Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
   4 * Copyright (C) 2005-2006 Thomas Gleixner
   5 *
   6 * This file contains driver APIs to the irq subsystem.
   7 */
   8
   9#define pr_fmt(fmt) "genirq: " fmt
  10
  11#include <linux/irq.h>
  12#include <linux/kthread.h>
  13#include <linux/module.h>
  14#include <linux/random.h>
  15#include <linux/interrupt.h>
  16#include <linux/irqdomain.h>
  17#include <linux/slab.h>
  18#include <linux/sched.h>
  19#include <linux/sched/rt.h>
  20#include <linux/sched/task.h>
  21#include <linux/sched/isolation.h>
  22#include <uapi/linux/sched/types.h>
  23#include <linux/task_work.h>
  24
  25#include "internals.h"
  26
  27#if defined(CONFIG_IRQ_FORCED_THREADING) && !defined(CONFIG_PREEMPT_RT)
  28DEFINE_STATIC_KEY_FALSE(force_irqthreads_key);
 
  29
  30static int __init setup_forced_irqthreads(char *arg)
  31{
  32	static_branch_enable(&force_irqthreads_key);
  33	return 0;
  34}
  35early_param("threadirqs", setup_forced_irqthreads);
  36#endif
  37
  38static void __synchronize_hardirq(struct irq_desc *desc, bool sync_chip)
  39{
  40	struct irq_data *irqd = irq_desc_get_irq_data(desc);
  41	bool inprogress;
  42
  43	do {
  44		unsigned long flags;
  45
  46		/*
  47		 * Wait until we're out of the critical section.  This might
  48		 * give the wrong answer due to the lack of memory barriers.
  49		 */
  50		while (irqd_irq_inprogress(&desc->irq_data))
  51			cpu_relax();
  52
  53		/* Ok, that indicated we're done: double-check carefully. */
  54		raw_spin_lock_irqsave(&desc->lock, flags);
  55		inprogress = irqd_irq_inprogress(&desc->irq_data);
  56
  57		/*
  58		 * If requested and supported, check at the chip whether it
  59		 * is in flight at the hardware level, i.e. already pending
  60		 * in a CPU and waiting for service and acknowledge.
  61		 */
  62		if (!inprogress && sync_chip) {
  63			/*
  64			 * Ignore the return code. inprogress is only updated
  65			 * when the chip supports it.
  66			 */
  67			__irq_get_irqchip_state(irqd, IRQCHIP_STATE_ACTIVE,
  68						&inprogress);
  69		}
  70		raw_spin_unlock_irqrestore(&desc->lock, flags);
  71
  72		/* Oops, that failed? */
  73	} while (inprogress);
  74}
  75
  76/**
  77 *	synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
  78 *	@irq: interrupt number to wait for
  79 *
  80 *	This function waits for any pending hard IRQ handlers for this
  81 *	interrupt to complete before returning. If you use this
  82 *	function while holding a resource the IRQ handler may need you
  83 *	will deadlock. It does not take associated threaded handlers
  84 *	into account.
  85 *
  86 *	Do not use this for shutdown scenarios where you must be sure
  87 *	that all parts (hardirq and threaded handler) have completed.
  88 *
  89 *	Returns: false if a threaded handler is active.
  90 *
  91 *	This function may be called - with care - from IRQ context.
  92 *
  93 *	It does not check whether there is an interrupt in flight at the
  94 *	hardware level, but not serviced yet, as this might deadlock when
  95 *	called with interrupts disabled and the target CPU of the interrupt
  96 *	is the current CPU.
  97 */
  98bool synchronize_hardirq(unsigned int irq)
  99{
 100	struct irq_desc *desc = irq_to_desc(irq);
 101
 102	if (desc) {
 103		__synchronize_hardirq(desc, false);
 104		return !atomic_read(&desc->threads_active);
 105	}
 106
 107	return true;
 108}
 109EXPORT_SYMBOL(synchronize_hardirq);
 110
 111/**
 112 *	synchronize_irq - wait for pending IRQ handlers (on other CPUs)
 113 *	@irq: interrupt number to wait for
 114 *
 115 *	This function waits for any pending IRQ handlers for this interrupt
 116 *	to complete before returning. If you use this function while
 117 *	holding a resource the IRQ handler may need you will deadlock.
 118 *
 119 *	Can only be called from preemptible code as it might sleep when
 120 *	an interrupt thread is associated to @irq.
 121 *
 122 *	It optionally makes sure (when the irq chip supports that method)
 123 *	that the interrupt is not pending in any CPU and waiting for
 124 *	service.
 125 */
 126void synchronize_irq(unsigned int irq)
 127{
 128	struct irq_desc *desc = irq_to_desc(irq);
 129
 130	if (desc) {
 131		__synchronize_hardirq(desc, true);
 132		/*
 133		 * We made sure that no hardirq handler is
 134		 * running. Now verify that no threaded handlers are
 135		 * active.
 136		 */
 137		wait_event(desc->wait_for_threads,
 138			   !atomic_read(&desc->threads_active));
 139	}
 140}
 141EXPORT_SYMBOL(synchronize_irq);
 142
 143#ifdef CONFIG_SMP
 144cpumask_var_t irq_default_affinity;
 145
 146static bool __irq_can_set_affinity(struct irq_desc *desc)
 147{
 148	if (!desc || !irqd_can_balance(&desc->irq_data) ||
 149	    !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
 150		return false;
 151	return true;
 152}
 153
 154/**
 155 *	irq_can_set_affinity - Check if the affinity of a given irq can be set
 156 *	@irq:		Interrupt to check
 157 *
 158 */
 159int irq_can_set_affinity(unsigned int irq)
 160{
 161	return __irq_can_set_affinity(irq_to_desc(irq));
 162}
 163
 164/**
 165 * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
 166 * @irq:	Interrupt to check
 167 *
 168 * Like irq_can_set_affinity() above, but additionally checks for the
 169 * AFFINITY_MANAGED flag.
 170 */
 171bool irq_can_set_affinity_usr(unsigned int irq)
 172{
 173	struct irq_desc *desc = irq_to_desc(irq);
 174
 175	return __irq_can_set_affinity(desc) &&
 176		!irqd_affinity_is_managed(&desc->irq_data);
 177}
 178
 179/**
 180 *	irq_set_thread_affinity - Notify irq threads to adjust affinity
 181 *	@desc:		irq descriptor which has affinity changed
 182 *
 183 *	We just set IRQTF_AFFINITY and delegate the affinity setting
 184 *	to the interrupt thread itself. We can not call
 185 *	set_cpus_allowed_ptr() here as we hold desc->lock and this
 186 *	code can be called from hard interrupt context.
 187 */
 188void irq_set_thread_affinity(struct irq_desc *desc)
 189{
 190	struct irqaction *action;
 191
 192	for_each_action_of_desc(desc, action)
 193		if (action->thread)
 194			set_bit(IRQTF_AFFINITY, &action->thread_flags);
 195}
 196
 197#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
 198static void irq_validate_effective_affinity(struct irq_data *data)
 199{
 200	const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
 201	struct irq_chip *chip = irq_data_get_irq_chip(data);
 202
 203	if (!cpumask_empty(m))
 204		return;
 205	pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
 206		     chip->name, data->irq);
 207}
 
 
 
 
 
 
 208#else
 209static inline void irq_validate_effective_affinity(struct irq_data *data) { }
 
 
 210#endif
 211
 212int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
 213			bool force)
 214{
 215	struct irq_desc *desc = irq_data_to_desc(data);
 216	struct irq_chip *chip = irq_data_get_irq_chip(data);
 217	const struct cpumask  *prog_mask;
 218	int ret;
 219
 220	static DEFINE_RAW_SPINLOCK(tmp_mask_lock);
 221	static struct cpumask tmp_mask;
 222
 223	if (!chip || !chip->irq_set_affinity)
 224		return -EINVAL;
 225
 226	raw_spin_lock(&tmp_mask_lock);
 227	/*
 228	 * If this is a managed interrupt and housekeeping is enabled on
 229	 * it check whether the requested affinity mask intersects with
 230	 * a housekeeping CPU. If so, then remove the isolated CPUs from
 231	 * the mask and just keep the housekeeping CPU(s). This prevents
 232	 * the affinity setter from routing the interrupt to an isolated
 233	 * CPU to avoid that I/O submitted from a housekeeping CPU causes
 234	 * interrupts on an isolated one.
 235	 *
 236	 * If the masks do not intersect or include online CPU(s) then
 237	 * keep the requested mask. The isolated target CPUs are only
 238	 * receiving interrupts when the I/O operation was submitted
 239	 * directly from them.
 240	 *
 241	 * If all housekeeping CPUs in the affinity mask are offline, the
 242	 * interrupt will be migrated by the CPU hotplug code once a
 243	 * housekeeping CPU which belongs to the affinity mask comes
 244	 * online.
 245	 */
 246	if (irqd_affinity_is_managed(data) &&
 247	    housekeeping_enabled(HK_TYPE_MANAGED_IRQ)) {
 248		const struct cpumask *hk_mask;
 249
 250		hk_mask = housekeeping_cpumask(HK_TYPE_MANAGED_IRQ);
 
 251
 
 
 
 252		cpumask_and(&tmp_mask, mask, hk_mask);
 253		if (!cpumask_intersects(&tmp_mask, cpu_online_mask))
 254			prog_mask = mask;
 255		else
 256			prog_mask = &tmp_mask;
 
 
 257	} else {
 258		prog_mask = mask;
 259	}
 260
 261	/*
 262	 * Make sure we only provide online CPUs to the irqchip,
 263	 * unless we are being asked to force the affinity (in which
 264	 * case we do as we are told).
 265	 */
 266	cpumask_and(&tmp_mask, prog_mask, cpu_online_mask);
 267	if (!force && !cpumask_empty(&tmp_mask))
 268		ret = chip->irq_set_affinity(data, &tmp_mask, force);
 269	else if (force)
 270		ret = chip->irq_set_affinity(data, mask, force);
 271	else
 272		ret = -EINVAL;
 273
 274	raw_spin_unlock(&tmp_mask_lock);
 275
 276	switch (ret) {
 277	case IRQ_SET_MASK_OK:
 278	case IRQ_SET_MASK_OK_DONE:
 279		cpumask_copy(desc->irq_common_data.affinity, mask);
 280		fallthrough;
 281	case IRQ_SET_MASK_OK_NOCOPY:
 282		irq_validate_effective_affinity(data);
 283		irq_set_thread_affinity(desc);
 284		ret = 0;
 285	}
 286
 287	return ret;
 288}
 289
 290#ifdef CONFIG_GENERIC_PENDING_IRQ
 291static inline int irq_set_affinity_pending(struct irq_data *data,
 292					   const struct cpumask *dest)
 293{
 294	struct irq_desc *desc = irq_data_to_desc(data);
 295
 296	irqd_set_move_pending(data);
 297	irq_copy_pending(desc, dest);
 298	return 0;
 299}
 300#else
 301static inline int irq_set_affinity_pending(struct irq_data *data,
 302					   const struct cpumask *dest)
 303{
 304	return -EBUSY;
 305}
 306#endif
 307
 308static int irq_try_set_affinity(struct irq_data *data,
 309				const struct cpumask *dest, bool force)
 310{
 311	int ret = irq_do_set_affinity(data, dest, force);
 312
 313	/*
 314	 * In case that the underlying vector management is busy and the
 315	 * architecture supports the generic pending mechanism then utilize
 316	 * this to avoid returning an error to user space.
 317	 */
 318	if (ret == -EBUSY && !force)
 319		ret = irq_set_affinity_pending(data, dest);
 320	return ret;
 321}
 322
 323static bool irq_set_affinity_deactivated(struct irq_data *data,
 324					 const struct cpumask *mask)
 325{
 326	struct irq_desc *desc = irq_data_to_desc(data);
 327
 328	/*
 329	 * Handle irq chips which can handle affinity only in activated
 330	 * state correctly
 331	 *
 332	 * If the interrupt is not yet activated, just store the affinity
 333	 * mask and do not call the chip driver at all. On activation the
 334	 * driver has to make sure anyway that the interrupt is in a
 335	 * usable state so startup works.
 336	 */
 337	if (!IS_ENABLED(CONFIG_IRQ_DOMAIN_HIERARCHY) ||
 338	    irqd_is_activated(data) || !irqd_affinity_on_activate(data))
 339		return false;
 340
 341	cpumask_copy(desc->irq_common_data.affinity, mask);
 342	irq_data_update_effective_affinity(data, mask);
 343	irqd_set(data, IRQD_AFFINITY_SET);
 344	return true;
 345}
 346
 347int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 348			    bool force)
 349{
 350	struct irq_chip *chip = irq_data_get_irq_chip(data);
 351	struct irq_desc *desc = irq_data_to_desc(data);
 352	int ret = 0;
 353
 354	if (!chip || !chip->irq_set_affinity)
 355		return -EINVAL;
 356
 357	if (irq_set_affinity_deactivated(data, mask))
 358		return 0;
 359
 360	if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
 361		ret = irq_try_set_affinity(data, mask, force);
 362	} else {
 363		irqd_set_move_pending(data);
 364		irq_copy_pending(desc, mask);
 365	}
 366
 367	if (desc->affinity_notify) {
 368		kref_get(&desc->affinity_notify->kref);
 369		if (!schedule_work(&desc->affinity_notify->work)) {
 370			/* Work was already scheduled, drop our extra ref */
 371			kref_put(&desc->affinity_notify->kref,
 372				 desc->affinity_notify->release);
 373		}
 374	}
 375	irqd_set(data, IRQD_AFFINITY_SET);
 376
 377	return ret;
 378}
 379
 380/**
 381 * irq_update_affinity_desc - Update affinity management for an interrupt
 382 * @irq:	The interrupt number to update
 383 * @affinity:	Pointer to the affinity descriptor
 384 *
 385 * This interface can be used to configure the affinity management of
 386 * interrupts which have been allocated already.
 387 *
 388 * There are certain limitations on when it may be used - attempts to use it
 389 * for when the kernel is configured for generic IRQ reservation mode (in
 390 * config GENERIC_IRQ_RESERVATION_MODE) will fail, as it may conflict with
 391 * managed/non-managed interrupt accounting. In addition, attempts to use it on
 392 * an interrupt which is already started or which has already been configured
 393 * as managed will also fail, as these mean invalid init state or double init.
 394 */
 395int irq_update_affinity_desc(unsigned int irq,
 396			     struct irq_affinity_desc *affinity)
 397{
 398	struct irq_desc *desc;
 399	unsigned long flags;
 400	bool activated;
 401	int ret = 0;
 402
 403	/*
 404	 * Supporting this with the reservation scheme used by x86 needs
 405	 * some more thought. Fail it for now.
 406	 */
 407	if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
 408		return -EOPNOTSUPP;
 409
 410	desc = irq_get_desc_buslock(irq, &flags, 0);
 411	if (!desc)
 412		return -EINVAL;
 413
 414	/* Requires the interrupt to be shut down */
 415	if (irqd_is_started(&desc->irq_data)) {
 416		ret = -EBUSY;
 417		goto out_unlock;
 418	}
 419
 420	/* Interrupts which are already managed cannot be modified */
 421	if (irqd_affinity_is_managed(&desc->irq_data)) {
 422		ret = -EBUSY;
 423		goto out_unlock;
 424	}
 425
 426	/*
 427	 * Deactivate the interrupt. That's required to undo
 428	 * anything an earlier activation has established.
 429	 */
 430	activated = irqd_is_activated(&desc->irq_data);
 431	if (activated)
 432		irq_domain_deactivate_irq(&desc->irq_data);
 433
 434	if (affinity->is_managed) {
 435		irqd_set(&desc->irq_data, IRQD_AFFINITY_MANAGED);
 436		irqd_set(&desc->irq_data, IRQD_MANAGED_SHUTDOWN);
 437	}
 438
 439	cpumask_copy(desc->irq_common_data.affinity, &affinity->mask);
 440
 441	/* Restore the activation state */
 442	if (activated)
 443		irq_domain_activate_irq(&desc->irq_data, false);
 444
 445out_unlock:
 446	irq_put_desc_busunlock(desc, flags);
 447	return ret;
 448}
 449
 450static int __irq_set_affinity(unsigned int irq, const struct cpumask *mask,
 451			      bool force)
 452{
 453	struct irq_desc *desc = irq_to_desc(irq);
 454	unsigned long flags;
 455	int ret;
 456
 457	if (!desc)
 458		return -EINVAL;
 459
 460	raw_spin_lock_irqsave(&desc->lock, flags);
 461	ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
 462	raw_spin_unlock_irqrestore(&desc->lock, flags);
 463	return ret;
 464}
 465
 466/**
 467 * irq_set_affinity - Set the irq affinity of a given irq
 468 * @irq:	Interrupt to set affinity
 469 * @cpumask:	cpumask
 470 *
 471 * Fails if cpumask does not contain an online CPU
 472 */
 473int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
 474{
 475	return __irq_set_affinity(irq, cpumask, false);
 476}
 477EXPORT_SYMBOL_GPL(irq_set_affinity);
 478
 479/**
 480 * irq_force_affinity - Force the irq affinity of a given irq
 481 * @irq:	Interrupt to set affinity
 482 * @cpumask:	cpumask
 483 *
 484 * Same as irq_set_affinity, but without checking the mask against
 485 * online cpus.
 486 *
 487 * Solely for low level cpu hotplug code, where we need to make per
 488 * cpu interrupts affine before the cpu becomes online.
 489 */
 490int irq_force_affinity(unsigned int irq, const struct cpumask *cpumask)
 491{
 492	return __irq_set_affinity(irq, cpumask, true);
 493}
 494EXPORT_SYMBOL_GPL(irq_force_affinity);
 495
 496int __irq_apply_affinity_hint(unsigned int irq, const struct cpumask *m,
 497			      bool setaffinity)
 498{
 499	unsigned long flags;
 500	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 501
 502	if (!desc)
 503		return -EINVAL;
 504	desc->affinity_hint = m;
 505	irq_put_desc_unlock(desc, flags);
 506	if (m && setaffinity)
 
