Linux Audio

Check our new training course

Loading...
v6.8
   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
 111static void __synchronize_irq(struct irq_desc *desc)
 112{
 113	__synchronize_hardirq(desc, true);
 114	/*
 115	 * We made sure that no hardirq handler is running. Now verify that no
 116	 * threaded handlers are active.
 117	 */
 118	wait_event(desc->wait_for_threads, !atomic_read(&desc->threads_active));
 119}
 120
 121/**
 122 *	synchronize_irq - wait for pending IRQ handlers (on other CPUs)
 123 *	@irq: interrupt number to wait for
 124 *
 125 *	This function waits for any pending IRQ handlers for this interrupt
 126 *	to complete before returning. If you use this function while
 127 *	holding a resource the IRQ handler may need you will deadlock.
 128 *
 129 *	Can only be called from preemptible code as it might sleep when
 130 *	an interrupt thread is associated to @irq.
 131 *
 132 *	It optionally makes sure (when the irq chip supports that method)
 133 *	that the interrupt is not pending in any CPU and waiting for
 134 *	service.
 135 */
 136void synchronize_irq(unsigned int irq)
 137{
 138	struct irq_desc *desc = irq_to_desc(irq);
 139
 140	if (desc)
 141		__synchronize_irq(desc);
 
 
 
 
 
 
 
 
 142}
 143EXPORT_SYMBOL(synchronize_irq);
 144
 145#ifdef CONFIG_SMP
 146cpumask_var_t irq_default_affinity;
 147
 148static bool __irq_can_set_affinity(struct irq_desc *desc)
 149{
 150	if (!desc || !irqd_can_balance(&desc->irq_data) ||
 151	    !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
 152		return false;
 153	return true;
 154}
 155
 156/**
 157 *	irq_can_set_affinity - Check if the affinity of a given irq can be set
 158 *	@irq:		Interrupt to check
 159 *
 160 */
 161int irq_can_set_affinity(unsigned int irq)
 162{
 163	return __irq_can_set_affinity(irq_to_desc(irq));
 164}
 165
 166/**
 167 * irq_can_set_affinity_usr - Check if affinity of a irq can be set from user space
 168 * @irq:	Interrupt to check
 169 *
 170 * Like irq_can_set_affinity() above, but additionally checks for the
 171 * AFFINITY_MANAGED flag.
 172 */
 173bool irq_can_set_affinity_usr(unsigned int irq)
 174{
 175	struct irq_desc *desc = irq_to_desc(irq);
 176
 177	return __irq_can_set_affinity(desc) &&
 178		!irqd_affinity_is_managed(&desc->irq_data);
 179}
 180
 181/**
 182 *	irq_set_thread_affinity - Notify irq threads to adjust affinity
 183 *	@desc:		irq descriptor which has affinity changed
 184 *
 185 *	We just set IRQTF_AFFINITY and delegate the affinity setting
 186 *	to the interrupt thread itself. We can not call
 187 *	set_cpus_allowed_ptr() here as we hold desc->lock and this
 188 *	code can be called from hard interrupt context.
 189 */
 190void irq_set_thread_affinity(struct irq_desc *desc)
 191{
 192	struct irqaction *action;
 193
 194	for_each_action_of_desc(desc, action) {
 195		if (action->thread)
 196			set_bit(IRQTF_AFFINITY, &action->thread_flags);
 197		if (action->secondary && action->secondary->thread)
 198			set_bit(IRQTF_AFFINITY, &action->secondary->thread_flags);
 199	}
 200}
 201
 202#ifdef CONFIG_GENERIC_IRQ_EFFECTIVE_AFF_MASK
 203static void irq_validate_effective_affinity(struct irq_data *data)
 204{
 205	const struct cpumask *m = irq_data_get_effective_affinity_mask(data);
 206	struct irq_chip *chip = irq_data_get_irq_chip(data);
 207
 208	if (!cpumask_empty(m))
 209		return;
 210	pr_warn_once("irq_chip %s did not update eff. affinity mask of irq %u\n",
 211		     chip->name, data->irq);
 212}
 
 
 
 
 
 
 213#else
 214static inline void irq_validate_effective_affinity(struct irq_data *data) { }
 
 
 215#endif
 216
 217int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
 218			bool force)
 219{
 220	struct irq_desc *desc = irq_data_to_desc(data);
 221	struct irq_chip *chip = irq_data_get_irq_chip(data);
 222	const struct cpumask  *prog_mask;
 223	int ret;
 224
 225	static DEFINE_RAW_SPINLOCK(tmp_mask_lock);
 226	static struct cpumask tmp_mask;
 227
 228	if (!chip || !chip->irq_set_affinity)
 229		return -EINVAL;
 230
 231	raw_spin_lock(&tmp_mask_lock);
 232	/*
 233	 * If this is a managed interrupt and housekeeping is enabled on
 234	 * it check whether the requested affinity mask intersects with
 235	 * a housekeeping CPU. If so, then remove the isolated CPUs from
 236	 * the mask and just keep the housekeeping CPU(s). This prevents
 237	 * the affinity setter from routing the interrupt to an isolated
 238	 * CPU to avoid that I/O submitted from a housekeeping CPU causes
 239	 * interrupts on an isolated one.
 240	 *
 241	 * If the masks do not intersect or include online CPU(s) then
 242	 * keep the requested mask. The isolated target CPUs are only
 243	 * receiving interrupts when the I/O operation was submitted
 244	 * directly from them.
 245	 *
 246	 * If all housekeeping CPUs in the affinity mask are offline, the
 247	 * interrupt will be migrated by the CPU hotplug code once a
 248	 * housekeeping CPU which belongs to the affinity mask comes
 249	 * online.
 250	 */
 251	if (irqd_affinity_is_managed(data) &&
 252	    housekeeping_enabled(HK_TYPE_MANAGED_IRQ)) {
 253		const struct cpumask *hk_mask;
 254
 255		hk_mask = housekeeping_cpumask(HK_TYPE_MANAGED_IRQ);
 
 256
 
 
 
 257		cpumask_and(&tmp_mask, mask, hk_mask);
 258		if (!cpumask_intersects(&tmp_mask, cpu_online_mask))
 259			prog_mask = mask;
 260		else
 261			prog_mask = &tmp_mask;
 
 
 262	} else {
 263		prog_mask = mask;
 264	}
 265
 266	/*
 267	 * Make sure we only provide online CPUs to the irqchip,
 268	 * unless we are being asked to force the affinity (in which
 269	 * case we do as we are told).
 270	 */
 271	cpumask_and(&tmp_mask, prog_mask, cpu_online_mask);
 272	if (!force && !cpumask_empty(&tmp_mask))
 273		ret = chip->irq_set_affinity(data, &tmp_mask, force);
 274	else if (force)
 275		ret = chip->irq_set_affinity(data, mask, force);
 276	else
 277		ret = -EINVAL;
 278
 279	raw_spin_unlock(&tmp_mask_lock);
 280
 281	switch (ret) {
 282	case IRQ_SET_MASK_OK:
 283	case IRQ_SET_MASK_OK_DONE:
 284		cpumask_copy(desc->irq_common_data.affinity, mask);
 285		fallthrough;
 286	case IRQ_SET_MASK_OK_NOCOPY:
 287		irq_validate_effective_affinity(data);
 288		irq_set_thread_affinity(desc);
 289		ret = 0;
 290	}
 291
 292	return ret;
 293}
 294
 295#ifdef CONFIG_GENERIC_PENDING_IRQ
 296static inline int irq_set_affinity_pending(struct irq_data *data,
 297					   const struct cpumask *dest)
 298{
 299	struct irq_desc *desc = irq_data_to_desc(data);
 300
 301	irqd_set_move_pending(data);
 302	irq_copy_pending(desc, dest);
 303	return 0;
 304}
 305#else
 306static inline int irq_set_affinity_pending(struct irq_data *data,
 307					   const struct cpumask *dest)
 308{
 309	return -EBUSY;
 310}
 311#endif
 312
 313static int irq_try_set_affinity(struct irq_data *data,
 314				const struct cpumask *dest, bool force)
 315{
 316	int ret = irq_do_set_affinity(data, dest, force);
 317
 318	/*
 319	 * In case that the underlying vector management is busy and the
 320	 * architecture supports the generic pending mechanism then utilize
 321	 * this to avoid returning an error to user space.
 322	 */
 323	if (ret == -EBUSY && !force)
 324		ret = irq_set_affinity_pending(data, dest);
 325	return ret;
 326}
 327
 328static bool irq_set_affinity_deactivated(struct irq_data *data,
 329					 const struct cpumask *mask)
 330{
 331	struct irq_desc *desc = irq_data_to_desc(data);
 332
 333	/*
 334	 * Handle irq chips which can handle affinity only in activated
 335	 * state correctly
 336	 *
 337	 * If the interrupt is not yet activated, just store the affinity
 338	 * mask and do not call the chip driver at all. On activation the
 339	 * driver has to make sure anyway that the interrupt is in a
 340	 * usable state so startup works.
 341	 */
 342	if (!IS_ENABLED(CONFIG_IRQ_DOMAIN_HIERARCHY) ||
 343	    irqd_is_activated(data) || !irqd_affinity_on_activate(data))
 344		return false;
 345
 346	cpumask_copy(desc->irq_common_data.affinity, mask);
 347	irq_data_update_effective_affinity(data, mask);
 348	irqd_set(data, IRQD_AFFINITY_SET);
 349	return true;
 350}
 351
 352int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 353			    bool force)
 354{
 355	struct irq_chip *chip = irq_data_get_irq_chip(data);
 356	struct irq_desc *desc = irq_data_to_desc(data);
 357	int ret = 0;
 358
 359	if (!chip || !chip->irq_set_affinity)
 360		return -EINVAL;
 361
 362	if (irq_set_affinity_deactivated(data, mask))
 363		return 0;
 364
 365	if (irq_can_move_pcntxt(data) && !irqd_is_setaffinity_pending(data)) {
 366		ret = irq_try_set_affinity(data, mask, force);
 367	} else {
 368		irqd_set_move_pending(data);
 369		irq_copy_pending(desc, mask);
 370	}
 371
 372	if (desc->affinity_notify) {
 373		kref_get(&desc->affinity_notify->kref);
 374		if (!schedule_work(&desc->affinity_notify->work)) {
 375			/* Work was already scheduled, drop our extra ref */
 376			kref_put(&desc->affinity_notify->kref,
 377				 desc->affinity_notify->release);
 378		}
 379	}
 380	irqd_set(data, IRQD_AFFINITY_SET);
 381
 382	return ret;
 383}
 384
 385/**
 386 * irq_update_affinity_desc - Update affinity management for an interrupt
 387 * @irq:	The interrupt number to update
 388 * @affinity:	Pointer to the affinity descriptor
 389 *
 390 * This interface can be used to configure the affinity management of
 391 * interrupts which have been allocated already.
 392 *
 393 * There are certain limitations on when it may be used - attempts to use it
 394 * for when the kernel is configured for generic IRQ reservation mode (in
 395 * config GENERIC_IRQ_RESERVATION_MODE) will fail, as it may conflict with
 396 * managed/non-managed interrupt accounting. In addition, attempts to use it on
 397 * an interrupt which is already started or which has already been configured
 398 * as managed will also fail, as these mean invalid init state or double init.
 399 */
 400int irq_update_affinity_desc(unsigned int irq,
 401			     struct irq_affinity_desc *affinity)
 402{
 403	struct irq_desc *desc;
 404	unsigned long flags;
 405	bool activated;
 406	int ret = 0;
 407
 408	/*
 409	 * Supporting this with the reservation scheme used by x86 needs
 410	 * some more thought. Fail it for now.
 411	 */
 412	if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
 413		return -EOPNOTSUPP;
 414
 415	desc = irq_get_desc_buslock(irq, &flags, 0);
 416	if (!desc)
 417		return -EINVAL;
 418
 419	/* Requires the interrupt to be shut down */
 420	if (irqd_is_started(&desc->irq_data)) {
 421		ret = -EBUSY;
 422		goto out_unlock;
 423	}
 424
 425	/* Interrupts which are already managed cannot be modified */
 426	if (irqd_affinity_is_managed(&desc->irq_data)) {
 427		ret = -EBUSY;
 428		goto out_unlock;
 429	}
 430
 431	/*
 432	 * Deactivate the interrupt. That's required to undo
 433	 * anything an earlier activation has established.
 434	 */
 435	activated = irqd_is_activated(&desc->irq_data);
 436	if (activated)
 437		irq_domain_deactivate_irq(&desc->irq_data);
 438
 439	if (affinity->is_managed) {
 440		irqd_set(&desc->irq_data, IRQD_AFFINITY_MANAGED);
 441		irqd_set(&desc->irq_data, IRQD_MANAGED_SHUTDOWN);
 442	}
 443
 444	cpumask_copy(desc->irq_common_data.affinity, &affinity->mask);
 445
 446	/* Restore the activation state */
 447	if (activated)
 448		irq_domain_activate_irq(&desc->irq_data, false);
 449
 450out_unlock:
 451	irq_put_desc_busunlock(desc, flags);
 452	return ret;
 453}
 454
 455static int __irq_set_affinity(unsigned int irq, const struct cpumask *mask,
 456			      bool force)
 457{
 458	struct irq_desc *desc = irq_to_desc(irq);
 459	unsigned long flags;
 460	int ret;
 461
 462	if (!desc)
 463		return -EINVAL;
 464
 465	raw_spin_lock_irqsave(&desc->lock, flags);
 466	ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
 467	raw_spin_unlock_irqrestore(&desc->lock, flags);
 468	return ret;
 469}
 470
 471/**
 472 * irq_set_affinity - Set the irq affinity of a given irq
 473 * @irq:	Interrupt to set affinity
 474 * @cpumask:	cpumask
 475 *
 476 * Fails if cpumask does not contain an online CPU
 477 */
 478int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
 479{
 480	return __irq_set_affinity(irq, cpumask, false);
 481}
 482EXPORT_SYMBOL_GPL(irq_set_affinity);
 483
 484/**
 485 * irq_force_affinity - Force the irq affinity of a given irq
 486 * @irq:	Interrupt to set affinity
 487 * @cpumask:	cpumask
 488 *
 489 * Same as irq_set_affinity, but without checking the mask against
 490 * online cpus.
 491 *
 492 * Solely for low level cpu hotplug code, where we need to make per
 493 * cpu interrupts affine before the cpu becomes online.
 494 */
 495int irq_force_affinity(unsigned int irq, const struct cpumask *cpumask)
 496{
 497	return __irq_set_affinity(irq, cpumask, true);
 498}
 499EXPORT_SYMBOL_GPL(irq_force_affinity);
 500
 501int __irq_apply_affinity_hint(unsigned int irq, const struct cpumask *m,
 502			      bool setaffinity)
 503{
 504	unsigned long flags;
 505	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 506
 507	if (!desc)
 508		return -EINVAL;
 509	desc->affinity_hint = m;
 510	irq_put_desc_unlock(desc, flags);
 511	if (m && setaffinity)
 
