linux/kernel/irq/manage.c
<<
>>
Prefs
   1/*
   2 * linux/kernel/irq/manage.c
   3 *
   4 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
   5 * Copyright (C) 2005-2006 Thomas Gleixner
   6 *
   7 * This file contains driver APIs to the irq subsystem.
   8 */
   9
  10#define pr_fmt(fmt) "genirq: " fmt
  11
  12#include <linux/irq.h>
  13#include <linux/kthread.h>
  14#include <linux/module.h>
  15#include <linux/random.h>
  16#include <linux/interrupt.h>
  17#include <linux/slab.h>
  18#include <linux/sched.h>
  19#include <linux/sched/rt.h>
  20#include <linux/task_work.h>
  21
  22#include "internals.h"
  23
  24#ifdef CONFIG_IRQ_FORCED_THREADING
  25__read_mostly bool force_irqthreads;
  26
  27static int __init setup_forced_irqthreads(char *arg)
  28{
  29        force_irqthreads = true;
  30        return 0;
  31}
  32early_param("threadirqs", setup_forced_irqthreads);
  33#endif
  34
  35static void __synchronize_hardirq(struct irq_desc *desc)
  36{
  37        bool inprogress;
  38
  39        do {
  40                unsigned long flags;
  41
  42                /*
  43                 * Wait until we're out of the critical section.  This might
  44                 * give the wrong answer due to the lack of memory barriers.
  45                 */
  46                while (irqd_irq_inprogress(&desc->irq_data))
  47                        cpu_relax();
  48
  49                /* Ok, that indicated we're done: double-check carefully. */
  50                raw_spin_lock_irqsave(&desc->lock, flags);
  51                inprogress = irqd_irq_inprogress(&desc->irq_data);
  52                raw_spin_unlock_irqrestore(&desc->lock, flags);
  53
  54                /* Oops, that failed? */
  55        } while (inprogress);
  56}
  57
  58/**
  59 *      synchronize_hardirq - wait for pending hard IRQ handlers (on other CPUs)
  60 *      @irq: interrupt number to wait for
  61 *
  62 *      This function waits for any pending hard IRQ handlers for this
  63 *      interrupt to complete before returning. If you use this
  64 *      function while holding a resource the IRQ handler may need you
  65 *      will deadlock. It does not take associated threaded handlers
  66 *      into account.
  67 *
  68 *      Do not use this for shutdown scenarios where you must be sure
  69 *      that all parts (hardirq and threaded handler) have completed.
  70 *
  71 *      Returns: false if a threaded handler is active.
  72 *
  73 *      This function may be called - with care - from IRQ context.
  74 */
  75bool synchronize_hardirq(unsigned int irq)
  76{
  77        struct irq_desc *desc = irq_to_desc(irq);
  78
  79        if (desc) {
  80                __synchronize_hardirq(desc);
  81                return !atomic_read(&desc->threads_active);
  82        }
  83
  84        return true;
  85}
  86EXPORT_SYMBOL(synchronize_hardirq);
  87
  88/**
  89 *      synchronize_irq - wait for pending IRQ handlers (on other CPUs)
  90 *      @irq: interrupt number to wait for
  91 *
  92 *      This function waits for any pending IRQ handlers for this interrupt
  93 *      to complete before returning. If you use this function while
  94 *      holding a resource the IRQ handler may need you will deadlock.
  95 *
  96 *      This function may be called - with care - from IRQ context.
  97 */
  98void synchronize_irq(unsigned int irq)
  99{
 100        struct irq_desc *desc = irq_to_desc(irq);
 101
 102        if (desc) {
 103                __synchronize_hardirq(desc);
 104                /*
 105                 * We made sure that no hardirq handler is
 106                 * running. Now verify that no threaded handlers are
 107                 * active.
 108                 */
 109                wait_event(desc->wait_for_threads,
 110                           !atomic_read(&desc->threads_active));
 111        }
 112}
 113EXPORT_SYMBOL(synchronize_irq);
 114
 115#ifdef CONFIG_SMP
 116cpumask_var_t irq_default_affinity;
 117
 118static int __irq_can_set_affinity(struct irq_desc *desc)
 119{
 120        if (!desc || !irqd_can_balance(&desc->irq_data) ||
 121            !desc->irq_data.chip || !desc->irq_data.chip->irq_set_affinity)
 122                return 0;
 123        return 1;
 124}
 125
 126/**
 127 *      irq_can_set_affinity - Check if the affinity of a given irq can be set
 128 *      @irq:           Interrupt to check
 129 *
 130 */
 131int irq_can_set_affinity(unsigned int irq)
 132{
 133        return __irq_can_set_affinity(irq_to_desc(irq));
 134}
 135
 136/**
 137 *      irq_set_thread_affinity - Notify irq threads to adjust affinity
 138 *      @desc:          irq descriptor which has affitnity changed
 139 *
 140 *      We just set IRQTF_AFFINITY and delegate the affinity setting
 141 *      to the interrupt thread itself. We can not call
 142 *      set_cpus_allowed_ptr() here as we hold desc->lock and this
 143 *      code can be called from hard interrupt context.
 144 */
 145void irq_set_thread_affinity(struct irq_desc *desc)
 146{
 147        struct irqaction *action = desc->action;
 148
 149        while (action) {
 150                if (action->thread)
 151                        set_bit(IRQTF_AFFINITY, &action->thread_flags);
 152                action = action->next;
 153        }
 154}
 155
 156#ifdef CONFIG_GENERIC_PENDING_IRQ
 157static inline bool irq_can_move_pcntxt(struct irq_data *data)
 158{
 159        return irqd_can_move_in_process_context(data);
 160}
 161static inline bool irq_move_pending(struct irq_data *data)
 162{
 163        return irqd_is_setaffinity_pending(data);
 164}
 165static inline void
 166irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask)
 167{
 168        cpumask_copy(desc->pending_mask, mask);
 169}
 170static inline void
 171irq_get_pending(struct cpumask *mask, struct irq_desc *desc)
 172{
 173        cpumask_copy(mask, desc->pending_mask);
 174}
 175#else
 176static inline bool irq_can_move_pcntxt(struct irq_data *data) { return true; }
 177static inline bool irq_move_pending(struct irq_data *data) { return false; }
 178static inline void
 179irq_copy_pending(struct irq_desc *desc, const struct cpumask *mask) { }
 180static inline void
 181irq_get_pending(struct cpumask *mask, struct irq_desc *desc) { }
 182#endif
 183
 184int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
 185                        bool force)
 186{
 187        struct irq_desc *desc = irq_data_to_desc(data);
 188        struct irq_chip *chip = irq_data_get_irq_chip(data);
 189        int ret;
 190
 191        ret = chip->irq_set_affinity(data, mask, force);
 192        switch (ret) {
 193        case IRQ_SET_MASK_OK:
 194        case IRQ_SET_MASK_OK_DONE:
 195                cpumask_copy(desc->irq_common_data.affinity, mask);
 196        case IRQ_SET_MASK_OK_NOCOPY:
 197                irq_set_thread_affinity(desc);
 198                ret = 0;
 199        }
 200
 201        return ret;
 202}
 203
 204int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 205                            bool force)
 206{
 207        struct irq_chip *chip = irq_data_get_irq_chip(data);
 208        struct irq_desc *desc = irq_data_to_desc(data);
 209        int ret = 0;
 210
 211        if (!chip || !chip->irq_set_affinity)
 212                return -EINVAL;
 213
 214        if (irq_can_move_pcntxt(data)) {
 215                ret = irq_do_set_affinity(data, mask, force);
 216        } else {
 217                irqd_set_move_pending(data);
 218                irq_copy_pending(desc, mask);
 219        }
 220
 221        if (desc->affinity_notify) {
 222                kref_get(&desc->affinity_notify->kref);
 223                schedule_work(&desc->affinity_notify->work);
 224        }
 225        irqd_set(data, IRQD_AFFINITY_SET);
 226
 227        return ret;
 228}
 229
 230int __irq_set_affinity(unsigned int irq, const struct cpumask *mask, bool force)
 231{
 232        struct irq_desc *desc = irq_to_desc(irq);
 233        unsigned long flags;
 234        int ret;
 235
 236        if (!desc)
 237                return -EINVAL;
 238
 239        raw_spin_lock_irqsave(&desc->lock, flags);
 240        ret = irq_set_affinity_locked(irq_desc_get_irq_data(desc), mask, force);
 241        raw_spin_unlock_irqrestore(&desc->lock, flags);
 242        return ret;
 243}
 244
 245int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
 246{
 247        unsigned long flags;
 248        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 249
 250        if (!desc)
 251                return -EINVAL;
 252        desc->affinity_hint = m;
 253        irq_put_desc_unlock(desc, flags);
 254        /* set the initial affinity to prevent every interrupt being on CPU0 */
 255        if (m)
 256                __irq_set_affinity(irq, m, false);
 257        return 0;
 258}
 259EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
 260
 261/**
 262 *      irq_set_vcpu_affinity - Set vcpu affinity for the interrupt
 263 *      @irq: interrupt number to set affinity
 264 *      @vcpu_info: vCPU specific data
 265 *
 266 *      This function uses the vCPU specific data to set the vCPU
 267 *      affinity for an irq. The vCPU specific data is passed from
 268 *      outside, such as KVM. One example code path is as below:
 269 *      KVM -> IOMMU -> irq_set_vcpu_affinity().