 507		__irq_set_affinity(irq, m, false);
 508	return 0;
 509}
 510EXPORT_SYMBOL_GPL(__irq_apply_affinity_hint);
 511
 512static void irq_affinity_notify(struct work_struct *work)
 513{
 514	struct irq_affinity_notify *notify =
 515		container_of(work, struct irq_affinity_notify, work);
 516	struct irq_desc *desc = irq_to_desc(notify->irq);
 517	cpumask_var_t cpumask;
 518	unsigned long flags;
 519
 520	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
 521		goto out;
 522
 523	raw_spin_lock_irqsave(&desc->lock, flags);
 524	if (irq_move_pending(&desc->irq_data))
 525		irq_get_pending(cpumask, desc);
 526	else
 527		cpumask_copy(cpumask, desc->irq_common_data.affinity);
 528	raw_spin_unlock_irqrestore(&desc->lock, flags);
 529
 530	notify->notify(notify, cpumask);
 531
 532	free_cpumask_var(cpumask);
 533out:
 534	kref_put(&notify->kref, notify->release);
 535}
 536
 537/**
 538 *	irq_set_affinity_notifier - control notification of IRQ affinity changes
 539 *	@irq:		Interrupt for which to enable/disable notification
 540 *	@notify:	Context for notification, or %NULL to disable
 541 *			notification.  Function pointers must be initialised;
 542 *			the other fields will be initialised by this function.
 543 *
 544 *	Must be called in process context.  Notification may only be enabled
 545 *	after the IRQ is allocated and must be disabled before the IRQ is
 546 *	freed using free_irq().
 547 */
 548int
 549irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 550{
 551	struct irq_desc *desc = irq_to_desc(irq);
 552	struct irq_affinity_notify *old_notify;
 553	unsigned long flags;
 554
 555	/* The release function is promised process context */
 556	might_sleep();
 557
 558	if (!desc || desc->istate & IRQS_NMI)
 559		return -EINVAL;
 560
 561	/* Complete initialisation of *notify */
 562	if (notify) {
 563		notify->irq = irq;
 564		kref_init(&notify->kref);
 565		INIT_WORK(&notify->work, irq_affinity_notify);
 566	}
 567
 568	raw_spin_lock_irqsave(&desc->lock, flags);
 569	old_notify = desc->affinity_notify;
 570	desc->affinity_notify = notify;
 571	raw_spin_unlock_irqrestore(&desc->lock, flags);
 572
 573	if (old_notify) {
 574		if (cancel_work_sync(&old_notify->work)) {
 575			/* Pending work had a ref, put that one too */
 576			kref_put(&old_notify->kref, old_notify->release);
 577		}
 578		kref_put(&old_notify->kref, old_notify->release);
 579	}
 580
 581	return 0;
 582}
 583EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
 584
 585#ifndef CONFIG_AUTO_IRQ_AFFINITY
 586/*
 587 * Generic version of the affinity autoselector.
 588 */
 589int irq_setup_affinity(struct irq_desc *desc)
 590{
 591	struct cpumask *set = irq_default_affinity;
 592	int ret, node = irq_desc_get_node(desc);
 593	static DEFINE_RAW_SPINLOCK(mask_lock);
 594	static struct cpumask mask;
 595
 596	/* Excludes PER_CPU and NO_BALANCE interrupts */
 597	if (!__irq_can_set_affinity(desc))
 598		return 0;
 599
 600	raw_spin_lock(&mask_lock);
 601	/*
 602	 * Preserve the managed affinity setting and a userspace affinity
 603	 * setup, but make sure that one of the targets is online.
 604	 */
 605	if (irqd_affinity_is_managed(&desc->irq_data) ||
 606	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 607		if (cpumask_intersects(desc->irq_common_data.affinity,
 608				       cpu_online_mask))
 609			set = desc->irq_common_data.affinity;
 610		else
 611			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
 612	}
 613
 614	cpumask_and(&mask, cpu_online_mask, set);
 615	if (cpumask_empty(&mask))
 616		cpumask_copy(&mask, cpu_online_mask);
 617
 618	if (node != NUMA_NO_NODE) {
 619		const struct cpumask *nodemask = cpumask_of_node(node);
 620
 621		/* make sure at least one of the cpus in nodemask is online */
 622		if (cpumask_intersects(&mask, nodemask))
 623			cpumask_and(&mask, &mask, nodemask);
 624	}
 625	ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
 626	raw_spin_unlock(&mask_lock);
 627	return ret;
 628}
 629#else
 630/* Wrapper for ALPHA specific affinity selector magic */
 631int irq_setup_affinity(struct irq_desc *desc)
 632{
 633	return irq_select_affinity(irq_desc_get_irq(desc));
 634}
 635#endif /* CONFIG_AUTO_IRQ_AFFINITY */
 636#endif /* CONFIG_SMP */
 637
 638
 639/**
 640 *	irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
 641 *	@irq: interrupt number to set affinity
 642 *	@vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
 643 *	            specific data for percpu_devid interrupts
 644 *
 645 *	This function uses the vCPU specific data to set the vCPU
 646 *	affinity for an irq. The vCPU specific data is passed from
 647 *	outside, such as KVM. One example code path is as below:
 648 *	KVM -> IOMMU -> irq_set_vcpu_affinity().
 649 */
 650int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
 651{
 652	unsigned long flags;
 653	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 654	struct irq_data *data;
 655	struct irq_chip *chip;
 656	int ret = -ENOSYS;
 657
 658	if (!desc)
 659		return -EINVAL;
 660
 661	data = irq_desc_get_irq_data(desc);
 662	do {
 663		chip = irq_data_get_irq_chip(data);
 664		if (chip && chip->irq_set_vcpu_affinity)
 665			break;
 666#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
 667		data = data->parent_data;
 668#else
 669		data = NULL;
 670#endif
 671	} while (data);
 672
 673	if (data)
 674		ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
 675	irq_put_desc_unlock(desc, flags);
 676
 677	return ret;
 678}
 679EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
 680
 681void __disable_irq(struct irq_desc *desc)
 682{
 683	if (!desc->depth++)
 684		irq_disable(desc);
 685}
 686
 687static int __disable_irq_nosync(unsigned int irq)
 688{
 689	unsigned long flags;
 690	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 691
 692	if (!desc)
 693		return -EINVAL;
 694	__disable_irq(desc);
 695	irq_put_desc_busunlock(desc, flags);
 696	return 0;
 697}
 698
 699/**
 700 *	disable_irq_nosync - disable an irq without waiting
 701 *	@irq: Interrupt to disable
 702 *
 703 *	Disable the selected interrupt line.  Disables and Enables are
 704 *	nested.
 705 *	Unlike disable_irq(), this function does not ensure existing
 706 *	instances of the IRQ handler have completed before returning.
 707 *
 708 *	This function may be called from IRQ context.
 709 */
 710void disable_irq_nosync(unsigned int irq)
 711{
 712	__disable_irq_nosync(irq);
 713}
 714EXPORT_SYMBOL(disable_irq_nosync);
 715
 716/**
 717 *	disable_irq - disable an irq and wait for completion
 718 *	@irq: Interrupt to disable
 719 *
 720 *	Disable the selected interrupt line.  Enables and Disables are
 721 *	nested.
 722 *	This function waits for any pending IRQ handlers for this interrupt
 723 *	to complete before returning. If you use this function while
 724 *	holding a resource the IRQ handler may need you will deadlock.
 725 *
 726 *	This function may be called - with care - from IRQ context.
 727 */
 728void disable_irq(unsigned int irq)
 729{
 730	if (!__disable_irq_nosync(irq))
 731		synchronize_irq(irq);
 732}
 733EXPORT_SYMBOL(disable_irq);
 734
 735/**
 736 *	disable_hardirq - disables an irq and waits for hardirq completion
 737 *	@irq: Interrupt to disable
 738 *
 739 *	Disable the selected interrupt line.  Enables and Disables are
 740 *	nested.
 741 *	This function waits for any pending hard IRQ handlers for this
 742 *	interrupt to complete before returning. If you use this function while
 743 *	holding a resource the hard IRQ handler may need you will deadlock.
 744 *
 745 *	When used to optimistically disable an interrupt from atomic context
 746 *	the return value must be checked.
 747 *
 748 *	Returns: false if a threaded handler is active.
 749 *
 750 *	This function may be called - with care - from IRQ context.
 751 */
 752bool disable_hardirq(unsigned int irq)
 753{
 754	if (!__disable_irq_nosync(irq))
 755		return synchronize_hardirq(irq);
 756
 757	return false;
 758}
 759EXPORT_SYMBOL_GPL(disable_hardirq);
 760
 761/**
 762 *	disable_nmi_nosync - disable an nmi without waiting
 763 *	@irq: Interrupt to disable
 764 *
 765 *	Disable the selected interrupt line. Disables and enables are
 766 *	nested.
 767 *	The interrupt to disable must have been requested through request_nmi.
 768 *	Unlike disable_nmi(), this function does not ensure existing
 769 *	instances of the IRQ handler have completed before returning.
 770 */
 771void disable_nmi_nosync(unsigned int irq)
 772{
 773	disable_irq_nosync(irq);
 774}
 775
 776void __enable_irq(struct irq_desc *desc)
 777{
 778	switch (desc->depth) {
 779	case 0:
 780 err_out:
 781		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
 782		     irq_desc_get_irq(desc));
 783		break;
 784	case 1: {
 785		if (desc->istate & IRQS_SUSPENDED)
 786			goto err_out;
 787		/* Prevent probing on this irq: */
 788		irq_settings_set_noprobe(desc);
 789		/*
 790		 * Call irq_startup() not irq_enable() here because the
 791		 * interrupt might be marked NOAUTOEN. So irq_startup()
 792		 * needs to be invoked when it gets enabled the first
 793		 * time. If it was already started up, then irq_startup()
 794		 * will invoke irq_enable() under the hood.
 795		 */
 796		irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
 797		break;
 798	}
 799	default:
 800		desc->depth--;
 801	}
 802}
 803
 804/**
 805 *	enable_irq - enable handling of an irq
 806 *	@irq: Interrupt to enable
 807 *
 808 *	Undoes the effect of one call to disable_irq().  If this
 809 *	matches the last disable, processing of interrupts on this
 810 *	IRQ line is re-enabled.
 811 *
 812 *	This function may be called from IRQ context only when
 813 *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
 814 */
 815void enable_irq(unsigned int irq)
 816{
 817	unsigned long flags;
 818	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 819
 820	if (!desc)
 821		return;
 822	if (WARN(!desc->irq_data.chip,
 823		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
 824		goto out;
 825
 826	__enable_irq(desc);
 827out:
 828	irq_put_desc_busunlock(desc, flags);
 829}
 830EXPORT_SYMBOL(enable_irq);
 831
 832/**
 833 *	enable_nmi - enable handling of an nmi
 834 *	@irq: Interrupt to enable
 835 *
 836 *	The interrupt to enable must have been requested through request_nmi.
 837 *	Undoes the effect of one call to disable_nmi(). If this
 838 *	matches the last disable, processing of interrupts on this
 839 *	IRQ line is re-enabled.
 840 */
 841void enable_nmi(unsigned int irq)
 842{
 843	enable_irq(irq);
 844}
 845
 846static int set_irq_wake_real(unsigned int irq, unsigned int on)
 847{
 848	struct irq_desc *desc = irq_to_desc(irq);
 849	int ret = -ENXIO;
 850
 851	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
 852		return 0;
 853
 854	if (desc->irq_data.chip->irq_set_wake)
 855		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
 856
 857	return ret;
 858}
 859
 860/**
 861 *	irq_set_irq_wake - control irq power management wakeup
 862 *	@irq:	interrupt to control
 863 *	@on:	enable/disable power management wakeup
 864 *
 865 *	Enable/disable power management wakeup mode, which is
 866 *	disabled by default.  Enables and disables must match,
 867 *	just as they match for non-wakeup mode support.
 868 *
 869 *	Wakeup mode lets this IRQ wake the system from sleep
 870 *	states like "suspend to RAM".
 871 *
 872 *	Note: irq enable/disable state is completely orthogonal
 873 *	to the enable/disable state of irq wake. An irq can be
 874 *	disabled with disable_irq() and still wake the system as
 875 *	long as the irq has wake enabled. If this does not hold,
 876 *	then the underlying irq chip and the related driver need
 877 *	to be investigated.
 878 */
 879int irq_set_irq_wake(unsigned int irq, unsigned int on)
 880{
 881	unsigned long flags;
 882	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 883	int ret = 0;
 884
 885	if (!desc)
 886		return -EINVAL;
 887
 888	/* Don't use NMIs as wake up interrupts please */
 889	if (desc->istate & IRQS_NMI) {
 890		ret = -EINVAL;
 891		goto out_unlock;
 892	}
 893
 894	/* wakeup-capable irqs can be shared between drivers that
 895	 * don't need to have the same sleep mode behaviors.
 896	 */
 897	if (on) {
 898		if (desc->wake_depth++ == 0) {
 899			ret = set_irq_wake_real(irq, on);
 900			if (ret)
 901				desc->wake_depth = 0;
 902			else
 903				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
 904		}
 905	} else {
 906		if (desc->wake_depth == 0) {
 907			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
 908		} else if (--desc->wake_depth == 0) {
 909			ret = set_irq_wake_real(irq, on);
 910			if (ret)
 911				desc->wake_depth = 1;
 912			else
 913				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
 914		}
 915	}
 916
 917out_unlock:
 918	irq_put_desc_busunlock(desc, flags);
 919	return ret;
 920}
 921EXPORT_SYMBOL(irq_set_irq_wake);
 922
 923/*
 924 * Internal function that tells the architecture code whether a
 925 * particular irq has been exclusively allocated or is available
 926 * for driver use.
 927 */
 928int can_request_irq(unsigned int irq, unsigned long irqflags)
 929{
 930	unsigned long flags;
 931	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 932	int canrequest = 0;
 933
 934	if (!desc)
 935		return 0;
 936
 937	if (irq_settings_can_request(desc)) {
 938		if (!desc->action ||
 939		    irqflags & desc->action->flags & IRQF_SHARED)
 940			canrequest = 1;
 941	}
 942	irq_put_desc_unlock(desc, flags);
 943	return canrequest;
 944}
 945
 946int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
 947{
 948	struct irq_chip *chip = desc->irq_data.chip;
 949	int ret, unmask = 0;
 950
 951	if (!chip || !chip->irq_set_type) {
 952		/*
 953		 * IRQF_TRIGGER_* but the PIC does not support multiple
 954		 * flow-types?
 955		 */
 956		pr_debug("No set_type function for IRQ %d (%s)\n",
 957			 irq_desc_get_irq(desc),
 958			 chip ? (chip->name ? : "unknown") : "unknown");
 959		return 0;
 960	}
 961
 962	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
 963		if (!irqd_irq_masked(&desc->irq_data))
 964			mask_irq(desc);
 965		if (!irqd_irq_disabled(&desc->irq_data))
 966			unmask = 1;
 967	}
 968
 969	/* Mask all flags except trigger mode */
 970	flags &= IRQ_TYPE_SENSE_MASK;
 971	ret = chip->irq_set_type(&desc->irq_data, flags);
 972
 973	switch (ret) {
 974	case IRQ_SET_MASK_OK:
 975	case IRQ_SET_MASK_OK_DONE:
 976		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
 977		irqd_set(&desc->irq_data, flags);
 978		fallthrough;
 979
 980	case IRQ_SET_MASK_OK_NOCOPY:
 981		flags = irqd_get_trigger_type(&desc->irq_data);
 982		irq_settings_set_trigger_mask(desc, flags);
 983		irqd_clear(&desc->irq_data, IRQD_LEVEL);
 984		irq_settings_clr_level(desc);
 985		if (flags & IRQ_TYPE_LEVEL_MASK) {
 986			irq_settings_set_level(desc);
 987			irqd_set(&desc->irq_data, IRQD_LEVEL);
 988		}
 989
 990		ret = 0;
 991		break;
 992	default:
 993		pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
 994		       flags, irq_desc_get_irq(desc), chip->irq_set_type);
 995	}
 996	if (unmask)
 997		unmask_irq(desc);
 998	return ret;
 999}
1000
1001#ifdef CONFIG_HARDIRQS_SW_RESEND
1002int irq_set_parent(int irq, int parent_irq)
1003{
1004	unsigned long flags;
1005	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
1006
1007	if (!desc)
1008		return -EINVAL;
1009
1010	desc->parent_irq = parent_irq;
1011
1012	irq_put_desc_unlock(desc, flags);
1013	return 0;
1014}
1015EXPORT_SYMBOL_GPL(irq_set_parent);
1016#endif
1017
1018/*
1019 * Default primary interrupt handler for threaded interrupts. Is
1020 * assigned as primary handler when request_threaded_irq is called
1021 * with handler == NULL. Useful for oneshot interrupts.
1022 */
1023static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
1024{
1025	return IRQ_WAKE_THREAD;
1026}
1027
1028/*
1029 * Primary handler for nested threaded interrupts. Should never be
1030 * called.
1031 */
1032static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
1033{
1034	WARN(1, "Primary handler called for nested irq %d\n", irq);
1035	return IRQ_NONE;
1036}
1037
1038static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
1039{
1040	WARN(1, "Secondary action handler called for irq %d\n", irq);
1041	return IRQ_NONE;
1042}
1043
1044static int irq_wait_for_interrupt(struct irqaction *action)
1045{
1046	for (;;) {
1047		set_current_state(TASK_INTERRUPTIBLE);
1048
1049		if (kthread_should_stop()) {
1050			/* may need to run one last time */
1051			if (test_and_clear_bit(IRQTF_RUNTHREAD,
1052					       &action->thread_flags)) {
1053				__set_current_state(TASK_RUNNING);
1054				return 0;
1055			}
1056			__set_current_state(TASK_RUNNING);
1057			return -1;
1058		}
1059
1060		if (test_and_clear_bit(IRQTF_RUNTHREAD,
1061				       &action->thread_flags)) {
1062			__set_current_state(TASK_RUNNING);
1063			return 0;
1064		}
1065		schedule();
1066	}
1067}
1068
1069/*
1070 * Oneshot interrupts keep the irq line masked until the threaded
1071 * handler finished. unmask if the interrupt has not been disabled and
1072 * is marked MASKED.
1073 */
1074static void irq_finalize_oneshot(struct irq_desc *desc,
1075				 struct irqaction *action)
1076{
1077	if (!(desc->istate & IRQS_ONESHOT) ||
1078	    action->handler == irq_forced_secondary_handler)
1079		return;
1080again:
1081	chip_bus_lock(desc);
1082	raw_spin_lock_irq(&desc->lock);
1083
1084	/*
1085	 * Implausible though it may be we need to protect us against
1086	 * the following scenario:
1087	 *
1088	 * The thread is faster done than the hard interrupt handler
1089	 * on the other CPU. If we unmask the irq line then the
1090	 * interrupt can come in again and masks the line, leaves due
1091	 * to IRQS_INPROGRESS and the irq line is masked forever.
1092	 *
1093	 * This also serializes the state of shared oneshot handlers
1094	 * versus "desc->threads_oneshot |= action->thread_mask;" in
1095	 * irq_wake_thread(). See the comment there which explains the
1096	 * serialization.
1097	 */
1098	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
1099		raw_spin_unlock_irq(&desc->lock);
1100		chip_bus_sync_unlock(desc);
1101		cpu_relax();
1102		goto again;
1103	}
1104
1105	/*
1106	 * Now check again, whether the thread should run. Otherwise
1107	 * we would clear the threads_oneshot bit of this thread which
1108	 * was just set.
1109	 */
1110	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1111		goto out_unlock;
1112
1113	desc->threads_oneshot &= ~action->thread_mask;
1114
1115	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
1116	    irqd_irq_masked(&desc->irq_data))
1117		unmask_threaded_irq(desc);
1118
1119out_unlock:
1120	raw_spin_unlock_irq(&desc->lock);
1121	chip_bus_sync_unlock(desc);
1122}
1123
1124#ifdef CONFIG_SMP
1125/*
1126 * Check whether we need to change the affinity of the interrupt thread.
1127 */
1128static void
1129irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
1130{
1131	cpumask_var_t mask;
1132	bool valid = true;
1133
1134	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
1135		return;
1136
1137	/*
1138	 * In case we are out of memory we set IRQTF_AFFINITY again and
1139	 * try again next time
1140	 */
1141	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1142		set_bit(IRQTF_AFFINITY, &action->thread_flags);
1143		return;
1144	}
1145
1146	raw_spin_lock_irq(&desc->lock);
1147	/*
1148	 * This code is triggered unconditionally. Check the affinity
1149	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
1150	 */
1151	if (cpumask_available(desc->irq_common_data.affinity)) {
1152		const struct cpumask *m;
1153
1154		m = irq_data_get_effective_affinity_mask(&desc->irq_data);
1155		cpumask_copy(mask, m);
1156	} else {
1157		valid = false;
1158	}
1159	raw_spin_unlock_irq(&desc->lock);
1160
1161	if (valid)
1162		set_cpus_allowed_ptr(current, mask);
1163	free_cpumask_var(mask);
1164}
1165#else
1166static inline void
1167irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
1168#endif
1169
1170/*
1171 * Interrupts which are not explicitly requested as threaded
1172 * interrupts rely on the implicit bh/preempt disable of the hard irq
1173 * context. So we need to disable bh here to avoid deadlocks and other
1174 * side effects.
1175 */
1176static irqreturn_t
1177irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
1178{
1179	irqreturn_t ret;
1180
1181	local_bh_disable();
1182	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1183		local_irq_disable();
1184	ret = action->thread_fn(action->irq, action->dev_id);
1185	if (ret == IRQ_HANDLED)
1186		atomic_inc(&desc->threads_handled);
1187
1188	irq_finalize_oneshot(desc, action);
1189	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1190		local_irq_enable();
1191	local_bh_enable();
1192	return ret;
1193}
1194
1195/*
1196 * Interrupts explicitly requested as threaded interrupts want to be
1197 * preemptible - many of them need to sleep and wait for slow busses to
1198 * complete.
1199 */
1200static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1201		struct irqaction *action)
1202{
1203	irqreturn_t ret;
1204
1205	ret = action->thread_fn(action->irq, action->dev_id);
1206	if (ret == IRQ_HANDLED)
1207		atomic_inc(&desc->threads_handled);
1208
1209	irq_finalize_oneshot(desc, action);
1210	return ret;
1211}
1212
1213static void wake_threads_waitq(struct irq_desc *desc)
1214{
1215	if (atomic_dec_and_test(&desc->threads_active))
1216		wake_up(&desc->wait_for_threads);
1217}
1218
1219static void irq_thread_dtor(struct callback_head *unused)
1220{
1221	struct task_struct *tsk = current;
1222	struct irq_desc *desc;
1223	struct irqaction *action;
1224
1225	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1226		return;
1227
1228	action = kthread_data(tsk);
1229
1230	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1231	       tsk->comm, tsk->pid, action->irq);
1232
1233
1234	desc = irq_to_desc(action->irq);
1235	/*
1236	 * If IRQTF_RUNTHREAD is set, we need to decrement
1237	 * desc->threads_active and wake possible waiters.
1238	 */
1239	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1240		wake_threads_waitq(desc);
1241
1242	/* Prevent a stale desc->threads_oneshot */
1243	irq_finalize_oneshot(desc, action);
1244}
1245
1246static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1247{
1248	struct irqaction *secondary = action->secondary;
1249
1250	if (WARN_ON_ONCE(!secondary))
1251		return;
1252
1253	raw_spin_lock_irq(&desc->lock);
1254	__irq_wake_thread(desc, secondary);
1255	raw_spin_unlock_irq(&desc->lock);
1256}
1257
1258/*
1259 * Internal function to notify that a interrupt thread is ready.
1260 */
1261static void irq_thread_set_ready(struct irq_desc *desc,
1262				 struct irqaction *action)
1263{
1264	set_bit(IRQTF_READY, &action->thread_flags);
1265	wake_up(&desc->wait_for_threads);
1266}
1267
1268/*
1269 * Internal function to wake up a interrupt thread and wait until it is
1270 * ready.
1271 */
1272static void wake_up_and_wait_for_irq_thread_ready(struct irq_desc *desc,
1273						  struct irqaction *action)
1274{
1275	if (!action || !action->thread)
1276		return;
1277
1278	wake_up_process(action->thread);
1279	wait_event(desc->wait_for_threads,
1280		   test_bit(IRQTF_READY, &action->thread_flags));
1281}
1282
1283/*
1284 * Interrupt handler thread
1285 */
1286static int irq_thread(void *data)
1287{
1288	struct callback_head on_exit_work;
1289	struct irqaction *action = data;
1290	struct irq_desc *desc = irq_to_desc(action->irq);
1291	irqreturn_t (*handler_fn)(struct irq_desc *desc,
1292			struct irqaction *action);
1293
1294	irq_thread_set_ready(desc, action);
1295
1296	sched_set_fifo(current);
1297
1298	if (force_irqthreads() && test_bit(IRQTF_FORCED_THREAD,
1299					   &action->thread_flags))
1300		handler_fn = irq_forced_thread_fn;
1301	else
1302		handler_fn = irq_thread_fn;
1303
1304	init_task_work(&on_exit_work, irq_thread_dtor);
1305	task_work_add(current, &on_exit_work, TWA_NONE);
1306
1307	irq_thread_check_affinity(desc, action);
1308
1309	while (!irq_wait_for_interrupt(action)) {
1310		irqreturn_t action_ret;
1311
1312		irq_thread_check_affinity(desc, action);
1313
1314		action_ret = handler_fn(desc, action);
1315		if (action_ret == IRQ_WAKE_THREAD)
1316			irq_wake_secondary(desc, action);
1317
1318		wake_threads_waitq(desc);
1319	}
1320
1321	/*
1322	 * This is the regular exit path. __free_irq() is stopping the
1323	 * thread via kthread_stop() after calling
1324	 * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1325	 * oneshot mask bit can be set.
1326	 */
1327	task_work_cancel(current, irq_thread_dtor);
1328	return 0;
1329}
1330
1331/**
1332 *	irq_wake_thread - wake the irq thread for the action identified by dev_id
1333 *	@irq:		Interrupt line
1334 *	@dev_id:	Device identity for which the thread should be woken
1335 *
1336 */
1337void irq_wake_thread(unsigned int irq, void *dev_id)
1338{
1339	struct irq_desc *desc = irq_to_desc(irq);
1340	struct irqaction *action;
1341	unsigned long flags;
1342
1343	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1344		return;
1345
1346	raw_spin_lock_irqsave(&desc->lock, flags);
1347	for_each_action_of_desc(desc, action) {
1348		if (action->dev_id == dev_id) {
1349			if (action->thread)
1350				__irq_wake_thread(desc, action);
1351			break;
1352		}
1353	}
1354	raw_spin_unlock_irqrestore(&desc->lock, flags);
1355}
1356EXPORT_SYMBOL_GPL(irq_wake_thread);
1357
1358static int irq_setup_forced_threading(struct irqaction *new)
1359{
1360	if (!force_irqthreads())
1361		return 0;
1362	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1363		return 0;
1364
1365	/*
1366	 * No further action required for interrupts which are requested as
1367	 * threaded interrupts already
1368	 */
1369	if (new->handler == irq_default_primary_handler)
1370		return 0;
1371
1372	new->flags |= IRQF_ONESHOT;
1373
1374	/*
1375	 * Handle the case where we have a real primary handler and a
1376	 * thread handler. We force thread them as well by creating a
1377	 * secondary action.
1378	 */
1379	if (new->handler && new->thread_fn) {
1380		/* Allocate the secondary action */
1381		new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1382		if (!new->secondary)
1383			return -ENOMEM;
1384		new->secondary->handler = irq_forced_secondary_handler;
1385		new->secondary->thread_fn = new->thread_fn;
1386		new->secondary->dev_id = new->dev_id;
1387		new->secondary->irq = new->irq;
1388		new->secondary->name = new->name;
1389	}
1390	/* Deal with the primary handler */
1391	set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1392	new->thread_fn = new->handler;
1393	new->handler = irq_default_primary_handler;
1394	return 0;
1395}
1396
1397static int irq_request_resources(struct irq_desc *desc)
1398{
1399	struct irq_data *d = &desc->irq_data;
1400	struct irq_chip *c = d->chip;
1401
1402	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1403}
1404
1405static void irq_release_resources(struct irq_desc *desc)
1406{
1407	struct irq_data *d = &desc->irq_data;
1408	struct irq_chip *c = d->chip;
1409
1410	if (c->irq_release_resources)
1411		c->irq_release_resources(d);
1412}
1413
1414static bool irq_supports_nmi(struct irq_desc *desc)
1415{
1416	struct irq_data *d = irq_desc_get_irq_data(desc);
1417
1418#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1419	/* Only IRQs directly managed by the root irqchip can be set as NMI */
1420	if (d->parent_data)
1421		return false;
1422#endif
1423	/* Don't support NMIs for chips behind a slow bus */
1424	if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1425		return false;
1426
1427	return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1428}
1429
1430static int irq_nmi_setup(struct irq_desc *desc)
1431{
1432	struct irq_data *d = irq_desc_get_irq_data(desc);
1433	struct irq_chip *c = d->chip;
1434
1435	return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1436}
1437
1438static void irq_nmi_teardown(struct irq_desc *desc)
1439{
1440	struct irq_data *d = irq_desc_get_irq_data(desc);
1441	struct irq_chip *c = d->chip;
1442
1443	if (c->irq_nmi_teardown)
1444		c->irq_nmi_teardown(d);
1445}
1446
1447static int
1448setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1449{
1450	struct task_struct *t;
1451
1452	if (!secondary) {
1453		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1454				   new->name);
1455	} else {
1456		t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1457				   new->name);
1458	}
1459
1460	if (IS_ERR(t))
1461		return PTR_ERR(t);
1462
 