 512		__irq_set_affinity(irq, m, false);
 513	return 0;
 514}
 515EXPORT_SYMBOL_GPL(__irq_apply_affinity_hint);
 516
 517static void irq_affinity_notify(struct work_struct *work)
 518{
 519	struct irq_affinity_notify *notify =
 520		container_of(work, struct irq_affinity_notify, work);
 521	struct irq_desc *desc = irq_to_desc(notify->irq);
 522	cpumask_var_t cpumask;
 523	unsigned long flags;
 524
 525	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
 526		goto out;
 527
 528	raw_spin_lock_irqsave(&desc->lock, flags);
 529	if (irq_move_pending(&desc->irq_data))
 530		irq_get_pending(cpumask, desc);
 531	else
 532		cpumask_copy(cpumask, desc->irq_common_data.affinity);
 533	raw_spin_unlock_irqrestore(&desc->lock, flags);
 534
 535	notify->notify(notify, cpumask);
 536
 537	free_cpumask_var(cpumask);
 538out:
 539	kref_put(&notify->kref, notify->release);
 540}
 541
 542/**
 543 *	irq_set_affinity_notifier - control notification of IRQ affinity changes
 544 *	@irq:		Interrupt for which to enable/disable notification
 545 *	@notify:	Context for notification, or %NULL to disable
 546 *			notification.  Function pointers must be initialised;
 547 *			the other fields will be initialised by this function.
 548 *
 549 *	Must be called in process context.  Notification may only be enabled
 550 *	after the IRQ is allocated and must be disabled before the IRQ is
 551 *	freed using free_irq().
 552 */
 553int
 554irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 555{
 556	struct irq_desc *desc = irq_to_desc(irq);
 557	struct irq_affinity_notify *old_notify;
 558	unsigned long flags;
 559
 560	/* The release function is promised process context */
 561	might_sleep();
 562
 563	if (!desc || desc->istate & IRQS_NMI)
 564		return -EINVAL;
 565
 566	/* Complete initialisation of *notify */
 567	if (notify) {
 568		notify->irq = irq;
 569		kref_init(&notify->kref);
 570		INIT_WORK(&notify->work, irq_affinity_notify);
 571	}
 572
 573	raw_spin_lock_irqsave(&desc->lock, flags);
 574	old_notify = desc->affinity_notify;
 575	desc->affinity_notify = notify;
 576	raw_spin_unlock_irqrestore(&desc->lock, flags);
 577
 578	if (old_notify) {
 579		if (cancel_work_sync(&old_notify->work)) {
 580			/* Pending work had a ref, put that one too */
 581			kref_put(&old_notify->kref, old_notify->release);
 582		}
 583		kref_put(&old_notify->kref, old_notify->release);
 584	}
 585
 586	return 0;
 587}
 588EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
 589
 590#ifndef CONFIG_AUTO_IRQ_AFFINITY
 591/*
 592 * Generic version of the affinity autoselector.
 593 */
 594int irq_setup_affinity(struct irq_desc *desc)
 595{
 596	struct cpumask *set = irq_default_affinity;
 597	int ret, node = irq_desc_get_node(desc);
 598	static DEFINE_RAW_SPINLOCK(mask_lock);
 599	static struct cpumask mask;
 600
 601	/* Excludes PER_CPU and NO_BALANCE interrupts */
 602	if (!__irq_can_set_affinity(desc))
 603		return 0;
 604
 605	raw_spin_lock(&mask_lock);
 606	/*
 607	 * Preserve the managed affinity setting and a userspace affinity
 608	 * setup, but make sure that one of the targets is online.
 609	 */
 610	if (irqd_affinity_is_managed(&desc->irq_data) ||
 611	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 612		if (cpumask_intersects(desc->irq_common_data.affinity,
 613				       cpu_online_mask))
 614			set = desc->irq_common_data.affinity;
 615		else
 616			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
 617	}
 618
 619	cpumask_and(&mask, cpu_online_mask, set);
 620	if (cpumask_empty(&mask))
 621		cpumask_copy(&mask, cpu_online_mask);
 622
 623	if (node != NUMA_NO_NODE) {
 624		const struct cpumask *nodemask = cpumask_of_node(node);
 625
 626		/* make sure at least one of the cpus in nodemask is online */
 627		if (cpumask_intersects(&mask, nodemask))
 628			cpumask_and(&mask, &mask, nodemask);
 629	}
 630	ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
 631	raw_spin_unlock(&mask_lock);
 632	return ret;
 633}
 634#else
 635/* Wrapper for ALPHA specific affinity selector magic */
 636int irq_setup_affinity(struct irq_desc *desc)
 637{
 638	return irq_select_affinity(irq_desc_get_irq(desc));
 639}
 640#endif /* CONFIG_AUTO_IRQ_AFFINITY */
 641#endif /* CONFIG_SMP */
 642
 643
 644/**
 645 *	irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
 646 *	@irq: interrupt number to set affinity
 647 *	@vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
 648 *	            specific data for percpu_devid interrupts
 649 *
 650 *	This function uses the vCPU specific data to set the vCPU
 651 *	affinity for an irq. The vCPU specific data is passed from
 652 *	outside, such as KVM. One example code path is as below:
 653 *	KVM -> IOMMU -> irq_set_vcpu_affinity().
 654 */
 655int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
 656{
 657	unsigned long flags;
 658	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 659	struct irq_data *data;
 660	struct irq_chip *chip;
 661	int ret = -ENOSYS;
 662
 663	if (!desc)
 664		return -EINVAL;
 665
 666	data = irq_desc_get_irq_data(desc);
 667	do {
 668		chip = irq_data_get_irq_chip(data);
 669		if (chip && chip->irq_set_vcpu_affinity)
 670			break;
 671#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
 672		data = data->parent_data;
 673#else
 674		data = NULL;
 675#endif
 676	} while (data);
 677
 678	if (data)
 679		ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
 680	irq_put_desc_unlock(desc, flags);
 681
 682	return ret;
 683}
 684EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
 685
 686void __disable_irq(struct irq_desc *desc)
 687{
 688	if (!desc->depth++)
 689		irq_disable(desc);
 690}
 691
 692static int __disable_irq_nosync(unsigned int irq)
 693{
 694	unsigned long flags;
 695	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 696
 697	if (!desc)
 698		return -EINVAL;
 699	__disable_irq(desc);
 700	irq_put_desc_busunlock(desc, flags);
 701	return 0;
 702}
 703
 704/**
 705 *	disable_irq_nosync - disable an irq without waiting
 706 *	@irq: Interrupt to disable
 707 *
 708 *	Disable the selected interrupt line.  Disables and Enables are
 709 *	nested.
 710 *	Unlike disable_irq(), this function does not ensure existing
 711 *	instances of the IRQ handler have completed before returning.
 712 *
 713 *	This function may be called from IRQ context.
 714 */
 715void disable_irq_nosync(unsigned int irq)
 716{
 717	__disable_irq_nosync(irq);
 718}
 719EXPORT_SYMBOL(disable_irq_nosync);
 720
 721/**
 722 *	disable_irq - disable an irq and wait for completion
 723 *	@irq: Interrupt to disable
 724 *
 725 *	Disable the selected interrupt line.  Enables and Disables are
 726 *	nested.
 727 *	This function waits for any pending IRQ handlers for this interrupt
 728 *	to complete before returning. If you use this function while
 729 *	holding a resource the IRQ handler may need you will deadlock.
 730 *
 731 *	Can only be called from preemptible code as it might sleep when
 732 *	an interrupt thread is associated to @irq.
 733 *
 734 */
 735void disable_irq(unsigned int irq)
 736{
 737	might_sleep();
 738	if (!__disable_irq_nosync(irq))
 739		synchronize_irq(irq);
 740}
 741EXPORT_SYMBOL(disable_irq);
 742
 743/**
 744 *	disable_hardirq - disables an irq and waits for hardirq completion
 745 *	@irq: Interrupt to disable
 746 *
 747 *	Disable the selected interrupt line.  Enables and Disables are
 748 *	nested.
 749 *	This function waits for any pending hard IRQ handlers for this
 750 *	interrupt to complete before returning. If you use this function while
 751 *	holding a resource the hard IRQ handler may need you will deadlock.
 752 *
 753 *	When used to optimistically disable an interrupt from atomic context
 754 *	the return value must be checked.
 755 *
 756 *	Returns: false if a threaded handler is active.
 757 *
 758 *	This function may be called - with care - from IRQ context.
 759 */
 760bool disable_hardirq(unsigned int irq)
 761{
 762	if (!__disable_irq_nosync(irq))
 763		return synchronize_hardirq(irq);
 764
 765	return false;
 766}
 767EXPORT_SYMBOL_GPL(disable_hardirq);
 768
 769/**
 770 *	disable_nmi_nosync - disable an nmi without waiting
 771 *	@irq: Interrupt to disable
 772 *
 773 *	Disable the selected interrupt line. Disables and enables are
 774 *	nested.
 775 *	The interrupt to disable must have been requested through request_nmi.
 776 *	Unlike disable_nmi(), this function does not ensure existing
 777 *	instances of the IRQ handler have completed before returning.
 778 */
 779void disable_nmi_nosync(unsigned int irq)
 780{
 781	disable_irq_nosync(irq);
 782}
 783
 784void __enable_irq(struct irq_desc *desc)
 785{
 786	switch (desc->depth) {
 787	case 0:
 788 err_out:
 789		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
 790		     irq_desc_get_irq(desc));
 791		break;
 792	case 1: {
 793		if (desc->istate & IRQS_SUSPENDED)
 794			goto err_out;
 795		/* Prevent probing on this irq: */
 796		irq_settings_set_noprobe(desc);
 797		/*
 798		 * Call irq_startup() not irq_enable() here because the
 799		 * interrupt might be marked NOAUTOEN. So irq_startup()
 800		 * needs to be invoked when it gets enabled the first
 801		 * time. If it was already started up, then irq_startup()
 802		 * will invoke irq_enable() under the hood.
 803		 */
 804		irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
 805		break;
 806	}
 807	default:
 808		desc->depth--;
 809	}
 810}
 811
 812/**
 813 *	enable_irq - enable handling of an irq
 814 *	@irq: Interrupt to enable
 815 *
 816 *	Undoes the effect of one call to disable_irq().  If this
 817 *	matches the last disable, processing of interrupts on this
 818 *	IRQ line is re-enabled.
 819 *
 820 *	This function may be called from IRQ context only when
 821 *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
 822 */
 823void enable_irq(unsigned int irq)
 824{
 825	unsigned long flags;
 826	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 827
 828	if (!desc)
 829		return;
 830	if (WARN(!desc->irq_data.chip,
 831		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
 832		goto out;
 833
 834	__enable_irq(desc);
 835out:
 836	irq_put_desc_busunlock(desc, flags);
 837}
 838EXPORT_SYMBOL(enable_irq);
 839
 840/**
 841 *	enable_nmi - enable handling of an nmi
 842 *	@irq: Interrupt to enable
 843 *
 844 *	The interrupt to enable must have been requested through request_nmi.
 845 *	Undoes the effect of one call to disable_nmi(). If this
 846 *	matches the last disable, processing of interrupts on this
 847 *	IRQ line is re-enabled.
 848 */
 849void enable_nmi(unsigned int irq)
 850{
 851	enable_irq(irq);
 852}
 853
 854static int set_irq_wake_real(unsigned int irq, unsigned int on)
 855{
 856	struct irq_desc *desc = irq_to_desc(irq);
 857	int ret = -ENXIO;
 858
 859	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
 860		return 0;
 861
 862	if (desc->irq_data.chip->irq_set_wake)
 863		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
 864
 865	return ret;
 866}
 867
 868/**
 869 *	irq_set_irq_wake - control irq power management wakeup
 870 *	@irq:	interrupt to control
 871 *	@on:	enable/disable power management wakeup
 872 *
 873 *	Enable/disable power management wakeup mode, which is
 874 *	disabled by default.  Enables and disables must match,
 875 *	just as they match for non-wakeup mode support.
 876 *
 877 *	Wakeup mode lets this IRQ wake the system from sleep
 878 *	states like "suspend to RAM".
 879 *
 880 *	Note: irq enable/disable state is completely orthogonal
 881 *	to the enable/disable state of irq wake. An irq can be
 882 *	disabled with disable_irq() and still wake the system as
 883 *	long as the irq has wake enabled. If this does not hold,
 884 *	then the underlying irq chip and the related driver need
 885 *	to be investigated.
 886 */
 887int irq_set_irq_wake(unsigned int irq, unsigned int on)
 888{
 889	unsigned long flags;
 890	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 891	int ret = 0;
 892
 893	if (!desc)
 894		return -EINVAL;
 895
 896	/* Don't use NMIs as wake up interrupts please */
 897	if (desc->istate & IRQS_NMI) {
 898		ret = -EINVAL;
 899		goto out_unlock;
 900	}
 901
 902	/* wakeup-capable irqs can be shared between drivers that
 903	 * don't need to have the same sleep mode behaviors.
 904	 */
 905	if (on) {
 906		if (desc->wake_depth++ == 0) {
 907			ret = set_irq_wake_real(irq, on);
 908			if (ret)
 909				desc->wake_depth = 0;
 910			else
 911				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
 912		}
 913	} else {
 914		if (desc->wake_depth == 0) {
 915			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
 916		} else if (--desc->wake_depth == 0) {
 917			ret = set_irq_wake_real(irq, on);
 918			if (ret)
 919				desc->wake_depth = 1;
 920			else
 921				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
 922		}
 923	}
 924
 925out_unlock:
 926	irq_put_desc_busunlock(desc, flags);
 927	return ret;
 928}
 929EXPORT_SYMBOL(irq_set_irq_wake);
 930
 931/*
 932 * Internal function that tells the architecture code whether a
 933 * particular irq has been exclusively allocated or is available
 934 * for driver use.
 935 */
 936int can_request_irq(unsigned int irq, unsigned long irqflags)
 937{
 938	unsigned long flags;
 939	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 940	int canrequest = 0;
 941
 942	if (!desc)
 943		return 0;
 944
 945	if (irq_settings_can_request(desc)) {
 946		if (!desc->action ||
 947		    irqflags & desc->action->flags & IRQF_SHARED)
 948			canrequest = 1;
 949	}
 950	irq_put_desc_unlock(desc, flags);
 951	return canrequest;
 952}
 953
 954int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
 955{
 956	struct irq_chip *chip = desc->irq_data.chip;
 957	int ret, unmask = 0;
 958
 959	if (!chip || !chip->irq_set_type) {
 960		/*
 961		 * IRQF_TRIGGER_* but the PIC does not support multiple
 962		 * flow-types?
 963		 */
 964		pr_debug("No set_type function for IRQ %d (%s)\n",
 965			 irq_desc_get_irq(desc),
 966			 chip ? (chip->name ? : "unknown") : "unknown");
 967		return 0;
 968	}
 969
 970	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
 971		if (!irqd_irq_masked(&desc->irq_data))
 972			mask_irq(desc);
 973		if (!irqd_irq_disabled(&desc->irq_data))
 974			unmask = 1;
 975	}
 976
 977	/* Mask all flags except trigger mode */
 978	flags &= IRQ_TYPE_SENSE_MASK;
 979	ret = chip->irq_set_type(&desc->irq_data, flags);
 980
 981	switch (ret) {
 982	case IRQ_SET_MASK_OK:
 983	case IRQ_SET_MASK_OK_DONE:
 984		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
 985		irqd_set(&desc->irq_data, flags);
 986		fallthrough;
 987
 988	case IRQ_SET_MASK_OK_NOCOPY:
 989		flags = irqd_get_trigger_type(&desc->irq_data);
 990		irq_settings_set_trigger_mask(desc, flags);
 991		irqd_clear(&desc->irq_data, IRQD_LEVEL);
 992		irq_settings_clr_level(desc);
 993		if (flags & IRQ_TYPE_LEVEL_MASK) {
 994			irq_settings_set_level(desc);
 995			irqd_set(&desc->irq_data, IRQD_LEVEL);
 996		}
 997
 998		ret = 0;
 999		break;
1000	default:
1001		pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
1002		       flags, irq_desc_get_irq(desc), chip->irq_set_type);
1003	}
1004	if (unmask)
1005		unmask_irq(desc);
1006	return ret;
1007}
1008
1009#ifdef CONFIG_HARDIRQS_SW_RESEND
1010int irq_set_parent(int irq, int parent_irq)
1011{
1012	unsigned long flags;
1013	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
1014
1015	if (!desc)
1016		return -EINVAL;
1017
1018	desc->parent_irq = parent_irq;
1019
1020	irq_put_desc_unlock(desc, flags);
1021	return 0;
1022}
1023EXPORT_SYMBOL_GPL(irq_set_parent);
1024#endif
1025
1026/*
1027 * Default primary interrupt handler for threaded interrupts. Is
1028 * assigned as primary handler when request_threaded_irq is called
1029 * with handler == NULL. Useful for oneshot interrupts.
1030 */
1031static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
1032{
1033	return IRQ_WAKE_THREAD;
1034}
1035
1036/*
1037 * Primary handler for nested threaded interrupts. Should never be
1038 * called.
1039 */
1040static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
1041{
1042	WARN(1, "Primary handler called for nested irq %d\n", irq);
1043	return IRQ_NONE;
1044}
1045
1046static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
1047{
1048	WARN(1, "Secondary action handler called for irq %d\n", irq);
1049	return IRQ_NONE;
1050}
1051
1052static int irq_wait_for_interrupt(struct irqaction *action)
1053{
1054	for (;;) {
1055		set_current_state(TASK_INTERRUPTIBLE);
1056
1057		if (kthread_should_stop()) {
1058			/* may need to run one last time */
1059			if (test_and_clear_bit(IRQTF_RUNTHREAD,
1060					       &action->thread_flags)) {
1061				__set_current_state(TASK_RUNNING);
1062				return 0;
1063			}
1064			__set_current_state(TASK_RUNNING);
1065			return -1;
1066		}
1067
1068		if (test_and_clear_bit(IRQTF_RUNTHREAD,
1069				       &action->thread_flags)) {
1070			__set_current_state(TASK_RUNNING);
1071			return 0;
1072		}
1073		schedule();
1074	}
1075}
1076
1077/*
1078 * Oneshot interrupts keep the irq line masked until the threaded
1079 * handler finished. unmask if the interrupt has not been disabled and
1080 * is marked MASKED.
1081 */
1082static void irq_finalize_oneshot(struct irq_desc *desc,
1083				 struct irqaction *action)
1084{
1085	if (!(desc->istate & IRQS_ONESHOT) ||
1086	    action->handler == irq_forced_secondary_handler)
1087		return;
1088again:
1089	chip_bus_lock(desc);
1090	raw_spin_lock_irq(&desc->lock);
1091
1092	/*
1093	 * Implausible though it may be we need to protect us against
1094	 * the following scenario:
1095	 *
1096	 * The thread is faster done than the hard interrupt handler
1097	 * on the other CPU. If we unmask the irq line then the
1098	 * interrupt can come in again and masks the line, leaves due
1099	 * to IRQS_INPROGRESS and the irq line is masked forever.
1100	 *
1101	 * This also serializes the state of shared oneshot handlers
1102	 * versus "desc->threads_oneshot |= action->thread_mask;" in
1103	 * irq_wake_thread(). See the comment there which explains the
1104	 * serialization.
1105	 */
1106	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
1107		raw_spin_unlock_irq(&desc->lock);
1108		chip_bus_sync_unlock(desc);
1109		cpu_relax();
1110		goto again;
1111	}
1112
1113	/*
1114	 * Now check again, whether the thread should run. Otherwise
1115	 * we would clear the threads_oneshot bit of this thread which
1116	 * was just set.
1117	 */
1118	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1119		goto out_unlock;
1120
1121	desc->threads_oneshot &= ~action->thread_mask;
1122
1123	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
1124	    irqd_irq_masked(&desc->irq_data))
1125		unmask_threaded_irq(desc);
1126
1127out_unlock:
1128	raw_spin_unlock_irq(&desc->lock);
1129	chip_bus_sync_unlock(desc);
1130}
1131
1132#ifdef CONFIG_SMP
1133/*
1134 * Check whether we need to change the affinity of the interrupt thread.
1135 */
1136static void
1137irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
1138{
1139	cpumask_var_t mask;
1140	bool valid = true;
1141
1142	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
1143		return;
1144
1145	/*
1146	 * In case we are out of memory we set IRQTF_AFFINITY again and
1147	 * try again next time
1148	 */
1149	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1150		set_bit(IRQTF_AFFINITY, &action->thread_flags);
1151		return;
1152	}
1153
1154	raw_spin_lock_irq(&desc->lock);
1155	/*
1156	 * This code is triggered unconditionally. Check the affinity
1157	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
1158	 */
1159	if (cpumask_available(desc->irq_common_data.affinity)) {
1160		const struct cpumask *m;
1161
1162		m = irq_data_get_effective_affinity_mask(&desc->irq_data);
1163		cpumask_copy(mask, m);
1164	} else {
1165		valid = false;
1166	}
1167	raw_spin_unlock_irq(&desc->lock);
1168
1169	if (valid)
1170		set_cpus_allowed_ptr(current, mask);
1171	free_cpumask_var(mask);
1172}
1173#else
1174static inline void
1175irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
1176#endif
1177
1178/*
1179 * Interrupts which are not explicitly requested as threaded
1180 * interrupts rely on the implicit bh/preempt disable of the hard irq
1181 * context. So we need to disable bh here to avoid deadlocks and other
1182 * side effects.
1183 */
1184static irqreturn_t
1185irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
1186{
1187	irqreturn_t ret;
1188
1189	local_bh_disable();
1190	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1191		local_irq_disable();
1192	ret = action->thread_fn(action->irq, action->dev_id);
1193	if (ret == IRQ_HANDLED)
1194		atomic_inc(&desc->threads_handled);
1195
1196	irq_finalize_oneshot(desc, action);
1197	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1198		local_irq_enable();
1199	local_bh_enable();
1200	return ret;
1201}
1202
1203/*
1204 * Interrupts explicitly requested as threaded interrupts want to be
1205 * preemptible - many of them need to sleep and wait for slow busses to
1206 * complete.
1207 */
1208static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1209		struct irqaction *action)
1210{
1211	irqreturn_t ret;
1212
1213	ret = action->thread_fn(action->irq, action->dev_id);
1214	if (ret == IRQ_HANDLED)
1215		atomic_inc(&desc->threads_handled);
1216
1217	irq_finalize_oneshot(desc, action);
1218	return ret;
1219}
1220
1221void wake_threads_waitq(struct irq_desc *desc)
1222{
1223	if (atomic_dec_and_test(&desc->threads_active))
1224		wake_up(&desc->wait_for_threads);
1225}
1226
1227static void irq_thread_dtor(struct callback_head *unused)
1228{
1229	struct task_struct *tsk = current;
1230	struct irq_desc *desc;
1231	struct irqaction *action;
1232
1233	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1234		return;
1235
1236	action = kthread_data(tsk);
1237
1238	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1239	       tsk->comm, tsk->pid, action->irq);
1240
1241
1242	desc = irq_to_desc(action->irq);
1243	/*
1244	 * If IRQTF_RUNTHREAD is set, we need to decrement
1245	 * desc->threads_active and wake possible waiters.
1246	 */
1247	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1248		wake_threads_waitq(desc);
1249
1250	/* Prevent a stale desc->threads_oneshot */
1251	irq_finalize_oneshot(desc, action);
1252}
1253
1254static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1255{
1256	struct irqaction *secondary = action->secondary;
1257
1258	if (WARN_ON_ONCE(!secondary))
1259		return;
1260
1261	raw_spin_lock_irq(&desc->lock);
1262	__irq_wake_thread(desc, secondary);
1263	raw_spin_unlock_irq(&desc->lock);
1264}
1265
1266/*
1267 * Internal function to notify that a interrupt thread is ready.
1268 */
1269static void irq_thread_set_ready(struct irq_desc *desc,
1270				 struct irqaction *action)
1271{
1272	set_bit(IRQTF_READY, &action->thread_flags);
1273	wake_up(&desc->wait_for_threads);
1274}
1275
1276/*
1277 * Internal function to wake up a interrupt thread and wait until it is
1278 * ready.
1279 */
1280static void wake_up_and_wait_for_irq_thread_ready(struct irq_desc *desc,
1281						  struct irqaction *action)
1282{
1283	if (!action || !action->thread)
1284		return;
1285
1286	wake_up_process(action->thread);
1287	wait_event(desc->wait_for_threads,
1288		   test_bit(IRQTF_READY, &action->thread_flags));
1289}
1290
1291/*
1292 * Interrupt handler thread
1293 */
1294static int irq_thread(void *data)
1295{
1296	struct callback_head on_exit_work;
1297	struct irqaction *action = data;
1298	struct irq_desc *desc = irq_to_desc(action->irq);
1299	irqreturn_t (*handler_fn)(struct irq_desc *desc,
1300			struct irqaction *action);
1301
1302	irq_thread_set_ready(desc, action);
1303
1304	sched_set_fifo(current);
1305
1306	if (force_irqthreads() && test_bit(IRQTF_FORCED_THREAD,
1307					   &action->thread_flags))
1308		handler_fn = irq_forced_thread_fn;
1309	else
1310		handler_fn = irq_thread_fn;
1311
1312	init_task_work(&on_exit_work, irq_thread_dtor);
1313	task_work_add(current, &on_exit_work, TWA_NONE);
1314
1315	irq_thread_check_affinity(desc, action);
1316
1317	while (!irq_wait_for_interrupt(action)) {
1318		irqreturn_t action_ret;
1319
1320		irq_thread_check_affinity(desc, action);
1321
1322		action_ret = handler_fn(desc, action);
1323		if (action_ret == IRQ_WAKE_THREAD)
1324			irq_wake_secondary(desc, action);
1325
1326		wake_threads_waitq(desc);
1327	}
1328
1329	/*
1330	 * This is the regular exit path. __free_irq() is stopping the
1331	 * thread via kthread_stop() after calling
1332	 * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1333	 * oneshot mask bit can be set.
1334	 */
1335	task_work_cancel(current, irq_thread_dtor);
1336	return 0;
1337}
1338
1339/**
1340 *	irq_wake_thread - wake the irq thread for the action identified by dev_id
1341 *	@irq:		Interrupt line
1342 *	@dev_id:	Device identity for which the thread should be woken
1343 *
1344 */
1345void irq_wake_thread(unsigned int irq, void *dev_id)
1346{
1347	struct irq_desc *desc = irq_to_desc(irq);
1348	struct irqaction *action;
1349	unsigned long flags;
1350
1351	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1352		return;
1353
1354	raw_spin_lock_irqsave(&desc->lock, flags);
1355	for_each_action_of_desc(desc, action) {
1356		if (action->dev_id == dev_id) {
1357			if (action->thread)
1358				__irq_wake_thread(desc, action);
1359			break;
1360		}
1361	}
1362	raw_spin_unlock_irqrestore(&desc->lock, flags);
1363}
1364EXPORT_SYMBOL_GPL(irq_wake_thread);
1365
1366static int irq_setup_forced_threading(struct irqaction *new)
1367{
1368	if (!force_irqthreads())
1369		return 0;
1370	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1371		return 0;
1372
1373	/*
1374	 * No further action required for interrupts which are requested as
1375	 * threaded interrupts already
1376	 */
1377	if (new->handler == irq_default_primary_handler)
1378		return 0;
1379
1380	new->flags |= IRQF_ONESHOT;
1381
1382	/*
1383	 * Handle the case where we have a real primary handler and a
1384	 * thread handler. We force thread them as well by creating a
1385	 * secondary action.
1386	 */
1387	if (new->handler && new->thread_fn) {
1388		/* Allocate the secondary action */
1389		new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1390		if (!new->secondary)
1391			return -ENOMEM;
1392		new->secondary->handler = irq_forced_secondary_handler;
1393		new->secondary->thread_fn = new->thread_fn;
1394		new->secondary->dev_id = new->dev_id;
1395		new->secondary->irq = new->irq;
1396		new->secondary->name = new->name;
1397	}
1398	/* Deal with the primary handler */
1399	set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1400	new->thread_fn = new->handler;
1401	new->handler = irq_default_primary_handler;
1402	return 0;
1403}
1404
1405static int irq_request_resources(struct irq_desc *desc)
1406{
1407	struct irq_data *d = &desc->irq_data;
1408	struct irq_chip *c = d->chip;
1409
1410	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1411}
1412
1413static void irq_release_resources(struct irq_desc *desc)
1414{
1415	struct irq_data *d = &desc->irq_data;
1416	struct irq_chip *c = d->chip;
1417
1418	if (c->irq_release_resources)
1419		c->irq_release_resources(d);
1420}
1421
1422static bool irq_supports_nmi(struct irq_desc *desc)
1423{
1424	struct irq_data *d = irq_desc_get_irq_data(desc);
1425
1426#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1427	/* Only IRQs directly managed by the root irqchip can be set as NMI */
1428	if (d->parent_data)
1429		return false;
1430#endif
1431	/* Don't support NMIs for chips behind a slow bus */
1432	if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1433		return false;
1434
1435	return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1436}
1437
1438static int irq_nmi_setup(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	return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1444}
1445
1446static void irq_nmi_teardown(struct irq_desc *desc)
1447{
1448	struct irq_data *d = irq_desc_get_irq_data(desc);
1449	struct irq_chip *c = d->chip;
1450
1451	if (c->irq_nmi_teardown)
1452		c->irq_nmi_teardown(d);
1453}
1454
1455static int
1456setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1457{
1458	struct task_struct *t;
1459
1460	if (!secondary) {
1461		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1462				   new->name);
1463	} else {
1464		t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1465				   new->name);
1466	}
1467
1468	if (IS_ERR(t))
1469		return PTR_ERR(t);
1470
 