 270 */
 271int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info)
 272{
 273        unsigned long flags;
 274        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 275        struct irq_data *data;
 276        struct irq_chip *chip;
 277        int ret = -ENOSYS;
 278
 279        if (!desc)
 280                return -EINVAL;
 281
 282        data = irq_desc_get_irq_data(desc);
 283        chip = irq_data_get_irq_chip(data);
 284        if (chip && chip->irq_set_vcpu_affinity)
 285                ret = chip->irq_set_vcpu_affinity(data, vcpu_info);
 286        irq_put_desc_unlock(desc, flags);
 287
 288        return ret;
 289}
 290EXPORT_SYMBOL_GPL(irq_set_vcpu_affinity);
 291
 292static void irq_affinity_notify(struct work_struct *work)
 293{
 294        struct irq_affinity_notify *notify =
 295                container_of(work, struct irq_affinity_notify, work);
 296        struct irq_desc *desc = irq_to_desc(notify->irq);
 297        cpumask_var_t cpumask;
 298        unsigned long flags;
 299
 300        if (!desc || !alloc_cpumask_var(&cpumask, GFP_KERNEL))
 301                goto out;
 302
 303        raw_spin_lock_irqsave(&desc->lock, flags);
 304        if (irq_move_pending(&desc->irq_data))
 305                irq_get_pending(cpumask, desc);
 306        else
 307                cpumask_copy(cpumask, desc->irq_common_data.affinity);
 308        raw_spin_unlock_irqrestore(&desc->lock, flags);
 309
 310        notify->notify(notify, cpumask);
 311
 312        free_cpumask_var(cpumask);
 313out:
 314        kref_put(&notify->kref, notify->release);
 315}
 316
 317/**
 318 *      irq_set_affinity_notifier - control notification of IRQ affinity changes
 319 *      @irq:           Interrupt for which to enable/disable notification
 320 *      @notify:        Context for notification, or %NULL to disable
 321 *                      notification.  Function pointers must be initialised;
 322 *                      the other fields will be initialised by this function.
 323 *
 324 *      Must be called in process context.  Notification may only be enabled
 325 *      after the IRQ is allocated and must be disabled before the IRQ is
 326 *      freed using free_irq().
 327 */
 328int
 329irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 330{
 331        struct irq_desc *desc = irq_to_desc(irq);
 332        struct irq_affinity_notify *old_notify;
 333        unsigned long flags;
 334
 335        /* The release function is promised process context */
 336        might_sleep();
 337
 338        if (!desc)
 339                return -EINVAL;
 340
 341        /* Complete initialisation of *notify */
 342        if (notify) {
 343                notify->irq = irq;
 344                kref_init(&notify->kref);
 345                INIT_WORK(&notify->work, irq_affinity_notify);
 346        }
 347
 348        raw_spin_lock_irqsave(&desc->lock, flags);
 349        old_notify = desc->affinity_notify;
 350        desc->affinity_notify = notify;
 351        raw_spin_unlock_irqrestore(&desc->lock, flags);
 352
 353        if (old_notify)
 354                kref_put(&old_notify->kref, old_notify->release);
 355
 356        return 0;
 357}
 358EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
 359
 360#ifndef CONFIG_AUTO_IRQ_AFFINITY
 361/*
 362 * Generic version of the affinity autoselector.
 363 */
 364static int setup_affinity(struct irq_desc *desc, struct cpumask *mask)
 365{
 366        struct cpumask *set = irq_default_affinity;
 367        int node = irq_desc_get_node(desc);
 368
 369        /* Excludes PER_CPU and NO_BALANCE interrupts */
 370        if (!__irq_can_set_affinity(desc))
 371                return 0;
 372
 373        /*
 374         * Preserve an userspace affinity setup, but make sure that
 375         * one of the targets is online.
 376         */
 377        if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 378                if (cpumask_intersects(desc->irq_common_data.affinity,
 379                                       cpu_online_mask))
 380                        set = desc->irq_common_data.affinity;
 381                else
 382                        irqd_clear(&desc->irq_data, IRQD_AFFINITY_SET);
 383        }
 384
 385        cpumask_and(mask, cpu_online_mask, set);
 386        if (node != NUMA_NO_NODE) {
 387                const struct cpumask *nodemask = cpumask_of_node(node);
 388
 389                /* make sure at least one of the cpus in nodemask is online */
 390                if (cpumask_intersects(mask, nodemask))
 391                        cpumask_and(mask, mask, nodemask);
 392        }
 393        irq_do_set_affinity(&desc->irq_data, mask, false);
 394        return 0;
 395}
 396#else
 397/* Wrapper for ALPHA specific affinity selector magic */
 398static inline int setup_affinity(struct irq_desc *d, struct cpumask *mask)
 399{
 400        return irq_select_affinity(irq_desc_get_irq(d));
 401}
 402#endif
 403
 404/*
 405 * Called when affinity is set via /proc/irq
 406 */
 407int irq_select_affinity_usr(unsigned int irq, struct cpumask *mask)
 408{
 409        struct irq_desc *desc = irq_to_desc(irq);
 410        unsigned long flags;
 411        int ret;
 412
 413        raw_spin_lock_irqsave(&desc->lock, flags);
 414        ret = setup_affinity(desc, mask);
 415        raw_spin_unlock_irqrestore(&desc->lock, flags);
 416        return ret;
 417}
 418
 419#else
 420static inline int
 421setup_affinity(struct irq_desc *desc, struct cpumask *mask)
 422{
 423        return 0;
 424}
 425#endif
 426
 427void __disable_irq(struct irq_desc *desc)
 428{
 429        if (!desc->depth++)
 430                irq_disable(desc);
 431}
 432
 433static int __disable_irq_nosync(unsigned int irq)
 434{
 435        unsigned long flags;
 436        struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 437
 438        if (!desc)
 439                return -EINVAL;
 440        __disable_irq(desc);
 441        irq_put_desc_busunlock(desc, flags);
 442        return 0;
 443}
 444
 445/**
 446 *      disable_irq_nosync - disable an irq without waiting
 447 *      @irq: Interrupt to disable
 448 *
 449 *      Disable the selected interrupt line.  Disables and Enables are
 450 *      nested.
 451 *      Unlike disable_irq(), this function does not ensure existing
 452 *      instances of the IRQ handler have completed before returning.
 453 *
 454 *      This function may be called from IRQ context.
 455 */
 456void disable_irq_nosync(unsigned int irq)
 457{
 458        __disable_irq_nosync(irq);
 459}
 460EXPORT_SYMBOL(disable_irq_nosync);
 461
 462/**
 463 *      disable_irq - disable an irq and wait for completion
 464 *      @irq: Interrupt to disable
 465 *
 466 *      Disable the selected interrupt line.  Enables and Disables are
 467 *      nested.
 468 *      This function waits for any pending IRQ handlers for this interrupt
 469 *      to complete before returning. If you use this function while
 470 *      holding a resource the IRQ handler may need you will deadlock.
 471 *
 472 *      This function may be called - with care - from IRQ context.
 473 */
 474void disable_irq(unsigned int irq)
 475{
 476        if (!__disable_irq_nosync(irq))
 477                synchronize_irq(irq);
 478}
 479EXPORT_SYMBOL(disable_irq);
 480
 481/**
 482 *      disable_hardirq - disables an irq and waits for hardirq completion
 483 *      @irq: Interrupt to disable
 484 *
 485 *      Disable the selected interrupt line.  Enables and Disables are
 486 *      nested.