 
1463	/*
1464	 * We keep the reference to the task struct even if
1465	 * the thread dies to avoid that the interrupt code
1466	 * references an already freed task_struct.
1467	 */
1468	new->thread = get_task_struct(t);
1469	/*
1470	 * Tell the thread to set its affinity. This is
1471	 * important for shared interrupt handlers as we do
1472	 * not invoke setup_affinity() for the secondary
1473	 * handlers as everything is already set up. Even for
1474	 * interrupts marked with IRQF_NO_BALANCE this is
1475	 * correct as we want the thread to move to the cpu(s)
1476	 * on which the requesting code placed the interrupt.
1477	 */
1478	set_bit(IRQTF_AFFINITY, &new->thread_flags);
1479	return 0;
1480}
1481
1482/*
1483 * Internal function to register an irqaction - typically used to
1484 * allocate special interrupts that are part of the architecture.
1485 *
1486 * Locking rules:
1487 *
1488 * desc->request_mutex	Provides serialization against a concurrent free_irq()
1489 *   chip_bus_lock	Provides serialization for slow bus operations
1490 *     desc->lock	Provides serialization against hard interrupts
1491 *
1492 * chip_bus_lock and desc->lock are sufficient for all other management and
1493 * interrupt related functions. desc->request_mutex solely serializes
1494 * request/free_irq().
1495 */
1496static int
1497__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1498{
1499	struct irqaction *old, **old_ptr;
1500	unsigned long flags, thread_mask = 0;
1501	int ret, nested, shared = 0;
1502
1503	if (!desc)
1504		return -EINVAL;
1505
1506	if (desc->irq_data.chip == &no_irq_chip)
1507		return -ENOSYS;
1508	if (!try_module_get(desc->owner))
1509		return -ENODEV;
1510
1511	new->irq = irq;
1512
1513	/*
1514	 * If the trigger type is not specified by the caller,
1515	 * then use the default for this interrupt.
1516	 */
1517	if (!(new->flags & IRQF_TRIGGER_MASK))
1518		new->flags |= irqd_get_trigger_type(&desc->irq_data);
1519
1520	/*
1521	 * Check whether the interrupt nests into another interrupt
1522	 * thread.
1523	 */
1524	nested = irq_settings_is_nested_thread(desc);
1525	if (nested) {
1526		if (!new->thread_fn) {
1527			ret = -EINVAL;
1528			goto out_mput;
1529		}
1530		/*
1531		 * Replace the primary handler which was provided from
1532		 * the driver for non nested interrupt handling by the
1533		 * dummy function which warns when called.
1534		 */
1535		new->handler = irq_nested_primary_handler;
1536	} else {
1537		if (irq_settings_can_thread(desc)) {
1538			ret = irq_setup_forced_threading(new);
1539			if (ret)
1540				goto out_mput;
1541		}
1542	}
1543
1544	/*
1545	 * Create a handler thread when a thread function is supplied
1546	 * and the interrupt does not nest into another interrupt
1547	 * thread.
1548	 */
1549	if (new->thread_fn && !nested) {
1550		ret = setup_irq_thread(new, irq, false);
1551		if (ret)
1552			goto out_mput;
1553		if (new->secondary) {
1554			ret = setup_irq_thread(new->secondary, irq, true);
1555			if (ret)
1556				goto out_thread;
1557		}
1558	}
1559
1560	/*
1561	 * Drivers are often written to work w/o knowledge about the
1562	 * underlying irq chip implementation, so a request for a
1563	 * threaded irq without a primary hard irq context handler
1564	 * requires the ONESHOT flag to be set. Some irq chips like
1565	 * MSI based interrupts are per se one shot safe. Check the
1566	 * chip flags, so we can avoid the unmask dance at the end of
1567	 * the threaded handler for those.
1568	 */
1569	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1570		new->flags &= ~IRQF_ONESHOT;
1571
1572	/*
1573	 * Protects against a concurrent __free_irq() call which might wait
1574	 * for synchronize_hardirq() to complete without holding the optional
1575	 * chip bus lock and desc->lock. Also protects against handing out
1576	 * a recycled oneshot thread_mask bit while it's still in use by
1577	 * its previous owner.
1578	 */
1579	mutex_lock(&desc->request_mutex);
1580
1581	/*
1582	 * Acquire bus lock as the irq_request_resources() callback below
1583	 * might rely on the serialization or the magic power management
1584	 * functions which are abusing the irq_bus_lock() callback,
1585	 */
1586	chip_bus_lock(desc);
1587
1588	/* First installed action requests resources. */
1589	if (!desc->action) {
1590		ret = irq_request_resources(desc);
1591		if (ret) {
1592			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1593			       new->name, irq, desc->irq_data.chip->name);
1594			goto out_bus_unlock;
1595		}
1596	}
1597
1598	/*
1599	 * The following block of code has to be executed atomically
1600	 * protected against a concurrent interrupt and any of the other
1601	 * management calls which are not serialized via
1602	 * desc->request_mutex or the optional bus lock.
1603	 */
1604	raw_spin_lock_irqsave(&desc->lock, flags);
1605	old_ptr = &desc->action;
1606	old = *old_ptr;
1607	if (old) {
1608		/*
1609		 * Can't share interrupts unless both agree to and are
1610		 * the same type (level, edge, polarity). So both flag
1611		 * fields must have IRQF_SHARED set and the bits which
1612		 * set the trigger type must match. Also all must
1613		 * agree on ONESHOT.
1614		 * Interrupt lines used for NMIs cannot be shared.
1615		 */
1616		unsigned int oldtype;
1617
1618		if (desc->istate & IRQS_NMI) {
1619			pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1620				new->name, irq, desc->irq_data.chip->name);
1621			ret = -EINVAL;
1622			goto out_unlock;
1623		}
1624
1625		/*
1626		 * If nobody did set the configuration before, inherit
1627		 * the one provided by the requester.
1628		 */
1629		if (irqd_trigger_type_was_set(&desc->irq_data)) {
1630			oldtype = irqd_get_trigger_type(&desc->irq_data);
1631		} else {
1632			oldtype = new->flags & IRQF_TRIGGER_MASK;
1633			irqd_set_trigger_type(&desc->irq_data, oldtype);
1634		}
1635
1636		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1637		    (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1638		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1639			goto mismatch;
1640
1641		/* All handlers must agree on per-cpuness */
1642		if ((old->flags & IRQF_PERCPU) !=
1643		    (new->flags & IRQF_PERCPU))
1644			goto mismatch;
1645
1646		/* add new interrupt at end of irq queue */
1647		do {
1648			/*
1649			 * Or all existing action->thread_mask bits,
1650			 * so we can find the next zero bit for this
1651			 * new action.
1652			 */
1653			thread_mask |= old->thread_mask;
1654			old_ptr = &old->next;
1655			old = *old_ptr;
1656		} while (old);
1657		shared = 1;
1658	}
1659
1660	/*
1661	 * Setup the thread mask for this irqaction for ONESHOT. For
1662	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1663	 * conditional in irq_wake_thread().
1664	 */
1665	if (new->flags & IRQF_ONESHOT) {
1666		/*
1667		 * Unlikely to have 32 resp 64 irqs sharing one line,
1668		 * but who knows.
1669		 */
1670		if (thread_mask == ~0UL) {
1671			ret = -EBUSY;
1672			goto out_unlock;
1673		}
1674		/*
1675		 * The thread_mask for the action is or'ed to
1676		 * desc->thread_active to indicate that the
1677		 * IRQF_ONESHOT thread handler has been woken, but not
1678		 * yet finished. The bit is cleared when a thread
1679		 * completes. When all threads of a shared interrupt
1680		 * line have completed desc->threads_active becomes
1681		 * zero and the interrupt line is unmasked. See
1682		 * handle.c:irq_wake_thread() for further information.
1683		 *
1684		 * If no thread is woken by primary (hard irq context)
1685		 * interrupt handlers, then desc->threads_active is
1686		 * also checked for zero to unmask the irq line in the
1687		 * affected hard irq flow handlers
1688		 * (handle_[fasteoi|level]_irq).
1689		 *
1690		 * The new action gets the first zero bit of
1691		 * thread_mask assigned. See the loop above which or's
1692		 * all existing action->thread_mask bits.
1693		 */
1694		new->thread_mask = 1UL << ffz(thread_mask);
1695
1696	} else if (new->handler == irq_default_primary_handler &&
1697		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1698		/*
1699		 * The interrupt was requested with handler = NULL, so
1700		 * we use the default primary handler for it. But it
1701		 * does not have the oneshot flag set. In combination
1702		 * with level interrupts this is deadly, because the
1703		 * default primary handler just wakes the thread, then
1704		 * the irq lines is reenabled, but the device still
1705		 * has the level irq asserted. Rinse and repeat....
1706		 *
1707		 * While this works for edge type interrupts, we play
1708		 * it safe and reject unconditionally because we can't
1709		 * say for sure which type this interrupt really
1710		 * has. The type flags are unreliable as the
1711		 * underlying chip implementation can override them.
1712		 */
1713		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for %s (irq %d)\n",
1714		       new->name, irq);
1715		ret = -EINVAL;
1716		goto out_unlock;
1717	}
1718
1719	if (!shared) {
 
 
1720		/* Setup the type (level, edge polarity) if configured: */
1721		if (new->flags & IRQF_TRIGGER_MASK) {
1722			ret = __irq_set_trigger(desc,
1723						new->flags & IRQF_TRIGGER_MASK);
1724
1725			if (ret)
1726				goto out_unlock;
1727		}
1728
1729		/*
1730		 * Activate the interrupt. That activation must happen
1731		 * independently of IRQ_NOAUTOEN. request_irq() can fail
1732		 * and the callers are supposed to handle
1733		 * that. enable_irq() of an interrupt requested with
1734		 * IRQ_NOAUTOEN is not supposed to fail. The activation
1735		 * keeps it in shutdown mode, it merily associates
1736		 * resources if necessary and if that's not possible it
1737		 * fails. Interrupts which are in managed shutdown mode
1738		 * will simply ignore that activation request.
1739		 */
1740		ret = irq_activate(desc);
1741		if (ret)
1742			goto out_unlock;
1743
1744		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1745				  IRQS_ONESHOT | IRQS_WAITING);
1746		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1747
1748		if (new->flags & IRQF_PERCPU) {
1749			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1750			irq_settings_set_per_cpu(desc);
1751			if (new->flags & IRQF_NO_DEBUG)
1752				irq_settings_set_no_debug(desc);
1753		}
1754
1755		if (noirqdebug)
1756			irq_settings_set_no_debug(desc);
1757
1758		if (new->flags & IRQF_ONESHOT)
1759			desc->istate |= IRQS_ONESHOT;
1760
1761		/* Exclude IRQ from balancing if requested */
1762		if (new->flags & IRQF_NOBALANCING) {
1763			irq_settings_set_no_balancing(desc);
1764			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1765		}
1766
1767		if (!(new->flags & IRQF_NO_AUTOEN) &&
1768		    irq_settings_can_autoenable(desc)) {
1769			irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1770		} else {
1771			/*
1772			 * Shared interrupts do not go well with disabling
1773			 * auto enable. The sharing interrupt might request
1774			 * it while it's still disabled and then wait for
1775			 * interrupts forever.
1776			 */
1777			WARN_ON_ONCE(new->flags & IRQF_SHARED);
1778			/* Undo nested disables: */
1779			desc->depth = 1;
1780		}
1781
1782	} else if (new->flags & IRQF_TRIGGER_MASK) {
1783		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1784		unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1785
1786		if (nmsk != omsk)
1787			/* hope the handler works with current  trigger mode */
1788			pr_warn("irq %d uses trigger mode %u; requested %u\n",
1789				irq, omsk, nmsk);
1790	}
1791
1792	*old_ptr = new;
1793
1794	irq_pm_install_action(desc, new);
1795
1796	/* Reset broken irq detection when installing new handler */
1797	desc->irq_count = 0;
1798	desc->irqs_unhandled = 0;
1799
1800	/*
1801	 * Check whether we disabled the irq via the spurious handler
1802	 * before. Reenable it and give it another chance.
1803	 */
1804	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1805		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1806		__enable_irq(desc);
1807	}
1808
1809	raw_spin_unlock_irqrestore(&desc->lock, flags);
1810	chip_bus_sync_unlock(desc);
1811	mutex_unlock(&desc->request_mutex);
1812
1813	irq_setup_timings(desc, new);
1814
1815	wake_up_and_wait_for_irq_thread_ready(desc, new);
1816	wake_up_and_wait_for_irq_thread_ready(desc, new->secondary);
 
 
 
 
 