 
1471	/*
1472	 * We keep the reference to the task struct even if
1473	 * the thread dies to avoid that the interrupt code
1474	 * references an already freed task_struct.
1475	 */
1476	new->thread = get_task_struct(t);
1477	/*
1478	 * Tell the thread to set its affinity. This is
1479	 * important for shared interrupt handlers as we do
1480	 * not invoke setup_affinity() for the secondary
1481	 * handlers as everything is already set up. Even for
1482	 * interrupts marked with IRQF_NO_BALANCE this is
1483	 * correct as we want the thread to move to the cpu(s)
1484	 * on which the requesting code placed the interrupt.
1485	 */
1486	set_bit(IRQTF_AFFINITY, &new->thread_flags);
1487	return 0;
1488}
1489
1490/*
1491 * Internal function to register an irqaction - typically used to
1492 * allocate special interrupts that are part of the architecture.
1493 *
1494 * Locking rules:
1495 *
1496 * desc->request_mutex	Provides serialization against a concurrent free_irq()
1497 *   chip_bus_lock	Provides serialization for slow bus operations
1498 *     desc->lock	Provides serialization against hard interrupts
1499 *
1500 * chip_bus_lock and desc->lock are sufficient for all other management and
1501 * interrupt related functions. desc->request_mutex solely serializes
1502 * request/free_irq().
1503 */
1504static int
1505__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1506{
1507	struct irqaction *old, **old_ptr;
1508	unsigned long flags, thread_mask = 0;
1509	int ret, nested, shared = 0;
1510
1511	if (!desc)
1512		return -EINVAL;
1513
1514	if (desc->irq_data.chip == &no_irq_chip)
1515		return -ENOSYS;
1516	if (!try_module_get(desc->owner))
1517		return -ENODEV;
1518
1519	new->irq = irq;
1520
1521	/*
1522	 * If the trigger type is not specified by the caller,
1523	 * then use the default for this interrupt.
1524	 */
1525	if (!(new->flags & IRQF_TRIGGER_MASK))
1526		new->flags |= irqd_get_trigger_type(&desc->irq_data);
1527
1528	/*
1529	 * Check whether the interrupt nests into another interrupt
1530	 * thread.
1531	 */
1532	nested = irq_settings_is_nested_thread(desc);
1533	if (nested) {
1534		if (!new->thread_fn) {
1535			ret = -EINVAL;
1536			goto out_mput;
1537		}
1538		/*
1539		 * Replace the primary handler which was provided from
1540		 * the driver for non nested interrupt handling by the
1541		 * dummy function which warns when called.
1542		 */
1543		new->handler = irq_nested_primary_handler;
1544	} else {
1545		if (irq_settings_can_thread(desc)) {
1546			ret = irq_setup_forced_threading(new);
1547			if (ret)
1548				goto out_mput;
1549		}
1550	}
1551
1552	/*
1553	 * Create a handler thread when a thread function is supplied
1554	 * and the interrupt does not nest into another interrupt
1555	 * thread.
1556	 */
1557	if (new->thread_fn && !nested) {
1558		ret = setup_irq_thread(new, irq, false);
1559		if (ret)
1560			goto out_mput;
1561		if (new->secondary) {
1562			ret = setup_irq_thread(new->secondary, irq, true);
1563			if (ret)
1564				goto out_thread;
1565		}
1566	}
1567
1568	/*
1569	 * Drivers are often written to work w/o knowledge about the
1570	 * underlying irq chip implementation, so a request for a
1571	 * threaded irq without a primary hard irq context handler
1572	 * requires the ONESHOT flag to be set. Some irq chips like
1573	 * MSI based interrupts are per se one shot safe. Check the
1574	 * chip flags, so we can avoid the unmask dance at the end of
1575	 * the threaded handler for those.
1576	 */
1577	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1578		new->flags &= ~IRQF_ONESHOT;
1579
1580	/*
1581	 * Protects against a concurrent __free_irq() call which might wait
1582	 * for synchronize_hardirq() to complete without holding the optional
1583	 * chip bus lock and desc->lock. Also protects against handing out
1584	 * a recycled oneshot thread_mask bit while it's still in use by
1585	 * its previous owner.
1586	 */
1587	mutex_lock(&desc->request_mutex);
1588
1589	/*
1590	 * Acquire bus lock as the irq_request_resources() callback below
1591	 * might rely on the serialization or the magic power management
1592	 * functions which are abusing the irq_bus_lock() callback,
1593	 */
1594	chip_bus_lock(desc);
1595
1596	/* First installed action requests resources. */
1597	if (!desc->action) {
1598		ret = irq_request_resources(desc);
1599		if (ret) {
1600			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1601			       new->name, irq, desc->irq_data.chip->name);
1602			goto out_bus_unlock;
1603		}
1604	}
1605
1606	/*
1607	 * The following block of code has to be executed atomically
1608	 * protected against a concurrent interrupt and any of the other
1609	 * management calls which are not serialized via
1610	 * desc->request_mutex or the optional bus lock.
1611	 */
1612	raw_spin_lock_irqsave(&desc->lock, flags);
1613	old_ptr = &desc->action;
1614	old = *old_ptr;
1615	if (old) {
1616		/*
1617		 * Can't share interrupts unless both agree to and are
1618		 * the same type (level, edge, polarity). So both flag
1619		 * fields must have IRQF_SHARED set and the bits which
1620		 * set the trigger type must match. Also all must
1621		 * agree on ONESHOT.
1622		 * Interrupt lines used for NMIs cannot be shared.
1623		 */
1624		unsigned int oldtype;
1625
1626		if (desc->istate & IRQS_NMI) {
1627			pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1628				new->name, irq, desc->irq_data.chip->name);
1629			ret = -EINVAL;
1630			goto out_unlock;
1631		}
1632
1633		/*
1634		 * If nobody did set the configuration before, inherit
1635		 * the one provided by the requester.
1636		 */
1637		if (irqd_trigger_type_was_set(&desc->irq_data)) {
1638			oldtype = irqd_get_trigger_type(&desc->irq_data);
1639		} else {
1640			oldtype = new->flags & IRQF_TRIGGER_MASK;
1641			irqd_set_trigger_type(&desc->irq_data, oldtype);
1642		}
1643
1644		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1645		    (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1646		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1647			goto mismatch;
1648
1649		/* All handlers must agree on per-cpuness */
1650		if ((old->flags & IRQF_PERCPU) !=
1651		    (new->flags & IRQF_PERCPU))
1652			goto mismatch;
1653
1654		/* add new interrupt at end of irq queue */
1655		do {
1656			/*
1657			 * Or all existing action->thread_mask bits,
1658			 * so we can find the next zero bit for this
1659			 * new action.
1660			 */
1661			thread_mask |= old->thread_mask;
1662			old_ptr = &old->next;
1663			old = *old_ptr;
1664		} while (old);
1665		shared = 1;
1666	}
1667
1668	/*
1669	 * Setup the thread mask for this irqaction for ONESHOT. For
1670	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1671	 * conditional in irq_wake_thread().
1672	 */
1673	if (new->flags & IRQF_ONESHOT) {
1674		/*
1675		 * Unlikely to have 32 resp 64 irqs sharing one line,
1676		 * but who knows.
1677		 */
1678		if (thread_mask == ~0UL) {
1679			ret = -EBUSY;
1680			goto out_unlock;
1681		}
1682		/*
1683		 * The thread_mask for the action is or'ed to
1684		 * desc->thread_active to indicate that the
1685		 * IRQF_ONESHOT thread handler has been woken, but not
1686		 * yet finished. The bit is cleared when a thread
1687		 * completes. When all threads of a shared interrupt
1688		 * line have completed desc->threads_active becomes
1689		 * zero and the interrupt line is unmasked. See
1690		 * handle.c:irq_wake_thread() for further information.
1691		 *
1692		 * If no thread is woken by primary (hard irq context)
1693		 * interrupt handlers, then desc->threads_active is
1694		 * also checked for zero to unmask the irq line in the
1695		 * affected hard irq flow handlers
1696		 * (handle_[fasteoi|level]_irq).
1697		 *
1698		 * The new action gets the first zero bit of
1699		 * thread_mask assigned. See the loop above which or's
1700		 * all existing action->thread_mask bits.
1701		 */
1702		new->thread_mask = 1UL << ffz(thread_mask);
1703
1704	} else if (new->handler == irq_default_primary_handler &&
1705		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1706		/*
1707		 * The interrupt was requested with handler = NULL, so
1708		 * we use the default primary handler for it. But it
1709		 * does not have the oneshot flag set. In combination
1710		 * with level interrupts this is deadly, because the
1711		 * default primary handler just wakes the thread, then
1712		 * the irq lines is reenabled, but the device still
1713		 * has the level irq asserted. Rinse and repeat....
1714		 *
1715		 * While this works for edge type interrupts, we play
1716		 * it safe and reject unconditionally because we can't
1717		 * say for sure which type this interrupt really
1718		 * has. The type flags are unreliable as the
1719		 * underlying chip implementation can override them.
1720		 */
1721		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for %s (irq %d)\n",
1722		       new->name, irq);
1723		ret = -EINVAL;
1724		goto out_unlock;
1725	}
1726
1727	if (!shared) {
 
 
1728		/* Setup the type (level, edge polarity) if configured: */
1729		if (new->flags & IRQF_TRIGGER_MASK) {
1730			ret = __irq_set_trigger(desc,
1731						new->flags & IRQF_TRIGGER_MASK);
1732
1733			if (ret)
1734				goto out_unlock;
1735		}
1736
1737		/*
1738		 * Activate the interrupt. That activation must happen
1739		 * independently of IRQ_NOAUTOEN. request_irq() can fail
1740		 * and the callers are supposed to handle
1741		 * that. enable_irq() of an interrupt requested with
1742		 * IRQ_NOAUTOEN is not supposed to fail. The activation
1743		 * keeps it in shutdown mode, it merily associates
1744		 * resources if necessary and if that's not possible it
1745		 * fails. Interrupts which are in managed shutdown mode
1746		 * will simply ignore that activation request.
1747		 */
1748		ret = irq_activate(desc);
1749		if (ret)
1750			goto out_unlock;
1751
1752		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1753				  IRQS_ONESHOT | IRQS_WAITING);
1754		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1755
1756		if (new->flags & IRQF_PERCPU) {
1757			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1758			irq_settings_set_per_cpu(desc);
1759			if (new->flags & IRQF_NO_DEBUG)
1760				irq_settings_set_no_debug(desc);
1761		}
1762
1763		if (noirqdebug)
1764			irq_settings_set_no_debug(desc);
1765
1766		if (new->flags & IRQF_ONESHOT)
1767			desc->istate |= IRQS_ONESHOT;
1768
1769		/* Exclude IRQ from balancing if requested */
1770		if (new->flags & IRQF_NOBALANCING) {
1771			irq_settings_set_no_balancing(desc);
1772			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1773		}
1774
1775		if (!(new->flags & IRQF_NO_AUTOEN) &&
1776		    irq_settings_can_autoenable(desc)) {
1777			irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1778		} else {
1779			/*
1780			 * Shared interrupts do not go well with disabling
1781			 * auto enable. The sharing interrupt might request
1782			 * it while it's still disabled and then wait for
1783			 * interrupts forever.
1784			 */
1785			WARN_ON_ONCE(new->flags & IRQF_SHARED);
1786			/* Undo nested disables: */
1787			desc->depth = 1;
1788		}
1789
1790	} else if (new->flags & IRQF_TRIGGER_MASK) {
1791		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1792		unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1793
1794		if (nmsk != omsk)
1795			/* hope the handler works with current  trigger mode */
1796			pr_warn("irq %d uses trigger mode %u; requested %u\n",
1797				irq, omsk, nmsk);
1798	}
1799
1800	*old_ptr = new;
1801
1802	irq_pm_install_action(desc, new);
1803
1804	/* Reset broken irq detection when installing new handler */
1805	desc->irq_count = 0;
1806	desc->irqs_unhandled = 0;
1807
1808	/*
1809	 * Check whether we disabled the irq via the spurious handler
1810	 * before. Reenable it and give it another chance.
1811	 */
1812	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1813		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1814		__enable_irq(desc);
1815	}
1816
1817	raw_spin_unlock_irqrestore(&desc->lock, flags);
1818	chip_bus_sync_unlock(desc);
1819	mutex_unlock(&desc->request_mutex);
1820
1821	irq_setup_timings(desc, new);
1822
1823	wake_up_and_wait_for_irq_thread_ready(desc, new);
1824	wake_up_and_wait_for_irq_thread_ready(desc, new->secondary);
 
 
 
 
 
 
1825
1826	register_irq_proc(irq, desc);
1827	new->dir = NULL;
1828	register_handler_proc(irq, new);
1829	return 0;
1830
1831mismatch:
1832	if (!(new->flags & IRQF_PROBE_SHARED)) {
1833		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1834		       irq, new->flags, new->name, old->flags, old->name);
1835#ifdef CONFIG_DEBUG_SHIRQ
1836		dump_stack();
1837#endif
1838	}
1839	ret = -EBUSY;
1840
1841out_unlock:
1842	raw_spin_unlock_irqrestore(&desc->lock, flags);
1843
1844	if (!desc->action)
1845		irq_release_resources(desc);
1846out_bus_unlock:
1847	chip_bus_sync_unlock(desc);
1848	mutex_unlock(&desc->request_mutex);
1849
1850out_thread:
1851	if (new->thread) {
1852		struct task_struct *t = new->thread;
1853
1854		new->thread = NULL;
1855		kthread_stop_put(t);
 
1856	}
1857	if (new->secondary && new->secondary->thread) {
1858		struct task_struct *t = new->secondary->thread;
1859
1860		new->secondary->thread = NULL;
1861		kthread_stop_put(t);
 
1862	}
1863out_mput:
1864	module_put(desc->owner);
1865	return ret;
1866}
1867
1868/*
1869 * Internal function to unregister an irqaction - used to free
1870 * regular and special interrupts that are part of the architecture.
1871 */
1872static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1873{
1874	unsigned irq = desc->irq_data.irq;
1875	struct irqaction *action, **action_ptr;
1876	unsigned long flags;
1877
1878	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1879
1880	mutex_lock(&desc->request_mutex);
1881	chip_bus_lock(desc);
1882	raw_spin_lock_irqsave(&desc->lock, flags);
1883
1884	/*
1885	 * There can be multiple actions per IRQ descriptor, find the right
1886	 * one based on the dev_id:
1887	 */
1888	action_ptr = &desc->action;
1889	for (;;) {
1890		action = *action_ptr;
1891
1892		if (!action) {
1893			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1894			raw_spin_unlock_irqrestore(&desc->lock, flags);
1895			chip_bus_sync_unlock(desc);
1896			mutex_unlock(&desc->request_mutex);
1897			return NULL;
1898		}
1899
1900		if (action->dev_id == dev_id)
1901			break;
1902		action_ptr = &action->next;
1903	}
1904
1905	/* Found it - now remove it from the list of entries: */
1906	*action_ptr = action->next;
1907
1908	irq_pm_remove_action(desc, action);
1909
1910	/* If this was the last handler, shut down the IRQ line: */
1911	if (!desc->action) {
1912		irq_settings_clr_disable_unlazy(desc);
1913		/* Only shutdown. Deactivate after synchronize_hardirq() */
1914		irq_shutdown(desc);
1915	}
1916
1917#ifdef CONFIG_SMP
1918	/* make sure affinity_hint is cleaned up */
1919	if (WARN_ON_ONCE(desc->affinity_hint))
1920		desc->affinity_hint = NULL;
1921#endif
1922
1923	raw_spin_unlock_irqrestore(&desc->lock, flags);
1924	/*
1925	 * Drop bus_lock here so the changes which were done in the chip
1926	 * callbacks above are synced out to the irq chips which hang
1927	 * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1928	 *
1929	 * Aside of that the bus_lock can also be taken from the threaded
1930	 * handler in irq_finalize_oneshot() which results in a deadlock
1931	 * because kthread_stop() would wait forever for the thread to
1932	 * complete, which is blocked on the bus lock.
1933	 *
1934	 * The still held desc->request_mutex() protects against a
1935	 * concurrent request_irq() of this irq so the release of resources
1936	 * and timing data is properly serialized.
1937	 */
1938	chip_bus_sync_unlock(desc);
1939
1940	unregister_handler_proc(irq, action);
1941
1942	/*
1943	 * Make sure it's not being used on another CPU and if the chip
1944	 * supports it also make sure that there is no (not yet serviced)
1945	 * interrupt in flight at the hardware level.
1946	 */
1947	__synchronize_irq(desc);
1948
1949#ifdef CONFIG_DEBUG_SHIRQ
1950	/*
1951	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1952	 * event to happen even now it's being freed, so let's make sure that
1953	 * is so by doing an extra call to the handler ....
1954	 *
1955	 * ( We do this after actually deregistering it, to make sure that a
1956	 *   'real' IRQ doesn't run in parallel with our fake. )
1957	 */
1958	if (action->flags & IRQF_SHARED) {
1959		local_irq_save(flags);
1960		action->handler(irq, dev_id);
1961		local_irq_restore(flags);
1962	}
1963#endif
1964
1965	/*
1966	 * The action has already been removed above, but the thread writes
1967	 * its oneshot mask bit when it completes. Though request_mutex is
1968	 * held across this which prevents __setup_irq() from handing out
1969	 * the same bit to a newly requested action.
1970	 */
1971	if (action->thread) {
1972		kthread_stop_put(action->thread);
1973		if (action->secondary && action->secondary->thread)
1974			kthread_stop_put(action->secondary->thread);
 
 
 