 487 *      This function waits for any pending hard IRQ handlers for this
 488 *      interrupt to complete before returning. If you use this function while
 489 *      holding a resource the hard IRQ handler may need you will deadlock.
 490 *
 491 *      When used to optimistically disable an interrupt from atomic context
 492 *      the return value must be checked.
 493 *
 494 *      Returns: false if a threaded handler is active.
 495 *
 496 *      This function may be called - with care - from IRQ context.
 497 */
 498bool disable_hardirq(unsigned int irq)
 499{
 500        if (!__disable_irq_nosync(irq))
 501                return synchronize_hardirq(irq);
 502
 503        return false;
 504}
 505EXPORT_SYMBOL_GPL(disable_hardirq);
 506
 507void __enable_irq(struct irq_desc *desc)
 508{
 509        switch (desc->depth) {
 510        case 0:
 511 err_out:
 512                WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n",
 513                     irq_desc_get_irq(desc));
 514                break;
 515        case 1: {
 516                if (desc->istate & IRQS_SUSPENDED)
 517                        goto err_out;
 518                /* Prevent probing on this irq: */
 519                irq_settings_set_noprobe(desc);
 520                irq_enable(desc);
 521                check_irq_resend(desc);
 522                /* fall-through */
 523        }
 524        default:
 525                desc->depth--;
 526        }
 527}
 528
 529/**
 530 *      enable_irq - enable handling of an irq
 531 *      @irq: Interrupt to enable
 532 *
 533 *      Undoes the effect of one call to disable_irq().  If this
 534 *      matches the last disable, processing of interrupts on this
 535 *      IRQ line is re-enabled.
 536 *
 537 *      This function may be called from IRQ context only when
 538 *      desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
 539 */
 540void enable_irq(unsigned int irq)
 541{
 542        unsigned long flags;
 543        struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 544
 545        if (!desc)
 546                return;
 547        if (WARN(!desc->irq_data.chip,
 548                 KERN_ERR "enable_irq before setup/request_irq: irq %u\n", irq))
 549                goto out;
 550
 551        __enable_irq(desc);
 552out:
 553        irq_put_desc_busunlock(desc, flags);
 554}
 555EXPORT_SYMBOL(enable_irq);
 556
 557static int set_irq_wake_real(unsigned int irq, unsigned int on)
 558{
 559        struct irq_desc *desc = irq_to_desc(irq);
 560        int ret = -ENXIO;
 561
 562        if (irq_desc_get_chip(desc)->flags &  IRQCHIP_SKIP_SET_WAKE)
 563                return 0;
 564
 565        if (desc->irq_data.chip->irq_set_wake)
 566                ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
 567
 568        return ret;
 569}
 570
 571/**
 572 *      irq_set_irq_wake - control irq power management wakeup
 573 *      @irq:   interrupt to control
 574 *      @on:    enable/disable power management wakeup
 575 *
 576 *      Enable/disable power management wakeup mode, which is
 577 *      disabled by default.  Enables and disables must match,
 578 *      just as they match for non-wakeup mode support.
 579 *
 580 *      Wakeup mode lets this IRQ wake the system from sleep
 581 *      states like "suspend to RAM".
 582 */
 583int irq_set_irq_wake(unsigned int irq, unsigned int on)
 584{
 585        unsigned long flags;
 586        struct irq_desc *desc = irq_get_desc_buslock(irq, &flags, IRQ_GET_DESC_CHECK_GLOBAL);
 587        int ret = 0;
 588
 589        if (!desc)
 590                return -EINVAL;
 591
 592        /* wakeup-capable irqs can be shared between drivers that
 593         * don't need to have the same sleep mode behaviors.
 594         */
 595        if (on) {
 596                if (desc->wake_depth++ == 0) {
 597                        ret = set_irq_wake_real(irq, on);
 598                        if (ret)
 599                                desc->wake_depth = 0;
 600                        else
 601                                irqd_set(&desc->irq_data, IRQD_WAKEUP_STATE);
 602                }
 603        } else {
 604                if (desc->wake_depth == 0) {
 605                        WARN(1, "Unbalanced IRQ %d wake disable\n", irq);
 606                } else if (--desc->wake_depth == 0) {
 607                        ret = set_irq_wake_real(irq, on);
 608                        if (ret)
 609                                desc->wake_depth = 1;
 610                        else
 611                                irqd_clear(&desc->irq_data, IRQD_WAKEUP_STATE);
 612                }
 613        }
 614        irq_put_desc_busunlock(desc, flags);
 615        return ret;
 616}
 617EXPORT_SYMBOL(irq_set_irq_wake);
 618
 619/*
 620 * Internal function that tells the architecture code whether a
 621 * particular irq has been exclusively allocated or is available
 622 * for driver use.
 623 */
 624int can_request_irq(unsigned int irq, unsigned long irqflags)
 625{
 626        unsigned long flags;
 627        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 628        int canrequest = 0;
 629
 630        if (!desc)
 631                return 0;
 632
 633        if (irq_settings_can_request(desc)) {
 634                if (!desc->action ||
 635                    irqflags & desc->action->flags & IRQF_SHARED)
 636                        canrequest = 1;
 637        }
 638        irq_put_desc_unlock(desc, flags);
 639        return canrequest;
 640}
 641
 642int __irq_set_trigger(struct irq_desc *desc, unsigned long flags)
 643{
 644        struct irq_chip *chip = desc->irq_data.chip;
 645        int ret, unmask = 0;
 646
 647        if (!chip || !chip->irq_set_type) {
 648                /*
 649                 * IRQF_TRIGGER_* but the PIC does not support multiple
 650                 * flow-types?
 651                 */
 652                pr_debug("No set_type function for IRQ %d (%s)\n",
 653                         irq_desc_get_irq(desc),
 654                         chip ? (chip->name ? : "unknown") : "unknown");
 655                return 0;
 656        }
 657
 658        flags &= IRQ_TYPE_SENSE_MASK;
 659
 660        if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
 661                if (!irqd_irq_masked(&desc->irq_data))
 662                        mask_irq(desc);
 663                if (!irqd_irq_disabled(&desc->irq_data))
 664                        unmask = 1;
 665        }
 666
 667        /* caller masked out all except trigger mode flags */
 668        ret = chip->irq_set_type(&desc->irq_data, flags);
 669
 670        switch (ret) {
 671        case IRQ_SET_MASK_OK:
 672        case IRQ_SET_MASK_OK_DONE:
 673                irqd_clear(&desc->irq_data, IRQD_TRIGGER_MASK);
 674                irqd_set(&desc->irq_data, flags);
 675
 676        case IRQ_SET_MASK_OK_NOCOPY:
 677                flags = irqd_get_trigger_type(&desc->irq_data);
 678                irq_settings_set_trigger_mask(desc, flags);
 679                irqd_clear(&desc->irq_data, IRQD_LEVEL);
 680                irq_settings_clr_level(desc);
 681                if (flags & IRQ_TYPE_LEVEL_MASK) {
 682                        irq_settings_set_level(desc);
 683                        irqd_set(&desc->irq_data, IRQD_LEVEL);
 684                }
 685
 686                ret = 0;
 687                break;
 688        default:
 689                pr_err("Setting trigger mode %lu for irq %u failed (%pF)\n",
 690                       flags, irq_desc_get_irq(desc), chip->irq_set_type);
 691        }
 692        if (unmask)
 693                unmask_irq(desc);
 694        return ret;
 695}
 696
 697#ifdef CONFIG_HARDIRQS_SW_RESEND
 698int irq_set_parent(int irq, int parent_irq)
 699{
 700        unsigned long flags;
 701        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, 0);
 702
 703        if (!desc)
 704                return -EINVAL;
 705
 706        desc->parent_irq = parent_irq;
 707
 708        irq_put_desc_unlock(desc, flags);
 709        return 0;
 710}
 711#endif
 712
 713/*
 714 * Default primary interrupt handler for threaded interrupts. Is
 715 * assigned as primary handler when request_threaded_irq is called
 716 * with handler == NULL. Useful for oneshot interrupts.
 717 */
 718static irqreturn_t irq_default_primary_handler(int irq, void *dev_id)
 719{
 720        return IRQ_WAKE_THREAD;
 721}
 722
 723/*
 724 * Primary handler for nested threaded interrupts. Should never be
 725 * called.