 
1817
1818	register_irq_proc(irq, desc);
1819	new->dir = NULL;
1820	register_handler_proc(irq, new);
1821	return 0;
1822
1823mismatch:
1824	if (!(new->flags & IRQF_PROBE_SHARED)) {
1825		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1826		       irq, new->flags, new->name, old->flags, old->name);
1827#ifdef CONFIG_DEBUG_SHIRQ
1828		dump_stack();
1829#endif
1830	}
1831	ret = -EBUSY;
1832
1833out_unlock:
1834	raw_spin_unlock_irqrestore(&desc->lock, flags);
1835
1836	if (!desc->action)
1837		irq_release_resources(desc);
1838out_bus_unlock:
1839	chip_bus_sync_unlock(desc);
1840	mutex_unlock(&desc->request_mutex);
1841
1842out_thread:
1843	if (new->thread) {
1844		struct task_struct *t = new->thread;
1845
1846		new->thread = NULL;
1847		kthread_stop(t);
1848		put_task_struct(t);
1849	}
1850	if (new->secondary && new->secondary->thread) {
1851		struct task_struct *t = new->secondary->thread;
1852
1853		new->secondary->thread = NULL;
1854		kthread_stop(t);
1855		put_task_struct(t);
1856	}
1857out_mput:
1858	module_put(desc->owner);
1859	return ret;
1860}
1861
1862/*
1863 * Internal function to unregister an irqaction - used to free
1864 * regular and special interrupts that are part of the architecture.
1865 */
1866static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1867{
1868	unsigned irq = desc->irq_data.irq;
1869	struct irqaction *action, **action_ptr;
1870	unsigned long flags;
1871
1872	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1873
1874	mutex_lock(&desc->request_mutex);
1875	chip_bus_lock(desc);
1876	raw_spin_lock_irqsave(&desc->lock, flags);
1877
1878	/*
1879	 * There can be multiple actions per IRQ descriptor, find the right
1880	 * one based on the dev_id:
1881	 */
1882	action_ptr = &desc->action;
1883	for (;;) {
1884		action = *action_ptr;
1885
1886		if (!action) {
1887			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1888			raw_spin_unlock_irqrestore(&desc->lock, flags);
1889			chip_bus_sync_unlock(desc);
1890			mutex_unlock(&desc->request_mutex);
1891			return NULL;
1892		}
1893
1894		if (action->dev_id == dev_id)
1895			break;
1896		action_ptr = &action->next;
1897	}
1898
1899	/* Found it - now remove it from the list of entries: */
1900	*action_ptr = action->next;
1901
1902	irq_pm_remove_action(desc, action);
1903
1904	/* If this was the last handler, shut down the IRQ line: */
1905	if (!desc->action) {
1906		irq_settings_clr_disable_unlazy(desc);
1907		/* Only shutdown. Deactivate after synchronize_hardirq() */
1908		irq_shutdown(desc);
1909	}
1910
1911#ifdef CONFIG_SMP
1912	/* make sure affinity_hint is cleaned up */
1913	if (WARN_ON_ONCE(desc->affinity_hint))
1914		desc->affinity_hint = NULL;
1915#endif
1916
1917	raw_spin_unlock_irqrestore(&desc->lock, flags);
1918	/*
1919	 * Drop bus_lock here so the changes which were done in the chip
1920	 * callbacks above are synced out to the irq chips which hang
1921	 * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1922	 *
1923	 * Aside of that the bus_lock can also be taken from the threaded
1924	 * handler in irq_finalize_oneshot() which results in a deadlock
1925	 * because kthread_stop() would wait forever for the thread to
1926	 * complete, which is blocked on the bus lock.
1927	 *
1928	 * The still held desc->request_mutex() protects against a
1929	 * concurrent request_irq() of this irq so the release of resources
1930	 * and timing data is properly serialized.
1931	 */
1932	chip_bus_sync_unlock(desc);
1933
1934	unregister_handler_proc(irq, action);
1935
1936	/*
1937	 * Make sure it's not being used on another CPU and if the chip
1938	 * supports it also make sure that there is no (not yet serviced)
1939	 * interrupt in flight at the hardware level.
1940	 */
1941	__synchronize_hardirq(desc, true);
1942
1943#ifdef CONFIG_DEBUG_SHIRQ
1944	/*
1945	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1946	 * event to happen even now it's being freed, so let's make sure that
1947	 * is so by doing an extra call to the handler ....
1948	 *
1949	 * ( We do this after actually deregistering it, to make sure that a
1950	 *   'real' IRQ doesn't run in parallel with our fake. )
1951	 */
1952	if (action->flags & IRQF_SHARED) {
1953		local_irq_save(flags);
1954		action->handler(irq, dev_id);
1955		local_irq_restore(flags);
1956	}
1957#endif
1958
1959	/*
1960	 * The action has already been removed above, but the thread writes
1961	 * its oneshot mask bit when it completes. Though request_mutex is
1962	 * held across this which prevents __setup_irq() from handing out
1963	 * the same bit to a newly requested action.
1964	 */
1965	if (action->thread) {
1966		kthread_stop(action->thread);
1967		put_task_struct(action->thread);
1968		if (action->secondary && action->secondary->thread) {
1969			kthread_stop(action->secondary->thread);
1970			put_task_struct(action->secondary->thread);
1971		}
1972	}
1973
1974	/* Last action releases resources */
1975	if (!desc->action) {
1976		/*
1977		 * Reacquire bus lock as irq_release_resources() might
1978		 * require it to deallocate resources over the slow bus.
1979		 */
1980		chip_bus_lock(desc);
1981		/*
1982		 * There is no interrupt on the fly anymore. Deactivate it
1983		 * completely.
1984		 */
1985		raw_spin_lock_irqsave(&desc->lock, flags);
1986		irq_domain_deactivate_irq(&desc->irq_data);
1987		raw_spin_unlock_irqrestore(&desc->lock, flags);
1988
1989		irq_release_resources(desc);
1990		chip_bus_sync_unlock(desc);
1991		irq_remove_timings(desc);
1992	}
1993
1994	mutex_unlock(&desc->request_mutex);
1995
1996	irq_chip_pm_put(&desc->irq_data);
1997	module_put(desc->owner);
1998	kfree(action->secondary);
1999	return action;
2000}
2001
2002/**
2003 *	free_irq - free an interrupt allocated with request_irq
2004 *	@irq: Interrupt line to free
2005 *	@dev_id: Device identity to free
2006 *
2007 *	Remove an interrupt handler. The handler is removed and if the
2008 *	interrupt line is no longer in use by any driver it is disabled.
2009 *	On a shared IRQ the caller must ensure the interrupt is disabled
2010 *	on the card it drives before calling this function. The function
2011 *	does not return until any executing interrupts for this IRQ
2012 *	have completed.
2013 *
2014 *	This function must not be called from interrupt context.
2015 *
2016 *	Returns the devname argument passed to request_irq.
2017 */
2018const void *free_irq(unsigned int irq, void *dev_id)
2019{
2020	struct irq_desc *desc = irq_to_desc(irq);
2021	struct irqaction *action;
2022	const char *devname;
2023
2024	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2025		return NULL;
2026
2027#ifdef CONFIG_SMP
2028	if (WARN_ON(desc->affinity_notify))
2029		desc->affinity_notify = NULL;
2030#endif
2031
2032	action = __free_irq(desc, dev_id);
2033
2034	if (!action)
2035		return NULL;
2036
2037	devname = action->name;
2038	kfree(action);
2039	return devname;
2040}
2041EXPORT_SYMBOL(free_irq);
2042
2043/* This function must be called with desc->lock held */
2044static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
2045{
2046	const char *devname = NULL;
2047
2048	desc->istate &= ~IRQS_NMI;
2049
2050	if (!WARN_ON(desc->action == NULL)) {
2051		irq_pm_remove_action(desc, desc->action);
2052		devname = desc->action->name;
2053		unregister_handler_proc(irq, desc->action);
2054
2055		kfree(desc->action);
2056		desc->action = NULL;
2057	}
2058
2059	irq_settings_clr_disable_unlazy(desc);
2060	irq_shutdown_and_deactivate(desc);
2061
2062	irq_release_resources(desc);
2063
2064	irq_chip_pm_put(&desc->irq_data);
2065	module_put(desc->owner);
2066
2067	return devname;
2068}
2069
2070const void *free_nmi(unsigned int irq, void *dev_id)
2071{
2072	struct irq_desc *desc = irq_to_desc(irq);
2073	unsigned long flags;
2074	const void *devname;
2075
2076	if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
2077		return NULL;
2078
2079	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2080		return NULL;
2081
2082	/* NMI still enabled */
2083	if (WARN_ON(desc->depth == 0))
2084		disable_nmi_nosync(irq);
2085
2086	raw_spin_lock_irqsave(&desc->lock, flags);
2087
2088	irq_nmi_teardown(desc);
2089	devname = __cleanup_nmi(irq, desc);
2090
2091	raw_spin_unlock_irqrestore(&desc->lock, flags);
2092
2093	return devname;
2094}
2095
2096/**
2097 *	request_threaded_irq - allocate an interrupt line
2098 *	@irq: Interrupt line to allocate
2099 *	@handler: Function to be called when the IRQ occurs.
2100 *		  Primary handler for threaded interrupts.
2101 *		  If handler is NULL and thread_fn != NULL
2102 *		  the default primary handler is installed.
2103 *	@thread_fn: Function called from the irq handler thread
2104 *		    If NULL, no irq thread is created
2105 *	@irqflags: Interrupt type flags
2106 *	@devname: An ascii name for the claiming device
2107 *	@dev_id: A cookie passed back to the handler function
2108 *
2109 *	This call allocates interrupt resources and enables the
2110 *	interrupt line and IRQ handling. From the point this
2111 *	call is made your handler function may be invoked. Since
2112 *	your handler function must clear any interrupt the board
2113 *	raises, you must take care both to initialise your hardware
2114 *	and to set up the interrupt handler in the right order.
2115 *
2116 *	If you want to set up a threaded irq handler for your device
2117 *	then you need to supply @handler and @thread_fn. @handler is
2118 *	still called in hard interrupt context and has to check
2119 *	whether the interrupt originates from the device. If yes it
2120 *	needs to disable the interrupt on the device and return
2121 *	IRQ_WAKE_THREAD which will wake up the handler thread and run
2122 *	@thread_fn. This split handler design is necessary to support
2123 *	shared interrupts.
2124 *
2125 *	Dev_id must be globally unique. Normally the address of the
2126 *	device data structure is used as the cookie. Since the handler
2127 *	receives this value it makes sense to use it.
2128 *
2129 *	If your interrupt is shared you must pass a non NULL dev_id
2130 *	as this is required when freeing the interrupt.
2131 *
2132 *	Flags:
2133 *
2134 *	IRQF_SHARED		Interrupt is shared
2135 *	IRQF_TRIGGER_*		Specify active edge(s) or level
2136 *	IRQF_ONESHOT		Run thread_fn with interrupt line masked
2137 */
2138int request_threaded_irq(unsigned int irq, irq_handler_t handler,
2139			 irq_handler_t thread_fn, unsigned long irqflags,
2140			 const char *devname, void *dev_id)
2141{
2142	struct irqaction *action;
2143	struct irq_desc *desc;
2144	int retval;
2145
2146	if (irq == IRQ_NOTCONNECTED)
2147		return -ENOTCONN;
2148
2149	/*
2150	 * Sanity-check: shared interrupts must pass in a real dev-ID,
2151	 * otherwise we'll have trouble later trying to figure out
2152	 * which interrupt is which (messes up the interrupt freeing
2153	 * logic etc).
2154	 *
2155	 * Also shared interrupts do not go well with disabling auto enable.
2156	 * The sharing interrupt might request it while it's still disabled
2157	 * and then wait for interrupts forever.
2158	 *
2159	 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
2160	 * it cannot be set along with IRQF_NO_SUSPEND.
2161	 */
2162	if (((irqflags & IRQF_SHARED) && !dev_id) ||
2163	    ((irqflags & IRQF_SHARED) && (irqflags & IRQF_NO_AUTOEN)) ||
2164	    (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
2165	    ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
2166		return -EINVAL;
2167
2168	desc = irq_to_desc(irq);
2169	if (!desc)
2170		return -EINVAL;
2171
2172	if (!irq_settings_can_request(desc) ||
2173	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2174		return -EINVAL;
2175
2176	if (!handler) {
2177		if (!thread_fn)
2178			return -EINVAL;
2179		handler = irq_default_primary_handler;
2180	}
2181
2182	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2183	if (!action)
2184		return -ENOMEM;
2185
2186	action->handler = handler;
2187	action->thread_fn = thread_fn;
2188	action->flags = irqflags;
2189	action->name = devname;
2190	action->dev_id = dev_id;
2191
2192	retval = irq_chip_pm_get(&desc->irq_data);
2193	if (retval < 0) {
2194		kfree(action);
2195		return retval;
2196	}
2197
2198	retval = __setup_irq(irq, desc, action);
2199
2200	if (retval) {
2201		irq_chip_pm_put(&desc->irq_data);
2202		kfree(action->secondary);
2203		kfree(action);
2204	}
2205
2206#ifdef CONFIG_DEBUG_SHIRQ_FIXME
2207	if (!retval && (irqflags & IRQF_SHARED)) {
2208		/*
2209		 * It's a shared IRQ -- the driver ought to be prepared for it
2210		 * to happen immediately, so let's make sure....
2211		 * We disable the irq to make sure that a 'real' IRQ doesn't
2212		 * run in parallel with our fake.
2213		 */
2214		unsigned long flags;
2215
2216		disable_irq(irq);
2217		local_irq_save(flags);
2218
2219		handler(irq, dev_id);
2220
2221		local_irq_restore(flags);
2222		enable_irq(irq);
2223	}
2224#endif
2225	return retval;
2226}
2227EXPORT_SYMBOL(request_threaded_irq);
2228
2229/**
2230 *	request_any_context_irq - allocate an interrupt line
2231 *	@irq: Interrupt line to allocate
2232 *	@handler: Function to be called when the IRQ occurs.
2233 *		  Threaded handler for threaded interrupts.
2234 *	@flags: Interrupt type flags
2235 *	@name: An ascii name for the claiming device
2236 *	@dev_id: A cookie passed back to the handler function
2237 *
2238 *	This call allocates interrupt resources and enables the
2239 *	interrupt line and IRQ handling. It selects either a
2240 *	hardirq or threaded handling method depending on the
2241 *	context.
2242 *
2243 *	On failure, it returns a negative value. On success,
2244 *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2245 */
2246int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2247			    unsigned long flags, const char *name, void *dev_id)
2248{
2249	struct irq_desc *desc;
2250	int ret;
2251
2252	if (irq == IRQ_NOTCONNECTED)
2253		return -ENOTCONN;
2254
2255	desc = irq_to_desc(irq);
2256	if (!desc)
2257		return -EINVAL;
2258
2259	if (irq_settings_is_nested_thread(desc)) {
2260		ret = request_threaded_irq(irq, NULL, handler,
2261					   flags, name, dev_id);
2262		return !ret ? IRQC_IS_NESTED : ret;
2263	}
2264
2265	ret = request_irq(irq, handler, flags, name, dev_id);
2266	return !ret ? IRQC_IS_HARDIRQ : ret;
2267}
2268EXPORT_SYMBOL_GPL(request_any_context_irq);
2269
2270/**
2271 *	request_nmi - allocate an interrupt line for NMI delivery
2272 *	@irq: Interrupt line to allocate
2273 *	@handler: Function to be called when the IRQ occurs.
2274 *		  Threaded handler for threaded interrupts.
2275 *	@irqflags: Interrupt type flags
2276 *	@name: An ascii name for the claiming device
2277 *	@dev_id: A cookie passed back to the handler function
2278 *
2279 *	This call allocates interrupt resources and enables the
2280 *	interrupt line and IRQ handling. It sets up the IRQ line
2281 *	to be handled as an NMI.
2282 *
2283 *	An interrupt line delivering NMIs cannot be shared and IRQ handling
2284 *	cannot be threaded.
2285 *
2286 *	Interrupt lines requested for NMI delivering must produce per cpu
2287 *	interrupts and have auto enabling setting disabled.
2288 *
2289 *	Dev_id must be globally unique. Normally the address of the
2290 *	device data structure is used as the cookie. Since the handler
2291 *	receives this value it makes sense to use it.
2292 *
2293 *	If the interrupt line cannot be used to deliver NMIs, function
2294 *	will fail and return a negative value.
2295 */
2296int request_nmi(unsigned int irq, irq_handler_t handler,
2297		unsigned long irqflags, const char *name, void *dev_id)
2298{
2299	struct irqaction *action;
2300	struct irq_desc *desc;
2301	unsigned long flags;
2302	int retval;
2303
2304	if (irq == IRQ_NOTCONNECTED)
2305		return -ENOTCONN;
2306
2307	/* NMI cannot be shared, used for Polling */
2308	if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2309		return -EINVAL;
2310
2311	if (!(irqflags & IRQF_PERCPU))
2312		return -EINVAL;
2313
2314	if (!handler)
2315		return -EINVAL;
2316
2317	desc = irq_to_desc(irq);
2318
2319	if (!desc || (irq_settings_can_autoenable(desc) &&
2320	    !(irqflags & IRQF_NO_AUTOEN)) ||
2321	    !irq_settings_can_request(desc) ||
2322	    WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2323	    !irq_supports_nmi(desc))
2324		return -EINVAL;
2325
2326	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2327	if (!action)
2328		return -ENOMEM;
2329
2330	action->handler = handler;
2331	action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2332	action->name = name;
2333	action->dev_id = dev_id;
2334
2335	retval = irq_chip_pm_get(&desc->irq_data);
2336	if (retval < 0)
2337		goto err_out;
2338
2339	retval = __setup_irq(irq, desc, action);
2340	if (retval)
2341		goto err_irq_setup;
2342
2343	raw_spin_lock_irqsave(&desc->lock, flags);
2344
2345	/* Setup NMI state */
2346	desc->istate |= IRQS_NMI;
2347	retval = irq_nmi_setup(desc);
2348	if (retval) {
2349		__cleanup_nmi(irq, desc);
2350		raw_spin_unlock_irqrestore(&desc->lock, flags);
2351		return -EINVAL;
2352	}
2353
2354	raw_spin_unlock_irqrestore(&desc->lock, flags);
2355
2356	return 0;
2357
2358err_irq_setup:
2359	irq_chip_pm_put(&desc->irq_data);
2360err_out:
2361	kfree(action);
2362
2363	return retval;
2364}
2365
2366void enable_percpu_irq(unsigned int irq, unsigned int type)
2367{
2368	unsigned int cpu = smp_processor_id();
2369	unsigned long flags;
2370	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2371
2372	if (!desc)
2373		return;
2374
2375	/*
2376	 * If the trigger type is not specified by the caller, then
2377	 * use the default for this interrupt.
2378	 */
2379	type &= IRQ_TYPE_SENSE_MASK;
2380	if (type == IRQ_TYPE_NONE)
2381		type = irqd_get_trigger_type(&desc->irq_data);
2382
2383	if (type != IRQ_TYPE_NONE) {
2384		int ret;
2385
2386		ret = __irq_set_trigger(desc, type);
2387
2388		if (ret) {
2389			WARN(1, "failed to set type for IRQ%d\n", irq);
2390			goto out;
2391		}
2392	}
2393
2394	irq_percpu_enable(desc, cpu);
2395out:
2396	irq_put_desc_unlock(desc, flags);
2397}
2398EXPORT_SYMBOL_GPL(enable_percpu_irq);
2399
2400void enable_percpu_nmi(unsigned int irq, unsigned int type)
2401{
2402	enable_percpu_irq(irq, type);
2403}
2404
2405/**
2406 * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2407 * @irq:	Linux irq number to check for
2408 *
2409 * Must be called from a non migratable context. Returns the enable
2410 * state of a per cpu interrupt on the current cpu.
2411 */
2412bool irq_percpu_is_enabled(unsigned int irq)
2413{
2414	unsigned int cpu = smp_processor_id();
2415	struct irq_desc *desc;
2416	unsigned long flags;
2417	bool is_enabled;
2418
2419	desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2420	if (!desc)
2421		return false;
2422
2423	is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2424	irq_put_desc_unlock(desc, flags);
2425
2426	return is_enabled;
2427}
2428EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2429
2430void disable_percpu_irq(unsigned int irq)
2431{
2432	unsigned int cpu = smp_processor_id();
2433	unsigned long flags;
2434	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2435
2436	if (!desc)
2437		return;
2438
2439	irq_percpu_disable(desc, cpu);
2440	irq_put_desc_unlock(desc, flags);
2441}
2442EXPORT_SYMBOL_GPL(disable_percpu_irq);
2443
2444void disable_percpu_nmi(unsigned int irq)
2445{
2446	disable_percpu_irq(irq);
2447}
2448
2449/*
2450 * Internal function to unregister a percpu irqaction.
2451 */
2452static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2453{
2454	struct irq_desc *desc = irq_to_desc(irq);
2455	struct irqaction *action;
2456	unsigned long flags;
2457
2458	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2459
2460	if (!desc)
2461		return NULL;
2462
2463	raw_spin_lock_irqsave(&desc->lock, flags);
2464
2465	action = desc->action;
2466	if (!action || action->percpu_dev_id != dev_id) {
2467		WARN(1, "Trying to free already-free IRQ %d\n", irq);
2468		goto bad;
2469	}
2470
2471	if (!cpumask_empty(desc->percpu_enabled)) {
2472		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2473		     irq, cpumask_first(desc->percpu_enabled));
2474		goto bad;
2475	}
2476
2477	/* Found it - now remove it from the list of entries: */
2478	desc->action = NULL;
2479
2480	desc->istate &= ~IRQS_NMI;
2481
2482	raw_spin_unlock_irqrestore(&desc->lock, flags);
2483
2484	unregister_handler_proc(irq, action);
2485
2486	irq_chip_pm_put(&desc->irq_data);
2487	module_put(desc->owner);
2488	return action;
2489
2490bad:
2491	raw_spin_unlock_irqrestore(&desc->lock, flags);
2492	return NULL;
2493}
2494
2495/**
2496 *	remove_percpu_irq - free a per-cpu interrupt
2497 *	@irq: Interrupt line to free
2498 *	@act: irqaction for the interrupt
2499 *
2500 * Used to remove interrupts statically setup by the early boot process.
2501 */
2502void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2503{
2504	struct irq_desc *desc = irq_to_desc(irq);
2505
2506	if (desc && irq_settings_is_per_cpu_devid(desc))
2507	    __free_percpu_irq(irq, act->percpu_dev_id);
2508}
2509
2510/**
2511 *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
2512 *	@irq: Interrupt line to free
2513 *	@dev_id: Device identity to free
2514 *
2515 *	Remove a percpu interrupt handler. The handler is removed, but
2516 *	the interrupt line is not disabled. This must be done on each
2517 *	CPU before calling this function. The function does not return
2518 *	until any executing interrupts for this IRQ have completed.
2519 *
2520 *	This function must not be called from interrupt context.
2521 */
2522void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2523{
2524	struct irq_desc *desc = irq_to_desc(irq);
2525
2526	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2527		return;
2528
2529	chip_bus_lock(desc);
2530	kfree(__free_percpu_irq(irq, dev_id));
2531	chip_bus_sync_unlock(desc);
2532}
2533EXPORT_SYMBOL_GPL(free_percpu_irq);
2534
2535void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2536{
2537	struct irq_desc *desc = irq_to_desc(irq);
2538
2539	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2540		return;
2541
2542	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2543		return;
2544
2545	kfree(__free_percpu_irq(irq, dev_id));
2546}
2547
2548/**
2549 *	setup_percpu_irq - setup a per-cpu interrupt
2550 *	@irq: Interrupt line to setup
2551 *	@act: irqaction for the interrupt
2552 *
2553 * Used to statically setup per-cpu interrupts in the early boot process.
2554 */
2555int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2556{
2557	struct irq_desc *desc = irq_to_desc(irq);
2558	int retval;
2559
2560	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2561		return -EINVAL;
2562
2563	retval = irq_chip_pm_get(&desc->irq_data);
2564	if (retval < 0)
2565		return retval;
2566
2567	retval = __setup_irq(irq, desc, act);
2568
2569	if (retval)
2570		irq_chip_pm_put(&desc->irq_data);
2571
2572	return retval;
2573}
2574
2575/**
2576 *	__request_percpu_irq - allocate a percpu interrupt line
2577 *	@irq: Interrupt line to allocate
2578 *	@handler: Function to be called when the IRQ occurs.
2579 *	@flags: Interrupt type flags (IRQF_TIMER only)
2580 *	@devname: An ascii name for the claiming device
2581 *	@dev_id: A percpu cookie passed back to the handler function
2582 *
2583 *	This call allocates interrupt resources and enables the
2584 *	interrupt on the local CPU. If the interrupt is supposed to be
2585 *	enabled on other CPUs, it has to be done on each CPU using
2586 *	enable_percpu_irq().
2587 *
2588 *	Dev_id must be globally unique. It is a per-cpu variable, and
2589 *	the handler gets called with the interrupted CPU's instance of
2590 *	that variable.
2591 */
2592int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2593			 unsigned long flags, const char *devname,
2594			 void __percpu *dev_id)
2595{
2596	struct irqaction *action;
2597	struct irq_desc *desc;
2598	int retval;
2599
2600	if (!dev_id)
2601		return -EINVAL;
2602
2603	desc = irq_to_desc(irq);
2604	if (!desc || !irq_settings_can_request(desc) ||
2605	    !irq_settings_is_per_cpu_devid(desc))
2606		return -EINVAL;
2607
2608	if (flags && flags != IRQF_TIMER)
2609		return -EINVAL;
2610
2611	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2612	if (!action)
2613		return -ENOMEM;
2614
2615	action->handler = handler;
2616	action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2617	action->name = devname;
2618	action->percpu_dev_id = dev_id;
2619
2620	retval = irq_chip_pm_get(&desc->irq_data);
2621	if (retval < 0) {
2622		kfree(action);
2623		return retval;
2624	}
2625
2626	retval = __setup_irq(irq, desc, action);
2627
2628	if (retval) {
2629		irq_chip_pm_put(&desc->irq_data);
2630		kfree(action);
2631	}
2632
2633	return retval;
2634}
2635EXPORT_SYMBOL_GPL(__request_percpu_irq);
2636
2637/**
2638 *	request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2639 *	@irq: Interrupt line to allocate
2640 *	@handler: Function to be called when the IRQ occurs.
2641 *	@name: An ascii name for the claiming device
2642 *	@dev_id: A percpu cookie passed back to the handler function
2643 *
2644 *	This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2645 *	have to be setup on each CPU by calling prepare_percpu_nmi() before
2646 *	being enabled on the same CPU by using enable_percpu_nmi().
2647 *
2648 *	Dev_id must be globally unique. It is a per-cpu variable, and
2649 *	the handler gets called with the interrupted CPU's instance of
2650 *	that variable.
2651 *
2652 *	Interrupt lines requested for NMI delivering should have auto enabling
2653 *	setting disabled.
2654 *
2655 *	If the interrupt line cannot be used to deliver NMIs, function
2656 *	will fail returning a negative value.
2657 */
2658int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2659		       const char *name, void __percpu *dev_id)
2660{
2661	struct irqaction *action;
2662	struct irq_desc *desc;
2663	unsigned long flags;
2664	int retval;
2665
2666	if (!handler)
2667		return -EINVAL;
2668
2669	desc = irq_to_desc(irq);
2670
2671	if (!desc || !irq_settings_can_request(desc) ||
2672	    !irq_settings_is_per_cpu_devid(desc) ||
2673	    irq_settings_can_autoenable(desc) ||
2674	    !irq_supports_nmi(desc))
2675		return -EINVAL;
2676
2677	/* The line cannot already be NMI */
2678	if (desc->istate & IRQS_NMI)
2679		return -EINVAL;
2680
2681	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2682	if (!action)
2683		return -ENOMEM;
2684
2685	action->handler = handler;
2686	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2687		| IRQF_NOBALANCING;
2688	action->name = name;
2689	action->percpu_dev_id = dev_id;
2690
2691	retval = irq_chip_pm_get(&desc->irq_data);
2692	if (retval < 0)
2693		goto err_out;
2694
2695	retval = __setup_irq(irq, desc, action);
2696	if (retval)
2697		goto err_irq_setup;
2698
2699	raw_spin_lock_irqsave(&desc->lock, flags);
2700	desc->istate |= IRQS_NMI;
2701	raw_spin_unlock_irqrestore(&desc->lock, flags);
2702
2703	return 0;
2704
2705err_irq_setup:
2706	irq_chip_pm_put(&desc->irq_data);
2707err_out:
2708	kfree(action);
2709
2710	return retval;
2711}
2712
2713/**
2714 *	prepare_percpu_nmi - performs CPU local setup for NMI delivery
2715 *	@irq: Interrupt line to prepare for NMI delivery
2716 *
2717 *	This call prepares an interrupt line to deliver NMI on the current CPU,
2718 *	before that interrupt line gets enabled with enable_percpu_nmi().
2719 *
2720 *	As a CPU local operation, this should be called from non-preemptible
2721 *	context.
2722 *
2723 *	If the interrupt line cannot be used to deliver NMIs, function
2724 *	will fail returning a negative value.
2725 */
2726int prepare_percpu_nmi(unsigned int irq)
2727{
2728	unsigned long flags;
2729	struct irq_desc *desc;
2730	int ret = 0;
2731
2732	WARN_ON(preemptible());
2733
2734	desc = irq_get_desc_lock(irq, &flags,
2735				 IRQ_GET_DESC_CHECK_PERCPU);
2736	if (!desc)
2737		return -EINVAL;
2738
2739	if (WARN(!(desc->istate & IRQS_NMI),
2740		 KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2741		 irq)) {
2742		ret = -EINVAL;
2743		goto out;
2744	}
2745
2746	ret = irq_nmi_setup(desc);
2747	if (ret) {
2748		pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2749		goto out;
2750	}
2751
2752out:
2753	irq_put_desc_unlock(desc, flags);
2754	return ret;
2755}
2756
2757/**
2758 *	teardown_percpu_nmi - undoes NMI setup of IRQ line
2759 *	@irq: Interrupt line from which CPU local NMI configuration should be
2760 *	      removed
2761 *
2762 *	This call undoes the setup done by prepare_percpu_nmi().
2763 *
2764 *	IRQ line should not be enabled for the current CPU.
2765 *
2766 *	As a CPU local operation, this should be called from non-preemptible
2767 *	context.
2768 */
2769void teardown_percpu_nmi(unsigned int irq)
2770{
2771	unsigned long flags;
2772	struct irq_desc *desc;
2773
2774	WARN_ON(preemptible());
2775
2776	desc = irq_get_desc_lock(irq, &flags,
2777				 IRQ_GET_DESC_CHECK_PERCPU);
2778	if (!desc)
2779		return;
2780
2781	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2782		goto out;
2783
2784	irq_nmi_teardown(desc);
2785out:
2786	irq_put_desc_unlock(desc, flags);
2787}
2788
2789int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2790			    bool *state)
2791{
2792	struct irq_chip *chip;
2793	int err = -EINVAL;
2794
2795	do {
2796		chip = irq_data_get_irq_chip(data);
2797		if (WARN_ON_ONCE(!chip))
2798			return -ENODEV;
2799		if (chip->irq_get_irqchip_state)
2800			break;
2801#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2802		data = data->parent_data;
2803#else
2804		data = NULL;
2805#endif
2806	} while (data);
2807
2808	if (data)
2809		err = chip->irq_get_irqchip_state(data, which, state);
2810	return err;
2811}
2812
2813/**
2814 *	irq_get_irqchip_state - returns the irqchip state of a interrupt.
2815 *	@irq: Interrupt line that is forwarded to a VM
2816 *	@which: One of IRQCHIP_STATE_* the caller wants to know about
2817 *	@state: a pointer to a boolean where the state is to be stored
2818 *
2819 *	This call snapshots the internal irqchip state of an
2820 *	interrupt, returning into @state the bit corresponding to
2821 *	stage @which
2822 *
2823 *	This function should be called with preemption disabled if the
2824 *	interrupt controller has per-cpu registers.
2825 */
2826int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2827			  bool *state)
2828{
2829	struct irq_desc *desc;
2830	struct irq_data *data;
2831	unsigned long flags;
2832	int err = -EINVAL;
2833
2834	desc = irq_get_desc_buslock(irq, &flags, 0);
2835	if (!desc)
2836		return err;
2837
2838	data = irq_desc_get_irq_data(desc);
2839
2840	err = __irq_get_irqchip_state(data, which, state);
2841
2842	irq_put_desc_busunlock(desc, flags);
2843	return err;
2844}
2845EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2846
2847/**
2848 *	irq_set_irqchip_state - set the state of a forwarded interrupt.
2849 *	@irq: Interrupt line that is forwarded to a VM
2850 *	@which: State to be restored (one of IRQCHIP_STATE_*)
2851 *	@val: Value corresponding to @which
2852 *
2853 *	This call sets the internal irqchip state of an interrupt,
2854 *	depending on the value of @which.
2855 *
2856 *	This function should be called with migration disabled if the
2857 *	interrupt controller has per-cpu registers.
2858 */
2859int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2860			  bool val)
2861{
2862	struct irq_desc *desc;
2863	struct irq_data *data;
2864	struct irq_chip *chip;
2865	unsigned long flags;
2866	int err = -EINVAL;
2867
2868	desc = irq_get_desc_buslock(irq, &flags, 0);
2869	if (!desc)
2870		return err;
2871
2872	data = irq_desc_get_irq_data(desc);
2873
2874	do {
2875		chip = irq_data_get_irq_chip(data);
2876		if (WARN_ON_ONCE(!chip)) {
2877			err = -ENODEV;
2878			goto out_unlock;
2879		}
2880		if (chip->irq_set_irqchip_state)
2881			break;
2882#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2883		data = data->parent_data;
2884#else
2885		data = NULL;
2886#endif
2887	} while (data);
2888
2889	if (data)
2890		err = chip->irq_set_irqchip_state(data, which, val);
2891
2892out_unlock:
2893	irq_put_desc_busunlock(desc, flags);
2894	return err;
2895}
2896EXPORT_SYMBOL_GPL(irq_set_irqchip_state);
2897
2898/**
2899 * irq_has_action - Check whether an interrupt is requested
2900 * @irq:	The linux irq number
2901 *
2902 * Returns: A snapshot of the current state
2903 */
2904bool irq_has_action(unsigned int irq)
2905{
2906	bool res;
2907
2908	rcu_read_lock();
2909	res = irq_desc_has_action(irq_to_desc(irq));
2910	rcu_read_unlock();
2911	return res;
2912}
2913EXPORT_SYMBOL_GPL(irq_has_action);
2914
2915/**
2916 * irq_check_status_bit - Check whether bits in the irq descriptor status are set
2917 * @irq:	The linux irq number
2918 * @bitmask:	The bitmask to evaluate
2919 *
2920 * Returns: True if one of the bits in @bitmask is set
2921 */
2922bool irq_check_status_bit(unsigned int irq, unsigned int bitmask)
2923{
2924	struct irq_desc *desc;
2925	bool res = false;
2926
2927	rcu_read_lock();
2928	desc = irq_to_desc(irq);
2929	if (desc)
2930		res = !!(desc->status_use_accessors & bitmask);
2931	rcu_read_unlock();
2932	return res;
2933}
2934EXPORT_SYMBOL_GPL(irq_check_status_bit);
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
   4 * Copyright (C) 2005-2006 Thomas Gleixner
   5 *
   6 * This file contains driver APIs to the irq subsystem.
   7 */
   8
   9#define pr_fmt(fmt) "genirq: " fmt
  10
  11#include <linux/irq.h>
  12#include <linux/kthread.h>
  13#include <linux/module.h>
  14#include <linux/random.h>
  15#include <linux/interrupt.h>
  16#include <linux/irqdomain.h>
  17#include <linux/slab.h>
  18#include <linux/sched.h>
  19#include <linux/sched/rt.h>
  20#include <linux/sched/task.h>
  21#include <linux/sched/isolation.h>
  22#include <uapi/linux/sched/types.h>
  23#include <linux/task_work.h>
  24
  25#include "internals.h"
  26
  27#if defined(CONFIG_IRQ_FORCED_THREADING) && !defined(CONFIG_PREEMPT_RT)
  28__read_mostly bool force_irqthreads;
  29EXPORT_SYMBOL_GPL(force_irqthreads);
  30
  31static int __init setup_forced_irqthreads(char *arg)
  32{
  33	force_irqthreads = true;
  34	return 0;
  35}
  36early_param("threadirqs", setup_forced_irqthreads);
  37#endif
  38
  39static void __synchronize_hardirq(struct irq_desc *desc, bool sync_chip)
  40{
  41	struct irq_data *irqd = irq_desc_get_irq_data(desc);
  42	bool inprogress;
  43
  44	do {
  45		unsigned long flags;
  46
  47		/*
  48		 * Wait until we're out of the critical section.  This might
  49		 * give the wrong answer due to the lack of memory barriers.
  50		 */
  51		while (irqd_irq_inprogress(&desc->irq_data))
  52			cpu_relax();
  53
  54		/* Ok, that indicated we're done: double-check carefully. */
  55		raw_spin_lock_irqsave(&desc->lock, flags);
  56		inprogress = irqd_irq_inprogress(&desc->irq_data);
  57
  58		/*
  59		 * If requested and supported, check at the chip whether it
  60		 * is in flight at the hardware level, i.e. already pending
  61		 * in a CPU and waiting for service and acknowledge.
  62		 */
  63		if (!inprogress && sync_chip) {
  64			/*
  65			 * Ignore the return code. inprogress is only updated
  66			 * when the chip supports it.
  67			 */
  68			__irq_get_irqchip_state(irqd, IRQCHIP_STATE_ACTIVE,
  69						&inprogress);
  70		}
  71		raw_spin_unlock_irqrestore(&desc->lock, flags);
  72
  73		/* Oops, that failed? */
  74	} while (inprogress);
  75}
  76
  77/**
  78 *	synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
  79 *	@irq: interrupt number to wait for
  80 *
  81 *	This function waits for any pending hard IRQ handlers for this
  82 *	interrupt to complete before returning. If you use this
  83 *	function while holding a resource the IRQ handler may need you
  84 *	will deadlock. It does not take associated threaded handlers
  85 *	into account.
  86 *
  87 *	Do not use this for shutdown scenarios where you must be sure
  88 *	that all parts (hardirq and threaded handler) have completed.
  89 *
  90 *	Returns: false if a threaded handler is active.
  91 *
  92 *	This function may be called - with care - from IRQ context.
  93 *
  94 *	It does not check whether there is an interrupt in flight at the
  95 *	hardware level, but not serviced yet, as this might deadlock when
  96 *	called with interrupts disabled and the target CPU of the interrupt
  97 *	is the current CPU.
  98 */
  99bool synchronize_hardirq(unsigned int irq)
 100{
 101	struct irq_desc *desc = irq_to_desc(irq);
 102
 103	if (desc) {
 104		__synchronize_hardirq(desc, false);
 105		return !atomic_read(&desc->threads_active);
 106	}
 107
 108	return true;
 109}
 110EXPORT_SYMBOL(synchronize_hardirq);
 111
 112/**
 113 *	synchronize_irq - wait for pending IRQ handlers (on other CPUs)
 114 *	@irq: interrupt number to wait for
 115 *
 116 *	This function waits for any pending IRQ handlers for this interrupt
 117 *	to complete before returning. If you use this function while
 118 *	holding a resource the IRQ handler may need you will deadlock.
 119 *
 120 *	Can only be called from preemptible code as it might sleep when
 121 *	an interrupt thread is associated to @irq.
 122 *
 123 *	It optionally makes sure (when the irq chip supports that method)
 124 *	that the interrupt is not pending in any CPU and waiting for
 125 *	service.
 126 */
 127void synchronize_irq(unsigned int irq)
 128{
 129	struct irq_desc *desc = irq_to_desc(irq);
 130
 131	if (desc) {
 132		__synchronize_hardirq(desc, true);
 133		/*
 134		 * We made sure that no hardirq handler is
 135		 * running. Now verify that no threaded handlers are
 136		 * active.
 137		 */
 138		wait_event(desc->wait_for_threads,
 139			   !atomic_read(&desc->threads_active));
 140	}
 141}
 142EXPORT_SYMBOL(synchronize_irq);
 143
 144#ifdef CONFIG_SMP
 145cpumask_var_t irq_default_affinity;
 146
 147static bool __irq_can_set_affinity(struct irq_desc *desc)
 148{
 149	if (!desc || !irqd_can_balance(&desc->irq_data) ||
 150	    !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
 151		return false;
 152	return true;
 153}
 154
 155/**
 156 *	irq_can_set_affinity - Check if the affinity of a given irq can be set
 157 *	@irq:		Interrupt to check
 158 *
 159 */
 160int irq_can_set_affinity(unsigned int irq)
 161{
 162	return __irq_can_set_affinity(irq_to_desc(irq));
 163}
 164
 165/**
 166 * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
 167 * @irq:	Interrupt to check
 168 *
 169 * Like irq_can_set_affinity() above, but additionally checks for the
 170 * AFFINITY_MANAGED flag.
 171 */
 172bool irq_can_set_affinity_usr(unsigned int irq)
 173{
 174	struct irq_desc *desc = irq_to_desc(irq);
 175
 176	return __irq_can_set_affinity(desc) &&
 177		!irqd_affinity_is_managed(&desc->irq_data);
 178}
 179
 180/**
 181 *	irq_set_thread_affinity - Notify irq threads to adjust affinity
 182 *	@desc:		irq descriptor which has affitnity changed
 183 *
 184 *	We just set IRQTF_AFFINITY and delegate the affinity setting
 185 *	to the interrupt thread itself. We can not call
 186 *	set_cpus_allowed_ptr() here as we hold desc->lock and this
 187 *	code can be called from hard interrupt context.
 188 */
 189void irq_set_thread_affinity(struct irq_desc *desc)
 190{
 191	struct irqaction *action;
 192
 193	for_each_action_of_desc(desc, action)
 194		if (action->thread)
 195			set_bit(IRQTF_AFFINITY, &action->thread_flags);
 196}
 197
 198#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
 199static void irq_validate_effective_affinity(struct irq_data *data)
 200{
 201	const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
 202	struct irq_chip *chip = irq_data_get_irq_chip(data);
 203
 204	if (!cpumask_empty(m))
 205		return;
 206	pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
 207		     chip->name, data->irq);
 208}
 209
 210static inline void irq_init_effective_affinity(struct irq_data *data,
 211					       const struct cpumask *mask)
 212{
 213	cpumask_copy(irq_data_get_effective_affinity_mask(data), mask);
 214}
 215#else
 216static inline void irq_validate_effective_affinity(struct irq_data *data) { }
 217static inline void irq_init_effective_affinity(struct irq_data *data,
 218					       const struct cpumask *mask) { }
 219#endif
 220
 221int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
 222			bool force)
 223{
 224	struct irq_desc *desc = irq_data_to_desc(data);
 225	struct irq_chip *chip = irq_data_get_irq_chip(data);
 