1975	}
1976
1977	/* Last action releases resources */
1978	if (!desc->action) {
1979		/*
1980		 * Reacquire bus lock as irq_release_resources() might
1981		 * require it to deallocate resources over the slow bus.
1982		 */
1983		chip_bus_lock(desc);
1984		/*
1985		 * There is no interrupt on the fly anymore. Deactivate it
1986		 * completely.
1987		 */
1988		raw_spin_lock_irqsave(&desc->lock, flags);
1989		irq_domain_deactivate_irq(&desc->irq_data);
1990		raw_spin_unlock_irqrestore(&desc->lock, flags);
1991
1992		irq_release_resources(desc);
1993		chip_bus_sync_unlock(desc);
1994		irq_remove_timings(desc);
1995	}
1996
1997	mutex_unlock(&desc->request_mutex);
1998
1999	irq_chip_pm_put(&desc->irq_data);
2000	module_put(desc->owner);
2001	kfree(action->secondary);
2002	return action;
2003}
2004
2005/**
2006 *	free_irq - free an interrupt allocated with request_irq
2007 *	@irq: Interrupt line to free
2008 *	@dev_id: Device identity to free
2009 *
2010 *	Remove an interrupt handler. The handler is removed and if the
2011 *	interrupt line is no longer in use by any driver it is disabled.
2012 *	On a shared IRQ the caller must ensure the interrupt is disabled
2013 *	on the card it drives before calling this function. The function
2014 *	does not return until any executing interrupts for this IRQ
2015 *	have completed.
2016 *
2017 *	This function must not be called from interrupt context.
2018 *
2019 *	Returns the devname argument passed to request_irq.
2020 */
2021const void *free_irq(unsigned int irq, void *dev_id)
2022{
2023	struct irq_desc *desc = irq_to_desc(irq);
2024	struct irqaction *action;
2025	const char *devname;
2026
2027	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2028		return NULL;
2029
2030#ifdef CONFIG_SMP
2031	if (WARN_ON(desc->affinity_notify))
2032		desc->affinity_notify = NULL;
2033#endif
2034
2035	action = __free_irq(desc, dev_id);
2036
2037	if (!action)
2038		return NULL;
2039
2040	devname = action->name;
2041	kfree(action);
2042	return devname;
2043}
2044EXPORT_SYMBOL(free_irq);
2045
2046/* This function must be called with desc->lock held */
2047static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
2048{
2049	const char *devname = NULL;
2050
2051	desc->istate &= ~IRQS_NMI;
2052
2053	if (!WARN_ON(desc->action == NULL)) {
2054		irq_pm_remove_action(desc, desc->action);
2055		devname = desc->action->name;
2056		unregister_handler_proc(irq, desc->action);
2057
2058		kfree(desc->action);
2059		desc->action = NULL;
2060	}
2061
2062	irq_settings_clr_disable_unlazy(desc);
2063	irq_shutdown_and_deactivate(desc);
2064
2065	irq_release_resources(desc);
2066
2067	irq_chip_pm_put(&desc->irq_data);
2068	module_put(desc->owner);
2069
2070	return devname;
2071}
2072
2073const void *free_nmi(unsigned int irq, void *dev_id)
2074{
2075	struct irq_desc *desc = irq_to_desc(irq);
2076	unsigned long flags;
2077	const void *devname;
2078
2079	if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
2080		return NULL;
2081
2082	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2083		return NULL;
2084
2085	/* NMI still enabled */
2086	if (WARN_ON(desc->depth == 0))
2087		disable_nmi_nosync(irq);
2088
2089	raw_spin_lock_irqsave(&desc->lock, flags);
2090
2091	irq_nmi_teardown(desc);
2092	devname = __cleanup_nmi(irq, desc);
2093
2094	raw_spin_unlock_irqrestore(&desc->lock, flags);
2095
2096	return devname;
2097}
2098
2099/**
2100 *	request_threaded_irq - allocate an interrupt line
2101 *	@irq: Interrupt line to allocate
2102 *	@handler: Function to be called when the IRQ occurs.
2103 *		  Primary handler for threaded interrupts.
2104 *		  If handler is NULL and thread_fn != NULL
2105 *		  the default primary handler is installed.
2106 *	@thread_fn: Function called from the irq handler thread
2107 *		    If NULL, no irq thread is created
2108 *	@irqflags: Interrupt type flags
2109 *	@devname: An ascii name for the claiming device
2110 *	@dev_id: A cookie passed back to the handler function
2111 *
2112 *	This call allocates interrupt resources and enables the
2113 *	interrupt line and IRQ handling. From the point this
2114 *	call is made your handler function may be invoked. Since
2115 *	your handler function must clear any interrupt the board
2116 *	raises, you must take care both to initialise your hardware
2117 *	and to set up the interrupt handler in the right order.
2118 *
2119 *	If you want to set up a threaded irq handler for your device
2120 *	then you need to supply @handler and @thread_fn. @handler is
2121 *	still called in hard interrupt context and has to check
2122 *	whether the interrupt originates from the device. If yes it
2123 *	needs to disable the interrupt on the device and return
2124 *	IRQ_WAKE_THREAD which will wake up the handler thread and run
2125 *	@thread_fn. This split handler design is necessary to support
2126 *	shared interrupts.
2127 *
2128 *	Dev_id must be globally unique. Normally the address of the
2129 *	device data structure is used as the cookie. Since the handler
2130 *	receives this value it makes sense to use it.
2131 *
2132 *	If your interrupt is shared you must pass a non NULL dev_id
2133 *	as this is required when freeing the interrupt.
2134 *
2135 *	Flags:
2136 *
2137 *	IRQF_SHARED		Interrupt is shared
2138 *	IRQF_TRIGGER_*		Specify active edge(s) or level
2139 *	IRQF_ONESHOT		Run thread_fn with interrupt line masked
2140 */
2141int request_threaded_irq(unsigned int irq, irq_handler_t handler,
2142			 irq_handler_t thread_fn, unsigned long irqflags,
2143			 const char *devname, void *dev_id)
2144{
2145	struct irqaction *action;
2146	struct irq_desc *desc;
2147	int retval;
2148
2149	if (irq == IRQ_NOTCONNECTED)
2150		return -ENOTCONN;
2151
2152	/*
2153	 * Sanity-check: shared interrupts must pass in a real dev-ID,
2154	 * otherwise we'll have trouble later trying to figure out
2155	 * which interrupt is which (messes up the interrupt freeing
2156	 * logic etc).
2157	 *
2158	 * Also shared interrupts do not go well with disabling auto enable.
2159	 * The sharing interrupt might request it while it's still disabled
2160	 * and then wait for interrupts forever.
2161	 *
2162	 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
2163	 * it cannot be set along with IRQF_NO_SUSPEND.
2164	 */
2165	if (((irqflags & IRQF_SHARED) && !dev_id) ||
2166	    ((irqflags & IRQF_SHARED) && (irqflags & IRQF_NO_AUTOEN)) ||
2167	    (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
2168	    ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
2169		return -EINVAL;
2170
2171	desc = irq_to_desc(irq);
2172	if (!desc)
2173		return -EINVAL;
2174
2175	if (!irq_settings_can_request(desc) ||
2176	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2177		return -EINVAL;
2178
2179	if (!handler) {
2180		if (!thread_fn)
2181			return -EINVAL;
2182		handler = irq_default_primary_handler;
2183	}
2184
2185	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2186	if (!action)
2187		return -ENOMEM;
2188
2189	action->handler = handler;
2190	action->thread_fn = thread_fn;
2191	action->flags = irqflags;
2192	action->name = devname;
2193	action->dev_id = dev_id;
2194
2195	retval = irq_chip_pm_get(&desc->irq_data);
2196	if (retval < 0) {
2197		kfree(action);
2198		return retval;
2199	}
2200
2201	retval = __setup_irq(irq, desc, action);
2202
2203	if (retval) {
2204		irq_chip_pm_put(&desc->irq_data);
2205		kfree(action->secondary);
2206		kfree(action);
2207	}
2208
2209#ifdef CONFIG_DEBUG_SHIRQ_FIXME
2210	if (!retval && (irqflags & IRQF_SHARED)) {
2211		/*
2212		 * It's a shared IRQ -- the driver ought to be prepared for it
2213		 * to happen immediately, so let's make sure....
2214		 * We disable the irq to make sure that a 'real' IRQ doesn't
2215		 * run in parallel with our fake.
2216		 */
2217		unsigned long flags;
2218
2219		disable_irq(irq);
2220		local_irq_save(flags);
2221
2222		handler(irq, dev_id);
2223
2224		local_irq_restore(flags);
2225		enable_irq(irq);
2226	}
2227#endif
2228	return retval;
2229}
2230EXPORT_SYMBOL(request_threaded_irq);
2231
2232/**
2233 *	request_any_context_irq - allocate an interrupt line
2234 *	@irq: Interrupt line to allocate
2235 *	@handler: Function to be called when the IRQ occurs.
2236 *		  Threaded handler for threaded interrupts.
2237 *	@flags: Interrupt type flags
2238 *	@name: An ascii name for the claiming device
2239 *	@dev_id: A cookie passed back to the handler function
2240 *
2241 *	This call allocates interrupt resources and enables the
2242 *	interrupt line and IRQ handling. It selects either a
2243 *	hardirq or threaded handling method depending on the
2244 *	context.
2245 *
2246 *	On failure, it returns a negative value. On success,
2247 *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2248 */
2249int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2250			    unsigned long flags, const char *name, void *dev_id)
2251{
2252	struct irq_desc *desc;
2253	int ret;
2254
2255	if (irq == IRQ_NOTCONNECTED)
2256		return -ENOTCONN;
2257
2258	desc = irq_to_desc(irq);
2259	if (!desc)
2260		return -EINVAL;
2261
2262	if (irq_settings_is_nested_thread(desc)) {
2263		ret = request_threaded_irq(irq, NULL, handler,
2264					   flags, name, dev_id);
2265		return !ret ? IRQC_IS_NESTED : ret;
2266	}
2267
2268	ret = request_irq(irq, handler, flags, name, dev_id);
2269	return !ret ? IRQC_IS_HARDIRQ : ret;
2270}
2271EXPORT_SYMBOL_GPL(request_any_context_irq);
2272
2273/**
2274 *	request_nmi - allocate an interrupt line for NMI delivery
2275 *	@irq: Interrupt line to allocate
2276 *	@handler: Function to be called when the IRQ occurs.
2277 *		  Threaded handler for threaded interrupts.
2278 *	@irqflags: Interrupt type flags
2279 *	@name: An ascii name for the claiming device
2280 *	@dev_id: A cookie passed back to the handler function
2281 *
2282 *	This call allocates interrupt resources and enables the
2283 *	interrupt line and IRQ handling. It sets up the IRQ line
2284 *	to be handled as an NMI.
2285 *
2286 *	An interrupt line delivering NMIs cannot be shared and IRQ handling
2287 *	cannot be threaded.
2288 *
2289 *	Interrupt lines requested for NMI delivering must produce per cpu
2290 *	interrupts and have auto enabling setting disabled.
2291 *
2292 *	Dev_id must be globally unique. Normally the address of the
2293 *	device data structure is used as the cookie. Since the handler
2294 *	receives this value it makes sense to use it.
2295 *
2296 *	If the interrupt line cannot be used to deliver NMIs, function
2297 *	will fail and return a negative value.
2298 */
2299int request_nmi(unsigned int irq, irq_handler_t handler,
2300		unsigned long irqflags, const char *name, void *dev_id)
2301{
2302	struct irqaction *action;
2303	struct irq_desc *desc;
2304	unsigned long flags;
2305	int retval;
2306
2307	if (irq == IRQ_NOTCONNECTED)
2308		return -ENOTCONN;
2309
2310	/* NMI cannot be shared, used for Polling */
2311	if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2312		return -EINVAL;
2313
2314	if (!(irqflags & IRQF_PERCPU))
2315		return -EINVAL;
2316
2317	if (!handler)
2318		return -EINVAL;
2319
2320	desc = irq_to_desc(irq);
2321
2322	if (!desc || (irq_settings_can_autoenable(desc) &&
2323	    !(irqflags & IRQF_NO_AUTOEN)) ||
2324	    !irq_settings_can_request(desc) ||
2325	    WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2326	    !irq_supports_nmi(desc))
2327		return -EINVAL;
2328
2329	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2330	if (!action)
2331		return -ENOMEM;
2332
2333	action->handler = handler;
2334	action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2335	action->name = name;
2336	action->dev_id = dev_id;
2337
2338	retval = irq_chip_pm_get(&desc->irq_data);
2339	if (retval < 0)
2340		goto err_out;
2341
2342	retval = __setup_irq(irq, desc, action);
2343	if (retval)
2344		goto err_irq_setup;
2345
2346	raw_spin_lock_irqsave(&desc->lock, flags);
2347
2348	/* Setup NMI state */
2349	desc->istate |= IRQS_NMI;
2350	retval = irq_nmi_setup(desc);
2351	if (retval) {
2352		__cleanup_nmi(irq, desc);
2353		raw_spin_unlock_irqrestore(&desc->lock, flags);
2354		return -EINVAL;
2355	}
2356
2357	raw_spin_unlock_irqrestore(&desc->lock, flags);
2358
2359	return 0;
2360
2361err_irq_setup:
2362	irq_chip_pm_put(&desc->irq_data);
2363err_out:
2364	kfree(action);
2365
2366	return retval;
2367}
2368
2369void enable_percpu_irq(unsigned int irq, unsigned int type)
2370{
2371	unsigned int cpu = smp_processor_id();
2372	unsigned long flags;
2373	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2374
2375	if (!desc)
2376		return;
2377
2378	/*
2379	 * If the trigger type is not specified by the caller, then
2380	 * use the default for this interrupt.
2381	 */
2382	type &= IRQ_TYPE_SENSE_MASK;
2383	if (type == IRQ_TYPE_NONE)
2384		type = irqd_get_trigger_type(&desc->irq_data);
2385
2386	if (type != IRQ_TYPE_NONE) {
2387		int ret;
2388
2389		ret = __irq_set_trigger(desc, type);
2390
2391		if (ret) {
2392			WARN(1, "failed to set type for IRQ%d\n", irq);
2393			goto out;
2394		}
2395	}
2396
2397	irq_percpu_enable(desc, cpu);
2398out:
2399	irq_put_desc_unlock(desc, flags);
2400}
2401EXPORT_SYMBOL_GPL(enable_percpu_irq);
2402
2403void enable_percpu_nmi(unsigned int irq, unsigned int type)
2404{
2405	enable_percpu_irq(irq, type);
2406}
2407
2408/**
2409 * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2410 * @irq:	Linux irq number to check for
2411 *
2412 * Must be called from a non migratable context. Returns the enable
2413 * state of a per cpu interrupt on the current cpu.
2414 */
2415bool irq_percpu_is_enabled(unsigned int irq)
2416{
2417	unsigned int cpu = smp_processor_id();
2418	struct irq_desc *desc;
2419	unsigned long flags;
2420	bool is_enabled;
2421
2422	desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2423	if (!desc)
2424		return false;
2425
2426	is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2427	irq_put_desc_unlock(desc, flags);
2428
2429	return is_enabled;
2430}
2431EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2432
2433void disable_percpu_irq(unsigned int irq)
2434{
2435	unsigned int cpu = smp_processor_id();
2436	unsigned long flags;
2437	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2438
2439	if (!desc)
2440		return;
2441
2442	irq_percpu_disable(desc, cpu);
2443	irq_put_desc_unlock(desc, flags);
2444}
2445EXPORT_SYMBOL_GPL(disable_percpu_irq);
2446
2447void disable_percpu_nmi(unsigned int irq)
2448{
2449	disable_percpu_irq(irq);
2450}
2451
2452/*
2453 * Internal function to unregister a percpu irqaction.
2454 */
2455static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2456{
2457	struct irq_desc *desc = irq_to_desc(irq);
2458	struct irqaction *action;
2459	unsigned long flags;
2460
2461	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2462
2463	if (!desc)
2464		return NULL;
2465
2466	raw_spin_lock_irqsave(&desc->lock, flags);
2467
2468	action = desc->action;
2469	if (!action || action->percpu_dev_id != dev_id) {
2470		WARN(1, "Trying to free already-free IRQ %d\n", irq);
2471		goto bad;
2472	}
2473
2474	if (!cpumask_empty(desc->percpu_enabled)) {
2475		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2476		     irq, cpumask_first(desc->percpu_enabled));
2477		goto bad;
2478	}
2479
2480	/* Found it - now remove it from the list of entries: */
2481	desc->action = NULL;
2482
2483	desc->istate &= ~IRQS_NMI;
2484
2485	raw_spin_unlock_irqrestore(&desc->lock, flags);
2486
2487	unregister_handler_proc(irq, action);
2488
2489	irq_chip_pm_put(&desc->irq_data);
2490	module_put(desc->owner);
2491	return action;
2492
2493bad:
2494	raw_spin_unlock_irqrestore(&desc->lock, flags);
2495	return NULL;
2496}
2497
2498/**
2499 *	remove_percpu_irq - free a per-cpu interrupt
2500 *	@irq: Interrupt line to free
2501 *	@act: irqaction for the interrupt
2502 *
2503 * Used to remove interrupts statically setup by the early boot process.
2504 */
2505void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2506{
2507	struct irq_desc *desc = irq_to_desc(irq);
2508
2509	if (desc && irq_settings_is_per_cpu_devid(desc))
2510	    __free_percpu_irq(irq, act->percpu_dev_id);
2511}
2512
2513/**
2514 *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
2515 *	@irq: Interrupt line to free
2516 *	@dev_id: Device identity to free
2517 *
2518 *	Remove a percpu interrupt handler. The handler is removed, but
2519 *	the interrupt line is not disabled. This must be done on each
2520 *	CPU before calling this function. The function does not return
2521 *	until any executing interrupts for this IRQ have completed.
2522 *
2523 *	This function must not be called from interrupt context.
2524 */
2525void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2526{
2527	struct irq_desc *desc = irq_to_desc(irq);
2528
2529	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2530		return;
2531
2532	chip_bus_lock(desc);
2533	kfree(__free_percpu_irq(irq, dev_id));
2534	chip_bus_sync_unlock(desc);
2535}
2536EXPORT_SYMBOL_GPL(free_percpu_irq);
2537
2538void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2539{
2540	struct irq_desc *desc = irq_to_desc(irq);
2541
2542	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2543		return;
2544
2545	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2546		return;
2547
2548	kfree(__free_percpu_irq(irq, dev_id));
2549}
2550
2551/**
2552 *	setup_percpu_irq - setup a per-cpu interrupt
2553 *	@irq: Interrupt line to setup
2554 *	@act: irqaction for the interrupt
2555 *
2556 * Used to statically setup per-cpu interrupts in the early boot process.
2557 */
2558int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2559{
2560	struct irq_desc *desc = irq_to_desc(irq);
2561	int retval;
2562
2563	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2564		return -EINVAL;
2565
2566	retval = irq_chip_pm_get(&desc->irq_data);
2567	if (retval < 0)
2568		return retval;
2569
2570	retval = __setup_irq(irq, desc, act);
2571
2572	if (retval)
2573		irq_chip_pm_put(&desc->irq_data);
2574
2575	return retval;
2576}
2577
2578/**
2579 *	__request_percpu_irq - allocate a percpu interrupt line
2580 *	@irq: Interrupt line to allocate
2581 *	@handler: Function to be called when the IRQ occurs.
2582 *	@flags: Interrupt type flags (IRQF_TIMER only)
2583 *	@devname: An ascii name for the claiming device
2584 *	@dev_id: A percpu cookie passed back to the handler function
2585 *
2586 *	This call allocates interrupt resources and enables the
2587 *	interrupt on the local CPU. If the interrupt is supposed to be
2588 *	enabled on other CPUs, it has to be done on each CPU using
2589 *	enable_percpu_irq().
2590 *
2591 *	Dev_id must be globally unique. It is a per-cpu variable, and
2592 *	the handler gets called with the interrupted CPU's instance of
2593 *	that variable.
2594 */
2595int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2596			 unsigned long flags, const char *devname,
2597			 void __percpu *dev_id)
2598{
2599	struct irqaction *action;
2600	struct irq_desc *desc;
2601	int retval;
2602
2603	if (!dev_id)
2604		return -EINVAL;
2605
2606	desc = irq_to_desc(irq);
2607	if (!desc || !irq_settings_can_request(desc) ||
2608	    !irq_settings_is_per_cpu_devid(desc))
2609		return -EINVAL;
2610
2611	if (flags && flags != IRQF_TIMER)
2612		return -EINVAL;
2613
2614	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2615	if (!action)
2616		return -ENOMEM;
2617
2618	action->handler = handler;
2619	action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2620	action->name = devname;
2621	action->percpu_dev_id = dev_id;
2622
2623	retval = irq_chip_pm_get(&desc->irq_data);
2624	if (retval < 0) {
2625		kfree(action);
2626		return retval;
2627	}
2628
2629	retval = __setup_irq(irq, desc, action);
2630
2631	if (retval) {
2632		irq_chip_pm_put(&desc->irq_data);
2633		kfree(action);
2634	}
2635
2636	return retval;
2637}
2638EXPORT_SYMBOL_GPL(__request_percpu_irq);
2639
2640/**
2641 *	request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2642 *	@irq: Interrupt line to allocate
2643 *	@handler: Function to be called when the IRQ occurs.
2644 *	@name: An ascii name for the claiming device
2645 *	@dev_id: A percpu cookie passed back to the handler function
2646 *
2647 *	This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2648 *	have to be setup on each CPU by calling prepare_percpu_nmi() before
2649 *	being enabled on the same CPU by using enable_percpu_nmi().
2650 *
2651 *	Dev_id must be globally unique. It is a per-cpu variable, and
2652 *	the handler gets called with the interrupted CPU's instance of
2653 *	that variable.
2654 *
2655 *	Interrupt lines requested for NMI delivering should have auto enabling
2656 *	setting disabled.
2657 *
2658 *	If the interrupt line cannot be used to deliver NMIs, function
2659 *	will fail returning a negative value.
2660 */
2661int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2662		       const char *name, void __percpu *dev_id)
2663{
2664	struct irqaction *action;
2665	struct irq_desc *desc;
2666	unsigned long flags;
2667	int retval;
2668
2669	if (!handler)
2670		return -EINVAL;
2671
2672	desc = irq_to_desc(irq);
2673
2674	if (!desc || !irq_settings_can_request(desc) ||
2675	    !irq_settings_is_per_cpu_devid(desc) ||
2676	    irq_settings_can_autoenable(desc) ||
2677	    !irq_supports_nmi(desc))
2678		return -EINVAL;
2679
2680	/* The line cannot already be NMI */
2681	if (desc->istate & IRQS_NMI)
2682		return -EINVAL;
2683
2684	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2685	if (!action)
2686		return -ENOMEM;
2687
2688	action->handler = handler;
2689	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2690		| IRQF_NOBALANCING;
2691	action->name = name;
2692	action->percpu_dev_id = dev_id;
2693
2694	retval = irq_chip_pm_get(&desc->irq_data);
2695	if (retval < 0)
2696		goto err_out;
2697
2698	retval = __setup_irq(irq, desc, action);
2699	if (retval)
2700		goto err_irq_setup;
2701
2702	raw_spin_lock_irqsave(&desc->lock, flags);
2703	desc->istate |= IRQS_NMI;
2704	raw_spin_unlock_irqrestore(&desc->lock, flags);
2705
2706	return 0;
2707
2708err_irq_setup:
2709	irq_chip_pm_put(&desc->irq_data);
2710err_out:
2711	kfree(action);
2712
2713	return retval;
2714}
2715
2716/**
2717 *	prepare_percpu_nmi - performs CPU local setup for NMI delivery
2718 *	@irq: Interrupt line to prepare for NMI delivery
2719 *
2720 *	This call prepares an interrupt line to deliver NMI on the current CPU,
2721 *	before that interrupt line gets enabled with enable_percpu_nmi().
2722 *
2723 *	As a CPU local operation, this should be called from non-preemptible
2724 *	context.
2725 *
2726 *	If the interrupt line cannot be used to deliver NMIs, function
2727 *	will fail returning a negative value.
2728 */
2729int prepare_percpu_nmi(unsigned int irq)
2730{
2731	unsigned long flags;
2732	struct irq_desc *desc;
2733	int ret = 0;
2734
2735	WARN_ON(preemptible());
2736
2737	desc = irq_get_desc_lock(irq, &flags,
2738				 IRQ_GET_DESC_CHECK_PERCPU);
2739	if (!desc)
2740		return -EINVAL;
2741
2742	if (WARN(!(desc->istate & IRQS_NMI),
2743		 KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2744		 irq)) {
2745		ret = -EINVAL;
2746		goto out;
2747	}
2748
2749	ret = irq_nmi_setup(desc);
2750	if (ret) {
2751		pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2752		goto out;
2753	}
2754
2755out:
2756	irq_put_desc_unlock(desc, flags);
2757	return ret;
2758}
2759
2760/**
2761 *	teardown_percpu_nmi - undoes NMI setup of IRQ line
2762 *	@irq: Interrupt line from which CPU local NMI configuration should be
2763 *	      removed
2764 *
2765 *	This call undoes the setup done by prepare_percpu_nmi().
2766 *
2767 *	IRQ line should not be enabled for the current CPU.
2768 *
2769 *	As a CPU local operation, this should be called from non-preemptible
2770 *	context.
2771 */
2772void teardown_percpu_nmi(unsigned int irq)
2773{
2774	unsigned long flags;
2775	struct irq_desc *desc;
2776
2777	WARN_ON(preemptible());
2778
2779	desc = irq_get_desc_lock(irq, &flags,
2780				 IRQ_GET_DESC_CHECK_PERCPU);
2781	if (!desc)
2782		return;
2783
2784	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2785		goto out;
2786
2787	irq_nmi_teardown(desc);
2788out:
2789	irq_put_desc_unlock(desc, flags);
2790}
2791
2792int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2793			    bool *state)
2794{
2795	struct irq_chip *chip;
2796	int err = -EINVAL;
2797
2798	do {
2799		chip = irq_data_get_irq_chip(data);
2800		if (WARN_ON_ONCE(!chip))
2801			return -ENODEV;
2802		if (chip->irq_get_irqchip_state)
2803			break;
2804#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2805		data = data->parent_data;
2806#else
2807		data = NULL;
2808#endif
2809	} while (data);
2810
2811	if (data)
2812		err = chip->irq_get_irqchip_state(data, which, state);
2813	return err;
2814}
2815
2816/**
2817 *	irq_get_irqchip_state - returns the irqchip state of a interrupt.
2818 *	@irq: Interrupt line that is forwarded to a VM
2819 *	@which: One of IRQCHIP_STATE_* the caller wants to know about
2820 *	@state: a pointer to a boolean where the state is to be stored
2821 *
2822 *	This call snapshots the internal irqchip state of an
2823 *	interrupt, returning into @state the bit corresponding to
2824 *	stage @which
2825 *
2826 *	This function should be called with preemption disabled if the
2827 *	interrupt controller has per-cpu registers.
2828 */
2829int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2830			  bool *state)
2831{
2832	struct irq_desc *desc;
2833	struct irq_data *data;
2834	unsigned long flags;
2835	int err = -EINVAL;
2836
2837	desc = irq_get_desc_buslock(irq, &flags, 0);
2838	if (!desc)
2839		return err;
2840
2841	data = irq_desc_get_irq_data(desc);
2842
2843	err = __irq_get_irqchip_state(data, which, state);
2844
2845	irq_put_desc_busunlock(desc, flags);
2846	return err;
2847}
2848EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2849
2850/**
2851 *	irq_set_irqchip_state - set the state of a forwarded interrupt.
2852 *	@irq: Interrupt line that is forwarded to a VM
2853 *	@which: State to be restored (one of IRQCHIP_STATE_*)
2854 *	@val: Value corresponding to @which
2855 *
2856 *	This call sets the internal irqchip state of an interrupt,
2857 *	depending on the value of @which.
2858 *
2859 *	This function should be called with migration disabled if the
2860 *	interrupt controller has per-cpu registers.
2861 */
2862int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2863			  bool val)
2864{
2865	struct irq_desc *desc;
2866	struct irq_data *data;
2867	struct irq_chip *chip;
2868	unsigned long flags;
2869	int err = -EINVAL;
2870
2871	desc = irq_get_desc_buslock(irq, &flags, 0);
2872	if (!desc)
2873		return err;
2874
2875	data = irq_desc_get_irq_data(desc);
2876
2877	do {
2878		chip = irq_data_get_irq_chip(data);
2879		if (WARN_ON_ONCE(!chip)) {
2880			err = -ENODEV;
2881			goto out_unlock;
2882		}
2883		if (chip->irq_set_irqchip_state)
2884			break;
2885#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2886		data = data->parent_data;
2887#else
2888		data = NULL;
2889#endif
2890	} while (data);
2891
2892	if (data)
2893		err = chip->irq_set_irqchip_state(data, which, val);
2894
2895out_unlock:
2896	irq_put_desc_busunlock(desc, flags);
2897	return err;
2898}
2899EXPORT_SYMBOL_GPL(irq_set_irqchip_state);
2900
2901/**
2902 * irq_has_action - Check whether an interrupt is requested
2903 * @irq:	The linux irq number
2904 *
2905 * Returns: A snapshot of the current state
2906 */
2907bool irq_has_action(unsigned int irq)
2908{
2909	bool res;
2910
2911	rcu_read_lock();
2912	res = irq_desc_has_action(irq_to_desc(irq));
2913	rcu_read_unlock();
2914	return res;
2915}
2916EXPORT_SYMBOL_GPL(irq_has_action);
2917
2918/**
2919 * irq_check_status_bit - Check whether bits in the irq descriptor status are set
2920 * @irq:	The linux irq number
2921 * @bitmask:	The bitmask to evaluate
2922 *
2923 * Returns: True if one of the bits in @bitmask is set
2924 */
2925bool irq_check_status_bit(unsigned int irq, unsigned int bitmask)
2926{
2927	struct irq_desc *desc;
2928	bool res = false;
2929
2930	rcu_read_lock();
2931	desc = irq_to_desc(irq);
2932	if (desc)
2933		res = !!(desc->status_use_accessors & bitmask);
2934	rcu_read_unlock();
2935	return res;
2936}
2937EXPORT_SYMBOL_GPL(irq_check_status_bit);
v5.14.15
   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 affinity 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	 * usable 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
 374/**
 375 * irq_update_affinity_desc - Update affinity management for an interrupt
 376 * @irq:	The interrupt number to update
 377 * @affinity:	Pointer to the affinity descriptor
 378 *
 379 * This interface can be used to configure the affinity management of
 380 * interrupts which have been allocated already.
 381 *
 382 * There are certain limitations on when it may be used - attempts to use it
 383 * for when the kernel is configured for generic IRQ reservation mode (in
 384 * config GENERIC_IRQ_RESERVATION_MODE) will fail, as it may conflict with
 385 * managed/non-managed interrupt accounting. In addition, attempts to use it on
 386 * an interrupt which is already started or which has already been configured
 387 * as managed will also fail, as these mean invalid init state or double init.
 388 */
 389int irq_update_affinity_desc(unsigned int irq,
 390			     struct irq_affinity_desc *affinity)
 391{
 392	struct irq_desc *desc;
 393	unsigned long flags;
 394	bool activated;
 395	int ret = 0;
 396
 397	/*
 398	 * Supporting this with the reservation scheme used by x86 needs
 399	 * some more thought. Fail it for now.
 400	 */
 401	if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
 402		return -EOPNOTSUPP;
 403
 404	desc = irq_get_desc_buslock(irq, &flags, 0);
 405	if (!desc)
 406		return -EINVAL;
 407
 408	/* Requires the interrupt to be shut down */
 409	if (irqd_is_started(&desc->irq_data)) {
 410		ret = -EBUSY;
 411		goto out_unlock;
 412	}
 413
 414	/* Interrupts which are already managed cannot be modified */
 415	if (irqd_affinity_is_managed(&desc->irq_data)) {
 416		ret = -EBUSY;
 417		goto out_unlock;
 418	}
 419
 420	/*
 421	 * Deactivate the interrupt. That's required to undo
 422	 * anything an earlier activation has established.
 423	 */
 424	activated = irqd_is_activated(&desc->irq_data);
 425	if (activated)
 426		irq_domain_deactivate_irq(&desc->irq_data);
 427
 428	if (affinity->is_managed) {
 429		irqd_set(&desc->irq_data, IRQD_AFFINITY_MANAGED);
 430		irqd_set(&desc->irq_data, IRQD_MANAGED_SHUTDOWN);
 431	}
 432
 433	cpumask_copy(desc->irq_common_data.affinity, &affinity->mask);
 434
 435	/* Restore the activation state */
 436	if (activated)
 437		irq_domain_activate_irq(&desc->irq_data, false);
 438
 439out_unlock:
 440	irq_put_desc_busunlock(desc, flags);
 441	return ret;
 442}
 443
 444static int __irq_set_affinity(unsigned int irq, const struct cpumask *mask,
 445			      bool force)
 446{
 447	struct irq_desc *desc = irq_to_desc(irq);
 448	unsigned long flags;
 449	int ret;
 450
 451	if (!desc)
 452		return -EINVAL;
 453
 454	raw_spin_lock_irqsave(&desc->lock, flags);
 455	ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
 456	raw_spin_unlock_irqrestore(&desc->lock, flags);
 457	return ret;
 458}
 459
 460/**
 461 * irq_set_affinity - Set the irq affinity of a given irq
 462 * @irq:	Interrupt to set affinity
 463 * @cpumask:	cpumask
 464 *
 465 * Fails if cpumask does not contain an online CPU
 466 */
 467int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
 468{
 469	return __irq_set_affinity(irq, cpumask, false);
 470}
 471EXPORT_SYMBOL_GPL(irq_set_affinity);
 472
 473/**
 474 * irq_force_affinity - Force the irq affinity of a given irq
 475 * @irq:	Interrupt to set affinity
 476 * @cpumask:	cpumask
 477 *
 478 * Same as irq_set_affinity, but without checking the mask against
 479 * online cpus.
 480 *
 481 * Solely for low level cpu hotplug code, where we need to make per
 482 * cpu interrupts affine before the cpu becomes online.
 483 */
 484int irq_force_affinity(unsigned int irq, const struct cpumask *cpumask)
 485{
 486	return __irq_set_affinity(irq, cpumask, true);
 487}
 488EXPORT_SYMBOL_GPL(irq_force_affinity);
 489
 490int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
 
