 
1436 /**
1437  *	request_threaded_irq - allocate an interrupt line
1438  *	@irq: Interrupt line to allocate
1439  *	@handler: Function to be called when the IRQ occurs.
1440  *		  Primary handler for threaded interrupts
1441  *		  If NULL and thread_fn != NULL the default
1442  *		  primary handler is installed
1443  *	@thread_fn: Function called from the irq handler thread
1444  *		    If NULL, no irq thread is created
1445  *	@irqflags: Interrupt type flags
1446  *	@devname: An ascii name for the claiming device
1447  *	@dev_id: A cookie passed back to the handler function
1448  *
1449  *	This call allocates interrupt resources and enables the
1450  *	interrupt line and IRQ handling. From the point this
1451  *	call is made your handler function may be invoked. Since
1452  *	your handler function must clear any interrupt the board
1453  *	raises, you must take care both to initialise your hardware
1454  *	and to set up the interrupt handler in the right order.
1455  *
1456  *	If you want to set up a threaded irq handler for your device
1457  *	then you need to supply @handler and @thread_fn. @handler is
1458  *	still called in hard interrupt context and has to check
1459  *	whether the interrupt originates from the device. If yes it
1460  *	needs to disable the interrupt on the device and return
1461  *	IRQ_WAKE_THREAD which will wake up the handler thread and run
1462  *	@thread_fn. This split handler design is necessary to support
1463  *	shared interrupts.
1464  *
1465  *	Dev_id must be globally unique. Normally the address of the
1466  *	device data structure is used as the cookie. Since the handler
1467  *	receives this value it makes sense to use it.
1468  *
1469  *	If your interrupt is shared you must pass a non NULL dev_id
1470  *	as this is required when freeing the interrupt.
1471  *
1472  *	Flags:
1473  *
1474  *	IRQF_SHARED		Interrupt is shared
1475  *	IRQF_TRIGGER_*		Specify active edge(s) or level
1476  *
1477  */
1478 int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1479 			 irq_handler_t thread_fn, unsigned long irqflags,
1480 			 const char *devname, void *dev_id)
1481 {
1482 	struct irqaction *action;
1483 	struct irq_desc *desc;
1484 	int retval;
1485 
1486 	/*
1487 	 * Sanity-check: shared interrupts must pass in a real dev-ID,
1488 	 * otherwise we'll have trouble later trying to figure out
1489 	 * which interrupt is which (messes up the interrupt freeing
1490 	 * logic etc).
1491 	 */
1492 	if ((irqflags & IRQF_SHARED) && !dev_id)
1493 		return -EINVAL;
1494 
1495 	desc = irq_to_desc(irq);
1496 	if (!desc)
1497 		return -EINVAL;
1498 
1499 	if (!irq_settings_can_request(desc) ||
1500 	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1501 		return -EINVAL;
1502 
1503 	if (!handler) {
1504 		if (!thread_fn)
1505 			return -EINVAL;
1506 		handler = irq_default_primary_handler;
1507 	}
1508 
1509 	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1510 	if (!action)
1511 		return -ENOMEM;
1512 
1513 	action->handler = handler;
1514 	action->thread_fn = thread_fn;
1515 	action->flags = irqflags;
1516 	action->name = devname;
1517 	action->dev_id = dev_id;
1518 
1519 	chip_bus_lock(desc);
1520 	retval = __setup_irq(irq, desc, action);
1521 	chip_bus_sync_unlock(desc);
1522 
1523 	if (retval)
1524 		kfree(action);
1525 
1526 #ifdef CONFIG_DEBUG_SHIRQ_FIXME
1527 	if (!retval && (irqflags & IRQF_SHARED)) {
1528 		/*
1529 		 * It's a shared IRQ -- the driver ought to be prepared for it
1530 		 * to happen immediately, so let's make sure....
1531 		 * We disable the irq to make sure that a 'real' IRQ doesn't
1532 		 * run in parallel with our fake.