 226	int ret;
 227
 
 
 
 228	if (!chip || !chip->irq_set_affinity)
 229		return -EINVAL;
 230
 
 231	/*
 232	 * If this is a managed interrupt and housekeeping is enabled on
 233	 * it check whether the requested affinity mask intersects with
 234	 * a housekeeping CPU. If so, then remove the isolated CPUs from
 235	 * the mask and just keep the housekeeping CPU(s). This prevents
 236	 * the affinity setter from routing the interrupt to an isolated
 237	 * CPU to avoid that I/O submitted from a housekeeping CPU causes
 238	 * interrupts on an isolated one.
 239	 *
 240	 * If the masks do not intersect or include online CPU(s) then
 241	 * keep the requested mask. The isolated target CPUs are only
 242	 * receiving interrupts when the I/O operation was submitted
 243	 * directly from them.
 244	 *
 245	 * If all housekeeping CPUs in the affinity mask are offline, the
 246	 * interrupt will be migrated by the CPU hotplug code once a
 247	 * housekeeping CPU which belongs to the affinity mask comes
 248	 * online.
 249	 */
 250	if (irqd_affinity_is_managed(data) &&
 251	    housekeeping_enabled(HK_FLAG_MANAGED_IRQ)) {
 252		const struct cpumask *hk_mask, *prog_mask;
 253
 254		static DEFINE_RAW_SPINLOCK(tmp_mask_lock);
 255		static struct cpumask tmp_mask;
 256
 257		hk_mask = housekeeping_cpumask(HK_FLAG_MANAGED_IRQ);
 258
 259		raw_spin_lock(&tmp_mask_lock);
 260		cpumask_and(&tmp_mask, mask, hk_mask);
 261		if (!cpumask_intersects(&tmp_mask, cpu_online_mask))
 262			prog_mask = mask;
 263		else
 264			prog_mask = &tmp_mask;
 265		ret = chip->irq_set_affinity(data, prog_mask, force);
 266		raw_spin_unlock(&tmp_mask_lock);
 267	} else {
 
 
 
 
 
 
 
 
 
 
 
 
 268		ret = chip->irq_set_affinity(data, mask, force);
 269	}
 
 
 