 491{
 492	unsigned long flags;
 493	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 494
 495	if (!desc)
 496		return -EINVAL;
 497	desc->affinity_hint = m;
 498	irq_put_desc_unlock(desc, flags);
 499	/* set the initial affinity to prevent every interrupt being on CPU0 */
 500	if (m)
 501		__irq_set_affinity(irq, m, false);
 502	return 0;
 503}
 504EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
 505
 506static void irq_affinity_notify(struct work_struct *work)
 507{
 508	struct irq_affinity_notify *notify =
 509		container_of(work, struct irq_affinity_notify, work);
 510	struct irq_desc *desc = irq_to_desc(notify->irq);
 511	cpumask_var_t cpumask;
 512	unsigned long flags;
 513
 514	if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
 515		goto out;
 516
 517	raw_spin_lock_irqsave(&desc->lock, flags);
 518	if (irq_move_pending(&desc->irq_data))
 519		irq_get_pending(cpumask, desc);
 520	else
 521		cpumask_copy(cpumask, desc->irq_common_data.affinity);
 522	raw_spin_unlock_irqrestore(&desc->lock, flags);
 523
 524	notify->notify(notify, cpumask);
 525
 526	free_cpumask_var(cpumask);
 527out:
 528	kref_put(&notify->kref, notify->release);
 529}
 530
 531/**
 532 *	irq_set_affinity_notifier - control notification of IRQ affinity changes
 533 *	@irq:		Interrupt for which to enable/disable notification
 534 *	@notify:	Context for notification, or %NULL to disable
 535 *			notification.  Function pointers must be initialised;
 536 *			the other fields will be initialised by this function.
 537 *
 538 *	Must be called in process context.  Notification may only be enabled
 539 *	after the IRQ is allocated and must be disabled before the IRQ is
 540 *	freed using free_irq().
 541 */
 542int
 543irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 544{
 545	struct irq_desc *desc = irq_to_desc(irq);
 546	struct irq_affinity_notify *old_notify;
 547	unsigned long flags;
 548
 549	/* The release function is promised process context */
 550	might_sleep();
 551
 552	if (!desc || desc->istate & IRQS_NMI)
 553		return -EINVAL;
 554
 555	/* Complete initialisation of *notify */
 556	if (notify) {
 557		notify->irq = irq;
 558		kref_init(&notify->kref);
 559		INIT_WORK(&notify->work, irq_affinity_notify);
 560	}
 561
 562	raw_spin_lock_irqsave(&desc->lock, flags);
 563	old_notify = desc->affinity_notify;
 564	desc->affinity_notify = notify;
 565	raw_spin_unlock_irqrestore(&desc->lock, flags);
 566
 567	if (old_notify) {
 568		if (cancel_work_sync(&old_notify->work)) {
 569			/* Pending work had a ref, put that one too */
 570			kref_put(&old_notify->kref, old_notify->release);
 571		}
 572		kref_put(&old_notify->kref, old_notify->release);
 573	}
 574
 575	return 0;
 576}
 577EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
 578
 579#ifndef CONFIG_AUTO_IRQ_AFFINITY
 580/*
 581 * Generic version of the affinity autoselector.
 582 */
 583int irq_setup_affinity(struct irq_desc *desc)
 584{
 585	struct cpumask *set = irq_default_affinity;
 586	int ret, node = irq_desc_get_node(desc);
 587	static DEFINE_RAW_SPINLOCK(mask_lock);
 588	static struct cpumask mask;
 589
 590	/* Excludes PER_CPU and NO_BALANCE interrupts */
 591	if (!__irq_can_set_affinity(desc))
 592		return 0;
 593
 594	raw_spin_lock(&mask_lock);
 595	/*
 596	 * Preserve the managed affinity setting and a userspace affinity
 597	 * setup, but make sure that one of the targets is online.
 598	 */
 599	if (irqd_affinity_is_managed(&desc->irq_data) ||
 600	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 601		if (cpumask_intersects(desc->irq_common_data.affinity,
 602				       cpu_online_mask))
 603			set = desc->irq_common_data.affinity;
 604		else
 605			irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
 606	}
 607
 608	cpumask_and(&mask, cpu_online_mask, set);
 609	if (cpumask_empty(&mask))
 610		cpumask_copy(&mask, cpu_online_mask);
 611
 612	if (node != NUMA_NO_NODE) {
 613		const struct cpumask *nodemask = cpumask_of_node(node);
 614
 615		/* make sure at least one of the cpus in nodemask is online */
 616		if (cpumask_intersects(&mask, nodemask))
 617			cpumask_and(&mask, &mask, nodemask);
 618	}
 619	ret = irq_do_set_affinity(&desc->irq_data, &mask, false);
 620	raw_spin_unlock(&mask_lock);
 621	return ret;
 622}
 623#else
 624/* Wrapper for ALPHA specific affinity selector magic */
 625int irq_setup_affinity(struct irq_desc *desc)
 626{
 627	return irq_select_affinity(irq_desc_get_irq(desc));
 628}
 629#endif /* CONFIG_AUTO_IRQ_AFFINITY */
 630#endif /* CONFIG_SMP */
 631
 632
 633/**
 634 *	irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
 635 *	@irq: interrupt number to set affinity
 636 *	@vcpu_info: vCPU specific data or pointer to a percpu array of vCPU
 637 *	            specific data for percpu_devid interrupts
 638 *
 639 *	This function uses the vCPU specific data to set the vCPU
 640 *	affinity for an irq. The vCPU specific data is passed from
 641 *	outside, such as KVM. One example code path is as below:
 642 *	KVM -> IOMMU -> irq_set_vcpu_affinity().
 643 */
 644int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
 645{
 646	unsigned long flags;
 647	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 648	struct irq_data *data;
 649	struct irq_chip *chip;
 650	int ret = -ENOSYS;
 651
 652	if (!desc)
 653		return -EINVAL;
 654
 655	data = irq_desc_get_irq_data(desc);
 656	do {
 657		chip = irq_data_get_irq_chip(data);
 658		if (chip && chip->irq_set_vcpu_affinity)
 659			break;
 660#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
 661		data = data->parent_data;
 662#else
 663		data = NULL;
 664#endif
 665	} while (data);
 666
 667	if (data)
 668		ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
 669	irq_put_desc_unlock(desc, flags);
 670
 671	return ret;
 672}
 673EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
 674
 675void __disable_irq(struct irq_desc *desc)
 676{
 677	if (!desc->depth++)
 678		irq_disable(desc);
 679}
 680
 681static int __disable_irq_nosync(unsigned int irq)
 682{
 683	unsigned long flags;
 684	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 685
 686	if (!desc)
 687		return -EINVAL;
 688	__disable_irq(desc);
 689	irq_put_desc_busunlock(desc, flags);
 690	return 0;
 691}
 692
 693/**
 694 *	disable_irq_nosync - disable an irq without waiting
 695 *	@irq: Interrupt to disable
 696 *
 697 *	Disable the selected interrupt line.  Disables and Enables are
 698 *	nested.
 699 *	Unlike disable_irq(), this function does not ensure existing
 700 *	instances of the IRQ handler have completed before returning.
 701 *
 702 *	This function may be called from IRQ context.
 703 */
 704void disable_irq_nosync(unsigned int irq)
 705{
 706	__disable_irq_nosync(irq);
 707}
 708EXPORT_SYMBOL(disable_irq_nosync);
 709
 710/**
 711 *	disable_irq - disable an irq and wait for completion
 712 *	@irq: Interrupt to disable
 713 *
 714 *	Disable the selected interrupt line.  Enables and Disables are
 715 *	nested.
 716 *	This function waits for any pending IRQ handlers for this interrupt
 717 *	to complete before returning. If you use this function while
 718 *	holding a resource the IRQ handler may need you will deadlock.
 719 *
 720 *	This function may be called - with care - from IRQ context.
 