 726 */
 727static irqreturn_t irq_nested_primary_handler(int irq, void *dev_id)
 728{
 729        WARN(1, "Primary handler called for nested irq %d\n", irq);
 730        return IRQ_NONE;
 731}
 732
 733static int irq_wait_for_interrupt(struct irqaction *action)
 734{
 735        set_current_state(TASK_INTERRUPTIBLE);
 736
 737        while (!kthread_should_stop()) {
 738
 739                if (test_and_clear_bit(IRQTF_RUNTHREAD,
 740                                       &action->thread_flags)) {
 741                        __set_current_state(TASK_RUNNING);
 742                        return 0;
 743                }
 744                schedule();
 745                set_current_state(TASK_INTERRUPTIBLE);
 746        }
 747        __set_current_state(TASK_RUNNING);
 748        return -1;
 749}
 750
 751/*
 752 * Oneshot interrupts keep the irq line masked until the threaded
 753 * handler finished. unmask if the interrupt has not been disabled and
 754 * is marked MASKED.
 755 */
 756static void irq_finalize_oneshot(struct irq_desc *desc,
 757                                 struct irqaction *action)
 758{
 759        if (!(desc->istate & IRQS_ONESHOT))
 760                return;
 761again:
 762        chip_bus_lock(desc);
 763        raw_spin_lock_irq(&desc->lock);
 764
 765        /*
 766         * Implausible though it may be we need to protect us against
 767         * the following scenario:
 768         *
 769         * The thread is faster done than the hard interrupt handler
 770         * on the other CPU. If we unmask the irq line then the
 771         * interrupt can come in again and masks the line, leaves due
 772         * to IRQS_INPROGRESS and the irq line is masked forever.
 773         *
 774         * This also serializes the state of shared oneshot handlers
 775         * versus "desc->threads_onehsot |= action->thread_mask;" in
 776         * irq_wake_thread(). See the comment there which explains the
 777         * serialization.
 778         */
 779        if (unlikely(irqd_irq_inprogress(&desc->irq_data))) {
 780                raw_spin_unlock_irq(&desc->lock);
 781                chip_bus_sync_unlock(desc);
 782                cpu_relax();
 783                goto again;
 784        }
 785
 786        /*
 787         * Now check again, whether the thread should run. Otherwise
 788         * we would clear the threads_oneshot bit of this thread which
 789         * was just set.
 790         */
 791        if (test_bit(IRQTF_RUNTHREAD, &action->thread_flags))
 792                goto out_unlock;
 793
 794        desc->threads_oneshot &= ~action->thread_mask;
 795
 796        if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
 797            irqd_irq_masked(&desc->irq_data))
 798                unmask_threaded_irq(desc);
 799
 800out_unlock:
 801        raw_spin_unlock_irq(&desc->lock);
 802        chip_bus_sync_unlock(desc);
 803}
 804
 805#ifdef CONFIG_SMP
 806/*
 807 * Check whether we need to change the affinity of the interrupt thread.
 808 */
 809static void
 810irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
 811{
 812        cpumask_var_t mask;
 813        bool valid = true;
 814
 815        if (!test_and_clear_bit(IRQTF_AFFINITY, &action->thread_flags))
 816                return;
 817
 818        /*
 819         * In case we are out of memory we set IRQTF_AFFINITY again and
 820         * try again next time
 821         */
 822        if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
 823                set_bit(IRQTF_AFFINITY, &action->thread_flags);
 824                return;
 825        }
 826
 827        raw_spin_lock_irq(&desc->lock);
 828        /*
 829         * This code is triggered unconditionally. Check the affinity
 830         * mask pointer. For CPU_MASK_OFFSTACK=n this is optimized out.
 831         */
 832        if (desc->irq_common_data.affinity)
 833                cpumask_copy(mask, desc->irq_common_data.affinity);
 834        else
 835                valid = false;
 836        raw_spin_unlock_irq(&desc->lock);
 837
 838        if (valid)
 839                set_cpus_allowed_ptr(current, mask);
 840        free_cpumask_var(mask);
 841}
 842#else
 843static inline void
 844irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action) { }
 845#endif
 846
 847/*
 848 * Interrupts which are not explicitely requested as threaded
 849 * interrupts rely on the implicit bh/preempt disable of the hard irq
 850 * context. So we need to disable bh here to avoid deadlocks and other
 851 * side effects.
 852 */
 853static irqreturn_t
 854irq_forced_thread_fn(struct irq_desc *desc, struct irqaction *action)
 855{
 856        irqreturn_t ret;
 857
 858        local_bh_disable();
 859        ret = action->thread_fn(action->irq, action->dev_id);
 860        irq_finalize_oneshot(desc, action);
 861        local_bh_enable();
 862        return ret;
 863}
 864
 865/*
 866 * Interrupts explicitly requested as threaded interrupts want to be
 867 * preemtible - many of them need to sleep and wait for slow busses to
 868 * complete.
 869 */
 870static irqreturn_t irq_thread_fn(struct irq_desc *desc,
 871                struct irqaction *action)
 872{
 873        irqreturn_t ret;
 874
 875        ret = action->thread_fn(action->irq, action->dev_id);
 876        irq_finalize_oneshot(desc, action);
 877        return ret;
 878}
 879
 880static void wake_threads_waitq(struct irq_desc *desc)
 881{
 882        if (atomic_dec_and_test(&desc->threads_active))
 883                wake_up(&desc->wait_for_threads);
 884}
 885
 886static void irq_thread_dtor(struct callback_head *unused)
 887{
 888        struct task_struct *tsk = current;
 889        struct irq_desc *desc;
 890        struct irqaction *action;
 891
 892        if (WARN_ON_ONCE(!(current->flags & PF_EXITING)))
 893                return;
 894
 895        action = kthread_data(tsk);
 896
 897        pr_err("exiting task \"%s\" (%d) is an active IRQ thread (irq %d)\n",
 898               tsk->comm, tsk->pid, action->irq);
 899
 900
 901        desc = irq_to_desc(action->irq);
 902        /*
 903         * If IRQTF_RUNTHREAD is set, we need to decrement
 904         * desc->threads_active and wake possible waiters.
 905         */
 906        if (test_and_clear_bit(IRQTF_RUNTHREAD, &action->thread_flags))
 907                wake_threads_waitq(desc);
 908
 909        /* Prevent a stale desc->threads_oneshot */
 910        irq_finalize_oneshot(desc, action);
 911}
 912
 913/*
 914 * Interrupt handler thread
 915 */
 916static int irq_thread(void *data)
 917{
 918        struct callback_head on_exit_work;
 919        struct irqaction *action = data;
 920        struct irq_desc *desc = irq_to_desc(action->irq);
 921        irqreturn_t (*handler_fn)(struct irq_desc *desc,
 922                        struct irqaction *action);
 923
 924        if (force_irqthreads && test_bit(IRQTF_FORCED_THREAD,
 925                                        &action->thread_flags))
 926                handler_fn = irq_forced_thread_fn;
 927        else
 928                handler_fn = irq_thread_fn;
 929
 930        init_task_work(&on_exit_work, irq_thread_dtor);
 931        task_work_add(current, &on_exit_work, false);
 932
 933        irq_thread_check_affinity(desc, action);
 934
 935        while (!irq_wait_for_interrupt(action)) {
 936                irqreturn_t action_ret;
 937
 938                irq_thread_check_affinity(desc, action);
 939
 940                action_ret = handler_fn(desc, action);
 941                if (action_ret == IRQ_HANDLED)
 942                        atomic_inc(&desc->threads_handled);
 943
 944                wake_threads_waitq(desc);
 945        }
 946
 947        /*
 948         * This is the regular exit path. __free_irq() is stopping the
 949         * thread via kthread_stop() after calling
 950         * synchronize_irq(). So neither IRQTF_RUNTHREAD nor the
 951         * oneshot mask bit can be set. We cannot verify that as we
 952         * cannot touch the oneshot mask at this point anymore as
 953         * __setup_irq() might have given out currents thread_mask
 954         * again.