 
 270	switch (ret) {
 271	case IRQ_SET_MASK_OK:
 272	case IRQ_SET_MASK_OK_DONE:
 273		cpumask_copy(desc->irq_common_data.affinity, mask);
 274		fallthrough;
 275	case IRQ_SET_MASK_OK_NOCOPY:
 276		irq_validate_effective_affinity(data);
 277		irq_set_thread_affinity(desc);
 278		ret = 0;
 279	}
 280
 281	return ret;
 282}
 283
 284#ifdef CONFIG_GENERIC_PENDING_IRQ
 285static inline int irq_set_affinity_pending(struct irq_data *data,
 286					   const struct cpumask *dest)
 287{
 288	struct irq_desc *desc = irq_data_to_desc(data);
 289
 290	irqd_set_move_pending(data);
 291	irq_copy_pending(desc, dest);
 292	return 0;
 293}
 294#else
 295static inline int irq_set_affinity_pending(struct irq_data *data,
 296					   const struct cpumask *dest)
 297{
 298	return -EBUSY;
 299}
 300#endif
 301
 302static int irq_try_set_affinity(struct irq_data *data,
 303				const struct cpumask *dest, bool force)
 304{
 305	int ret = irq_do_set_affinity(data, dest, force);
 306
 307	/*
 308	 * In case that the underlying vector management is busy and the
 309	 * architecture supports the generic pending mechanism then utilize
 310	 * this to avoid returning an error to user space.
 311	 */
 312	if (ret == -EBUSY && !force)
 313		ret = irq_set_affinity_pending(data, dest);
 314	return ret;
 315}
 316
 317static bool irq_set_affinity_deactivated(struct irq_data *data,
 318					 const struct cpumask *mask, bool force)
 319{
 320	struct irq_desc *desc = irq_data_to_desc(data);
 321
 322	/*
 323	 * Handle irq chips which can handle affinity only in activated
 324	 * state correctly
 325	 *
 326	 * If the interrupt is not yet activated, just store the affinity
 327	 * mask and do not call the chip driver at all. On activation the
 328	 * driver has to make sure anyway that the interrupt is in a
 329	 * useable state so startup works.
 330	 */
 331	if (!IS_ENABLED(CONFIG_IRQ_DOMAIN_HIERARCHY) ||
 332	    irqd_is_activated(data) || !irqd_affinity_on_activate(data))
 333		return false;
 334
 335	cpumask_copy(desc->irq_common_data.affinity, mask);
 336	irq_init_effective_affinity(data, mask);
 337	irqd_set(data, IRQD_AFFINITY_SET);
 338	return true;
 339}
 340
 341int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 342			    bool force)
 343{
 344	struct irq_chip *chip = irq_data_get_irq_chip(data);
 345	struct irq_desc *desc = irq_data_to_desc(data);
 346	int ret = 0;
 347
 348	if (!chip || !chip->irq_set_affinity)
 349		return -EINVAL;
 350
 351	if (irq_set_affinity_deactivated(data, mask, force))
 352		return 0;
 353
 354	if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
 355		ret = irq_try_set_affinity(data, mask, force);
 356	} else {
 357		irqd_set_move_pending(data);
 358		irq_copy_pending(desc, mask);
 359	}
 360
 361	if (desc->affinity_notify) {
 362		kref_get(&desc->affinity_notify->kref);
 363		if (!schedule_work(&desc->affinity_notify->work)) {
 364			/* Work was already scheduled, drop our extra ref */
 365			kref_put(&desc->affinity_notify->kref,
 366				 desc->affinity_notify->release);
 367		}
 368	}
 369	irqd_set(data, IRQD_AFFINITY_SET);
 370
 371	return ret;
 372}
 373
 374int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 375{
 376	struct irq_desc *desc = irq_to_desc(irq);
 377	unsigned long flags;
 378	int ret;
 379
 380	if (!desc)
 381		return -EINVAL;
 382
 383	raw_spin_lock_irqsave(&desc->lock, flags);
 384	ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
 385	raw_spin_unlock_irqrestore(&desc->lock, flags);
 386	return ret;
 387}
 388
 389int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 390{
 391	unsigned long flags;
 392	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 393
 394	if (!desc)
 395		return -EINVAL;
 396	desc->affinity_hint = m;
 397	irq_put_desc_unlock(desc, flags);
 398	/* set the initial affinity to prevent every interrupt being on CPU0 */
 399	if (m)
 400		__irq_set_affinity(irq, m, false);
 401	return 0;
 402}
 403EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
 404
 405static void irq_affinity_notify(struct work_struct *work)
 406{
 407	struct irq_affinity_notify *notify =
 408		container_of(work, struct irq_affinity_notify, work);
 409	struct irq_desc *desc = irq_to_desc(notify->irq);
 410	cpumask_var_t cpumask;
 411	unsigned long flags;
 412
 413	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
 414		goto out;
 415
 416	raw_spin_lock_irqsave(&desc->lock, flags);
 417	if (irq_move_pending(&desc->irq_data))
 418		irq_get_pending(cpumask, desc);
 419	else
 420		cpumask_copy(cpumask, desc->irq_common_data.affinity);
 421	raw_spin_unlock_irqrestore(&desc->lock, flags);
 422
 423	notify->notify(notify, cpumask);
 424
 425	free_cpumask_var(cpumask);
 426out:
 427	kref_put(&notify->kref, notify->release);
 428}
 429
 430/**
 431 *	irq_set_affinity_notifier - control notification of IRQ affinity changes
 432 *	@irq:		Interrupt for which to enable/disable notification
 433 *	@notify:	Context for notification, or %NULL to disable
 434 *			notification.  Function pointers must be initialised;
 435 *			the other fields will be initialised by this function.
 436 *
 437 *	Must be called in process context.  Notification may only be enabled
 438 *	after the IRQ is allocated and must be disabled before the IRQ is
 439 *	freed using free_irq().
 440 */
 441int
 442irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 443{
 444	struct irq_desc *desc = irq_to_desc(irq);
 445	struct irq_affinity_notify *old_notify;
 446	unsigned long flags;
 447
 448	/* The release function is promised process context */
 449	might_sleep();
 450
 451	if (!desc || desc->istate & IRQS_NMI)
 452		return -EINVAL;
 453
 454	/* Complete initialisation of *notify */
 455	if (notify) {
 456		notify->irq = irq;
 457		kref_init(&notify->kref);
 458		INIT_WORK(&notify->work, irq_affinity_notify);
 459	}
 460
 461	raw_spin_lock_irqsave(&desc->lock, flags);
 462	old_notify = desc->affinity_notify;
 463	desc->affinity_notify = notify;
 464	raw_spin_unlock_irqrestore(&desc->lock, flags);
 465
 466	if (old_notify) {
 467		if (cancel_work_sync(&old_notify->work)) {
 468			/* Pending work had a ref, put that one too */
 469			kref_put(&old_notify->kref, old_notify->release);
 470		}
 471		kref_put(&old_notify->kref, old_notify->release);
 472	}
 473
 474	return 0;
 475}
 476EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
 477
 478#ifndef CONFIG_AUTO_IRQ_AFFINITY
 479/*
 480 * Generic version of the affinity autoselector.
 481 */
 482int irq_setup_affinity(struct irq_desc *desc)
 483{
 484	struct cpumask *set = irq_default_affinity;
 485	int ret, node = irq_desc_get_node(desc);
 486	static DEFINE_RAW_SPINLOCK(mask_lock);
 487	static struct cpumask mask;
 488
 489	/* Excludes PER_CPU and NO_BALANCE interrupts */
 490	if (!__irq_can_set_affinity(desc))
 491		return 0;
 492
 493	raw_spin_lock(&mask_lock);
 494	/*
 495	 * Preserve the managed affinity setting and a userspace affinity
 496	 * setup, but make sure that one of the targets is online.
 497	 */
 498	if (irqd_affinity_is_managed(&desc->irq_data) ||
 499	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 500		if (cpumask_intersects(desc->irq_common_data.affinity,
 501				       cpu_online_mask))
 502			set = desc->irq_common_data.affinity;
 503		else
 504			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
 505	}
 506
 507	cpumask_and(&mask, cpu_online_mask, set);
 508	if (cpumask_empty(&mask))
 509		cpumask_copy(&mask, cpu_online_mask);
 510
 511	if (node != NUMA_NO_NODE) {
 512		const struct cpumask *nodemask = cpumask_of_node(node);
 513
 514		/* make sure at least one of the cpus in nodemask is online */
 515		if (cpumask_intersects(&mask, nodemask))
 516			cpumask_and(&mask, &mask, nodemask);
 517	}
 518	ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
 519	raw_spin_unlock(&mask_lock);
 520	return ret;
 521}
 522#else
 523/* Wrapper for ALPHA specific affinity selector magic */
 524int irq_setup_affinity(struct irq_desc *desc)
 525{
 526	return irq_select_affinity(irq_desc_get_irq(desc));
 527}
 528#endif /* CONFIG_AUTO_IRQ_AFFINITY */
 529#endif /* CONFIG_SMP */
 530
 531
 532/**
 533 *	irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
 534 *	@irq: interrupt number to set affinity
 535 *	@vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
 536 *	            specific data for percpu_devid interrupts
 537 *
 538 *	This function uses the vCPU specific data to set the vCPU
 539 *	affinity for an irq. The vCPU specific data is passed from
 540 *	outside, such as KVM. One example code path is as below:
 541 *	KVM -> IOMMU -> irq_set_vcpu_affinity().
 542 */
 543int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
 544{
 545	unsigned long flags;
 546	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 547	struct irq_data *data;
 548	struct irq_chip *chip;
 549	int ret = -ENOSYS;
 550
 551	if (!desc)
 552		return -EINVAL;
 553
 554	data = irq_desc_get_irq_data(desc);
 555	do {
 556		chip = irq_data_get_irq_chip(data);
 557		if (chip && chip->irq_set_vcpu_affinity)
 558			break;
 559#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
 560		data = data->parent_data;
 561#else
 562		data = NULL;
 563#endif
 564	} while (data);
 565
 566	if (data)
 567		ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
 568	irq_put_desc_unlock(desc, flags);
 569
 570	return ret;
 571}
 572EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
 573
 574void __disable_irq(struct irq_desc *desc)
 575{
 576	if (!desc->depth++)
 577		irq_disable(desc);
 578}
 579
 580static int __disable_irq_nosync(unsigned int irq)
 581{
 582	unsigned long flags;
 583	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 584
 585	if (!desc)
 586		return -EINVAL;
 587	__disable_irq(desc);
 588	irq_put_desc_busunlock(desc, flags);
 589	return 0;
 590}
 591
 592/**
 593 *	disable_irq_nosync - disable an irq without waiting
 594 *	@irq: Interrupt to disable
 595 *
 596 *	Disable the selected interrupt line.  Disables and Enables are
 597 *	nested.
 598 *	Unlike disable_irq(), this function does not ensure existing
 599 *	instances of the IRQ handler have completed before returning.
 600 *
 601 *	This function may be called from IRQ context.
 602 */
 603void disable_irq_nosync(unsigned int irq)
 604{
 605	__disable_irq_nosync(irq);
 606}
 607EXPORT_SYMBOL(disable_irq_nosync);
 608
 609/**
 610 *	disable_irq - disable an irq and wait for completion
 611 *	@irq: Interrupt to disable
 612 *
 613 *	Disable the selected interrupt line.  Enables and Disables are
 614 *	nested.
 615 *	This function waits for any pending IRQ handlers for this interrupt
 616 *	to complete before returning. If you use this function while
 617 *	holding a resource the IRQ handler may need you will deadlock.
 618 *
 619 *	This function may be called - with care - from IRQ context.
 620 */
 621void disable_irq(unsigned int irq)
 622{
 623	if (!__disable_irq_nosync(irq))
 624		synchronize_irq(irq);
 625}
 626EXPORT_SYMBOL(disable_irq);
 627
 628/**
 629 *	disable_hardirq - disables an irq and waits for hardirq completion
 630 *	@irq: Interrupt to disable
 631 *
 632 *	Disable the selected interrupt line.  Enables and Disables are
 633 *	nested.
 634 *	This function waits for any pending hard IRQ handlers for this
 635 *	interrupt to complete before returning. If you use this function while
 636 *	holding a resource the hard IRQ handler may need you will deadlock.
 637 *
 638 *	When used to optimistically disable an interrupt from atomic context
 639 *	the return value must be checked.
 640 *
 641 *	Returns: false if a threaded handler is active.
 642 *
 643 *	This function may be called - with care - from IRQ context.
 644 */
 645bool disable_hardirq(unsigned int irq)
 646{
 647	if (!__disable_irq_nosync(irq))
 648		return synchronize_hardirq(irq);
 649
 650	return false;
 651}
 652EXPORT_SYMBOL_GPL(disable_hardirq);
 653
 654/**
 655 *	disable_nmi_nosync - disable an nmi without waiting
 656 *	@irq: Interrupt to disable
 657 *
 658 *	Disable the selected interrupt line. Disables and enables are
 659 *	nested.
 660 *	The interrupt to disable must have been requested through request_nmi.
 661 *	Unlike disable_nmi(), this function does not ensure existing
 662 *	instances of the IRQ handler have completed before returning.
 663 */
 664void disable_nmi_nosync(unsigned int irq)
 665{
 666	disable_irq_nosync(irq);
 667}
 668
 669void __enable_irq(struct irq_desc *desc)
 670{
 671	switch (desc->depth) {
 672	case 0:
 673 err_out:
 674		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
 675		     irq_desc_get_irq(desc));
 676		break;
 677	case 1: {
 678		if (desc->istate & IRQS_SUSPENDED)
 679			goto err_out;
 680		/* Prevent probing on this irq: */
 681		irq_settings_set_noprobe(desc);
 682		/*
 683		 * Call irq_startup() not irq_enable() here because the
 684		 * interrupt might be marked NOAUTOEN. So irq_startup()
 685		 * needs to be invoked when it gets enabled the first
 686		 * time. If it was already started up, then irq_startup()
 687		 * will invoke irq_enable() under the hood.
 688		 */
 689		irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
 690		break;
 691	}
 692	default:
 693		desc->depth--;
 694	}
 695}
 696
 697/**
 698 *	enable_irq - enable handling of an irq
 699 *	@irq: Interrupt to enable
 700 *
 701 *	Undoes the effect of one call to disable_irq().  If this
 702 *	matches the last disable, processing of interrupts on this
 703 *	IRQ line is re-enabled.
 704 *
 705 *	This function may be called from IRQ context only when
 706 *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
 707 */
 708void enable_irq(unsigned int irq)
 709{
 710	unsigned long flags;
 711	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 712
 713	if (!desc)
 714		return;
 715	if (WARN(!desc->irq_data.chip,
 716		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
 717		goto out;
 718
 719	__enable_irq(desc);
 720out:
 721	irq_put_desc_busunlock(desc, flags);
 722}
 723EXPORT_SYMBOL(enable_irq);
 724
 725/**
 726 *	enable_nmi - enable handling of an nmi
 727 *	@irq: Interrupt to enable
 728 *
 729 *	The interrupt to enable must have been requested through request_nmi.
 730 *	Undoes the effect of one call to disable_nmi(). If this
 731 *	matches the last disable, processing of interrupts on this
 732 *	IRQ line is re-enabled.
 733 */
 734void enable_nmi(unsigned int irq)
 735{
 736	enable_irq(irq);
 737}
 738
 739static int set_irq_wake_real(unsigned int irq, unsigned int on)
 740{
 741	struct irq_desc *desc = irq_to_desc(irq);
 742	int ret = -ENXIO;
 743
 744	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
 745		return 0;
 746
 747	if (desc->irq_data.chip->irq_set_wake)
 748		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
 749
 750	return ret;
 751}
 752
 753/**
 754 *	irq_set_irq_wake - control irq power management wakeup
 755 *	@irq:	interrupt to control
 756 *	@on:	enable/disable power management wakeup
 757 *
 758 *	Enable/disable power management wakeup mode, which is
 759 *	disabled by default.  Enables and disables must match,
 760 *	just as they match for non-wakeup mode support.
 761 *
 762 *	Wakeup mode lets this IRQ wake the system from sleep
 763 *	states like "suspend to RAM".
 764 *
 765 *	Note: irq enable/disable state is completely orthogonal
 766 *	to the enable/disable state of irq wake. An irq can be
 767 *	disabled with disable_irq() and still wake the system as
 768 *	long as the irq has wake enabled. If this does not hold,
 769 *	then the underlying irq chip and the related driver need
 770 *	to be investigated.
 771 */
 772int irq_set_irq_wake(unsigned int irq, unsigned int on)
 773{
 774	unsigned long flags;
 775	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 776	int ret = 0;
 777
 778	if (!desc)
 779		return -EINVAL;
 780
 781	/* Don't use NMIs as wake up interrupts please */
 782	if (desc->istate & IRQS_NMI) {
 783		ret = -EINVAL;
 784		goto out_unlock;
 785	}
 786
 787	/* wakeup-capable irqs can be shared between drivers that
 788	 * don't need to have the same sleep mode behaviors.
 789	 */
 790	if (on) {
 791		if (desc->wake_depth++ == 0) {
 792			ret = set_irq_wake_real(irq, on);
 793			if (ret)
 794				desc->wake_depth = 0;
 795			else
 796				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
 797		}
 798	} else {
 799		if (desc->wake_depth == 0) {
 800			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
 801		} else if (--desc->wake_depth == 0) {
 802			ret = set_irq_wake_real(irq, on);
 803			if (ret)
 804				desc->wake_depth = 1;
 805			else
 806				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
 807		}
 808	}
 809
 810out_unlock:
 811	irq_put_desc_busunlock(desc, flags);
 812	return ret;
 813}
 814EXPORT_SYMBOL(irq_set_irq_wake);
 815
 816/*
 817 * Internal function that tells the architecture code whether a
 818 * particular irq has been exclusively allocated or is available
 819 * for driver use.
 820 */
 821int can_request_irq(unsigned int irq, unsigned long irqflags)
 822{
 823	unsigned long flags;
 824	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 825	int canrequest = 0;
 826
 827	if (!desc)
 828		return 0;
 829
 830	if (irq_settings_can_request(desc)) {
 831		if (!desc->action ||
 832		    irqflags & desc->action->flags & IRQF_SHARED)
 833			canrequest = 1;
 834	}
 835	irq_put_desc_unlock(desc, flags);
 836	return canrequest;
 837}
 838
 839int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
 840{
 841	struct irq_chip *chip = desc->irq_data.chip;
 842	int ret, unmask = 0;
 843
 844	if (!chip || !chip->irq_set_type) {
 845		/*
 846		 * IRQF_TRIGGER_* but the PIC does not support multiple
 847		 * flow-types?
 848		 */
 849		pr_debug("No set_type function for IRQ %d (%s)\n",
 850			 irq_desc_get_irq(desc),
 851			 chip ? (chip->name ? : "unknown") : "unknown");
 852		return 0;
 853	}
 854
 855	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
 856		if (!irqd_irq_masked(&desc->irq_data))
 857			mask_irq(desc);
 858		if (!irqd_irq_disabled(&desc->irq_data))
 859			unmask = 1;
 860	}
 861
 862	/* Mask all flags except trigger mode */
 863	flags &= IRQ_TYPE_SENSE_MASK;
 864	ret = chip->irq_set_type(&desc->irq_data, flags);
 865
 866	switch (ret) {
 867	case IRQ_SET_MASK_OK:
 868	case IRQ_SET_MASK_OK_DONE:
 869		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
 870		irqd_set(&desc->irq_data, flags);
 871		fallthrough;
 872
 873	case IRQ_SET_MASK_OK_NOCOPY:
 874		flags = irqd_get_trigger_type(&desc->irq_data);
 875		irq_settings_set_trigger_mask(desc, flags);
 876		irqd_clear(&desc->irq_data, IRQD_LEVEL);
 877		irq_settings_clr_level(desc);
 878		if (flags & IRQ_TYPE_LEVEL_MASK) {
 879			irq_settings_set_level(desc);
 880			irqd_set(&desc->irq_data, IRQD_LEVEL);
 881		}
 882
 883		ret = 0;
 884		break;
 885	default:
 886		pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
 887		       flags, irq_desc_get_irq(desc), chip->irq_set_type);
 888	}
 889	if (unmask)
 890		unmask_irq(desc);
 891	return ret;
 892}
 893
 894#ifdef CONFIG_HARDIRQS_SW_RESEND
 895int irq_set_parent(int irq, int parent_irq)
 896{
 897	unsigned long flags;
 898	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 899
 900	if (!desc)
 901		return -EINVAL;
 902
 903	desc->parent_irq = parent_irq;
 904
 905	irq_put_desc_unlock(desc, flags);
 906	return 0;
 907}
 908EXPORT_SYMBOL_GPL(irq_set_parent);
 909#endif
 910
 911/*
 912 * Default primary interrupt handler for threaded interrupts. Is
 913 * assigned as primary handler when request_threaded_irq is called
 914 * with handler == NULL. Useful for oneshot interrupts.
 915 */
 916static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
 917{
 918	return IRQ_WAKE_THREAD;
 919}
 920
 921/*
 922 * Primary handler for nested threaded interrupts. Should never be
 923 * called.
 924 */
 925static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
 926{
 927	WARN(1, "Primary handler called for nested irq %d\n", irq);
 928	return IRQ_NONE;
 929}
 930
 931static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
 932{
 933	WARN(1, "Secondary action handler called for irq %d\n", irq);
 934	return IRQ_NONE;
 935}
 936
 937static int irq_wait_for_interrupt(struct irqaction *action)
 938{
 939	for (;;) {
 940		set_current_state(TASK_INTERRUPTIBLE);
 941
 942		if (kthread_should_stop()) {
 943			/* may need to run one last time */
 944			if (test_and_clear_bit(IRQTF_RUNTHREAD,
 945					       &action->thread_flags)) {
 946				__set_current_state(TASK_RUNNING);
 947				return 0;
 948			}
 949			__set_current_state(TASK_RUNNING);
 950			return -1;
 951		}
 952
 953		if (test_and_clear_bit(IRQTF_RUNTHREAD,
 954				       &action->thread_flags)) {
 955			__set_current_state(TASK_RUNNING);
 956			return 0;
 957		}
 958		schedule();
 959	}
 960}
 961
 962/*
 963 * Oneshot interrupts keep the irq line masked until the threaded
 964 * handler finished. unmask if the interrupt has not been disabled and
 965 * is marked MASKED.
 966 */
 967static void irq_finalize_oneshot(struct irq_desc *desc,
 968				 struct irqaction *action)
 969{
 970	if (!(desc->istate & IRQS_ONESHOT) ||
 971	    action->handler == irq_forced_secondary_handler)
 972		return;
 973again:
 974	chip_bus_lock(desc);
 975	raw_spin_lock_irq(&desc->lock);
 976
 977	/*
 978	 * Implausible though it may be we need to protect us against
 979	 * the following scenario:
 980	 *
 981	 * The thread is faster done than the hard interrupt handler
 982	 * on the other CPU. If we unmask the irq line then the
 983	 * interrupt can come in again and masks the line, leaves due
 984	 * to IRQS_INPROGRESS and the irq line is masked forever.
 985	 *
 986	 * This also serializes the state of shared oneshot handlers
 987	 * versus "desc->threads_onehsot |= action->thread_mask;" in
 988	 * irq_wake_thread(). See the comment there which explains the
 989	 * serialization.
 990	 */
 991	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
 992		raw_spin_unlock_irq(&desc->lock);
 993		chip_bus_sync_unlock(desc);
 994		cpu_relax();
 995		goto again;
 996	}
 997
 998	/*
 999	 * Now check again, whether the thread should run. Otherwise
1000	 * we would clear the threads_oneshot bit of this thread which
1001	 * was just set.
1002	 */
1003	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1004		goto out_unlock;
1005
1006	desc->threads_oneshot &= ~action->thread_mask;
1007
1008	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
1009	    irqd_irq_masked(&desc->irq_data))
1010		unmask_threaded_irq(desc);
1011
1012out_unlock:
1013	raw_spin_unlock_irq(&desc->lock);
1014	chip_bus_sync_unlock(desc);
1015}
1016
1017#ifdef CONFIG_SMP
1018/*
1019 * Check whether we need to change the affinity of the interrupt thread.
1020 */
1021static void
1022irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
1023{
1024	cpumask_var_t mask;
1025	bool valid = true;
1026
1027	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
1028		return;
1029
1030	/*
1031	 * In case we are out of memory we set IRQTF_AFFINITY again and
1032	 * try again next time
1033	 */
1034	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1035		set_bit(IRQTF_AFFINITY, &action->thread_flags);
1036		return;
1037	}
1038
1039	raw_spin_lock_irq(&desc->lock);
1040	/*
1041	 * This code is triggered unconditionally. Check the affinity
1042	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
1043	 */
1044	if (cpumask_available(desc->irq_common_data.affinity)) {
1045		const struct cpumask *m;
1046
1047		m = irq_data_get_effective_affinity_mask(&desc->irq_data);
1048		cpumask_copy(mask, m);
1049	} else {
1050		valid = false;
1051	}
1052	raw_spin_unlock_irq(&desc->lock);
1053
1054	if (valid)
1055		set_cpus_allowed_ptr(current, mask);
1056	free_cpumask_var(mask);
1057}
1058#else
1059static inline void
1060irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
1061#endif
1062
1063/*
1064 * Interrupts which are not explicitly requested as threaded
1065 * interrupts rely on the implicit bh/preempt disable of the hard irq
1066 * context. So we need to disable bh here to avoid deadlocks and other
1067 * side effects.
1068 */
1069static irqreturn_t
1070irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
1071{
1072	irqreturn_t ret;
1073
1074	local_bh_disable();
 
 
1075	ret = action->thread_fn(action->irq, action->dev_id);
1076	if (ret == IRQ_HANDLED)
1077		atomic_inc(&desc->threads_handled);
1078
1079	irq_finalize_oneshot(desc, action);
 
 
1080	local_bh_enable();
1081	return ret;
1082}
1083
1084/*
1085 * Interrupts explicitly requested as threaded interrupts want to be
1086 * preemtible - many of them need to sleep and wait for slow busses to
1087 * complete.
1088 */
1089static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1090		struct irqaction *action)
1091{
1092	irqreturn_t ret;
1093
1094	ret = action->thread_fn(action->irq, action->dev_id);
1095	if (ret == IRQ_HANDLED)
1096		atomic_inc(&desc->threads_handled);
1097
1098	irq_finalize_oneshot(desc, action);
1099	return ret;
1100}
1101
1102static void wake_threads_waitq(struct irq_desc *desc)
1103{
1104	if (atomic_dec_and_test(&desc->threads_active))
1105		wake_up(&desc->wait_for_threads);
1106}
1107
1108static void irq_thread_dtor(struct callback_head *unused)
1109{
1110	struct task_struct *tsk = current;
1111	struct irq_desc *desc;
1112	struct irqaction *action;
1113
1114	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1115		return;
1116
1117	action = kthread_data(tsk);
1118
1119	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1120	       tsk->comm, tsk->pid, action->irq);
1121
1122
1123	desc = irq_to_desc(action->irq);
1124	/*
1125	 * If IRQTF_RUNTHREAD is set, we need to decrement
1126	 * desc->threads_active and wake possible waiters.
1127	 */
1128	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1129		wake_threads_waitq(desc);
1130
1131	/* Prevent a stale desc->threads_oneshot */
1132	irq_finalize_oneshot(desc, action);
1133}
1134
1135static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1136{
1137	struct irqaction *secondary = action->secondary;
1138
1139	if (WARN_ON_ONCE(!secondary))
1140		return;
1141
1142	raw_spin_lock_irq(&desc->lock);
1143	__irq_wake_thread(desc, secondary);
1144	raw_spin_unlock_irq(&desc->lock);
1145}
1146
1147/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1148 * Interrupt handler thread
1149 */
1150static int irq_thread(void *data)
1151{
1152	struct callback_head on_exit_work;
1153	struct irqaction *action = data;
1154	struct irq_desc *desc = irq_to_desc(action->irq);
1155	irqreturn_t (*handler_fn)(struct irq_desc *desc,
1156			struct irqaction *action);
1157
1158	if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
1159					&action->thread_flags))
 
 
 