 
 721 */
 722void disable_irq(unsigned int irq)
 723{
 
 724	if (!__disable_irq_nosync(irq))
 725		synchronize_irq(irq);
 726}
 727EXPORT_SYMBOL(disable_irq);
 728
 729/**
 730 *	disable_hardirq - disables an irq and waits for hardirq completion
 731 *	@irq: Interrupt to disable
 732 *
 733 *	Disable the selected interrupt line.  Enables and Disables are
 734 *	nested.
 735 *	This function waits for any pending hard IRQ handlers for this
 736 *	interrupt to complete before returning. If you use this function while
 737 *	holding a resource the hard IRQ handler may need you will deadlock.
 738 *
 739 *	When used to optimistically disable an interrupt from atomic context
 740 *	the return value must be checked.
 741 *
 742 *	Returns: false if a threaded handler is active.
 743 *
 744 *	This function may be called - with care - from IRQ context.
 745 */
 746bool disable_hardirq(unsigned int irq)
 747{
 748	if (!__disable_irq_nosync(irq))
 749		return synchronize_hardirq(irq);
 750
 751	return false;
 752}
 753EXPORT_SYMBOL_GPL(disable_hardirq);
 754
 755/**
 756 *	disable_nmi_nosync - disable an nmi without waiting
 757 *	@irq: Interrupt to disable
 758 *
 759 *	Disable the selected interrupt line. Disables and enables are
 760 *	nested.
 761 *	The interrupt to disable must have been requested through request_nmi.
 762 *	Unlike disable_nmi(), this function does not ensure existing
 763 *	instances of the IRQ handler have completed before returning.
 764 */
 765void disable_nmi_nosync(unsigned int irq)
 766{
 767	disable_irq_nosync(irq);
 768}
 769
 770void __enable_irq(struct irq_desc *desc)
 771{
 772	switch (desc->depth) {
 773	case 0:
 774 err_out:
 775		WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
 776		     irq_desc_get_irq(desc));
 777		break;
 778	case 1: {
 779		if (desc->istate & IRQS_SUSPENDED)
 780			goto err_out;
 781		/* Prevent probing on this irq: */
 782		irq_settings_set_noprobe(desc);
 783		/*
 784		 * Call irq_startup() not irq_enable() here because the
 785		 * interrupt might be marked NOAUTOEN. So irq_startup()
 786		 * needs to be invoked when it gets enabled the first
 787		 * time. If it was already started up, then irq_startup()
 788		 * will invoke irq_enable() under the hood.
 789		 */
 790		irq_startup(desc, IRQ_RESEND, IRQ_START_FORCE);
 791		break;
 792	}
 793	default:
 794		desc->depth--;
 795	}
 796}
 797
 798/**
 799 *	enable_irq - enable handling of an irq
 800 *	@irq: Interrupt to enable
 801 *
 802 *	Undoes the effect of one call to disable_irq().  If this
 803 *	matches the last disable, processing of interrupts on this
 804 *	IRQ line is re-enabled.
 805 *
 806 *	This function may be called from IRQ context only when
 807 *	desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
 808 */
 809void enable_irq(unsigned int irq)
 810{
 811	unsigned long flags;
 812	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 813
 814	if (!desc)
 815		return;
 816	if (WARN(!desc->irq_data.chip,
 817		 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
 818		goto out;
 819
 820	__enable_irq(desc);
 821out:
 822	irq_put_desc_busunlock(desc, flags);
 823}
 824EXPORT_SYMBOL(enable_irq);
 825
 826/**
 827 *	enable_nmi - enable handling of an nmi
 828 *	@irq: Interrupt to enable
 829 *
 830 *	The interrupt to enable must have been requested through request_nmi.
 831 *	Undoes the effect of one call to disable_nmi(). If this
 832 *	matches the last disable, processing of interrupts on this
 833 *	IRQ line is re-enabled.
 834 */
 835void enable_nmi(unsigned int irq)
 836{
 837	enable_irq(irq);
 838}
 839
 840static int set_irq_wake_real(unsigned int irq, unsigned int on)
 841{
 842	struct irq_desc *desc = irq_to_desc(irq);
 843	int ret = -ENXIO;
 844
 845	if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
 846		return 0;
 847
 848	if (desc->irq_data.chip->irq_set_wake)
 849		ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
 850
 851	return ret;
 852}
 853
 854/**
 855 *	irq_set_irq_wake - control irq power management wakeup
 856 *	@irq:	interrupt to control
 857 *	@on:	enable/disable power management wakeup
 858 *
 859 *	Enable/disable power management wakeup mode, which is
 860 *	disabled by default.  Enables and disables must match,
 861 *	just as they match for non-wakeup mode support.
 862 *
 863 *	Wakeup mode lets this IRQ wake the system from sleep
 864 *	states like "suspend to RAM".
 865 *
 866 *	Note: irq enable/disable state is completely orthogonal
 867 *	to the enable/disable state of irq wake. An irq can be
 868 *	disabled with disable_irq() and still wake the system as
 869 *	long as the irq has wake enabled. If this does not hold,
 870 *	then the underlying irq chip and the related driver need
 871 *	to be investigated.
 872 */
 873int irq_set_irq_wake(unsigned int irq, unsigned int on)
 874{
 875	unsigned long flags;
 876	struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 877	int ret = 0;
 878
 879	if (!desc)
 880		return -EINVAL;
 881
 882	/* Don't use NMIs as wake up interrupts please */
 883	if (desc->istate & IRQS_NMI) {
 884		ret = -EINVAL;
 885		goto out_unlock;
 886	}
 887
 888	/* wakeup-capable irqs can be shared between drivers that
 889	 * don't need to have the same sleep mode behaviors.
 890	 */
 891	if (on) {
 892		if (desc->wake_depth++ == 0) {
 893			ret = set_irq_wake_real(irq, on);
 894			if (ret)
 895				desc->wake_depth = 0;
 896			else
 897				irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
 898		}
 899	} else {
 900		if (desc->wake_depth == 0) {
 901			WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
 902		} else if (--desc->wake_depth == 0) {
 903			ret = set_irq_wake_real(irq, on);
 904			if (ret)
 905				desc->wake_depth = 1;
 906			else
 907				irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
 908		}
 909	}
 910
 911out_unlock:
 912	irq_put_desc_busunlock(desc, flags);
 913	return ret;
 914}
 915EXPORT_SYMBOL(irq_set_irq_wake);
 916
 917/*
 918 * Internal function that tells the architecture code whether a
 919 * particular irq has been exclusively allocated or is available
 920 * for driver use.
 921 */
 922int can_request_irq(unsigned int irq, unsigned long irqflags)
 923{
 924	unsigned long flags;
 925	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 926	int canrequest = 0;
 927
 928	if (!desc)
 929		return 0;
 930
 931	if (irq_settings_can_request(desc)) {
 932		if (!desc->action ||
 933		    irqflags & desc->action->flags & IRQF_SHARED)
 934			canrequest = 1;
 935	}
 936	irq_put_desc_unlock(desc, flags);
 937	return canrequest;
 938}
 939
 940int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
 941{
 942	struct irq_chip *chip = desc->irq_data.chip;
 943	int ret, unmask = 0;
 944
 945	if (!chip || !chip->irq_set_type) {
 946		/*
 947		 * IRQF_TRIGGER_* but the PIC does not support multiple
 948		 * flow-types?
 949		 */
 950		pr_debug("No set_type function for IRQ %d (%s)\n",
 951			 irq_desc_get_irq(desc),
 952			 chip ? (chip->name ? : "unknown") : "unknown");
 953		return 0;
 954	}
 955
 956	if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
 957		if (!irqd_irq_masked(&desc->irq_data))
 958			mask_irq(desc);
 959		if (!irqd_irq_disabled(&desc->irq_data))
 960			unmask = 1;
 961	}
 962
 963	/* Mask all flags except trigger mode */
 964	flags &= IRQ_TYPE_SENSE_MASK;
 965	ret = chip->irq_set_type(&desc->irq_data, flags);
 966
 967	switch (ret) {
 968	case IRQ_SET_MASK_OK:
 969	case IRQ_SET_MASK_OK_DONE:
 970		irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
 971		irqd_set(&desc->irq_data, flags);
 972		fallthrough;
 973
 974	case IRQ_SET_MASK_OK_NOCOPY:
 975		flags = irqd_get_trigger_type(&desc->irq_data);
 976		irq_settings_set_trigger_mask(desc, flags);
 977		irqd_clear(&desc->irq_data, IRQD_LEVEL);
 978		irq_settings_clr_level(desc);
 979		if (flags & IRQ_TYPE_LEVEL_MASK) {
 980			irq_settings_set_level(desc);
 981			irqd_set(&desc->irq_data, IRQD_LEVEL);
 982		}
 983
 984		ret = 0;
 985		break;
 986	default:
 987		pr_err("Setting trigger mode %lu for irq %u failed (%pS)\n",
 988		       flags, irq_desc_get_irq(desc), chip->irq_set_type);
 989	}
 990	if (unmask)
 991		unmask_irq(desc);
 992	return ret;
 993}
 994
 995#ifdef CONFIG_HARDIRQS_SW_RESEND
 996int irq_set_parent(int irq, int parent_irq)
 997{
 998	unsigned long flags;
 999	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
1000
1001	if (!desc)
1002		return -EINVAL;
1003
1004	desc->parent_irq = parent_irq;
1005
1006	irq_put_desc_unlock(desc, flags);
1007	return 0;
1008}
1009EXPORT_SYMBOL_GPL(irq_set_parent);
1010#endif
1011
1012/*
1013 * Default primary interrupt handler for threaded interrupts. Is
1014 * assigned as primary handler when request_threaded_irq is called
1015 * with handler == NULL. Useful for oneshot interrupts.
1016 */
1017static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
1018{
1019	return IRQ_WAKE_THREAD;
1020}
1021
1022/*
1023 * Primary handler for nested threaded interrupts. Should never be
1024 * called.
1025 */
1026static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
1027{
1028	WARN(1, "Primary handler called for nested irq %d\n", irq);
1029	return IRQ_NONE;
1030}
1031
1032static irqreturn_t irq_forced_secondary_handler(int irq, void *dev_id)
1033{
1034	WARN(1, "Secondary action handler called for irq %d\n", irq);
1035	return IRQ_NONE;
1036}
1037
1038static int irq_wait_for_interrupt(struct irqaction *action)
1039{
1040	for (;;) {
1041		set_current_state(TASK_INTERRUPTIBLE);
1042
1043		if (kthread_should_stop()) {
1044			/* may need to run one last time */
1045			if (test_and_clear_bit(IRQTF_RUNTHREAD,
1046					       &action->thread_flags)) {
1047				__set_current_state(TASK_RUNNING);
1048				return 0;
1049			}
1050			__set_current_state(TASK_RUNNING);
1051			return -1;
1052		}
1053
1054		if (test_and_clear_bit(IRQTF_RUNTHREAD,
1055				       &action->thread_flags)) {
1056			__set_current_state(TASK_RUNNING);
1057			return 0;
1058		}
1059		schedule();
1060	}
1061}
1062
1063/*
1064 * Oneshot interrupts keep the irq line masked until the threaded
1065 * handler finished. unmask if the interrupt has not been disabled and
1066 * is marked MASKED.
1067 */
1068static void irq_finalize_oneshot(struct irq_desc *desc,
1069				 struct irqaction *action)
1070{
1071	if (!(desc->istate & IRQS_ONESHOT) ||
1072	    action->handler == irq_forced_secondary_handler)
1073		return;
1074again:
1075	chip_bus_lock(desc);
1076	raw_spin_lock_irq(&desc->lock);
1077
1078	/*
1079	 * Implausible though it may be we need to protect us against
1080	 * the following scenario:
1081	 *
1082	 * The thread is faster done than the hard interrupt handler
1083	 * on the other CPU. If we unmask the irq line then the
1084	 * interrupt can come in again and masks the line, leaves due
1085	 * to IRQS_INPROGRESS and the irq line is masked forever.
1086	 *
1087	 * This also serializes the state of shared oneshot handlers
1088	 * versus "desc->threads_oneshot |= action->thread_mask;" in
1089	 * irq_wake_thread(). See the comment there which explains the
1090	 * serialization.
1091	 */
1092	if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
1093		raw_spin_unlock_irq(&desc->lock);
1094		chip_bus_sync_unlock(desc);
1095		cpu_relax();
1096		goto again;
1097	}
1098
1099	/*
1100	 * Now check again, whether the thread should run. Otherwise
1101	 * we would clear the threads_oneshot bit of this thread which
1102	 * was just set.
1103	 */
1104	if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1105		goto out_unlock;
1106
1107	desc->threads_oneshot &= ~action->thread_mask;
1108
1109	if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
1110	    irqd_irq_masked(&desc->irq_data))
1111		unmask_threaded_irq(desc);
1112
1113out_unlock:
1114	raw_spin_unlock_irq(&desc->lock);
1115	chip_bus_sync_unlock(desc);
1116}
1117
1118#ifdef CONFIG_SMP
1119/*
1120 * Check whether we need to change the affinity of the interrupt thread.
1121 */
1122static void
1123irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
1124{
1125	cpumask_var_t mask;
1126	bool valid = true;
1127
1128	if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
1129		return;
1130
1131	/*
1132	 * In case we are out of memory we set IRQTF_AFFINITY again and
1133	 * try again next time
1134	 */
1135	if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1136		set_bit(IRQTF_AFFINITY, &action->thread_flags);
1137		return;
1138	}
1139
1140	raw_spin_lock_irq(&desc->lock);
1141	/*
1142	 * This code is triggered unconditionally. Check the affinity
1143	 * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
1144	 */
1145	if (cpumask_available(desc->irq_common_data.affinity)) {
1146		const struct cpumask *m;
1147
1148		m = irq_data_get_effective_affinity_mask(&desc->irq_data);
1149		cpumask_copy(mask, m);
1150	} else {
1151		valid = false;
1152	}
1153	raw_spin_unlock_irq(&desc->lock);
1154
1155	if (valid)
1156		set_cpus_allowed_ptr(current, mask);
1157	free_cpumask_var(mask);
1158}
1159#else
1160static inline void
1161irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
1162#endif
1163
1164/*
1165 * Interrupts which are not explicitly requested as threaded
1166 * interrupts rely on the implicit bh/preempt disable of the hard irq
1167 * context. So we need to disable bh here to avoid deadlocks and other
1168 * side effects.
1169 */
1170static irqreturn_t
1171irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
1172{
1173	irqreturn_t ret;
1174
1175	local_bh_disable();
1176	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1177		local_irq_disable();
1178	ret = action->thread_fn(action->irq, action->dev_id);
1179	if (ret == IRQ_HANDLED)
1180		atomic_inc(&desc->threads_handled);
1181
1182	irq_finalize_oneshot(desc, action);
1183	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1184		local_irq_enable();
1185	local_bh_enable();
1186	return ret;
1187}
1188
1189/*
1190 * Interrupts explicitly requested as threaded interrupts want to be
1191 * preemptible - many of them need to sleep and wait for slow busses to
1192 * complete.
1193 */
1194static irqreturn_t irq_thread_fn(struct irq_desc *desc,
1195		struct irqaction *action)
1196{
1197	irqreturn_t ret;
1198
1199	ret = action->thread_fn(action->irq, action->dev_id);
1200	if (ret == IRQ_HANDLED)
1201		atomic_inc(&desc->threads_handled);
1202
1203	irq_finalize_oneshot(desc, action);
1204	return ret;
1205}
1206
1207static void wake_threads_waitq(struct irq_desc *desc)
1208{
1209	if (atomic_dec_and_test(&desc->threads_active))
1210		wake_up(&desc->wait_for_threads);
1211}
1212
1213static void irq_thread_dtor(struct callback_head *unused)
1214{
1215	struct task_struct *tsk = current;
1216	struct irq_desc *desc;
1217	struct irqaction *action;
1218
1219	if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
1220		return;
1221
1222	action = kthread_data(tsk);
1223
1224	pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
1225	       tsk->comm, tsk->pid, action->irq);
1226
1227
1228	desc = irq_to_desc(action->irq);
1229	/*
1230	 * If IRQTF_RUNTHREAD is set, we need to decrement
1231	 * desc->threads_active and wake possible waiters.
1232	 */
1233	if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
1234		wake_threads_waitq(desc);
1235
1236	/* Prevent a stale desc->threads_oneshot */
1237	irq_finalize_oneshot(desc, action);
1238}
1239
1240static void irq_wake_secondary(struct irq_desc *desc, struct irqaction *action)
1241{
1242	struct irqaction *secondary = action->secondary;
1243
1244	if (WARN_ON_ONCE(!secondary))
1245		return;
1246
1247	raw_spin_lock_irq(&desc->lock);
1248	__irq_wake_thread(desc, secondary);
1249	raw_spin_unlock_irq(&desc->lock);
1250}
1251
1252/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1253 * Interrupt handler thread
1254 */
1255static int irq_thread(void *data)
1256{
1257	struct callback_head on_exit_work;
1258	struct irqaction *action = data;
1259	struct irq_desc *desc = irq_to_desc(action->irq);
1260	irqreturn_t (*handler_fn)(struct irq_desc *desc,
1261			struct irqaction *action);
1262
1263	if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
1264					&action->thread_flags))
 
 
 
