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