 955         */
 956        task_work_cancel(current, irq_thread_dtor);
 957        return 0;
 958}
 959
 960/**
 961 *      irq_wake_thread - wake the irq thread for the action identified by dev_id
 962 *      @irq:           Interrupt line
 963 *      @dev_id:        Device identity for which the thread should be woken
 964 *
 965 */
 966void irq_wake_thread(unsigned int irq, void *dev_id)
 967{
 968        struct irq_desc *desc = irq_to_desc(irq);
 969        struct irqaction *action;
 970        unsigned long flags;
 971
 972        if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
 973                return;
 974
 975        raw_spin_lock_irqsave(&desc->lock, flags);
 976        for (action = desc->action; action; action = action->next) {
 977                if (action->dev_id == dev_id) {
 978                        if (action->thread)
 979                                __irq_wake_thread(desc, action);
 980                        break;
 981                }
 982        }
 983        raw_spin_unlock_irqrestore(&desc->lock, flags);
 984}
 985EXPORT_SYMBOL_GPL(irq_wake_thread);
 986
 987static void irq_setup_forced_threading(struct irqaction *new)
 988{
 989        if (!force_irqthreads)
 990                return;
 991        if (new->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT))
 992                return;
 993
 994        new->flags |= IRQF_ONESHOT;
 995
 996        if (!new->thread_fn) {
 997                set_bit(IRQTF_FORCED_THREAD, &new->thread_flags);
 998                new->thread_fn = new->handler;
 999                new->handler = irq_default_primary_handler;
1000        }
1001}
1002
1003static int irq_request_resources(struct irq_desc *desc)
1004{
1005        struct irq_data *d = &desc->irq_data;
1006        struct irq_chip *c = d->chip;
1007
1008        return c->irq_request_resources ? c->irq_request_resources(d) : 0;
1009}
1010
1011static void irq_release_resources(struct irq_desc *desc)
1012{
1013        struct irq_data *d = &desc->irq_data;
1014        struct irq_chip *c = d->chip;
1015
1016        if (c->irq_release_resources)
1017                c->irq_release_resources(d);
1018}
1019
1020/*
1021 * Internal function to register an irqaction - typically used to
1022 * allocate special interrupts that are part of the architecture.
1023 */
1024static int
1025__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
1026{
1027        struct irqaction *old, **old_ptr;
1028        unsigned long flags, thread_mask = 0;
1029        int ret, nested, shared = 0;
1030        cpumask_var_t mask;
1031
1032        if (!desc)
1033                return -EINVAL;
1034
1035        if (desc->irq_data.chip == &no_irq_chip)
1036                return -ENOSYS;
1037        if (!try_module_get(desc->owner))
1038                return -ENODEV;
1039
1040        /*
1041         * Check whether the interrupt nests into another interrupt
1042         * thread.
1043         */
1044        nested = irq_settings_is_nested_thread(desc);
1045        if (nested) {
1046                if (!new->thread_fn) {
1047                        ret = -EINVAL;
1048                        goto out_mput;
1049                }
1050                /*
1051                 * Replace the primary handler which was provided from
1052                 * the driver for non nested interrupt handling by the
1053                 * dummy function which warns when called.
1054                 */
1055                new->handler = irq_nested_primary_handler;
1056        } else {
1057                if (irq_settings_can_thread(desc))
1058                        irq_setup_forced_threading(new);
1059        }
1060
1061        /*
1062         * Create a handler thread when a thread function is supplied
1063         * and the interrupt does not nest into another interrupt
1064         * thread.
1065         */
1066        if (new->thread_fn && !nested) {
1067                struct task_struct *t;
1068                static const struct sched_param param = {
1069                        .sched_priority = MAX_USER_RT_PRIO/2,
1070                };
1071
1072                t = kthread_create(irq_thread, new, "irq/%d-%s", irq,
1073                                   new->name);
1074                if (IS_ERR(t)) {
1075                        ret = PTR_ERR(t);
1076                        goto out_mput;
1077                }
1078
1079                sched_setscheduler_nocheck(t, SCHED_FIFO, &param);
1080
1081                /*
1082                 * We keep the reference to the task struct even if
1083                 * the thread dies to avoid that the interrupt code
1084                 * references an already freed task_struct.
1085                 */
1086                get_task_struct(t);
1087                new->thread = t;
1088                /*
1089                 * Tell the thread to set its affinity. This is
1090                 * important for shared interrupt handlers as we do
1091                 * not invoke setup_affinity() for the secondary
1092                 * handlers as everything is already set up. Even for
1093                 * interrupts marked with IRQF_NO_BALANCE this is
1094                 * correct as we want the thread to move to the cpu(s)
1095                 * on which the requesting code placed the interrupt.
1096                 */
1097                set_bit(IRQTF_AFFINITY, &new->thread_flags);
1098        }
1099
1100        if (!alloc_cpumask_var(&mask, GFP_KERNEL)) {
1101                ret = -ENOMEM;
1102                goto out_thread;
1103        }
1104
1105        /*
1106         * Drivers are often written to work w/o knowledge about the
1107         * underlying irq chip implementation, so a request for a
1108         * threaded irq without a primary hard irq context handler
1109         * requires the ONESHOT flag to be set. Some irq chips like
1110         * MSI based interrupts are per se one shot safe. Check the
1111         * chip flags, so we can avoid the unmask dance at the end of
1112         * the threaded handler for those.
1113         */
1114        if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
1115                new->flags &= ~IRQF_ONESHOT;
1116
1117        /*
1118         * The following block of code has to be executed atomically
1119         */
1120        raw_spin_lock_irqsave(&desc->lock, flags);
1121        old_ptr = &desc->action;
1122        old = *old_ptr;
1123        if (old) {
1124                /*
1125                 * Can't share interrupts unless both agree to and are
1126                 * the same type (level, edge, polarity). So both flag
1127                 * fields must have IRQF_SHARED set and the bits which
1128                 * set the trigger type must match. Also all must
1129                 * agree on ONESHOT.
1130                 */
1131                if (!((old->flags & new->flags) & IRQF_SHARED) ||
1132                    ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
1133                    ((old->flags ^ new->flags) & IRQF_ONESHOT))
1134                        goto mismatch;
1135
1136                /* All handlers must agree on per-cpuness */
1137                if ((old->flags & IRQF_PERCPU) !=
1138                    (new->flags & IRQF_PERCPU))
1139                        goto mismatch;
1140
1141                /* add new interrupt at end of irq queue */
1142                do {
1143                        /*
1144                         * Or all existing action->thread_mask bits,
1145                         * so we can find the next zero bit for this
1146                         * new action.
1147                         */
1148                        thread_mask |= old->thread_mask;
1149                        old_ptr = &old->next;
1150                        old = *old_ptr;
1151                } while (old);
1152                shared = 1;
1153        }
1154
1155        /*
1156         * Setup the thread mask for this irqaction for ONESHOT. For
1157         * !ONESHOT irqs the thread mask is 0 so we can avoid a
1158         * conditional in irq_wake_thread().
1159         */
1160        if (new->flags & IRQF_ONESHOT) {
1161                /*
1162                 * Unlikely to have 32 resp 64 irqs sharing one line,
1163                 * but who knows.
1164                 */
1165                if (thread_mask == ~0UL) {
1166                        ret = -EBUSY;
1167                        goto out_mask;
1168                }
1169                /*
1170                 * The thread_mask for the action is or'ed to
1171                 * desc->thread_active to indicate that the
1172                 * IRQF_ONESHOT thread handler has been woken, but not
1173                 * yet finished. The bit is cleared when a thread
1174                 * completes. When all threads of a shared interrupt
1175                 * line have completed desc->threads_active becomes
1176                 * zero and the interrupt line is unmasked. See
1177                 * handle.c:irq_wake_thread() for further information.
1178                 *
1179                 * If no thread is woken by primary (hard irq context)
1180                 * interrupt handlers, then desc->threads_active is
1181                 * also checked for zero to unmask the irq line in the
1182                 * affected hard irq flow handlers
1183                 * (handle_[fasteoi|level]_irq).
1184                 *
1185                 * The new action gets the first zero bit of
1186                 * thread_mask assigned. See the loop above which or's
1187                 * all existing action->thread_mask bits.
1188                 */
1189                new->thread_mask = 1 << ffz(thread_mask);
1190
1191        } else if (new->handler == irq_default_primary_handler &&
1192                   !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
1193                /*
1194                 * The interrupt was requested with handler = NULL, so
1195                 * we use the default primary handler for it. But it
1196                 * does not have the oneshot flag set. In combination
1197                 * with level interrupts this is deadly, because the
1198                 * default primary handler just wakes the thread, then
1199                 * the irq lines is reenabled, but the device still
1200                 * has the level irq asserted. Rinse and repeat....
1201                 *
1202                 * While this works for edge type interrupts, we play
1203                 * it safe and reject unconditionally because we can't
1204                 * say for sure which type this interrupt really
1205                 * has. The type flags are unreliable as the
1206                 * underlying chip implementation can override them.