 
1265		handler_fn = irq_forced_thread_fn;
1266	else
1267		handler_fn = irq_thread_fn;
1268
1269	init_task_work(&on_exit_work, irq_thread_dtor);
1270	task_work_add(current, &on_exit_work, TWA_NONE);
1271
1272	irq_thread_check_affinity(desc, action);
1273
1274	while (!irq_wait_for_interrupt(action)) {
1275		irqreturn_t action_ret;
1276
1277		irq_thread_check_affinity(desc, action);
1278
1279		action_ret = handler_fn(desc, action);
1280		if (action_ret == IRQ_WAKE_THREAD)
1281			irq_wake_secondary(desc, action);
1282
1283		wake_threads_waitq(desc);
1284	}
1285
1286	/*
1287	 * This is the regular exit path. __free_irq() is stopping the
1288	 * thread via kthread_stop() after calling
1289	 * synchronize_hardirq(). So neither IRQTF_RUNTHREAD nor the
1290	 * oneshot mask bit can be set.
1291	 */
1292	task_work_cancel(current, irq_thread_dtor);
1293	return 0;
1294}
1295
1296/**
1297 *	irq_wake_thread - wake the irq thread for the action identified by dev_id
1298 *	@irq:		Interrupt line
1299 *	@dev_id:	Device identity for which the thread should be woken
1300 *
1301 */
1302void irq_wake_thread(unsigned int irq, void *dev_id)
1303{
1304	struct irq_desc *desc = irq_to_desc(irq);
1305	struct irqaction *action;
1306	unsigned long flags;
1307
1308	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1309		return;
1310
1311	raw_spin_lock_irqsave(&desc->lock, flags);
1312	for_each_action_of_desc(desc, action) {
1313		if (action->dev_id == dev_id) {
1314			if (action->thread)
1315				__irq_wake_thread(desc, action);
1316			break;
1317		}
1318	}
1319	raw_spin_unlock_irqrestore(&desc->lock, flags);
1320}
1321EXPORT_SYMBOL_GPL(irq_wake_thread);
1322
1323static int irq_setup_forced_threading(struct irqaction *new)
1324{
1325	if (!force_irqthreads)
1326		return 0;
1327	if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
1328		return 0;
1329
1330	/*
1331	 * No further action required for interrupts which are requested as
1332	 * threaded interrupts already
1333	 */
1334	if (new->handler == irq_default_primary_handler)
1335		return 0;
1336
1337	new->flags |= IRQF_ONESHOT;
1338
1339	/*
1340	 * Handle the case where we have a real primary handler and a
1341	 * thread handler. We force thread them as well by creating a
1342	 * secondary action.
1343	 */
1344	if (new->handler && new->thread_fn) {
1345		/* Allocate the secondary action */
1346		new->secondary = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1347		if (!new->secondary)
1348			return -ENOMEM;
1349		new->secondary->handler = irq_forced_secondary_handler;
1350		new->secondary->thread_fn = new->thread_fn;
1351		new->secondary->dev_id = new->dev_id;
1352		new->secondary->irq = new->irq;
1353		new->secondary->name = new->name;
1354	}
1355	/* Deal with the primary handler */
1356	set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
1357	new->thread_fn = new->handler;
1358	new->handler = irq_default_primary_handler;
1359	return 0;
1360}
1361
1362static int irq_request_resources(struct irq_desc *desc)
1363{
1364	struct irq_data *d = &desc->irq_data;
1365	struct irq_chip *c = d->chip;
1366
1367	return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1368}
1369
1370static void irq_release_resources(struct irq_desc *desc)
1371{
1372	struct irq_data *d = &desc->irq_data;
1373	struct irq_chip *c = d->chip;
1374
1375	if (c->irq_release_resources)
1376		c->irq_release_resources(d);
1377}
1378
1379static bool irq_supports_nmi(struct irq_desc *desc)
1380{
1381	struct irq_data *d = irq_desc_get_irq_data(desc);
1382
1383#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1384	/* Only IRQs directly managed by the root irqchip can be set as NMI */
1385	if (d->parent_data)
1386		return false;
1387#endif
1388	/* Don't support NMIs for chips behind a slow bus */
1389	if (d->chip->irq_bus_lock || d->chip->irq_bus_sync_unlock)
1390		return false;
1391
1392	return d->chip->flags & IRQCHIP_SUPPORTS_NMI;
1393}
1394
1395static int irq_nmi_setup(struct irq_desc *desc)
1396{
1397	struct irq_data *d = irq_desc_get_irq_data(desc);
1398	struct irq_chip *c = d->chip;
1399
1400	return c->irq_nmi_setup ? c->irq_nmi_setup(d) : -EINVAL;
1401}
1402
1403static void irq_nmi_teardown(struct irq_desc *desc)
1404{
1405	struct irq_data *d = irq_desc_get_irq_data(desc);
1406	struct irq_chip *c = d->chip;
1407
1408	if (c->irq_nmi_teardown)
1409		c->irq_nmi_teardown(d);
1410}
1411
1412static int
1413setup_irq_thread(struct irqaction *new, unsigned int irq, bool secondary)
1414{
1415	struct task_struct *t;
1416
1417	if (!secondary) {
1418		t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1419				   new->name);
1420	} else {
1421		t = kthread_create(irq_thread, new, "irq/%d-s-%s", irq,
1422				   new->name);
1423	}
1424
1425	if (IS_ERR(t))
1426		return PTR_ERR(t);
1427
1428	sched_set_fifo(t);
1429
1430	/*
1431	 * We keep the reference to the task struct even if
1432	 * the thread dies to avoid that the interrupt code
1433	 * references an already freed task_struct.
1434	 */
1435	new->thread = get_task_struct(t);
1436	/*
1437	 * Tell the thread to set its affinity. This is
1438	 * important for shared interrupt handlers as we do
1439	 * not invoke setup_affinity() for the secondary
1440	 * handlers as everything is already set up. Even for
1441	 * interrupts marked with IRQF_NO_BALANCE this is
1442	 * correct as we want the thread to move to the cpu(s)
1443	 * on which the requesting code placed the interrupt.
1444	 */
1445	set_bit(IRQTF_AFFINITY, &new->thread_flags);
1446	return 0;
1447}
1448
1449/*
1450 * Internal function to register an irqaction - typically used to
1451 * allocate special interrupts that are part of the architecture.
1452 *
1453 * Locking rules:
1454 *
1455 * desc->request_mutex	Provides serialization against a concurrent free_irq()
1456 *   chip_bus_lock	Provides serialization for slow bus operations
1457 *     desc->lock	Provides serialization against hard interrupts
1458 *
1459 * chip_bus_lock and desc->lock are sufficient for all other management and
1460 * interrupt related functions. desc->request_mutex solely serializes
1461 * request/free_irq().
1462 */
1463static int
1464__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1465{
1466	struct irqaction *old, **old_ptr;
1467	unsigned long flags, thread_mask = 0;
1468	int ret, nested, shared = 0;
1469
1470	if (!desc)
1471		return -EINVAL;
1472
1473	if (desc->irq_data.chip == &no_irq_chip)
1474		return -ENOSYS;
1475	if (!try_module_get(desc->owner))
1476		return -ENODEV;
1477
1478	new->irq = irq;
1479
1480	/*
1481	 * If the trigger type is not specified by the caller,
1482	 * then use the default for this interrupt.
1483	 */
1484	if (!(new->flags & IRQF_TRIGGER_MASK))
1485		new->flags |= irqd_get_trigger_type(&desc->irq_data);
1486
1487	/*
1488	 * Check whether the interrupt nests into another interrupt
1489	 * thread.
1490	 */
1491	nested = irq_settings_is_nested_thread(desc);
1492	if (nested) {
1493		if (!new->thread_fn) {
1494			ret = -EINVAL;
1495			goto out_mput;
1496		}
1497		/*
1498		 * Replace the primary handler which was provided from
1499		 * the driver for non nested interrupt handling by the
1500		 * dummy function which warns when called.
1501		 */
1502		new->handler = irq_nested_primary_handler;
1503	} else {
1504		if (irq_settings_can_thread(desc)) {
1505			ret = irq_setup_forced_threading(new);
1506			if (ret)
1507				goto out_mput;
1508		}
1509	}
1510
1511	/*
1512	 * Create a handler thread when a thread function is supplied
1513	 * and the interrupt does not nest into another interrupt
1514	 * thread.
1515	 */
1516	if (new->thread_fn && !nested) {
1517		ret = setup_irq_thread(new, irq, false);
1518		if (ret)
1519			goto out_mput;
1520		if (new->secondary) {
1521			ret = setup_irq_thread(new->secondary, irq, true);
1522			if (ret)
1523				goto out_thread;
1524		}
1525	}
1526
1527	/*
1528	 * Drivers are often written to work w/o knowledge about the
1529	 * underlying irq chip implementation, so a request for a
1530	 * threaded irq without a primary hard irq context handler
1531	 * requires the ONESHOT flag to be set. Some irq chips like
1532	 * MSI based interrupts are per se one shot safe. Check the
1533	 * chip flags, so we can avoid the unmask dance at the end of
1534	 * the threaded handler for those.
1535	 */
1536	if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1537		new->flags &= ~IRQF_ONESHOT;
1538
1539	/*
1540	 * Protects against a concurrent __free_irq() call which might wait
1541	 * for synchronize_hardirq() to complete without holding the optional
1542	 * chip bus lock and desc->lock. Also protects against handing out
1543	 * a recycled oneshot thread_mask bit while it's still in use by
1544	 * its previous owner.
1545	 */
1546	mutex_lock(&desc->request_mutex);
1547
1548	/*
1549	 * Acquire bus lock as the irq_request_resources() callback below
1550	 * might rely on the serialization or the magic power management
1551	 * functions which are abusing the irq_bus_lock() callback,
1552	 */
1553	chip_bus_lock(desc);
1554
1555	/* First installed action requests resources. */
1556	if (!desc->action) {
1557		ret = irq_request_resources(desc);
1558		if (ret) {
1559			pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1560			       new->name, irq, desc->irq_data.chip->name);
1561			goto out_bus_unlock;
1562		}
1563	}
1564
1565	/*
1566	 * The following block of code has to be executed atomically
1567	 * protected against a concurrent interrupt and any of the other
1568	 * management calls which are not serialized via
1569	 * desc->request_mutex or the optional bus lock.
1570	 */
1571	raw_spin_lock_irqsave(&desc->lock, flags);
1572	old_ptr = &desc->action;
1573	old = *old_ptr;
1574	if (old) {
1575		/*
1576		 * Can't share interrupts unless both agree to and are
1577		 * the same type (level, edge, polarity). So both flag
1578		 * fields must have IRQF_SHARED set and the bits which
1579		 * set the trigger type must match. Also all must
1580		 * agree on ONESHOT.
1581		 * Interrupt lines used for NMIs cannot be shared.
1582		 */
1583		unsigned int oldtype;
1584
1585		if (desc->istate & IRQS_NMI) {
1586			pr_err("Invalid attempt to share NMI for %s (irq %d) on irqchip %s.\n",
1587				new->name, irq, desc->irq_data.chip->name);
1588			ret = -EINVAL;
1589			goto out_unlock;
1590		}
1591
1592		/*
1593		 * If nobody did set the configuration before, inherit
1594		 * the one provided by the requester.
1595		 */
1596		if (irqd_trigger_type_was_set(&desc->irq_data)) {
1597			oldtype = irqd_get_trigger_type(&desc->irq_data);
1598		} else {
1599			oldtype = new->flags & IRQF_TRIGGER_MASK;
1600			irqd_set_trigger_type(&desc->irq_data, oldtype);
1601		}
1602
1603		if (!((old->flags & new->flags) & IRQF_SHARED) ||
1604		    (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
1605		    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1606			goto mismatch;
1607
1608		/* All handlers must agree on per-cpuness */
1609		if ((old->flags & IRQF_PERCPU) !=
1610		    (new->flags & IRQF_PERCPU))
1611			goto mismatch;
1612
1613		/* add new interrupt at end of irq queue */
1614		do {
1615			/*
1616			 * Or all existing action->thread_mask bits,
1617			 * so we can find the next zero bit for this
1618			 * new action.
1619			 */
1620			thread_mask |= old->thread_mask;
1621			old_ptr = &old->next;
1622			old = *old_ptr;
1623		} while (old);
1624		shared = 1;
1625	}
1626
1627	/*
1628	 * Setup the thread mask for this irqaction for ONESHOT. For
1629	 * !ONESHOT irqs the thread mask is 0 so we can avoid a
1630	 * conditional in irq_wake_thread().
1631	 */
1632	if (new->flags & IRQF_ONESHOT) {
1633		/*
1634		 * Unlikely to have 32 resp 64 irqs sharing one line,
1635		 * but who knows.
1636		 */
1637		if (thread_mask == ~0UL) {
1638			ret = -EBUSY;
1639			goto out_unlock;
1640		}
1641		/*
1642		 * The thread_mask for the action is or'ed to
1643		 * desc->thread_active to indicate that the
1644		 * IRQF_ONESHOT thread handler has been woken, but not
1645		 * yet finished. The bit is cleared when a thread
1646		 * completes. When all threads of a shared interrupt
1647		 * line have completed desc->threads_active becomes
1648		 * zero and the interrupt line is unmasked. See
1649		 * handle.c:irq_wake_thread() for further information.
1650		 *
1651		 * If no thread is woken by primary (hard irq context)
1652		 * interrupt handlers, then desc->threads_active is
1653		 * also checked for zero to unmask the irq line in the
1654		 * affected hard irq flow handlers
1655		 * (handle_[fasteoi|level]_irq).
1656		 *
1657		 * The new action gets the first zero bit of
1658		 * thread_mask assigned. See the loop above which or's
1659		 * all existing action->thread_mask bits.
1660		 */
1661		new->thread_mask = 1UL << ffz(thread_mask);
1662
1663	} else if (new->handler == irq_default_primary_handler &&
1664		   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1665		/*
1666		 * The interrupt was requested with handler = NULL, so
1667		 * we use the default primary handler for it. But it
1668		 * does not have the oneshot flag set. In combination
1669		 * with level interrupts this is deadly, because the
1670		 * default primary handler just wakes the thread, then
1671		 * the irq lines is reenabled, but the device still
1672		 * has the level irq asserted. Rinse and repeat....
1673		 *
1674		 * While this works for edge type interrupts, we play
1675		 * it safe and reject unconditionally because we can't
1676		 * say for sure which type this interrupt really
1677		 * has. The type flags are unreliable as the
1678		 * underlying chip implementation can override them.
1679		 */
1680		pr_err("Threaded irq requested with handler=NULL and !ONESHOT for %s (irq %d)\n",
1681		       new->name, irq);
1682		ret = -EINVAL;
1683		goto out_unlock;
1684	}
1685
1686	if (!shared) {
1687		init_waitqueue_head(&desc->wait_for_threads);
1688
1689		/* Setup the type (level, edge polarity) if configured: */
1690		if (new->flags & IRQF_TRIGGER_MASK) {
1691			ret = __irq_set_trigger(desc,
1692						new->flags & IRQF_TRIGGER_MASK);
1693
1694			if (ret)
1695				goto out_unlock;
1696		}
1697
1698		/*
1699		 * Activate the interrupt. That activation must happen
1700		 * independently of IRQ_NOAUTOEN. request_irq() can fail
1701		 * and the callers are supposed to handle
1702		 * that. enable_irq() of an interrupt requested with
1703		 * IRQ_NOAUTOEN is not supposed to fail. The activation
1704		 * keeps it in shutdown mode, it merily associates
1705		 * resources if necessary and if that's not possible it
1706		 * fails. Interrupts which are in managed shutdown mode
1707		 * will simply ignore that activation request.
1708		 */
1709		ret = irq_activate(desc);
1710		if (ret)
1711			goto out_unlock;
1712
1713		desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1714				  IRQS_ONESHOT | IRQS_WAITING);
1715		irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1716
1717		if (new->flags & IRQF_PERCPU) {
1718			irqd_set(&desc->irq_data, IRQD_PER_CPU);
1719			irq_settings_set_per_cpu(desc);
1720			if (new->flags & IRQF_NO_DEBUG)
1721				irq_settings_set_no_debug(desc);
1722		}
1723
1724		if (noirqdebug)
1725			irq_settings_set_no_debug(desc);
1726
1727		if (new->flags & IRQF_ONESHOT)
1728			desc->istate |= IRQS_ONESHOT;
1729
1730		/* Exclude IRQ from balancing if requested */
1731		if (new->flags & IRQF_NOBALANCING) {
1732			irq_settings_set_no_balancing(desc);
1733			irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1734		}
1735
1736		if (!(new->flags & IRQF_NO_AUTOEN) &&
1737		    irq_settings_can_autoenable(desc)) {
1738			irq_startup(desc, IRQ_RESEND, IRQ_START_COND);
1739		} else {
1740			/*
1741			 * Shared interrupts do not go well with disabling
1742			 * auto enable. The sharing interrupt might request
1743			 * it while it's still disabled and then wait for
1744			 * interrupts forever.
1745			 */
1746			WARN_ON_ONCE(new->flags & IRQF_SHARED);
1747			/* Undo nested disables: */
1748			desc->depth = 1;
1749		}
1750
1751	} else if (new->flags & IRQF_TRIGGER_MASK) {
1752		unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1753		unsigned int omsk = irqd_get_trigger_type(&desc->irq_data);
1754
1755		if (nmsk != omsk)
1756			/* hope the handler works with current  trigger mode */
1757			pr_warn("irq %d uses trigger mode %u; requested %u\n",
1758				irq, omsk, nmsk);
1759	}
1760
1761	*old_ptr = new;
1762
1763	irq_pm_install_action(desc, new);
1764
1765	/* Reset broken irq detection when installing new handler */
1766	desc->irq_count = 0;
1767	desc->irqs_unhandled = 0;
1768
1769	/*
1770	 * Check whether we disabled the irq via the spurious handler
1771	 * before. Reenable it and give it another chance.
1772	 */
1773	if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1774		desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1775		__enable_irq(desc);
1776	}
1777
1778	raw_spin_unlock_irqrestore(&desc->lock, flags);
1779	chip_bus_sync_unlock(desc);
1780	mutex_unlock(&desc->request_mutex);
1781
1782	irq_setup_timings(desc, new);
1783
1784	/*
1785	 * Strictly no need to wake it up, but hung_task complains
1786	 * when no hard interrupt wakes the thread up.
1787	 */
1788	if (new->thread)
1789		wake_up_process(new->thread);
1790	if (new->secondary)
1791		wake_up_process(new->secondary->thread);
1792
1793	register_irq_proc(irq, desc);
1794	new->dir = NULL;
1795	register_handler_proc(irq, new);
1796	return 0;
1797
1798mismatch:
1799	if (!(new->flags & IRQF_PROBE_SHARED)) {
1800		pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1801		       irq, new->flags, new->name, old->flags, old->name);
1802#ifdef CONFIG_DEBUG_SHIRQ
1803		dump_stack();
1804#endif
1805	}
1806	ret = -EBUSY;
1807
1808out_unlock:
1809	raw_spin_unlock_irqrestore(&desc->lock, flags);
1810
1811	if (!desc->action)
1812		irq_release_resources(desc);
1813out_bus_unlock:
1814	chip_bus_sync_unlock(desc);
1815	mutex_unlock(&desc->request_mutex);
1816
1817out_thread:
1818	if (new->thread) {
1819		struct task_struct *t = new->thread;
1820
1821		new->thread = NULL;
1822		kthread_stop(t);
1823		put_task_struct(t);
1824	}
1825	if (new->secondary && new->secondary->thread) {
1826		struct task_struct *t = new->secondary->thread;
1827
1828		new->secondary->thread = NULL;
1829		kthread_stop(t);
1830		put_task_struct(t);
1831	}
1832out_mput:
1833	module_put(desc->owner);
1834	return ret;
1835}
1836
1837/*
1838 * Internal function to unregister an irqaction - used to free
1839 * regular and special interrupts that are part of the architecture.
1840 */
1841static struct irqaction *__free_irq(struct irq_desc *desc, void *dev_id)
1842{
1843	unsigned irq = desc->irq_data.irq;
1844	struct irqaction *action, **action_ptr;
1845	unsigned long flags;
1846
1847	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1848
1849	mutex_lock(&desc->request_mutex);
1850	chip_bus_lock(desc);
1851	raw_spin_lock_irqsave(&desc->lock, flags);
1852
1853	/*
1854	 * There can be multiple actions per IRQ descriptor, find the right
1855	 * one based on the dev_id:
1856	 */
1857	action_ptr = &desc->action;
1858	for (;;) {
1859		action = *action_ptr;
1860
1861		if (!action) {
1862			WARN(1, "Trying to free already-free IRQ %d\n", irq);
1863			raw_spin_unlock_irqrestore(&desc->lock, flags);
1864			chip_bus_sync_unlock(desc);
1865			mutex_unlock(&desc->request_mutex);
1866			return NULL;
1867		}
1868
1869		if (action->dev_id == dev_id)
1870			break;
1871		action_ptr = &action->next;
1872	}
1873
1874	/* Found it - now remove it from the list of entries: */
1875	*action_ptr = action->next;
1876
1877	irq_pm_remove_action(desc, action);
1878
1879	/* If this was the last handler, shut down the IRQ line: */
1880	if (!desc->action) {
1881		irq_settings_clr_disable_unlazy(desc);
1882		/* Only shutdown. Deactivate after synchronize_hardirq() */
1883		irq_shutdown(desc);
1884	}
1885
1886#ifdef CONFIG_SMP
1887	/* make sure affinity_hint is cleaned up */
1888	if (WARN_ON_ONCE(desc->affinity_hint))
1889		desc->affinity_hint = NULL;
1890#endif
1891
1892	raw_spin_unlock_irqrestore(&desc->lock, flags);
1893	/*
1894	 * Drop bus_lock here so the changes which were done in the chip
1895	 * callbacks above are synced out to the irq chips which hang
1896	 * behind a slow bus (I2C, SPI) before calling synchronize_hardirq().
1897	 *
1898	 * Aside of that the bus_lock can also be taken from the threaded
1899	 * handler in irq_finalize_oneshot() which results in a deadlock
1900	 * because kthread_stop() would wait forever for the thread to
1901	 * complete, which is blocked on the bus lock.
1902	 *
1903	 * The still held desc->request_mutex() protects against a
1904	 * concurrent request_irq() of this irq so the release of resources
1905	 * and timing data is properly serialized.
1906	 */
1907	chip_bus_sync_unlock(desc);
1908
1909	unregister_handler_proc(irq, action);
1910
1911	/*
1912	 * Make sure it's not being used on another CPU and if the chip
1913	 * supports it also make sure that there is no (not yet serviced)
1914	 * interrupt in flight at the hardware level.
1915	 */
1916	__synchronize_hardirq(desc, true);
1917
1918#ifdef CONFIG_DEBUG_SHIRQ
1919	/*
1920	 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1921	 * event to happen even now it's being freed, so let's make sure that
1922	 * is so by doing an extra call to the handler ....
1923	 *
1924	 * ( We do this after actually deregistering it, to make sure that a
1925	 *   'real' IRQ doesn't run in parallel with our fake. )
1926	 */
1927	if (action->flags & IRQF_SHARED) {
1928		local_irq_save(flags);
1929		action->handler(irq, dev_id);
1930		local_irq_restore(flags);
1931	}
1932#endif
1933
1934	/*
1935	 * The action has already been removed above, but the thread writes
1936	 * its oneshot mask bit when it completes. Though request_mutex is
1937	 * held across this which prevents __setup_irq() from handing out
1938	 * the same bit to a newly requested action.
1939	 */
1940	if (action->thread) {
1941		kthread_stop(action->thread);
1942		put_task_struct(action->thread);
1943		if (action->secondary && action->secondary->thread) {
1944			kthread_stop(action->secondary->thread);
1945			put_task_struct(action->secondary->thread);
1946		}
1947	}
1948
1949	/* Last action releases resources */
1950	if (!desc->action) {
1951		/*
1952		 * Reacquire bus lock as irq_release_resources() might
1953		 * require it to deallocate resources over the slow bus.
1954		 */
1955		chip_bus_lock(desc);
1956		/*
1957		 * There is no interrupt on the fly anymore. Deactivate it
1958		 * completely.
1959		 */
1960		raw_spin_lock_irqsave(&desc->lock, flags);
1961		irq_domain_deactivate_irq(&desc->irq_data);
1962		raw_spin_unlock_irqrestore(&desc->lock, flags);
1963
1964		irq_release_resources(desc);
1965		chip_bus_sync_unlock(desc);
1966		irq_remove_timings(desc);
1967	}
1968
1969	mutex_unlock(&desc->request_mutex);
1970
1971	irq_chip_pm_put(&desc->irq_data);
1972	module_put(desc->owner);
1973	kfree(action->secondary);
1974	return action;
1975}
1976
1977/**
1978 *	free_irq - free an interrupt allocated with request_irq
1979 *	@irq: Interrupt line to free
1980 *	@dev_id: Device identity to free
1981 *
1982 *	Remove an interrupt handler. The handler is removed and if the
1983 *	interrupt line is no longer in use by any driver it is disabled.
1984 *	On a shared IRQ the caller must ensure the interrupt is disabled
1985 *	on the card it drives before calling this function. The function
1986 *	does not return until any executing interrupts for this IRQ
1987 *	have completed.
1988 *
1989 *	This function must not be called from interrupt context.
1990 *
1991 *	Returns the devname argument passed to request_irq.
1992 */
1993const void *free_irq(unsigned int irq, void *dev_id)
1994{
1995	struct irq_desc *desc = irq_to_desc(irq);
1996	struct irqaction *action;
1997	const char *devname;
1998
1999	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2000		return NULL;
2001
2002#ifdef CONFIG_SMP
2003	if (WARN_ON(desc->affinity_notify))
2004		desc->affinity_notify = NULL;
2005#endif
2006
2007	action = __free_irq(desc, dev_id);
2008
2009	if (!action)
2010		return NULL;
2011
2012	devname = action->name;
2013	kfree(action);
2014	return devname;
2015}
2016EXPORT_SYMBOL(free_irq);
2017
2018/* This function must be called with desc->lock held */
2019static const void *__cleanup_nmi(unsigned int irq, struct irq_desc *desc)
2020{
2021	const char *devname = NULL;
2022
2023	desc->istate &= ~IRQS_NMI;
2024
2025	if (!WARN_ON(desc->action == NULL)) {
2026		irq_pm_remove_action(desc, desc->action);
2027		devname = desc->action->name;
2028		unregister_handler_proc(irq, desc->action);
2029
2030		kfree(desc->action);
2031		desc->action = NULL;
2032	}
2033
2034	irq_settings_clr_disable_unlazy(desc);
2035	irq_shutdown_and_deactivate(desc);
2036
2037	irq_release_resources(desc);
2038
2039	irq_chip_pm_put(&desc->irq_data);
2040	module_put(desc->owner);
2041
2042	return devname;
2043}
2044
2045const void *free_nmi(unsigned int irq, void *dev_id)
2046{
2047	struct irq_desc *desc = irq_to_desc(irq);
2048	unsigned long flags;
2049	const void *devname;
2050
2051	if (!desc || WARN_ON(!(desc->istate & IRQS_NMI)))
2052		return NULL;
2053
2054	if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2055		return NULL;