 
1160		handler_fn = irq_forced_thread_fn;
1161	else
1162		handler_fn = irq_thread_fn;
1163
1164	init_task_work(&on_exit_work, irq_thread_dtor);
1165	task_work_add(current, &on_exit_work, false);
1166
1167	irq_thread_check_affinity(desc, action);
1168
1169	while (!irq_wait_for_interrupt(action)) {
1170		irqreturn_t action_ret;
1171
1172		irq_thread_check_affinity(desc, action);
1173
1174		action_ret = handler_fn(desc, action);
1175		if (action_ret == IRQ_WAKE_THREAD)
1176			irq_wake_secondary(desc, action);
1177
1178		wake_threads_waitq(desc);
1179	}
1180
1181	/*
1182	 * This is the regular exit path. __free_irq() is stopping the
1183	 * thread via kthread_stop() after calling
1184	 * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1185	 * oneshot mask bit can be set.
1186	 */
1187	task_work_cancel(current, irq_thread_dtor);
1188	return 0;
1189}
1190
1191/**
1192 *	irq_wake_thread - wake the irq thread for the action identified by dev_id
1193 *	@irq:		Interrupt line
1194 *	@dev_id:	Device identity for which the thread should be woken
1195 *
1196 */
1197void irq_wake_thread(unsigned int irq, void *dev_id)
1198{
1199	struct irq_desc *desc = irq_to_desc(irq);
1200	struct irqaction *action;
1201	unsigned long flags;
1202
1203	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1204		return;
1205
1206	raw_spin_lock_irqsave(&desc->lock, flags);
1207	for_each_action_of_desc(desc, action) {
1208		if (action->dev_id == dev_id) {
1209			if (action->thread)
1210				__irq_wake_thread(desc, action);
1211			break;
1212		}
1213	}
1214	raw_spin_unlock_irqrestore(&desc->lock, flags);
1215}
1216EXPORT_SYMBOL_GPL(irq_wake_thread);
1217
1218static int irq_setup_forced_threading(struct irqaction *new)
1219{
1220	if (!force_irqthreads)
1221		return 0;
1222	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1223		return 0;
1224
1225	/*
1226	 * No further action required for interrupts which are requested as
1227	 * threaded interrupts already
1228	 */
1229	if (new->handler == irq_default_primary_handler)
1230		return 0;
1231
1232	new->flags |= IRQF_ONESHOT;
1233
1234	/*
1235	 * Handle the case where we have a real primary handler and a
1236	 * thread handler. We force thread them as well by creating a
1237	 * secondary action.
1238	 */
1239	if (new->handler && new->thread_fn) {
1240		/* Allocate the secondary action */
1241		new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1242		if (!new->secondary)
1243			return -ENOMEM;
1244		new->secondary->handler = irq_forced_secondary_handler;
1245		new->secondary->thread_fn = new->thread_fn;
1246		new->secondary->dev_id = new->dev_id;
1247		new->secondary->irq = new->irq;
1248		new->secondary->name = new->name;
1249	}
1250	/* Deal with the primary handler */
1251	set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1252	new->thread_fn = new->handler;
1253	new->handler = irq_default_primary_handler;
1254	return 0;
1255}
1256
1257static int irq_request_resources(struct irq_desc *desc)
1258{
1259	struct irq_data *d = &desc->irq_data;
1260	struct irq_chip *c = d->chip;
1261
1262	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1263}
1264
1265static void irq_release_resources(struct irq_desc *desc)
1266{
1267	struct irq_data *d = &desc->irq_data;
1268	struct irq_chip *c = d->chip;
1269
1270	if (c->irq_release_resources)
1271		c->irq_release_resources(d);
1272}
1273
1274static bool irq_supports_nmi(struct irq_desc *desc)
1275{
1276	struct irq_data *d = irq_desc_get_irq_data(desc);
1277
1278#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1279	/* Only IRQs directly managed by the root irqchip can be set as NMI */
1280	if (d->parent_data)
1281		return false;
1282#endif
1283	/* Don't support NMIs for chips behind a slow bus */
1284	if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1285		return false;
1286
1287	return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1288}
1289
1290static int irq_nmi_setup(struct irq_desc *desc)
1291{
1292	struct irq_data *d = irq_desc_get_irq_data(desc);
1293	struct irq_chip *c = d->chip;
1294
1295	return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1296}
1297
1298static void irq_nmi_teardown(struct irq_desc *desc)
1299{
1300	struct irq_data *d = irq_desc_get_irq_data(desc);
1301	struct irq_chip *c = d->chip;
1302
1303	if (c->irq_nmi_teardown)
1304		c->irq_nmi_teardown(d);
1305}
1306
1307static int
1308setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1309{
1310	struct task_struct *t;
1311
1312	if (!secondary) {
1313		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1314				   new->name);
1315	} else {
1316		t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1317				   new->name);
1318	}
1319
1320	if (IS_ERR(t))
1321		return PTR_ERR(t);
1322
1323	sched_set_fifo(t);
1324
1325	/*
1326	 * We keep the reference to the task struct even if
1327	 * the thread dies to avoid that the interrupt code
1328	 * references an already freed task_struct.
1329	 */
1330	new->thread = get_task_struct(t);
1331	/*
1332	 * Tell the thread to set its affinity. This is
1333	 * important for shared interrupt handlers as we do
1334	 * not invoke setup_affinity() for the secondary
1335	 * handlers as everything is already set up. Even for
1336	 * interrupts marked with IRQF_NO_BALANCE this is
1337	 * correct as we want the thread to move to the cpu(s)
1338	 * on which the requesting code placed the interrupt.
1339	 */
1340	set_bit(IRQTF_AFFINITY, &new->thread_flags);
1341	return 0;
1342}
1343
1344/*
1345 * Internal function to register an irqaction - typically used to
1346 * allocate special interrupts that are part of the architecture.
1347 *
1348 * Locking rules:
1349 *
1350 * desc->request_mutex	Provides serialization against a concurrent free_irq()
1351 *   chip_bus_lock	Provides serialization for slow bus operations
1352 *     desc->lock	Provides serialization against hard interrupts
1353 *
1354 * chip_bus_lock and desc->lock are sufficient for all other management and
1355 * interrupt related functions. desc->request_mutex solely serializes
1356 * request/free_irq().
1357 */
1358static int
1359__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1360{
1361	struct irqaction *old, **old_ptr;
1362	unsigned long flags, thread_mask = 0;
1363	int ret, nested, shared = 0;
1364
1365	if (!desc)
1366		return -EINVAL;
1367
1368	if (desc->irq_data.chip == &no_irq_chip)
1369		return -ENOSYS;
1370	if (!try_module_get(desc->owner))
1371		return -ENODEV;
1372
1373	new->irq = irq;
1374
1375	/*
1376	 * If the trigger type is not specified by the caller,
1377	 * then use the default for this interrupt.
1378	 */
1379	if (!(new->flags & IRQF_TRIGGER_MASK))
1380		new->flags |= irqd_get_trigger_type(&desc->irq_data);
1381
1382	/*
1383	 * Check whether the interrupt nests into another interrupt
1384	 * thread.
1385	 */
1386	nested = irq_settings_is_nested_thread(desc);
1387	if (nested) {
1388		if (!new->thread_fn) {
1389			ret = -EINVAL;
1390			goto out_mput;
1391		}
1392		/*
1393		 * Replace the primary handler which was provided from
1394		 * the driver for non nested interrupt handling by the
1395		 * dummy function which warns when called.
1396		 */
1397		new->handler = irq_nested_primary_handler;
1398	} else {
1399		if (irq_settings_can_thread(desc)) {
1400			ret = irq_setup_forced_threading(new);
1401			if (ret)
1402				goto out_mput;
1403		}
1404	}
1405
1406	/*
1407	 * Create a handler thread when a thread function is supplied
1408	 * and the interrupt does not nest into another interrupt
1409	 * thread.
1410	 */
1411	if (new->thread_fn && !nested) {
1412		ret = setup_irq_thread(new, irq, false);
1413		if (ret)
1414			goto out_mput;
1415		if (new->secondary) {
1416			ret = setup_irq_thread(new->secondary, irq, true);
1417			if (ret)
1418				goto out_thread;
1419		}
1420	}
1421
1422	/*
1423	 * Drivers are often written to work w/o knowledge about the
1424	 * underlying irq chip implementation, so a request for a
1425	 * threaded irq without a primary hard irq context handler
1426	 * requires the ONESHOT flag to be set. Some irq chips like
1427	 * MSI based interrupts are per se one shot safe. Check the
1428	 * chip flags, so we can avoid the unmask dance at the end of
1429	 * the threaded handler for those.
1430	 */
1431	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1432		new->flags &= ~IRQF_ONESHOT;
1433
1434	/*
1435	 * Protects against a concurrent __free_irq() call which might wait
1436	 * for synchronize_hardirq() to complete without holding the optional
1437	 * chip bus lock and desc->lock. Also protects against handing out
1438	 * a recycled oneshot thread_mask bit while it's still in use by
1439	 * its previous owner.
1440	 */
1441	mutex_lock(&desc->request_mutex);
1442
1443	/*
1444	 * Acquire bus lock as the irq_request_resources() callback below
1445	 * might rely on the serialization or the magic power management
1446	 * functions which are abusing the irq_bus_lock() callback,
1447	 */
1448	chip_bus_lock(desc);
1449
1450	/* First installed action requests resources. */
1451	if (!desc->action) {
1452		ret = irq_request_resources(desc);
1453		if (ret) {
1454			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1455			       new->name, irq, desc->irq_data.chip->name);
1456			goto out_bus_unlock;
1457		}
1458	}
1459
1460	/*
1461	 * The following block of code has to be executed atomically
1462	 * protected against a concurrent interrupt and any of the other
1463	 * management calls which are not serialized via
1464	 * desc->request_mutex or the optional bus lock.
1465	 */
1466	raw_spin_lock_irqsave(&desc->lock, flags);
1467	old_ptr = &desc->action;
1468	old = *old_ptr;
1469	if (old) {
1470		/*
1471		 * Can't share interrupts unless both agree to and are
1472		 * the same type (level, edge, polarity). So both flag
1473		 * fields must have IRQF_SHARED set and the bits which
1474		 * set the trigger type must match. Also all must
1475		 * agree on ONESHOT.
1476		 * Interrupt lines used for NMIs cannot be shared.
1477		 */
1478		unsigned int oldtype;
1479
1480		if (desc->istate & IRQS_NMI) {
1481			pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1482				new->name, irq, desc->irq_data.chip->name);
1483			ret = -EINVAL;
1484			goto out_unlock;
1485		}
1486
1487		/*
1488		 * If nobody did set the configuration before, inherit
1489		 * the one provided by the requester.
1490		 */
1491		if (irqd_trigger_type_was_set(&desc->irq_data)) {
1492			oldtype = irqd_get_trigger_type(&desc->irq_data);
1493		} else {
1494			oldtype = new->flags & IRQF_TRIGGER_MASK;
1495			irqd_set_trigger_type(&desc->irq_data, oldtype);
1496		}
1497
1498		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1499		    (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1500		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1501			goto mismatch;
1502
1503		/* All handlers must agree on per-cpuness */
1504		if ((old->flags & IRQF_PERCPU) !=
1505		    (new->flags & IRQF_PERCPU))
1506			goto mismatch;
1507
1508		/* add new interrupt at end of irq queue */
1509		do {
1510			/*
1511			 * Or all existing action->thread_mask bits,
1512			 * so we can find the next zero bit for this
1513			 * new action.
1514			 */
1515			thread_mask |= old->thread_mask;
1516			old_ptr = &old->next;
1517			old = *old_ptr;
1518		} while (old);
1519		shared = 1;
1520	}
1521
1522	/*
1523	 * Setup the thread mask for this irqaction for ONESHOT. For
1524	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1525	 * conditional in irq_wake_thread().
1526	 */
1527	if (new->flags & IRQF_ONESHOT) {
1528		/*
1529		 * Unlikely to have 32 resp 64 irqs sharing one line,
1530		 * but who knows.
1531		 */
1532		if (thread_mask == ~0UL) {
1533			ret = -EBUSY;
1534			goto out_unlock;
1535		}
1536		/*
1537		 * The thread_mask for the action is or'ed to
1538		 * desc->thread_active to indicate that the
1539		 * IRQF_ONESHOT thread handler has been woken, but not
1540		 * yet finished. The bit is cleared when a thread
1541		 * completes. When all threads of a shared interrupt
1542		 * line have completed desc->threads_active becomes
1543		 * zero and the interrupt line is unmasked. See
1544		 * handle.c:irq_wake_thread() for further information.
1545		 *
1546		 * If no thread is woken by primary (hard irq context)
1547		 * interrupt handlers, then desc->threads_active is
1548		 * also checked for zero to unmask the irq line in the
1549		 * affected hard irq flow handlers
1550		 * (handle_[fasteoi|level]_irq).
1551		 *
1552		 * The new action gets the first zero bit of
1553		 * thread_mask assigned. See the loop above which or's
1554		 * all existing action->thread_mask bits.
1555		 */
1556		new->thread_mask = 1UL << ffz(thread_mask);
1557
1558	} else if (new->handler == irq_default_primary_handler &&
1559		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1560		/*
1561		 * The interrupt was requested with handler = NULL, so
1562		 * we use the default primary handler for it. But it
1563		 * does not have the oneshot flag set. In combination
1564		 * with level interrupts this is deadly, because the
1565		 * default primary handler just wakes the thread, then
1566		 * the irq lines is reenabled, but the device still
1567		 * has the level irq asserted. Rinse and repeat....
1568		 *
1569		 * While this works for edge type interrupts, we play
1570		 * it safe and reject unconditionally because we can't
1571		 * say for sure which type this interrupt really
1572		 * has. The type flags are unreliable as the
1573		 * underlying chip implementation can override them.
1574		 */
1575		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for %s (irq %d)\n",
1576		       new->name, irq);
1577		ret = -EINVAL;
1578		goto out_unlock;
1579	}
1580
1581	if (!shared) {
1582		init_waitqueue_head(&desc->wait_for_threads);
1583
1584		/* Setup the type (level, edge polarity) if configured: */
1585		if (new->flags & IRQF_TRIGGER_MASK) {
1586			ret = __irq_set_trigger(desc,
1587						new->flags & IRQF_TRIGGER_MASK);
1588
1589			if (ret)
1590				goto out_unlock;
1591		}
1592
1593		/*
1594		 * Activate the interrupt. That activation must happen
1595		 * independently of IRQ_NOAUTOEN. request_irq() can fail
1596		 * and the callers are supposed to handle
1597		 * that. enable_irq() of an interrupt requested with
1598		 * IRQ_NOAUTOEN is not supposed to fail. The activation
1599		 * keeps it in shutdown mode, it merily associates
1600		 * resources if necessary and if that's not possible it
1601		 * fails. Interrupts which are in managed shutdown mode
1602		 * will simply ignore that activation request.
1603		 */
1604		ret = irq_activate(desc);
1605		if (ret)
1606			goto out_unlock;
1607
1608		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1609				  IRQS_ONESHOT | IRQS_WAITING);
1610		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1611
1612		if (new->flags & IRQF_PERCPU) {
1613			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1614			irq_settings_set_per_cpu(desc);
 
 
1615		}
1616
 
 
 
1617		if (new->flags & IRQF_ONESHOT)
1618			desc->istate |= IRQS_ONESHOT;
1619
1620		/* Exclude IRQ from balancing if requested */
1621		if (new->flags & IRQF_NOBALANCING) {
1622			irq_settings_set_no_balancing(desc);
1623			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1624		}
1625
1626		if (irq_settings_can_autoenable(desc)) {
 
1627			irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1628		} else {
1629			/*
1630			 * Shared interrupts do not go well with disabling
1631			 * auto enable. The sharing interrupt might request
1632			 * it while it's still disabled and then wait for
1633			 * interrupts forever.
1634			 */
1635			WARN_ON_ONCE(new->flags & IRQF_SHARED);
1636			/* Undo nested disables: */
1637			desc->depth = 1;
1638		}
1639
1640	} else if (new->flags & IRQF_TRIGGER_MASK) {
1641		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1642		unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1643
1644		if (nmsk != omsk)
1645			/* hope the handler works with current  trigger mode */
1646			pr_warn("irq %d uses trigger mode %u; requested %u\n",
1647				irq, omsk, nmsk);
1648	}
1649
1650	*old_ptr = new;
1651
1652	irq_pm_install_action(desc, new);
1653
1654	/* Reset broken irq detection when installing new handler */
1655	desc->irq_count = 0;
1656	desc->irqs_unhandled = 0;
1657
1658	/*
1659	 * Check whether we disabled the irq via the spurious handler
1660	 * before. Reenable it and give it another chance.
1661	 */
1662	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1663		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1664		__enable_irq(desc);
1665	}
1666
1667	raw_spin_unlock_irqrestore(&desc->lock, flags);
1668	chip_bus_sync_unlock(desc);
1669	mutex_unlock(&desc->request_mutex);
1670
1671	irq_setup_timings(desc, new);
1672
1673	/*
1674	 * Strictly no need to wake it up, but hung_task complains
1675	 * when no hard interrupt wakes the thread up.
1676	 */
1677	if (new->thread)
1678		wake_up_process(new->thread);
1679	if (new->secondary)
1680		wake_up_process(new->secondary->thread);
1681
1682	register_irq_proc(irq, desc);
1683	new->dir = NULL;
1684	register_handler_proc(irq, new);
1685	return 0;
1686
1687mismatch:
1688	if (!(new->flags & IRQF_PROBE_SHARED)) {
1689		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1690		       irq, new->flags, new->name, old->flags, old->name);
1691#ifdef CONFIG_DEBUG_SHIRQ
1692		dump_stack();
1693#endif
1694	}
1695	ret = -EBUSY;
1696
1697out_unlock:
1698	raw_spin_unlock_irqrestore(&desc->lock, flags);
1699
1700	if (!desc->action)
1701		irq_release_resources(desc);
1702out_bus_unlock:
1703	chip_bus_sync_unlock(desc);
1704	mutex_unlock(&desc->request_mutex);
1705
1706out_thread:
1707	if (new->thread) {
1708		struct task_struct *t = new->thread;
1709
1710		new->thread = NULL;
1711		kthread_stop(t);
1712		put_task_struct(t);
1713	}
1714	if (new->secondary && new->secondary->thread) {
1715		struct task_struct *t = new->secondary->thread;
1716
1717		new->secondary->thread = NULL;
1718		kthread_stop(t);
1719		put_task_struct(t);
1720	}
1721out_mput:
1722	module_put(desc->owner);
1723	return ret;
1724}
1725
1726/*
1727 * Internal function to unregister an irqaction - used to free
1728 * regular and special interrupts that are part of the architecture.
1729 */
1730static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1731{
1732	unsigned irq = desc->irq_data.irq;
1733	struct irqaction *action, **action_ptr;
1734	unsigned long flags;
1735
1736	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1737
1738	mutex_lock(&desc->request_mutex);
1739	chip_bus_lock(desc);
1740	raw_spin_lock_irqsave(&desc->lock, flags);
1741
1742	/*
1743	 * There can be multiple actions per IRQ descriptor, find the right
1744	 * one based on the dev_id:
1745	 */
1746	action_ptr = &desc->action;
1747	for (;;) {
1748		action = *action_ptr;
1749
1750		if (!action) {
1751			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1752			raw_spin_unlock_irqrestore(&desc->lock, flags);
1753			chip_bus_sync_unlock(desc);
1754			mutex_unlock(&desc->request_mutex);
1755			return NULL;
1756		}
1757
1758		if (action->dev_id == dev_id)
1759			break;
1760		action_ptr = &action->next;
1761	}
1762
1763	/* Found it - now remove it from the list of entries: */
1764	*action_ptr = action->next;
1765
1766	irq_pm_remove_action(desc, action);
1767
1768	/* If this was the last handler, shut down the IRQ line: */
1769	if (!desc->action) {
1770		irq_settings_clr_disable_unlazy(desc);
1771		/* Only shutdown. Deactivate after synchronize_hardirq() */
1772		irq_shutdown(desc);
1773	}
1774
1775#ifdef CONFIG_SMP
1776	/* make sure affinity_hint is cleaned up */
1777	if (WARN_ON_ONCE(desc->affinity_hint))
1778		desc->affinity_hint = NULL;
1779#endif
1780
1781	raw_spin_unlock_irqrestore(&desc->lock, flags);
1782	/*
1783	 * Drop bus_lock here so the changes which were done in the chip
1784	 * callbacks above are synced out to the irq chips which hang
1785	 * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1786	 *
1787	 * Aside of that the bus_lock can also be taken from the threaded
1788	 * handler in irq_finalize_oneshot() which results in a deadlock
1789	 * because kthread_stop() would wait forever for the thread to
1790	 * complete, which is blocked on the bus lock.
1791	 *
1792	 * The still held desc->request_mutex() protects against a
1793	 * concurrent request_irq() of this irq so the release of resources
1794	 * and timing data is properly serialized.
1795	 */
1796	chip_bus_sync_unlock(desc);
1797
1798	unregister_handler_proc(irq, action);
1799
1800	/*
1801	 * Make sure it's not being used on another CPU and if the chip
1802	 * supports it also make sure that there is no (not yet serviced)
1803	 * interrupt in flight at the hardware level.
1804	 */
1805	__synchronize_hardirq(desc, true);
1806
1807#ifdef CONFIG_DEBUG_SHIRQ
1808	/*
1809	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1810	 * event to happen even now it's being freed, so let's make sure that
1811	 * is so by doing an extra call to the handler ....
1812	 *
1813	 * ( We do this after actually deregistering it, to make sure that a
1814	 *   'real' IRQ doesn't run in parallel with our fake. )
1815	 */
1816	if (action->flags & IRQF_SHARED) {
1817		local_irq_save(flags);
1818		action->handler(irq, dev_id);
1819		local_irq_restore(flags);
1820	}
1821#endif
1822
1823	/*
1824	 * The action has already been removed above, but the thread writes
1825	 * its oneshot mask bit when it completes. Though request_mutex is
1826	 * held across this which prevents __setup_irq() from handing out
1827	 * the same bit to a newly requested action.
1828	 */
1829	if (action->thread) {
1830		kthread_stop(action->thread);
1831		put_task_struct(action->thread);
1832		if (action->secondary && action->secondary->thread) {
1833			kthread_stop(action->secondary->thread);
1834			put_task_struct(action->secondary->thread);
1835		}
1836	}
1837
1838	/* Last action releases resources */
1839	if (!desc->action) {
1840		/*
1841		 * Reaquire bus lock as irq_release_resources() might
1842		 * require it to deallocate resources over the slow bus.
1843		 */
1844		chip_bus_lock(desc);
1845		/*
1846		 * There is no interrupt on the fly anymore. Deactivate it
1847		 * completely.
1848		 */
1849		raw_spin_lock_irqsave(&desc->lock, flags);
1850		irq_domain_deactivate_irq(&desc->irq_data);
1851		raw_spin_unlock_irqrestore(&desc->lock, flags);
1852
1853		irq_release_resources(desc);
1854		chip_bus_sync_unlock(desc);
1855		irq_remove_timings(desc);
1856	}
1857
1858	mutex_unlock(&desc->request_mutex);
1859
1860	irq_chip_pm_put(&desc->irq_data);
1861	module_put(desc->owner);
1862	kfree(action->secondary);
1863	return action;
1864}
1865
1866/**
1867 *	free_irq - free an interrupt allocated with request_irq
1868 *	@irq: Interrupt line to free
1869 *	@dev_id: Device identity to free
1870 *
1871 *	Remove an interrupt handler. The handler is removed and if the
1872 *	interrupt line is no longer in use by any driver it is disabled.
1873 *	On a shared IRQ the caller must ensure the interrupt is disabled
1874 *	on the card it drives before calling this function. The function
1875 *	does not return until any executing interrupts for this IRQ
1876 *	have completed.
1877 *
1878 *	This function must not be called from interrupt context.
1879 *
1880 *	Returns the devname argument passed to request_irq.
1881 */
1882const void *free_irq(unsigned int irq, void *dev_id)
1883{
1884	struct irq_desc *desc = irq_to_desc(irq);
1885	struct irqaction *action;
1886	const char *devname;
1887
1888	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1889		return NULL;
1890
1891#ifdef CONFIG_SMP
1892	if (WARN_ON(desc->affinity_notify))
1893		desc->affinity_notify = NULL;
1894#endif
1895
1896	action = __free_irq(desc, dev_id);
1897
1898	if (!action)
1899		return NULL;
1900
1901	devname = action->name;
1902	kfree(action);
1903	return devname;
1904}
1905EXPORT_SYMBOL(free_irq);
1906
1907/* This function must be called with desc->lock held */
1908static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
1909{
1910	const char *devname = NULL;
1911
1912	desc->istate &= ~IRQS_NMI;
1913
1914	if (!WARN_ON(desc->action == NULL)) {
1915		irq_pm_remove_action(desc, desc->action);
1916		devname = desc->action->name;
1917		unregister_handler_proc(irq, desc->action);
1918
1919		kfree(desc->action);
1920		desc->action = NULL;
1921	}
1922
1923	irq_settings_clr_disable_unlazy(desc);
1924	irq_shutdown_and_deactivate(desc);
1925
1926	irq_release_resources(desc);
1927
1928	irq_chip_pm_put(&desc->irq_data);
1929	module_put(desc->owner);
1930
1931	return devname;
1932}
1933
1934const void *free_nmi(unsigned int irq, void *dev_id)
1935{
1936	struct irq_desc *desc = irq_to_desc(irq);
1937	unsigned long flags;
1938	const void *devname;
1939
1940	if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
1941		return NULL;
1942
1943	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1944		return NULL;
1945
1946	/* NMI still enabled */
1947	if (WARN_ON(desc->depth == 0))
1948		disable_nmi_nosync(irq);
1949
1950	raw_spin_lock_irqsave(&desc->lock, flags);
1951
1952	irq_nmi_teardown(desc);
1953	devname = __cleanup_nmi(irq, desc);
1954
1955	raw_spin_unlock_irqrestore(&desc->lock, flags);
1956
1957	return devname;
1958}
1959
1960/**
1961 *	request_threaded_irq - allocate an interrupt line
1962 *	@irq: Interrupt line to allocate
1963 *	@handler: Function to be called when the IRQ occurs.
1964 *		  Primary handler for threaded interrupts
1965 *		  If NULL and thread_fn != NULL the default
1966 *		  primary handler is installed
1967 *	@thread_fn: Function called from the irq handler thread
1968 *		    If NULL, no irq thread is created
1969 *	@irqflags: Interrupt type flags
1970 *	@devname: An ascii name for the claiming device
1971 *	@dev_id: A cookie passed back to the handler function
1972 *
1973 *	This call allocates interrupt resources and enables the
1974 *	interrupt line and IRQ handling. From the point this
1975 *	call is made your handler function may be invoked. Since
1976 *	your handler function must clear any interrupt the board
1977 *	raises, you must take care both to initialise your hardware
1978 *	and to set up the interrupt handler in the right order.
1979 *
1980 *	If you want to set up a threaded irq handler for your device
1981 *	then you need to supply @handler and @thread_fn. @handler is
1982 *	still called in hard interrupt context and has to check
1983 *	whether the interrupt originates from the device. If yes it
1984 *	needs to disable the interrupt on the device and return
1985 *	IRQ_WAKE_THREAD which will wake up the handler thread and run
1986 *	@thread_fn. This split handler design is necessary to support
1987 *	shared interrupts.
1988 *
1989 *	Dev_id must be globally unique. Normally the address of the
1990 *	device data structure is used as the cookie. Since the handler
1991 *	receives this value it makes sense to use it.
1992 *
1993 *	If your interrupt is shared you must pass a non NULL dev_id
1994 *	as this is required when freeing the interrupt.
1995 *
1996 *	Flags:
1997 *
1998 *	IRQF_SHARED		Interrupt is shared
1999 *	IRQF_TRIGGER_*		Specify active edge(s) or level
2000 *
2001 */
2002int request_threaded_irq(unsigned int irq, irq_handler_t handler,
2003			 irq_handler_t thread_fn, unsigned long irqflags,
2004			 const char *devname, void *dev_id)
2005{
2006	struct irqaction *action;
2007	struct irq_desc *desc;
2008	int retval;
2009
2010	if (irq == IRQ_NOTCONNECTED)
2011		return -ENOTCONN;
2012
2013	/*
2014	 * Sanity-check: shared interrupts must pass in a real dev-ID,
2015	 * otherwise we'll have trouble later trying to figure out
2016	 * which interrupt is which (messes up the interrupt freeing
2017	 * logic etc).
2018	 *
 