1533 		 */
1534 		unsigned long flags;
1535 
1536 		disable_irq(irq);
1537 		local_irq_save(flags);
1538 
1539 		handler(irq, dev_id);
1540 
1541 		local_irq_restore(flags);
1542 		enable_irq(irq);
1543 	}
1544 #endif
1545 	return retval;
1546 }
1547 EXPORT_SYMBOL(request_threaded_irq);
1548 
1549 /**
1550  *	request_any_context_irq - allocate an interrupt line
1551  *	@irq: Interrupt line to allocate
1552  *	@handler: Function to be called when the IRQ occurs.
1553  *		  Threaded handler for threaded interrupts.
1554  *	@flags: Interrupt type flags
1555  *	@name: An ascii name for the claiming device
1556  *	@dev_id: A cookie passed back to the handler function
1557  *
1558  *	This call allocates interrupt resources and enables the
1559  *	interrupt line and IRQ handling. It selects either a
1560  *	hardirq or threaded handling method depending on the
1561  *	context.
1562  *
1563  *	On failure, it returns a negative value. On success,
1564  *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1565  */
1566 int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1567 			    unsigned long flags, const char *name, void *dev_id)
1568 {
1569 	struct irq_desc *desc = irq_to_desc(irq);
1570 	int ret;
1571 
1572 	if (!desc)
1573 		return -EINVAL;
1574 
1575 	if (irq_settings_is_nested_thread(desc)) {
1576 		ret = request_threaded_irq(irq, NULL, handler,
1577 					   flags, name, dev_id);
1578 		return !ret ? IRQC_IS_NESTED : ret;
1579 	}
1580 
1581 	ret = request_irq(irq, handler, flags, name, dev_id);
1582 	return !ret ? IRQC_IS_HARDIRQ : ret;
1583 }
1584 EXPORT_SYMBOL_GPL(request_any_context_irq);
1585 
1586 void enable_percpu_irq(unsigned int irq, unsigned int type)
1587 {
1588 	unsigned int cpu = smp_processor_id();
1589 	unsigned long flags;
1590 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1591 
1592 	if (!desc)
1593 		return;
1594 
1595 	type &= IRQ_TYPE_SENSE_MASK;
1596 	if (type != IRQ_TYPE_NONE) {
1597 		int ret;
1598 
1599 		ret = __irq_set_trigger(desc, irq, type);
1600 
1601 		if (ret) {
1602 			WARN(1, "failed to set type for IRQ%d\n", irq);
1603 			goto out;
1604 		}
1605 	}
1606 
1607 	irq_percpu_enable(desc, cpu);
1608 out:
1609 	irq_put_desc_unlock(desc, flags);
1610 }
1611 EXPORT_SYMBOL_GPL(enable_percpu_irq);
1612 
1613 void disable_percpu_irq(unsigned int irq)
1614 {
1615 	unsigned int cpu = smp_processor_id();
1616 	unsigned long flags;
1617 	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1618 
1619 	if (!desc)
1620 		return;
1621 
1622 	irq_percpu_disable(desc, cpu);
1623 	irq_put_desc_unlock(desc, flags);
1624 }
1625 EXPORT_SYMBOL_GPL(disable_percpu_irq);
1626 
1627 /*
1628  * Internal function to unregister a percpu irqaction.
1629  */
1630 static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1631 {
1632 	struct irq_desc *desc = irq_to_desc(irq);
1633 	struct irqaction *action;
1634 	unsigned long flags;
1635 
1636 	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1637 
1638 	if (!desc)
1639 		return NULL;
1640 
1641 	raw_spin_lock_irqsave(&desc->lock, flags);
1642 
1643 	action = desc->action;
1644 	if (!action || action->percpu_dev_id != dev_id) {
1645 		WARN(1, "Trying to free already-free IRQ %d\n", irq);
1646 		goto bad;
1647 	}
1648 
1649 	if (!cpumask_empty(desc->percpu_enabled)) {
1650 		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1651 		     irq, cpumask_first(desc->percpu_enabled));
1652 		goto bad;
1653 	}
1654 
1655 	/* Found it - now remove it from the list of entries: */
1656 	desc->action = NULL;
1657 
1658 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1659 
1660 	unregister_handler_proc(irq, action);
1661 
1662 	module_put(desc->owner);
1663 	return action;
1664 
1665 bad:
1666 	raw_spin_unlock_irqrestore(&desc->lock, flags);
1667 	return NULL;
1668 }
1669 
1670 /**
1671  *	remove_percpu_irq - free a per-cpu interrupt
1672  *	@irq: Interrupt line to free
1673  *	@act: irqaction for the interrupt
1674  *
1675  * Used to remove interrupts statically setup by the early boot process.