2056
2057	/* NMI still enabled */
2058	if (WARN_ON(desc->depth == 0))
2059		disable_nmi_nosync(irq);
2060
2061	raw_spin_lock_irqsave(&desc->lock, flags);
2062
2063	irq_nmi_teardown(desc);
2064	devname = __cleanup_nmi(irq, desc);
2065
2066	raw_spin_unlock_irqrestore(&desc->lock, flags);
2067
2068	return devname;
2069}
2070
2071/**
2072 *	request_threaded_irq - allocate an interrupt line
2073 *	@irq: Interrupt line to allocate
2074 *	@handler: Function to be called when the IRQ occurs.
2075 *		  Primary handler for threaded interrupts
2076 *		  If NULL and thread_fn != NULL the default
2077 *		  primary handler is installed
2078 *	@thread_fn: Function called from the irq handler thread
2079 *		    If NULL, no irq thread is created
2080 *	@irqflags: Interrupt type flags
2081 *	@devname: An ascii name for the claiming device
2082 *	@dev_id: A cookie passed back to the handler function
2083 *
2084 *	This call allocates interrupt resources and enables the
2085 *	interrupt line and IRQ handling. From the point this
2086 *	call is made your handler function may be invoked. Since
2087 *	your handler function must clear any interrupt the board
2088 *	raises, you must take care both to initialise your hardware
2089 *	and to set up the interrupt handler in the right order.
2090 *
2091 *	If you want to set up a threaded irq handler for your device
2092 *	then you need to supply @handler and @thread_fn. @handler is
2093 *	still called in hard interrupt context and has to check
2094 *	whether the interrupt originates from the device. If yes it
2095 *	needs to disable the interrupt on the device and return
2096 *	IRQ_WAKE_THREAD which will wake up the handler thread and run
2097 *	@thread_fn. This split handler design is necessary to support
2098 *	shared interrupts.
2099 *
2100 *	Dev_id must be globally unique. Normally the address of the
2101 *	device data structure is used as the cookie. Since the handler
2102 *	receives this value it makes sense to use it.
2103 *
2104 *	If your interrupt is shared you must pass a non NULL dev_id
2105 *	as this is required when freeing the interrupt.
2106 *
2107 *	Flags:
2108 *
2109 *	IRQF_SHARED		Interrupt is shared
2110 *	IRQF_TRIGGER_*		Specify active edge(s) or level
2111 *
2112 */
2113int request_threaded_irq(unsigned int irq, irq_handler_t handler,
2114			 irq_handler_t thread_fn, unsigned long irqflags,
2115			 const char *devname, void *dev_id)
2116{
2117	struct irqaction *action;
2118	struct irq_desc *desc;
2119	int retval;
2120
2121	if (irq == IRQ_NOTCONNECTED)
2122		return -ENOTCONN;
2123
2124	/*
2125	 * Sanity-check: shared interrupts must pass in a real dev-ID,
2126	 * otherwise we'll have trouble later trying to figure out
2127	 * which interrupt is which (messes up the interrupt freeing
2128	 * logic etc).
2129	 *
2130	 * Also shared interrupts do not go well with disabling auto enable.
2131	 * The sharing interrupt might request it while it's still disabled
2132	 * and then wait for interrupts forever.
2133	 *
2134	 * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
2135	 * it cannot be set along with IRQF_NO_SUSPEND.
2136	 */
2137	if (((irqflags & IRQF_SHARED) && !dev_id) ||
2138	    ((irqflags & IRQF_SHARED) && (irqflags & IRQF_NO_AUTOEN)) ||
2139	    (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
2140	    ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
2141		return -EINVAL;
2142
2143	desc = irq_to_desc(irq);
2144	if (!desc)
2145		return -EINVAL;
2146
2147	if (!irq_settings_can_request(desc) ||
2148	    WARN_ON(irq_settings_is_per_cpu_devid(desc)))
2149		return -EINVAL;
2150
2151	if (!handler) {
2152		if (!thread_fn)
2153			return -EINVAL;
2154		handler = irq_default_primary_handler;
2155	}
2156
2157	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2158	if (!action)
2159		return -ENOMEM;
2160
2161	action->handler = handler;
2162	action->thread_fn = thread_fn;
2163	action->flags = irqflags;
2164	action->name = devname;
2165	action->dev_id = dev_id;
2166
2167	retval = irq_chip_pm_get(&desc->irq_data);
2168	if (retval < 0) {
2169		kfree(action);
2170		return retval;
2171	}
2172
2173	retval = __setup_irq(irq, desc, action);
2174
2175	if (retval) {
2176		irq_chip_pm_put(&desc->irq_data);
2177		kfree(action->secondary);
2178		kfree(action);
2179	}
2180
2181#ifdef CONFIG_DEBUG_SHIRQ_FIXME
2182	if (!retval && (irqflags & IRQF_SHARED)) {
2183		/*
2184		 * It's a shared IRQ -- the driver ought to be prepared for it
2185		 * to happen immediately, so let's make sure....
2186		 * We disable the irq to make sure that a 'real' IRQ doesn't
2187		 * run in parallel with our fake.
2188		 */
2189		unsigned long flags;
2190
2191		disable_irq(irq);
2192		local_irq_save(flags);
2193
2194		handler(irq, dev_id);
2195
2196		local_irq_restore(flags);
2197		enable_irq(irq);
2198	}
2199#endif
2200	return retval;
2201}
2202EXPORT_SYMBOL(request_threaded_irq);
2203
2204/**
2205 *	request_any_context_irq - allocate an interrupt line
2206 *	@irq: Interrupt line to allocate
2207 *	@handler: Function to be called when the IRQ occurs.
2208 *		  Threaded handler for threaded interrupts.
2209 *	@flags: Interrupt type flags
2210 *	@name: An ascii name for the claiming device
2211 *	@dev_id: A cookie passed back to the handler function
2212 *
2213 *	This call allocates interrupt resources and enables the
2214 *	interrupt line and IRQ handling. It selects either a
2215 *	hardirq or threaded handling method depending on the
2216 *	context.
2217 *
2218 *	On failure, it returns a negative value. On success,
2219 *	it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
2220 */
2221int request_any_context_irq(unsigned int irq, irq_handler_t handler,
2222			    unsigned long flags, const char *name, void *dev_id)
2223{
2224	struct irq_desc *desc;
2225	int ret;
2226
2227	if (irq == IRQ_NOTCONNECTED)
2228		return -ENOTCONN;
2229
2230	desc = irq_to_desc(irq);
2231	if (!desc)
2232		return -EINVAL;
2233
2234	if (irq_settings_is_nested_thread(desc)) {
2235		ret = request_threaded_irq(irq, NULL, handler,
2236					   flags, name, dev_id);
2237		return !ret ? IRQC_IS_NESTED : ret;
2238	}
2239
2240	ret = request_irq(irq, handler, flags, name, dev_id);
2241	return !ret ? IRQC_IS_HARDIRQ : ret;
2242}
2243EXPORT_SYMBOL_GPL(request_any_context_irq);
2244
2245/**
2246 *	request_nmi - allocate an interrupt line for NMI delivery
2247 *	@irq: Interrupt line to allocate
2248 *	@handler: Function to be called when the IRQ occurs.
2249 *		  Threaded handler for threaded interrupts.
2250 *	@irqflags: Interrupt type flags
2251 *	@name: An ascii name for the claiming device
2252 *	@dev_id: A cookie passed back to the handler function
2253 *
2254 *	This call allocates interrupt resources and enables the
2255 *	interrupt line and IRQ handling. It sets up the IRQ line
2256 *	to be handled as an NMI.
2257 *
2258 *	An interrupt line delivering NMIs cannot be shared and IRQ handling
2259 *	cannot be threaded.
2260 *
2261 *	Interrupt lines requested for NMI delivering must produce per cpu
2262 *	interrupts and have auto enabling setting disabled.
2263 *
2264 *	Dev_id must be globally unique. Normally the address of the
2265 *	device data structure is used as the cookie. Since the handler
2266 *	receives this value it makes sense to use it.
2267 *
2268 *	If the interrupt line cannot be used to deliver NMIs, function
2269 *	will fail and return a negative value.
2270 */
2271int request_nmi(unsigned int irq, irq_handler_t handler,
2272		unsigned long irqflags, const char *name, void *dev_id)
2273{
2274	struct irqaction *action;
2275	struct irq_desc *desc;
2276	unsigned long flags;
2277	int retval;
2278
2279	if (irq == IRQ_NOTCONNECTED)
2280		return -ENOTCONN;
2281
2282	/* NMI cannot be shared, used for Polling */
2283	if (irqflags & (IRQF_SHARED | IRQF_COND_SUSPEND | IRQF_IRQPOLL))
2284		return -EINVAL;
2285
2286	if (!(irqflags & IRQF_PERCPU))
2287		return -EINVAL;
2288
2289	if (!handler)
2290		return -EINVAL;
2291
2292	desc = irq_to_desc(irq);
2293
2294	if (!desc || (irq_settings_can_autoenable(desc) &&
2295	    !(irqflags & IRQF_NO_AUTOEN)) ||
2296	    !irq_settings_can_request(desc) ||
2297	    WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
2298	    !irq_supports_nmi(desc))
2299		return -EINVAL;
2300
2301	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2302	if (!action)
2303		return -ENOMEM;
2304
2305	action->handler = handler;
2306	action->flags = irqflags | IRQF_NO_THREAD | IRQF_NOBALANCING;
2307	action->name = name;
2308	action->dev_id = dev_id;
2309
2310	retval = irq_chip_pm_get(&desc->irq_data);
2311	if (retval < 0)
2312		goto err_out;
2313
2314	retval = __setup_irq(irq, desc, action);
2315	if (retval)
2316		goto err_irq_setup;
2317
2318	raw_spin_lock_irqsave(&desc->lock, flags);
2319
2320	/* Setup NMI state */
2321	desc->istate |= IRQS_NMI;
2322	retval = irq_nmi_setup(desc);
2323	if (retval) {
2324		__cleanup_nmi(irq, desc);
2325		raw_spin_unlock_irqrestore(&desc->lock, flags);
2326		return -EINVAL;
2327	}
2328
2329	raw_spin_unlock_irqrestore(&desc->lock, flags);
2330
2331	return 0;
2332
2333err_irq_setup:
2334	irq_chip_pm_put(&desc->irq_data);
2335err_out:
2336	kfree(action);
2337
2338	return retval;
2339}
2340
2341void enable_percpu_irq(unsigned int irq, unsigned int type)
2342{
2343	unsigned int cpu = smp_processor_id();
2344	unsigned long flags;
2345	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2346
2347	if (!desc)
2348		return;
2349
2350	/*
2351	 * If the trigger type is not specified by the caller, then
2352	 * use the default for this interrupt.
2353	 */
2354	type &= IRQ_TYPE_SENSE_MASK;
2355	if (type == IRQ_TYPE_NONE)
2356		type = irqd_get_trigger_type(&desc->irq_data);
2357
2358	if (type != IRQ_TYPE_NONE) {
2359		int ret;
2360
2361		ret = __irq_set_trigger(desc, type);
2362
2363		if (ret) {
2364			WARN(1, "failed to set type for IRQ%d\n", irq);
2365			goto out;
2366		}
2367	}
2368
2369	irq_percpu_enable(desc, cpu);
2370out:
2371	irq_put_desc_unlock(desc, flags);
2372}
2373EXPORT_SYMBOL_GPL(enable_percpu_irq);
2374
2375void enable_percpu_nmi(unsigned int irq, unsigned int type)
2376{
2377	enable_percpu_irq(irq, type);
2378}
2379
2380/**
2381 * irq_percpu_is_enabled - Check whether the per cpu irq is enabled
2382 * @irq:	Linux irq number to check for
2383 *
2384 * Must be called from a non migratable context. Returns the enable
2385 * state of a per cpu interrupt on the current cpu.
2386 */
2387bool irq_percpu_is_enabled(unsigned int irq)
2388{
2389	unsigned int cpu = smp_processor_id();
2390	struct irq_desc *desc;
2391	unsigned long flags;
2392	bool is_enabled;
2393
2394	desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2395	if (!desc)
2396		return false;
2397
2398	is_enabled = cpumask_test_cpu(cpu, desc->percpu_enabled);
2399	irq_put_desc_unlock(desc, flags);
2400
2401	return is_enabled;
2402}
2403EXPORT_SYMBOL_GPL(irq_percpu_is_enabled);
2404
2405void disable_percpu_irq(unsigned int irq)
2406{
2407	unsigned int cpu = smp_processor_id();
2408	unsigned long flags;
2409	struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
2410
2411	if (!desc)
2412		return;
2413
2414	irq_percpu_disable(desc, cpu);
2415	irq_put_desc_unlock(desc, flags);
2416}
2417EXPORT_SYMBOL_GPL(disable_percpu_irq);
2418
2419void disable_percpu_nmi(unsigned int irq)
2420{
2421	disable_percpu_irq(irq);
2422}
2423
2424/*
2425 * Internal function to unregister a percpu irqaction.
2426 */
2427static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2428{
2429	struct irq_desc *desc = irq_to_desc(irq);
2430	struct irqaction *action;
2431	unsigned long flags;
2432
2433	WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
2434
2435	if (!desc)
2436		return NULL;
2437
2438	raw_spin_lock_irqsave(&desc->lock, flags);
2439
2440	action = desc->action;
2441	if (!action || action->percpu_dev_id != dev_id) {
2442		WARN(1, "Trying to free already-free IRQ %d\n", irq);
2443		goto bad;
2444	}
2445
2446	if (!cpumask_empty(desc->percpu_enabled)) {
2447		WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
2448		     irq, cpumask_first(desc->percpu_enabled));
2449		goto bad;
2450	}
2451
2452	/* Found it - now remove it from the list of entries: */
2453	desc->action = NULL;
2454
2455	desc->istate &= ~IRQS_NMI;
2456
2457	raw_spin_unlock_irqrestore(&desc->lock, flags);
2458
2459	unregister_handler_proc(irq, action);
2460
2461	irq_chip_pm_put(&desc->irq_data);
2462	module_put(desc->owner);
2463	return action;
2464
2465bad:
2466	raw_spin_unlock_irqrestore(&desc->lock, flags);
2467	return NULL;
2468}
2469
2470/**
2471 *	remove_percpu_irq - free a per-cpu interrupt
2472 *	@irq: Interrupt line to free
2473 *	@act: irqaction for the interrupt
2474 *
2475 * Used to remove interrupts statically setup by the early boot process.
2476 */
2477void remove_percpu_irq(unsigned int irq, struct irqaction *act)
2478{
2479	struct irq_desc *desc = irq_to_desc(irq);
2480
2481	if (desc && irq_settings_is_per_cpu_devid(desc))
2482	    __free_percpu_irq(irq, act->percpu_dev_id);
2483}
2484
2485/**
2486 *	free_percpu_irq - free an interrupt allocated with request_percpu_irq
2487 *	@irq: Interrupt line to free
2488 *	@dev_id: Device identity to free
2489 *
2490 *	Remove a percpu interrupt handler. The handler is removed, but
2491 *	the interrupt line is not disabled. This must be done on each
2492 *	CPU before calling this function. The function does not return
2493 *	until any executing interrupts for this IRQ have completed.
2494 *
2495 *	This function must not be called from interrupt context.
2496 */
2497void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
2498{
2499	struct irq_desc *desc = irq_to_desc(irq);
2500
2501	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2502		return;
2503
2504	chip_bus_lock(desc);
2505	kfree(__free_percpu_irq(irq, dev_id));
2506	chip_bus_sync_unlock(desc);
2507}
2508EXPORT_SYMBOL_GPL(free_percpu_irq);
2509
2510void free_percpu_nmi(unsigned int irq, void __percpu *dev_id)
2511{
2512	struct irq_desc *desc = irq_to_desc(irq);
2513
2514	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2515		return;
2516
2517	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2518		return;
2519
2520	kfree(__free_percpu_irq(irq, dev_id));
2521}
2522
2523/**
2524 *	setup_percpu_irq - setup a per-cpu interrupt
2525 *	@irq: Interrupt line to setup
2526 *	@act: irqaction for the interrupt
2527 *
2528 * Used to statically setup per-cpu interrupts in the early boot process.
2529 */
2530int setup_percpu_irq(unsigned int irq, struct irqaction *act)
2531{
2532	struct irq_desc *desc = irq_to_desc(irq);
2533	int retval;
2534
2535	if (!desc || !irq_settings_is_per_cpu_devid(desc))
2536		return -EINVAL;
2537
2538	retval = irq_chip_pm_get(&desc->irq_data);
2539	if (retval < 0)
2540		return retval;
2541
2542	retval = __setup_irq(irq, desc, act);
2543
2544	if (retval)
2545		irq_chip_pm_put(&desc->irq_data);
2546
2547	return retval;
2548}
2549
2550/**
2551 *	__request_percpu_irq - allocate a percpu interrupt line
2552 *	@irq: Interrupt line to allocate
2553 *	@handler: Function to be called when the IRQ occurs.
2554 *	@flags: Interrupt type flags (IRQF_TIMER only)
2555 *	@devname: An ascii name for the claiming device
2556 *	@dev_id: A percpu cookie passed back to the handler function
2557 *
2558 *	This call allocates interrupt resources and enables the
2559 *	interrupt on the local CPU. If the interrupt is supposed to be
2560 *	enabled on other CPUs, it has to be done on each CPU using
2561 *	enable_percpu_irq().
2562 *
2563 *	Dev_id must be globally unique. It is a per-cpu variable, and
2564 *	the handler gets called with the interrupted CPU's instance of
2565 *	that variable.
2566 */
2567int __request_percpu_irq(unsigned int irq, irq_handler_t handler,
2568			 unsigned long flags, const char *devname,
2569			 void __percpu *dev_id)
2570{
2571	struct irqaction *action;
2572	struct irq_desc *desc;
2573	int retval;
2574
2575	if (!dev_id)
2576		return -EINVAL;
2577
2578	desc = irq_to_desc(irq);
2579	if (!desc || !irq_settings_can_request(desc) ||
2580	    !irq_settings_is_per_cpu_devid(desc))
2581		return -EINVAL;
2582
2583	if (flags && flags != IRQF_TIMER)
2584		return -EINVAL;
2585
2586	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2587	if (!action)
2588		return -ENOMEM;
2589
2590	action->handler = handler;
2591	action->flags = flags | IRQF_PERCPU | IRQF_NO_SUSPEND;
2592	action->name = devname;
2593	action->percpu_dev_id = dev_id;
2594
2595	retval = irq_chip_pm_get(&desc->irq_data);
2596	if (retval < 0) {
2597		kfree(action);
2598		return retval;
2599	}
2600
2601	retval = __setup_irq(irq, desc, action);
2602
2603	if (retval) {
2604		irq_chip_pm_put(&desc->irq_data);
2605		kfree(action);
2606	}
2607
2608	return retval;
2609}
2610EXPORT_SYMBOL_GPL(__request_percpu_irq);
2611
2612/**
2613 *	request_percpu_nmi - allocate a percpu interrupt line for NMI delivery
2614 *	@irq: Interrupt line to allocate
2615 *	@handler: Function to be called when the IRQ occurs.
2616 *	@name: An ascii name for the claiming device
2617 *	@dev_id: A percpu cookie passed back to the handler function
2618 *
2619 *	This call allocates interrupt resources for a per CPU NMI. Per CPU NMIs
2620 *	have to be setup on each CPU by calling prepare_percpu_nmi() before
2621 *	being enabled on the same CPU by using enable_percpu_nmi().
2622 *
2623 *	Dev_id must be globally unique. It is a per-cpu variable, and
2624 *	the handler gets called with the interrupted CPU's instance of
2625 *	that variable.
2626 *
2627 *	Interrupt lines requested for NMI delivering should have auto enabling
2628 *	setting disabled.
2629 *
2630 *	If the interrupt line cannot be used to deliver NMIs, function
2631 *	will fail returning a negative value.
2632 */
2633int request_percpu_nmi(unsigned int irq, irq_handler_t handler,
2634		       const char *name, void __percpu *dev_id)
2635{
2636	struct irqaction *action;
2637	struct irq_desc *desc;
2638	unsigned long flags;
2639	int retval;
2640
2641	if (!handler)
2642		return -EINVAL;
2643
2644	desc = irq_to_desc(irq);
2645
2646	if (!desc || !irq_settings_can_request(desc) ||
2647	    !irq_settings_is_per_cpu_devid(desc) ||
2648	    irq_settings_can_autoenable(desc) ||
2649	    !irq_supports_nmi(desc))
2650		return -EINVAL;
2651
2652	/* The line cannot already be NMI */
2653	if (desc->istate & IRQS_NMI)
2654		return -EINVAL;
2655
2656	action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
2657	if (!action)
2658		return -ENOMEM;
2659
2660	action->handler = handler;
2661	action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND | IRQF_NO_THREAD
2662		| IRQF_NOBALANCING;
2663	action->name = name;
2664	action->percpu_dev_id = dev_id;
2665
2666	retval = irq_chip_pm_get(&desc->irq_data);
2667	if (retval < 0)
2668		goto err_out;
2669
2670	retval = __setup_irq(irq, desc, action);
2671	if (retval)
2672		goto err_irq_setup;
2673
2674	raw_spin_lock_irqsave(&desc->lock, flags);
2675	desc->istate |= IRQS_NMI;
2676	raw_spin_unlock_irqrestore(&desc->lock, flags);
2677
2678	return 0;
2679
2680err_irq_setup:
2681	irq_chip_pm_put(&desc->irq_data);
2682err_out:
2683	kfree(action);
2684
2685	return retval;
2686}
2687
2688/**
2689 *	prepare_percpu_nmi - performs CPU local setup for NMI delivery
2690 *	@irq: Interrupt line to prepare for NMI delivery
2691 *
2692 *	This call prepares an interrupt line to deliver NMI on the current CPU,
2693 *	before that interrupt line gets enabled with enable_percpu_nmi().
2694 *
2695 *	As a CPU local operation, this should be called from non-preemptible
2696 *	context.
2697 *
2698 *	If the interrupt line cannot be used to deliver NMIs, function
2699 *	will fail returning a negative value.
2700 */
2701int prepare_percpu_nmi(unsigned int irq)
2702{
2703	unsigned long flags;
2704	struct irq_desc *desc;
2705	int ret = 0;
2706
2707	WARN_ON(preemptible());
2708
2709	desc = irq_get_desc_lock(irq, &flags,
2710				 IRQ_GET_DESC_CHECK_PERCPU);
2711	if (!desc)
2712		return -EINVAL;
2713
2714	if (WARN(!(desc->istate & IRQS_NMI),
2715		 KERN_ERR "prepare_percpu_nmi called for a non-NMI interrupt: irq %u\n",
2716		 irq)) {
2717		ret = -EINVAL;
2718		goto out;
2719	}
2720
2721	ret = irq_nmi_setup(desc);
2722	if (ret) {
2723		pr_err("Failed to setup NMI delivery: irq %u\n", irq);
2724		goto out;
2725	}
2726
2727out:
2728	irq_put_desc_unlock(desc, flags);
2729	return ret;
2730}
2731
2732/**
2733 *	teardown_percpu_nmi - undoes NMI setup of IRQ line
2734 *	@irq: Interrupt line from which CPU local NMI configuration should be
2735 *	      removed
2736 *
2737 *	This call undoes the setup done by prepare_percpu_nmi().
2738 *
2739 *	IRQ line should not be enabled for the current CPU.
2740 *
2741 *	As a CPU local operation, this should be called from non-preemptible
2742 *	context.
2743 */
2744void teardown_percpu_nmi(unsigned int irq)
2745{
2746	unsigned long flags;
2747	struct irq_desc *desc;
2748
2749	WARN_ON(preemptible());
2750
2751	desc = irq_get_desc_lock(irq, &flags,
2752				 IRQ_GET_DESC_CHECK_PERCPU);
2753	if (!desc)
2754		return;
2755
2756	if (WARN_ON(!(desc->istate & IRQS_NMI)))
2757		goto out;
2758
2759	irq_nmi_teardown(desc);
2760out:
2761	irq_put_desc_unlock(desc, flags);
2762}
2763
2764int __irq_get_irqchip_state(struct irq_data *data, enum irqchip_irq_state which,
2765			    bool *state)
2766{
2767	struct irq_chip *chip;
2768	int err = -EINVAL;
2769
2770	do {
2771		chip = irq_data_get_irq_chip(data);
2772		if (WARN_ON_ONCE(!chip))
2773			return -ENODEV;
2774		if (chip->irq_get_irqchip_state)
2775			break;
2776#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2777		data = data->parent_data;
2778#else
2779		data = NULL;
2780#endif
2781	} while (data);
2782
2783	if (data)
2784		err = chip->irq_get_irqchip_state(data, which, state);
2785	return err;
2786}
2787
2788/**
2789 *	irq_get_irqchip_state - returns the irqchip state of a interrupt.
2790 *	@irq: Interrupt line that is forwarded to a VM
2791 *	@which: One of IRQCHIP_STATE_* the caller wants to know about
2792 *	@state: a pointer to a boolean where the state is to be stored
2793 *
2794 *	This call snapshots the internal irqchip state of an
2795 *	interrupt, returning into @state the bit corresponding to
2796 *	stage @which
2797 *
2798 *	This function should be called with preemption disabled if the
2799 *	interrupt controller has per-cpu registers.
2800 */
2801int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2802			  bool *state)
2803{
2804	struct irq_desc *desc;
2805	struct irq_data *data;
2806	unsigned long flags;
2807	int err = -EINVAL;
2808
2809	desc = irq_get_desc_buslock(irq, &flags, 0);
2810	if (!desc)
2811		return err;
2812
2813	data = irq_desc_get_irq_data(desc);
2814
2815	err = __irq_get_irqchip_state(data, which, state);
2816
2817	irq_put_desc_busunlock(desc, flags);
2818	return err;
2819}
2820EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
2821
2822/**
2823 *	irq_set_irqchip_state - set the state of a forwarded interrupt.
2824 *	@irq: Interrupt line that is forwarded to a VM
2825 *	@which: State to be restored (one of IRQCHIP_STATE_*)
2826 *	@val: Value corresponding to @which
2827 *
2828 *	This call sets the internal irqchip state of an interrupt,
2829 *	depending on the value of @which.
2830 *
2831 *	This function should be called with preemption disabled if the
2832 *	interrupt controller has per-cpu registers.
2833 */
2834int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
2835			  bool val)
2836{
2837	struct irq_desc *desc;
2838	struct irq_data *data;
2839	struct irq_chip *chip;
2840	unsigned long flags;
2841	int err = -EINVAL;
2842
2843	desc = irq_get_desc_buslock(irq, &flags, 0);
2844	if (!desc)
2845		return err;
2846
2847	data = irq_desc_get_irq_data(desc);
2848
2849	do {
2850		chip = irq_data_get_irq_chip(data);
2851		if (WARN_ON_ONCE(!chip)) {
2852			err = -ENODEV;
2853			goto out_unlock;
2854		}
2855		if (chip->irq_set_irqchip_state)
2856			break;
2857#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
2858		data = data->parent_data;
2859#else
2860		data = NULL;
2861#endif
2862	} while (data);
2863
2864	if (data)
2865		err = chip->irq_set_irqchip_state(data, which, val);
2866
2867out_unlock:
2868	irq_put_desc_busunlock(desc, flags);
2869	return err;
2870}
2871EXPORT_SYMBOL_GPL(irq_set_irqchip_state);
2872
2873/**
2874 * irq_has_action - Check whether an interrupt is requested
2875 * @irq:	The linux irq number
2876 *
2877 * Returns: A snapshot of the current state
2878 */
2879bool irq_has_action(unsigned int irq)
2880{
2881	bool res;
2882
2883	rcu_read_lock();
2884	res = irq_desc_has_action(irq_to_desc(irq));
2885	rcu_read_unlock();
2886	return res;
2887}
2888EXPORT_SYMBOL_GPL(irq_has_action);
2889
2890/**
2891 * irq_check_status_bit - Check whether bits in the irq descriptor status are set
2892 * @irq:	The linux irq number
2893 * @bitmask:	The bitmask to evaluate
2894 *
2895 * Returns: True if one of the bits in @bitmask is set
2896 */
2897bool irq_check_status_bit(unsigned int irq, unsigned int bitmask)
2898{
2899	struct irq_desc *desc;
2900	bool res = false;
2901
2902	rcu_read_lock();
2903	desc = irq_to_desc(irq);
2904	if (desc)
2905		res = !!(desc->status_use_accessors & bitmask);
2906	rcu_read_unlock();
2907	return res;
2908}
2909EXPORT_SYMBOL_GPL(irq_check_status_bit);