 
 
 
2019	 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
2020	 * it cannot be set along with IRQF_NO_SUSPEND.
2021	 */
2022	if (((irqflags & IRQF_SHARED) && !dev_id) ||
 
2023	    (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
2024	    ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
2025		return -EINVAL;
2026
2027	desc = irq_to_desc(irq);
2028	if (!desc)
2029		return -EINVAL;
2030
2031	if (!irq_settings_can_request(desc) ||
2032	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2033		return -EINVAL;
2034
2035	if (!handler) {
2036		if (!thread_fn)
2037			return -EINVAL;
2038		handler = irq_default_primary_handler;
2039	}
2040
2041	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2042	if (!action)
2043		return -ENOMEM;
2044
2045	action->handler = handler;
2046	action->thread_fn = thread_fn;
2047	action->flags = irqflags;
2048	action->name = devname;
2049	action->dev_id = dev_id;
2050
2051	retval = irq_chip_pm_get(&desc->irq_data);
2052	if (retval < 0) {
2053		kfree(action);
2054		return retval;
2055	}
2056
2057	retval = __setup_irq(irq, desc, action);
2058
2059	if (retval) {
2060		irq_chip_pm_put(&desc->irq_data);
2061		kfree(action->secondary);
2062		kfree(action);
2063	}
2064
2065#ifdef CONFIG_DEBUG_SHIRQ_FIXME
2066	if (!retval && (irqflags & IRQF_SHARED)) {
2067		/*
2068		 * It's a shared IRQ -- the driver ought to be prepared for it
2069		 * to happen immediately, so let's make sure....
2070		 * We disable the irq to make sure that a 'real' IRQ doesn't
2071		 * run in parallel with our fake.
2072		 */
2073		unsigned long flags;
2074
2075		disable_irq(irq);
2076		local_irq_save(flags);
2077
2078		handler(irq, dev_id);
2079
2080		local_irq_restore(flags);
2081		enable_irq(irq);
2082	}
2083#endif
2084	return retval;
2085}
2086EXPORT_SYMBOL(request_threaded_irq);
2087
2088/**
2089 *	request_any_context_irq - allocate an interrupt line
2090 *	@irq: Interrupt line to allocate
2091 *	@handler: Function to be called when the IRQ occurs.
2092 *		  Threaded handler for threaded interrupts.
2093 *	@flags: Interrupt type flags
2094 *	@name: An ascii name for the claiming device
2095 *	@dev_id: A cookie passed back to the handler function
2096 *
2097 *	This call allocates interrupt resources and enables the
2098 *	interrupt line and IRQ handling. It selects either a
2099 *	hardirq or threaded handling method depending on the
2100 *	context.
2101 *
2102 *	On failure, it returns a negative value. On success,
2103 *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2104 */
2105int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2106			    unsigned long flags, const char *name, void *dev_id)
2107{
2108	struct irq_desc *desc;
2109	int ret;
2110
2111	if (irq == IRQ_NOTCONNECTED)
2112		return -ENOTCONN;
2113
2114	desc = irq_to_desc(irq);
2115	if (!desc)
2116		return -EINVAL;
2117
2118	if (irq_settings_is_nested_thread(desc)) {
2119		ret = request_threaded_irq(irq, NULL, handler,
2120					   flags, name, dev_id);
2121		return !ret ? IRQC_IS_NESTED : ret;
2122	}
2123
2124	ret = request_irq(irq, handler, flags, name, dev_id);
2125	return !ret ? IRQC_IS_HARDIRQ : ret;
2126}
2127EXPORT_SYMBOL_GPL(request_any_context_irq);
2128
2129/**
2130 *	request_nmi - allocate an interrupt line for NMI delivery
2131 *	@irq: Interrupt line to allocate
2132 *	@handler: Function to be called when the IRQ occurs.
2133 *		  Threaded handler for threaded interrupts.
2134 *	@irqflags: Interrupt type flags
2135 *	@name: An ascii name for the claiming device
2136 *	@dev_id: A cookie passed back to the handler function
2137 *
2138 *	This call allocates interrupt resources and enables the
2139 *	interrupt line and IRQ handling. It sets up the IRQ line
2140 *	to be handled as an NMI.
2141 *
2142 *	An interrupt line delivering NMIs cannot be shared and IRQ handling
2143 *	cannot be threaded.
2144 *
2145 *	Interrupt lines requested for NMI delivering must produce per cpu
2146 *	interrupts and have auto enabling setting disabled.
2147 *
2148 *	Dev_id must be globally unique. Normally the address of the
2149 *	device data structure is used as the cookie. Since the handler
2150 *	receives this value it makes sense to use it.
2151 *
2152 *	If the interrupt line cannot be used to deliver NMIs, function
2153 *	will fail and return a negative value.
2154 */
2155int request_nmi(unsigned int irq, irq_handler_t handler,
2156		unsigned long irqflags, const char *name, void *dev_id)
2157{
2158	struct irqaction *action;
2159	struct irq_desc *desc;
2160	unsigned long flags;
2161	int retval;
2162
2163	if (irq == IRQ_NOTCONNECTED)
2164		return -ENOTCONN;
2165
2166	/* NMI cannot be shared, used for Polling */
2167	if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2168		return -EINVAL;
2169
2170	if (!(irqflags & IRQF_PERCPU))
2171		return -EINVAL;
2172
2173	if (!handler)
2174		return -EINVAL;
2175
2176	desc = irq_to_desc(irq);
2177
2178	if (!desc || irq_settings_can_autoenable(desc) ||
 
2179	    !irq_settings_can_request(desc) ||
2180	    WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2181	    !irq_supports_nmi(desc))
2182		return -EINVAL;
2183
2184	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2185	if (!action)
2186		return -ENOMEM;
2187
2188	action->handler = handler;
2189	action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2190	action->name = name;
2191	action->dev_id = dev_id;
2192
2193	retval = irq_chip_pm_get(&desc->irq_data);
2194	if (retval < 0)
2195		goto err_out;
2196
2197	retval = __setup_irq(irq, desc, action);
2198	if (retval)
2199		goto err_irq_setup;
2200
2201	raw_spin_lock_irqsave(&desc->lock, flags);
2202
2203	/* Setup NMI state */
2204	desc->istate |= IRQS_NMI;
2205	retval = irq_nmi_setup(desc);
2206	if (retval) {
2207		__cleanup_nmi(irq, desc);
2208		raw_spin_unlock_irqrestore(&desc->lock, flags);
2209		return -EINVAL;
2210	}
2211
2212	raw_spin_unlock_irqrestore(&desc->lock, flags);
2213
2214	return 0;
2215
2216err_irq_setup:
2217	irq_chip_pm_put(&desc->irq_data);
2218err_out:
2219	kfree(action);
2220
2221	return retval;
2222}
2223
2224void enable_percpu_irq(unsigned int irq, unsigned int type)
2225{
2226	unsigned int cpu = smp_processor_id();
2227	unsigned long flags;
2228	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2229
2230	if (!desc)
2231		return;
2232
2233	/*
2234	 * If the trigger type is not specified by the caller, then
2235	 * use the default for this interrupt.
2236	 */
2237	type &= IRQ_TYPE_SENSE_MASK;
2238	if (type == IRQ_TYPE_NONE)
2239		type = irqd_get_trigger_type(&desc->irq_data);
2240
2241	if (type != IRQ_TYPE_NONE) {
2242		int ret;
2243
2244		ret = __irq_set_trigger(desc, type);
2245
2246		if (ret) {
2247			WARN(1, "failed to set type for IRQ%d\n", irq);
2248			goto out;
2249		}
2250	}
2251
2252	irq_percpu_enable(desc, cpu);
2253out:
2254	irq_put_desc_unlock(desc, flags);
2255}
2256EXPORT_SYMBOL_GPL(enable_percpu_irq);
2257
2258void enable_percpu_nmi(unsigned int irq, unsigned int type)
2259{
2260	enable_percpu_irq(irq, type);
2261}
2262
2263/**
2264 * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2265 * @irq:	Linux irq number to check for
2266 *
2267 * Must be called from a non migratable context. Returns the enable
2268 * state of a per cpu interrupt on the current cpu.
2269 */
2270bool irq_percpu_is_enabled(unsigned int irq)
2271{
2272	unsigned int cpu = smp_processor_id();
2273	struct irq_desc *desc;
2274	unsigned long flags;
2275	bool is_enabled;
2276
2277	desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2278	if (!desc)
2279		return false;
2280
2281	is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2282	irq_put_desc_unlock(desc, flags);
2283
2284	return is_enabled;
2285}
2286EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2287
2288void disable_percpu_irq(unsigned int irq)
2289{
2290	unsigned int cpu = smp_processor_id();
2291	unsigned long flags;
2292	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2293
2294	if (!desc)
2295		return;
2296
2297	irq_percpu_disable(desc, cpu);
2298	irq_put_desc_unlock(desc, flags);
2299}
2300EXPORT_SYMBOL_GPL(disable_percpu_irq);
2301
2302void disable_percpu_nmi(unsigned int irq)
2303{
2304	disable_percpu_irq(irq);
2305}
2306
2307/*
2308 * Internal function to unregister a percpu irqaction.
2309 */
2310static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2311{
2312	struct irq_desc *desc = irq_to_desc(irq);
2313	struct irqaction *action;
2314	unsigned long flags;
2315
2316	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2317
2318	if (!desc)
2319		return NULL;
2320
2321	raw_spin_lock_irqsave(&desc->lock, flags);
2322
2323	action = desc->action;
2324	if (!action || action->percpu_dev_id != dev_id) {
2325		WARN(1, "Trying to free already-free IRQ %d\n", irq);
2326		goto bad;
2327	}
2328
2329	if (!cpumask_empty(desc->percpu_enabled)) {
2330		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2331		     irq, cpumask_first(desc->percpu_enabled));
2332		goto bad;
2333	}
2334
2335	/* Found it - now remove it from the list of entries: */
2336	desc->action = NULL;
2337
2338	desc->istate &= ~IRQS_NMI;
2339
2340	raw_spin_unlock_irqrestore(&desc->lock, flags);
2341
2342	unregister_handler_proc(irq, action);
2343
2344	irq_chip_pm_put(&desc->irq_data);
2345	module_put(desc->owner);
2346	return action;
2347
2348bad:
2349	raw_spin_unlock_irqrestore(&desc->lock, flags);
2350	return NULL;
2351}
2352
2353/**
2354 *	remove_percpu_irq - free a per-cpu interrupt
2355 *	@irq: Interrupt line to free
2356 *	@act: irqaction for the interrupt
2357 *
2358 * Used to remove interrupts statically setup by the early boot process.
2359 */
2360void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2361{
2362	struct irq_desc *desc = irq_to_desc(irq);
2363
2364	if (desc && irq_settings_is_per_cpu_devid(desc))
2365	    __free_percpu_irq(irq, act->percpu_dev_id);
2366}
2367
2368/**
2369 *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
2370 *	@irq: Interrupt line to free
2371 *	@dev_id: Device identity to free
2372 *
2373 *	Remove a percpu interrupt handler. The handler is removed, but
2374 *	the interrupt line is not disabled. This must be done on each
2375 *	CPU before calling this function. The function does not return
2376 *	until any executing interrupts for this IRQ have completed.
2377 *
2378 *	This function must not be called from interrupt context.
2379 */
2380void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2381{
2382	struct irq_desc *desc = irq_to_desc(irq);
2383
2384	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2385		return;
2386
2387	chip_bus_lock(desc);
2388	kfree(__free_percpu_irq(irq, dev_id));
2389	chip_bus_sync_unlock(desc);
2390}
2391EXPORT_SYMBOL_GPL(free_percpu_irq);
2392
2393void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2394{
2395	struct irq_desc *desc = irq_to_desc(irq);
2396
2397	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2398		return;
2399
2400	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2401		return;
2402
2403	kfree(__free_percpu_irq(irq, dev_id));
2404}
2405
2406/**
2407 *	setup_percpu_irq - setup a per-cpu interrupt
2408 *	@irq: Interrupt line to setup
2409 *	@act: irqaction for the interrupt
2410 *
2411 * Used to statically setup per-cpu interrupts in the early boot process.
2412 */
2413int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2414{
2415	struct irq_desc *desc = irq_to_desc(irq);
2416	int retval;
2417
2418	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2419		return -EINVAL;
2420
2421	retval = irq_chip_pm_get(&desc->irq_data);
2422	if (retval < 0)
2423		return retval;
2424
2425	retval = __setup_irq(irq, desc, act);
2426
2427	if (retval)
2428		irq_chip_pm_put(&desc->irq_data);
2429
2430	return retval;
2431}
2432
2433/**
2434 *	__request_percpu_irq - allocate a percpu interrupt line
2435 *	@irq: Interrupt line to allocate
2436 *	@handler: Function to be called when the IRQ occurs.
2437 *	@flags: Interrupt type flags (IRQF_TIMER only)
2438 *	@devname: An ascii name for the claiming device
2439 *	@dev_id: A percpu cookie passed back to the handler function
2440 *
2441 *	This call allocates interrupt resources and enables the
2442 *	interrupt on the local CPU. If the interrupt is supposed to be
2443 *	enabled on other CPUs, it has to be done on each CPU using
2444 *	enable_percpu_irq().
2445 *
2446 *	Dev_id must be globally unique. It is a per-cpu variable, and
2447 *	the handler gets called with the interrupted CPU's instance of
2448 *	that variable.
2449 */
2450int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2451			 unsigned long flags, const char *devname,
2452			 void __percpu *dev_id)
2453{
2454	struct irqaction *action;
2455	struct irq_desc *desc;
2456	int retval;
2457
2458	if (!dev_id)
2459		return -EINVAL;
2460
2461	desc = irq_to_desc(irq);
2462	if (!desc || !irq_settings_can_request(desc) ||
2463	    !irq_settings_is_per_cpu_devid(desc))
2464		return -EINVAL;
2465
2466	if (flags && flags != IRQF_TIMER)
2467		return -EINVAL;
2468
2469	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2470	if (!action)
2471		return -ENOMEM;
2472
2473	action->handler = handler;
2474	action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2475	action->name = devname;
2476	action->percpu_dev_id = dev_id;
2477
2478	retval = irq_chip_pm_get(&desc->irq_data);
2479	if (retval < 0) {
2480		kfree(action);
2481		return retval;
2482	}
2483
2484	retval = __setup_irq(irq, desc, action);
2485
2486	if (retval) {
2487		irq_chip_pm_put(&desc->irq_data);
2488		kfree(action);
2489	}
2490
2491	return retval;
2492}
2493EXPORT_SYMBOL_GPL(__request_percpu_irq);
2494
2495/**
2496 *	request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2497 *	@irq: Interrupt line to allocate
2498 *	@handler: Function to be called when the IRQ occurs.
2499 *	@name: An ascii name for the claiming device
2500 *	@dev_id: A percpu cookie passed back to the handler function
2501 *
2502 *	This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2503 *	have to be setup on each CPU by calling prepare_percpu_nmi() before
2504 *	being enabled on the same CPU by using enable_percpu_nmi().
2505 *
2506 *	Dev_id must be globally unique. It is a per-cpu variable, and
2507 *	the handler gets called with the interrupted CPU's instance of
2508 *	that variable.
2509 *
2510 *	Interrupt lines requested for NMI delivering should have auto enabling
2511 *	setting disabled.
2512 *
2513 *	If the interrupt line cannot be used to deliver NMIs, function
2514 *	will fail returning a negative value.
2515 */
2516int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2517		       const char *name, void __percpu *dev_id)
2518{
2519	struct irqaction *action;
2520	struct irq_desc *desc;
2521	unsigned long flags;
2522	int retval;
2523
2524	if (!handler)
2525		return -EINVAL;
2526
2527	desc = irq_to_desc(irq);
2528
2529	if (!desc || !irq_settings_can_request(desc) ||
2530	    !irq_settings_is_per_cpu_devid(desc) ||
2531	    irq_settings_can_autoenable(desc) ||
2532	    !irq_supports_nmi(desc))
2533		return -EINVAL;
2534
2535	/* The line cannot already be NMI */
2536	if (desc->istate & IRQS_NMI)
2537		return -EINVAL;
2538
2539	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2540	if (!action)
2541		return -ENOMEM;
2542
2543	action->handler = handler;
2544	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2545		| IRQF_NOBALANCING;
2546	action->name = name;
2547	action->percpu_dev_id = dev_id;
2548
2549	retval = irq_chip_pm_get(&desc->irq_data);
2550	if (retval < 0)
2551		goto err_out;
2552
2553	retval = __setup_irq(irq, desc, action);
2554	if (retval)
2555		goto err_irq_setup;
2556
2557	raw_spin_lock_irqsave(&desc->lock, flags);
2558	desc->istate |= IRQS_NMI;
2559	raw_spin_unlock_irqrestore(&desc->lock, flags);
2560
2561	return 0;
2562
2563err_irq_setup:
2564	irq_chip_pm_put(&desc->irq_data);
2565err_out:
2566	kfree(action);
2567
2568	return retval;
2569}
2570
2571/**
2572 *	prepare_percpu_nmi - performs CPU local setup for NMI delivery
2573 *	@irq: Interrupt line to prepare for NMI delivery
2574 *
2575 *	This call prepares an interrupt line to deliver NMI on the current CPU,
2576 *	before that interrupt line gets enabled with enable_percpu_nmi().
2577 *
2578 *	As a CPU local operation, this should be called from non-preemptible
2579 *	context.
2580 *
2581 *	If the interrupt line cannot be used to deliver NMIs, function
2582 *	will fail returning a negative value.
2583 */
2584int prepare_percpu_nmi(unsigned int irq)
2585{
2586	unsigned long flags;
2587	struct irq_desc *desc;
2588	int ret = 0;
2589
2590	WARN_ON(preemptible());
2591
2592	desc = irq_get_desc_lock(irq, &flags,
2593				 IRQ_GET_DESC_CHECK_PERCPU);
2594	if (!desc)
2595		return -EINVAL;
2596
2597	if (WARN(!(desc->istate & IRQS_NMI),
2598		 KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2599		 irq)) {
2600		ret = -EINVAL;
2601		goto out;
2602	}
2603
2604	ret = irq_nmi_setup(desc);
2605	if (ret) {
2606		pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2607		goto out;
2608	}
2609
2610out:
2611	irq_put_desc_unlock(desc, flags);
2612	return ret;
2613}
2614
2615/**
2616 *	teardown_percpu_nmi - undoes NMI setup of IRQ line
2617 *	@irq: Interrupt line from which CPU local NMI configuration should be
2618 *	      removed
2619 *
2620 *	This call undoes the setup done by prepare_percpu_nmi().
2621 *
2622 *	IRQ line should not be enabled for the current CPU.
2623 *
2624 *	As a CPU local operation, this should be called from non-preemptible
2625 *	context.
2626 */
2627void teardown_percpu_nmi(unsigned int irq)
2628{
2629	unsigned long flags;
2630	struct irq_desc *desc;
2631
2632	WARN_ON(preemptible());
2633
2634	desc = irq_get_desc_lock(irq, &flags,
2635				 IRQ_GET_DESC_CHECK_PERCPU);
2636	if (!desc)
2637		return;
2638
2639	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2640		goto out;
2641
2642	irq_nmi_teardown(desc);
2643out:
2644	irq_put_desc_unlock(desc, flags);
2645}
2646
2647int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2648			    bool *state)
2649{
2650	struct irq_chip *chip;
2651	int err = -EINVAL;
2652
2653	do {
2654		chip = irq_data_get_irq_chip(data);
2655		if (WARN_ON_ONCE(!chip))
2656			return -ENODEV;
2657		if (chip->irq_get_irqchip_state)
2658			break;
2659#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2660		data = data->parent_data;
2661#else
2662		data = NULL;
2663#endif
2664	} while (data);
2665
2666	if (data)
2667		err = chip->irq_get_irqchip_state(data, which, state);
2668	return err;
2669}
2670
2671/**
2672 *	irq_get_irqchip_state - returns the irqchip state of a interrupt.
2673 *	@irq: Interrupt line that is forwarded to a VM
2674 *	@which: One of IRQCHIP_STATE_* the caller wants to know about
2675 *	@state: a pointer to a boolean where the state is to be storeed
2676 *
2677 *	This call snapshots the internal irqchip state of an
2678 *	interrupt, returning into @state the bit corresponding to
2679 *	stage @which
2680 *
2681 *	This function should be called with preemption disabled if the
2682 *	interrupt controller has per-cpu registers.
2683 */
2684int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2685			  bool *state)
2686{
2687	struct irq_desc *desc;
2688	struct irq_data *data;
2689	unsigned long flags;
2690	int err = -EINVAL;
2691
2692	desc = irq_get_desc_buslock(irq, &flags, 0);
2693	if (!desc)
2694		return err;
2695
2696	data = irq_desc_get_irq_data(desc);
2697
2698	err = __irq_get_irqchip_state(data, which, state);
2699
2700	irq_put_desc_busunlock(desc, flags);
2701	return err;
2702}
2703EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2704
2705/**
2706 *	irq_set_irqchip_state - set the state of a forwarded interrupt.
2707 *	@irq: Interrupt line that is forwarded to a VM
2708 *	@which: State to be restored (one of IRQCHIP_STATE_*)
2709 *	@val: Value corresponding to @which
2710 *
2711 *	This call sets the internal irqchip state of an interrupt,
2712 *	depending on the value of @which.
2713 *
2714 *	This function should be called with preemption disabled if the
2715 *	interrupt controller has per-cpu registers.
2716 */
2717int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2718			  bool val)
2719{
2720	struct irq_desc *desc;
2721	struct irq_data *data;
2722	struct irq_chip *chip;
2723	unsigned long flags;
2724	int err = -EINVAL;
2725
2726	desc = irq_get_desc_buslock(irq, &flags, 0);
2727	if (!desc)
2728		return err;
2729
2730	data = irq_desc_get_irq_data(desc);
2731
2732	do {
2733		chip = irq_data_get_irq_chip(data);
2734		if (WARN_ON_ONCE(!chip)) {
2735			err = -ENODEV;
2736			goto out_unlock;
2737		}
2738		if (chip->irq_set_irqchip_state)
2739			break;
2740#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2741		data = data->parent_data;
2742#else
2743		data = NULL;
2744#endif
2745	} while (data);
2746
2747	if (data)
2748		err = chip->irq_set_irqchip_state(data, which, val);
2749
2750out_unlock:
2751	irq_put_desc_busunlock(desc, flags);
2752	return err;
2753}
2754EXPORT_SYMBOL_GPL(irq_set_irqchip_state);