1676  */
1677 void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1678 {
1679 	struct irq_desc *desc = irq_to_desc(irq);
1680 
1681 	if (desc && irq_settings_is_per_cpu_devid(desc))
1682 	    __free_percpu_irq(irq, act->percpu_dev_id);
1683 }
1684 
1685 /**
1686  *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
1687  *	@irq: Interrupt line to free
1688  *	@dev_id: Device identity to free
1689  *
1690  *	Remove a percpu interrupt handler. The handler is removed, but
1691  *	the interrupt line is not disabled. This must be done on each
1692  *	CPU before calling this function. The function does not return
1693  *	until any executing interrupts for this IRQ have completed.
1694  *
1695  *	This function must not be called from interrupt context.
1696  */
1697 void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1698 {
1699 	struct irq_desc *desc = irq_to_desc(irq);
1700 
1701 	if (!desc || !irq_settings_is_per_cpu_devid(desc))
1702 		return;
1703 
1704 	chip_bus_lock(desc);
1705 	kfree(__free_percpu_irq(irq, dev_id));
1706 	chip_bus_sync_unlock(desc);
1707 }
1708 
1709 /**
1710  *	setup_percpu_irq - setup a per-cpu interrupt
1711  *	@irq: Interrupt line to setup
1712  *	@act: irqaction for the interrupt
1713  *
1714  * Used to statically setup per-cpu interrupts in the early boot process.
1715  */
1716 int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1717 {
1718 	struct irq_desc *desc = irq_to_desc(irq);
1719 	int retval;
1720 
1721 	if (!desc || !irq_settings_is_per_cpu_devid(desc))
1722 		return -EINVAL;
1723 	chip_bus_lock(desc);
1724 	retval = __setup_irq(irq, desc, act);
1725 	chip_bus_sync_unlock(desc);
1726 
1727 	return retval;
1728 }
1729 
1730 /**
1731  *	request_percpu_irq - allocate a percpu interrupt line
1732  *	@irq: Interrupt line to allocate
1733  *	@handler: Function to be called when the IRQ occurs.
1734  *	@devname: An ascii name for the claiming device
1735  *	@dev_id: A percpu cookie passed back to the handler function
1736  *
1737  *	This call allocates interrupt resources, but doesn't
1738  *	automatically enable the interrupt. It has to be done on each
1739  *	CPU using enable_percpu_irq().
1740  *
1741  *	Dev_id must be globally unique. It is a per-cpu variable, and
1742  *	the handler gets called with the interrupted CPU's instance of
1743  *	that variable.
1744  */
1745 int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1746 		       const char *devname, void __percpu *dev_id)
1747 {
1748 	struct irqaction *action;
1749 	struct irq_desc *desc;
1750 	int retval;
1751 
1752 	if (!dev_id)
1753 		return -EINVAL;
1754 
1755 	desc = irq_to_desc(irq);
1756 	if (!desc || !irq_settings_can_request(desc) ||
1757 	    !irq_settings_is_per_cpu_devid(desc))
1758 		return -EINVAL;
1759 
1760 	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1761 	if (!action)
1762 		return -ENOMEM;
1763 
1764 	action->handler = handler;
1765 	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1766 	action->name = devname;
1767 	action->percpu_dev_id = dev_id;
1768 
1769 	chip_bus_lock(desc);
1770 	retval = __setup_irq(irq, desc, action);
1771 	chip_bus_sync_unlock(desc);
1772 
1773 	if (retval)
1774 		kfree(action);
1775 
1776 	return retval;
1777 }
1778