1207                 */
1208                pr_err("Threaded irq requested with handler=NULL and !ONESHOT for irq %d\n",
1209                       irq);
1210                ret = -EINVAL;
1211                goto out_mask;
1212        }
1213
1214        if (!shared) {
1215                ret = irq_request_resources(desc);
1216                if (ret) {
1217                        pr_err("Failed to request resources for %s (irq %d) on irqchip %s\n",
1218                               new->name, irq, desc->irq_data.chip->name);
1219                        goto out_mask;
1220                }
1221
1222                init_waitqueue_head(&desc->wait_for_threads);
1223
1224                /* Setup the type (level, edge polarity) if configured: */
1225                if (new->flags & IRQF_TRIGGER_MASK) {
1226                        ret = __irq_set_trigger(desc,
1227                                                new->flags & IRQF_TRIGGER_MASK);
1228
1229                        if (ret)
1230                                goto out_mask;
1231                }
1232
1233                desc->istate &= ~(IRQS_AUTODETECT | IRQS_SPURIOUS_DISABLED | \
1234                                  IRQS_ONESHOT | IRQS_WAITING);
1235                irqd_clear(&desc->irq_data, IRQD_IRQ_INPROGRESS);
1236
1237                if (new->flags & IRQF_PERCPU) {
1238                        irqd_set(&desc->irq_data, IRQD_PER_CPU);
1239                        irq_settings_set_per_cpu(desc);
1240                }
1241
1242                if (new->flags & IRQF_ONESHOT)
1243                        desc->istate |= IRQS_ONESHOT;
1244
1245                if (irq_settings_can_autoenable(desc))
1246                        irq_startup(desc, true);
1247                else
1248                        /* Undo nested disables: */
1249                        desc->depth = 1;
1250
1251                /* Exclude IRQ from balancing if requested */
1252                if (new->flags & IRQF_NOBALANCING) {
1253                        irq_settings_set_no_balancing(desc);
1254                        irqd_set(&desc->irq_data, IRQD_NO_BALANCING);
1255                }
1256
1257                /* Set default affinity mask once everything is setup */
1258                setup_affinity(desc, mask);
1259
1260        } else if (new->flags & IRQF_TRIGGER_MASK) {
1261                unsigned int nmsk = new->flags & IRQF_TRIGGER_MASK;
1262                unsigned int omsk = irq_settings_get_trigger_mask(desc);
1263
1264                if (nmsk != omsk)
1265                        /* hope the handler works with current  trigger mode */
1266                        pr_warning("irq %d uses trigger mode %u; requested %u\n",
1267                                   irq, nmsk, omsk);
1268        }
1269
1270        new->irq = irq;
1271        *old_ptr = new;
1272
1273        irq_pm_install_action(desc, new);
1274
1275        /* Reset broken irq detection when installing new handler */
1276        desc->irq_count = 0;
1277        desc->irqs_unhandled = 0;
1278
1279        /*
1280         * Check whether we disabled the irq via the spurious handler
1281         * before. Reenable it and give it another chance.
1282         */
1283        if (shared && (desc->istate & IRQS_SPURIOUS_DISABLED)) {
1284                desc->istate &= ~IRQS_SPURIOUS_DISABLED;
1285                __enable_irq(desc);
1286        }
1287
1288        raw_spin_unlock_irqrestore(&desc->lock, flags);
1289
1290        /*
1291         * Strictly no need to wake it up, but hung_task complains
1292         * when no hard interrupt wakes the thread up.
1293         */
1294        if (new->thread)
1295                wake_up_process(new->thread);
1296
1297        register_irq_proc(irq, desc);
1298        new->dir = NULL;
1299        register_handler_proc(irq, new);
1300        free_cpumask_var(mask);
1301
1302        return 0;
1303
1304mismatch:
1305        if (!(new->flags & IRQF_PROBE_SHARED)) {
1306                pr_err("Flags mismatch irq %d. %08x (%s) vs. %08x (%s)\n",
1307                       irq, new->flags, new->name, old->flags, old->name);
1308#ifdef CONFIG_DEBUG_SHIRQ
1309                dump_stack();
1310#endif
1311        }
1312        ret = -EBUSY;
1313
1314out_mask:
1315        raw_spin_unlock_irqrestore(&desc->lock, flags);
1316        free_cpumask_var(mask);
1317
1318out_thread:
1319        if (new->thread) {
1320                struct task_struct *t = new->thread;
1321
1322                new->thread = NULL;
1323                kthread_stop(t);
1324                put_task_struct(t);
1325        }
1326out_mput:
1327        module_put(desc->owner);
1328        return ret;
1329}
1330
1331/**
1332 *      setup_irq - setup an interrupt
1333 *      @irq: Interrupt line to setup
1334 *      @act: irqaction for the interrupt
1335 *
1336 * Used to statically setup interrupts in the early boot process.
1337 */
1338int setup_irq(unsigned int irq, struct irqaction *act)
1339{
1340        int retval;
1341        struct irq_desc *desc = irq_to_desc(irq);
1342
1343        if (WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1344                return -EINVAL;
1345        chip_bus_lock(desc);
1346        retval = __setup_irq(irq, desc, act);
1347        chip_bus_sync_unlock(desc);
1348
1349        return retval;
1350}
1351EXPORT_SYMBOL_GPL(setup_irq);
1352
1353/*
1354 * Internal function to unregister an irqaction - used to free
1355 * regular and special interrupts that are part of the architecture.
1356 */
1357static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
1358{
1359        struct irq_desc *desc = irq_to_desc(irq);
1360        struct irqaction *action, **action_ptr;
1361        unsigned long flags;
1362
1363        WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1364
1365        if (!desc)
1366                return NULL;
1367
1368        raw_spin_lock_irqsave(&desc->lock, flags);
1369
1370        /*
1371         * There can be multiple actions per IRQ descriptor, find the right
1372         * one based on the dev_id:
1373         */
1374        action_ptr = &desc->action;
1375        for (;;) {
1376                action = *action_ptr;
1377
1378                if (!action) {
1379                        WARN(1, "Trying to free already-free IRQ %d\n", irq);
1380                        raw_spin_unlock_irqrestore(&desc->lock, flags);
1381
1382                        return NULL;
1383                }
1384
1385                if (action->dev_id == dev_id)
1386                        break;
1387                action_ptr = &action->next;
1388        }
1389
1390        /* Found it - now remove it from the list of entries: */
1391        *action_ptr = action->next;
1392
1393        irq_pm_remove_action(desc, action);
1394
1395        /* If this was the last handler, shut down the IRQ line: */
1396        if (!desc->action) {
1397                irq_shutdown(desc);
1398                irq_release_resources(desc);
1399        }
1400
1401#ifdef CONFIG_SMP
1402        /* make sure affinity_hint is cleaned up */
1403        if (WARN_ON_ONCE(desc->affinity_hint))
1404                desc->affinity_hint = NULL;
1405#endif
1406
1407        raw_spin_unlock_irqrestore(&desc->lock, flags);
1408
1409        unregister_handler_proc(irq, action);
1410
1411        /* Make sure it's not being used on another CPU: */
1412        synchronize_irq(irq);
1413
1414#ifdef CONFIG_DEBUG_SHIRQ
1415        /*
1416         * It's a shared IRQ -- the driver ought to be prepared for an IRQ
1417         * event to happen even now it's being freed, so let's make sure that
1418         * is so by doing an extra call to the handler ....
1419         *
1420         * ( We do this after actually deregistering it, to make sure that a
1421         *   'real' IRQ doesn't run in * parallel with our fake. )
1422         */
1423        if (action->flags & IRQF_SHARED) {
1424                local_irq_save(flags);
1425                action->handler(irq, dev_id);
1426                local_irq_restore(flags);
1427        }
1428#endif
1429
1430        if (action->thread) {
1431                kthread_stop(action->thread);
1432                put_task_struct(action->thread);
1433        }
1434
1435        module_put(desc->owner);
1436        return action;
1437}
1438
1439/**
1440 *      remove_irq - free an interrupt
1441 *      @irq: Interrupt line to free
1442 *      @act: irqaction for the interrupt
1443 *
1444 * Used to remove interrupts statically setup by the early boot process.
1445 */
1446void remove_irq(unsigned int irq, struct irqaction *act)
1447{
1448        struct irq_desc *desc = irq_to_desc(irq);
1449
1450        if (desc && !WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1451            __free_irq(irq, act->dev_id);
1452}
1453EXPORT_SYMBOL_GPL(remove_irq);
1454
1455/**
1456 *      free_irq - free an interrupt allocated with request_irq
1457 *      @irq: Interrupt line to free
1458 *      @dev_id: Device identity to free
1459 *
1460 *      Remove an interrupt handler. The handler is removed and if the
1461 *      interrupt line is no longer in use by any driver it is disabled.
1462 *      On a shared IRQ the caller must ensure the interrupt is disabled
1463 *      on the card it drives before calling this function. The function
1464 *      does not return until any executing interrupts for this IRQ
1465 *      have completed.
1466 *
1467 *      This function must not be called from interrupt context.
1468 */
1469void free_irq(unsigned int irq, void *dev_id)
1470{
1471        struct irq_desc *desc = irq_to_desc(irq);
1472
1473        if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1474                return;
1475
1476#ifdef CONFIG_SMP
1477        if (WARN_ON(desc->affinity_notify))
1478                desc->affinity_notify = NULL;
1479#endif
1480
1481        chip_bus_lock(desc);
1482        kfree(__free_irq(irq, dev_id));
1483        chip_bus_sync_unlock(desc);
1484}
1485EXPORT_SYMBOL(free_irq);
1486
1487/**
1488 *      request_threaded_irq - allocate an interrupt line
1489 *      @irq: Interrupt line to allocate
1490 *      @handler: Function to be called when the IRQ occurs.
1491 *                Primary handler for threaded interrupts
1492 *                If NULL and thread_fn != NULL the default
1493 *                primary handler is installed
1494 *      @thread_fn: Function called from the irq handler thread
1495 *                  If NULL, no irq thread is created
1496 *      @irqflags: Interrupt type flags
1497 *      @devname: An ascii name for the claiming device
1498 *      @dev_id: A cookie passed back to the handler function
1499 *
1500 *      This call allocates interrupt resources and enables the
1501 *      interrupt line and IRQ handling. From the point this
1502 *      call is made your handler function may be invoked. Since
1503 *      your handler function must clear any interrupt the board
1504 *      raises, you must take care both to initialise your hardware
1505 *      and to set up the interrupt handler in the right order.
1506 *
1507 *      If you want to set up a threaded irq handler for your device
1508 *      then you need to supply @handler and @thread_fn. @handler is
1509 *      still called in hard interrupt context and has to check
1510 *      whether the interrupt originates from the device. If yes it
1511 *      needs to disable the interrupt on the device and return
1512 *      IRQ_WAKE_THREAD which will wake up the handler thread and run
1513 *      @thread_fn. This split handler design is necessary to support
1514 *      shared interrupts.
1515 *
1516 *      Dev_id must be globally unique. Normally the address of the
1517 *      device data structure is used as the cookie. Since the handler
1518 *      receives this value it makes sense to use it.
1519 *
1520 *      If your interrupt is shared you must pass a non NULL dev_id
1521 *      as this is required when freeing the interrupt.
1522 *
1523 *      Flags:
1524 *
1525 *      IRQF_SHARED             Interrupt is shared
1526 *      IRQF_TRIGGER_*          Specify active edge(s) or level
1527 *
1528 */
1529int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1530                         irq_handler_t thread_fn, unsigned long irqflags,
1531                         const char *devname, void *dev_id)
1532{
1533        struct irqaction *action;
1534        struct irq_desc *desc;
1535        int retval;
1536
1537        /*
1538         * Sanity-check: shared interrupts must pass in a real dev-ID,
1539         * otherwise we'll have trouble later trying to figure out
1540         * which interrupt is which (messes up the interrupt freeing
1541         * logic etc).
1542         *
1543         * Also IRQF_COND_SUSPEND only makes sense for shared interrupts and
1544         * it cannot be set along with IRQF_NO_SUSPEND.
1545         */
1546        if (((irqflags & IRQF_SHARED) && !dev_id) ||
1547            (!(irqflags & IRQF_SHARED) && (irqflags & IRQF_COND_SUSPEND)) ||
1548            ((irqflags & IRQF_NO_SUSPEND) && (irqflags & IRQF_COND_SUSPEND)))
1549                return -EINVAL;
1550
1551        desc = irq_to_desc(irq);
1552        if (!desc)
1553                return -EINVAL;
1554
1555        if (!irq_settings_can_request(desc) ||
1556            WARN_ON(irq_settings_is_per_cpu_devid(desc)))
1557                return -EINVAL;
1558
1559        if (!handler) {
1560                if (!thread_fn)
1561                        return -EINVAL;
1562                handler = irq_default_primary_handler;
1563        }
1564
1565        action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1566        if (!action)
1567                return -ENOMEM;
1568
1569        action->handler = handler;
1570        action->thread_fn = thread_fn;
1571        action->flags = irqflags;
1572        action->name = devname;
1573        action->dev_id = dev_id;
1574
1575        chip_bus_lock(desc);
1576        retval = __setup_irq(irq, desc, action);
1577        chip_bus_sync_unlock(desc);
1578
1579        if (retval)
1580                kfree(action);
1581
1582#ifdef CONFIG_DEBUG_SHIRQ_FIXME
1583        if (!retval && (irqflags & IRQF_SHARED)) {
1584                /*
1585                 * It's a shared IRQ -- the driver ought to be prepared for it
1586                 * to happen immediately, so let's make sure....
1587                 * We disable the irq to make sure that a 'real' IRQ doesn't
1588                 * run in parallel with our fake.
1589                 */
1590                unsigned long flags;
1591
1592                disable_irq(irq);
1593                local_irq_save(flags);
1594
1595                handler(irq, dev_id);
1596
1597                local_irq_restore(flags);
1598                enable_irq(irq);
1599        }
1600#endif
1601        return retval;
1602}
1603EXPORT_SYMBOL(request_threaded_irq);
1604
1605/**
1606 *      request_any_context_irq - allocate an interrupt line
1607 *      @irq: Interrupt line to allocate
1608 *      @handler: Function to be called when the IRQ occurs.
1609 *                Threaded handler for threaded interrupts.
1610 *      @flags: Interrupt type flags
1611 *      @name: An ascii name for the claiming device
1612 *      @dev_id: A cookie passed back to the handler function
1613 *
1614 *      This call allocates interrupt resources and enables the
1615 *      interrupt line and IRQ handling. It selects either a
1616 *      hardirq or threaded handling method depending on the
1617 *      context.
1618 *
1619 *      On failure, it returns a negative value. On success,
1620 *      it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
1621 */
1622int request_any_context_irq(unsigned int irq, irq_handler_t handler,
1623                            unsigned long flags, const char *name, void *dev_id)
1624{
1625        struct irq_desc *desc = irq_to_desc(irq);
1626        int ret;
1627
1628        if (!desc)
1629                return -EINVAL;
1630
1631        if (irq_settings_is_nested_thread(desc)) {
1632                ret = request_threaded_irq(irq, NULL, handler,
1633                                           flags, name, dev_id);
1634                return !ret ? IRQC_IS_NESTED : ret;
1635        }
1636
1637        ret = request_irq(irq, handler, flags, name, dev_id);
1638        return !ret ? IRQC_IS_HARDIRQ : ret;
1639}
1640EXPORT_SYMBOL_GPL(request_any_context_irq);
1641
1642void enable_percpu_irq(unsigned int irq, unsigned int type)
1643{
1644        unsigned int cpu = smp_processor_id();
1645        unsigned long flags;
1646        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1647
1648        if (!desc)
1649                return;
1650
1651        type &= IRQ_TYPE_SENSE_MASK;
1652        if (type != IRQ_TYPE_NONE) {
1653                int ret;
1654
1655                ret = __irq_set_trigger(desc, type);
1656
1657                if (ret) {
1658                        WARN(1, "failed to set type for IRQ%d\n", irq);
1659                        goto out;
1660                }
1661        }
1662
1663        irq_percpu_enable(desc, cpu);
1664out:
1665        irq_put_desc_unlock(desc, flags);
1666}
1667EXPORT_SYMBOL_GPL(enable_percpu_irq);
1668
1669void disable_percpu_irq(unsigned int irq)
1670{
1671        unsigned int cpu = smp_processor_id();
1672        unsigned long flags;
1673        struct irq_desc *desc = irq_get_desc_lock(irq, &flags, IRQ_GET_DESC_CHECK_PERCPU);
1674
1675        if (!desc)
1676                return;
1677
1678        irq_percpu_disable(desc, cpu);
1679        irq_put_desc_unlock(desc, flags);
1680}
1681EXPORT_SYMBOL_GPL(disable_percpu_irq);
1682
1683/*
1684 * Internal function to unregister a percpu irqaction.
1685 */
1686static struct irqaction *__free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1687{
1688        struct irq_desc *desc = irq_to_desc(irq);
1689        struct irqaction *action;
1690        unsigned long flags;
1691
1692        WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
1693
1694        if (!desc)
1695                return NULL;
1696
1697        raw_spin_lock_irqsave(&desc->lock, flags);
1698
1699        action = desc->action;
1700        if (!action || action->percpu_dev_id != dev_id) {
1701                WARN(1, "Trying to free already-free IRQ %d\n", irq);
1702                goto bad;
1703        }
1704
1705        if (!cpumask_empty(desc->percpu_enabled)) {
1706                WARN(1, "percpu IRQ %d still enabled on CPU%d!\n",
1707                     irq, cpumask_first(desc->percpu_enabled));
1708                goto bad;
1709        }
1710
1711        /* Found it - now remove it from the list of entries: */
1712        desc->action = NULL;
1713
1714        raw_spin_unlock_irqrestore(&desc->lock, flags);
1715
1716        unregister_handler_proc(irq, action);
1717
1718        module_put(desc->owner);
1719        return action;
1720
1721bad:
1722        raw_spin_unlock_irqrestore(&desc->lock, flags);
1723        return NULL;
1724}
1725
1726/**
1727 *      remove_percpu_irq - free a per-cpu interrupt
1728 *      @irq: Interrupt line to free
1729 *      @act: irqaction for the interrupt
1730 *
1731 * Used to remove interrupts statically setup by the early boot process.
1732 */
1733void remove_percpu_irq(unsigned int irq, struct irqaction *act)
1734{
1735        struct irq_desc *desc = irq_to_desc(irq);
1736
1737        if (desc && irq_settings_is_per_cpu_devid(desc))
1738            __free_percpu_irq(irq, act->percpu_dev_id);
1739}
1740
1741/**
1742 *      free_percpu_irq - free an interrupt allocated with request_percpu_irq
1743 *      @irq: Interrupt line to free
1744 *      @dev_id: Device identity to free
1745 *
1746 *      Remove a percpu interrupt handler. The handler is removed, but
1747 *      the interrupt line is not disabled. This must be done on each
1748 *      CPU before calling this function. The function does not return
1749 *      until any executing interrupts for this IRQ have completed.
1750 *
1751 *      This function must not be called from interrupt context.
1752 */
1753void free_percpu_irq(unsigned int irq, void __percpu *dev_id)
1754{
1755        struct irq_desc *desc = irq_to_desc(irq);
1756
1757        if (!desc || !irq_settings_is_per_cpu_devid(desc))
1758                return;
1759
1760        chip_bus_lock(desc);
1761        kfree(__free_percpu_irq(irq, dev_id));
1762        chip_bus_sync_unlock(desc);
1763}
1764
1765/**
1766 *      setup_percpu_irq - setup a per-cpu interrupt
1767 *      @irq: Interrupt line to setup
1768 *      @act: irqaction for the interrupt
1769 *
1770 * Used to statically setup per-cpu interrupts in the early boot process.
1771 */
1772int setup_percpu_irq(unsigned int irq, struct irqaction *act)
1773{
1774        struct irq_desc *desc = irq_to_desc(irq);
1775        int retval;
1776
1777        if (!desc || !irq_settings_is_per_cpu_devid(desc))
1778                return -EINVAL;
1779        chip_bus_lock(desc);
1780        retval = __setup_irq(irq, desc, act);
1781        chip_bus_sync_unlock(desc);
1782
1783        return retval;
1784}
1785
1786/**
1787 *      request_percpu_irq - allocate a percpu interrupt line
1788 *      @irq: Interrupt line to allocate
1789 *      @handler: Function to be called when the IRQ occurs.
1790 *      @devname: An ascii name for the claiming device
1791 *      @dev_id: A percpu cookie passed back to the handler function
1792 *
1793 *      This call allocates interrupt resources, but doesn't
1794 *      automatically enable the interrupt. It has to be done on each
1795 *      CPU using enable_percpu_irq().
1796 *
1797 *      Dev_id must be globally unique. It is a per-cpu variable, and
1798 *      the handler gets called with the interrupted CPU's instance of
1799 *      that variable.
1800 */
1801int request_percpu_irq(unsigned int irq, irq_handler_t handler,
1802                       const char *devname, void __percpu *dev_id)
1803{
1804        struct irqaction *action;
1805        struct irq_desc *desc;
1806        int retval;
1807
1808        if (!dev_id)
1809                return -EINVAL;
1810
1811        desc = irq_to_desc(irq);
1812        if (!desc || !irq_settings_can_request(desc) ||
1813            !irq_settings_is_per_cpu_devid(desc))
1814                return -EINVAL;
1815
1816        action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
1817        if (!action)
1818                return -ENOMEM;
1819
1820        action->handler = handler;
1821        action->flags = IRQF_PERCPU | IRQF_NO_SUSPEND;
1822        action->name = devname;
1823        action->percpu_dev_id = dev_id;
1824
1825        chip_bus_lock(desc);
1826        retval = __setup_irq(irq, desc, action);
1827        chip_bus_sync_unlock(desc);
1828
1829        if (retval)
1830                kfree(action);
1831
1832        return retval;
1833}
1834
1835/**
1836 *      irq_get_irqchip_state - returns the irqchip state of a interrupt.
1837 *      @irq: Interrupt line that is forwarded to a VM
1838 *      @which: One of IRQCHIP_STATE_* the caller wants to know about
1839 *      @state: a pointer to a boolean where the state is to be storeed
1840 *
1841 *      This call snapshots the internal irqchip state of an
1842 *      interrupt, returning into @state the bit corresponding to
1843 *      stage @which
1844 *
1845 *      This function should be called with preemption disabled if the
1846 *      interrupt controller has per-cpu registers.
1847 */
1848int irq_get_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1849                          bool *state)
1850{
1851        struct irq_desc *desc;
1852        struct irq_data *data;
1853        struct irq_chip *chip;
1854        unsigned long flags;
1855        int err = -EINVAL;
1856
1857        desc = irq_get_desc_buslock(irq, &flags, 0);
1858        if (!desc)
1859                return err;
1860
1861        data = irq_desc_get_irq_data(desc);
1862
1863        do {
1864                chip = irq_data_get_irq_chip(data);
1865                if (chip->irq_get_irqchip_state)
1866                        break;
1867#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1868                data = data->parent_data;
1869#else
1870                data = NULL;
1871#endif
1872        } while (data);
1873
1874        if (data)
1875                err = chip->irq_get_irqchip_state(data, which, state);
1876
1877        irq_put_desc_busunlock(desc, flags);
1878        return err;
1879}
1880EXPORT_SYMBOL_GPL(irq_get_irqchip_state);
1881
1882/**
1883 *      irq_set_irqchip_state - set the state of a forwarded interrupt.
1884 *      @irq: Interrupt line that is forwarded to a VM
1885 *      @which: State to be restored (one of IRQCHIP_STATE_*)
1886 *      @val: Value corresponding to @which
1887 *
1888 *      This call sets the internal irqchip state of an interrupt,
1889 *      depending on the value of @which.
1890 *
1891 *      This function should be called with preemption disabled if the
1892 *      interrupt controller has per-cpu registers.
1893 */
1894int irq_set_irqchip_state(unsigned int irq, enum irqchip_irq_state which,
1895                          bool val)
1896{
1897        struct irq_desc *desc;
1898        struct irq_data *data;
1899        struct irq_chip *chip;
1900        unsigned long flags;
1901        int err = -EINVAL;
1902
1903        desc = irq_get_desc_buslock(irq, &flags, 0);
1904        if (!desc)
1905                return err;
1906
1907        data = irq_desc_get_irq_data(desc);
1908
1909        do {
1910                chip = irq_data_get_irq_chip(data);
1911                if (chip->irq_set_irqchip_state)
1912                        break;
1913#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
1914                data = data->parent_data;
1915#else
1916                data = NULL;
1917#endif
1918        } while (data);
1919
1920        if (data)
1921                err = chip->irq_set_irqchip_state(data, which, val);
1922
1923        irq_put_desc_busunlock(desc, flags);
1924        return err;
1925}
1926EXPORT_SYMBOL_GPL(irq_set_irqchip_state);
1927