kernel/irq/handle.c

   1 /*
   2  * linux/kernel/irq/handle.c
   3  *
   4  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
   5  * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
   6  *
   7  * This file contains the core interrupt handling code.
   8  *
   9  * Detailed information is available in Documentation/DocBook/genericirq
  10  *
  11  */
  12
  13 #include <linux/irq.h>
  14 #include <linux/module.h>
  15 #include <linux/random.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/kernel_stat.h>
  18 #include <linux/rculist.h>
  19 #include <linux/hash.h>
  20 #include <trace/irq.h>
  21 #include <linux/bootmem.h>
  22
  23 #include "internals.h"
  24
  25 /*
  26  * lockdep: we want to handle all irq_desc locks as a single lock-class:
  27  */
  28 struct lock_class_key irq_desc_lock_class;
  29
  30 /**
  31  * handle_bad_irq - handle spurious and unhandled irqs
  32  * @irq:       the interrupt number
  33  * @desc:      description of the interrupt
  34  *
  35  * Handles spurious and unhandled IRQ's. It also prints a debugmessage.
  36  */
  37 void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
  38 {
  39         print_irq_desc(irq, desc);
  40         kstat_incr_irqs_this_cpu(irq, desc);
  41         ack_bad_irq(irq);
  42 }
  43
  44 #if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS)
  45 static void __init init_irq_default_affinity(void)
  46 {
  47         alloc_bootmem_cpumask_var(&irq_default_affinity);
  48         cpumask_setall(irq_default_affinity);
  49 }
  50 #else
  51 static void __init init_irq_default_affinity(void)
  52 {
  53 }
  54 #endif
  55
  56 /*
  57  * Linux has a controller-independent interrupt architecture.
  58  * Every controller has a 'controller-template', that is used
  59  * by the main code to do the right thing. Each driver-visible
  60  * interrupt source is transparently wired to the appropriate
  61  * controller. Thus drivers need not be aware of the
  62  * interrupt-controller.
  63  *
  64  * The code is designed to be easily extended with new/different
  65  * interrupt controllers, without having to do assembly magic or
  66  * having to touch the generic code.
  67  *
  68  * Controller mappings for all interrupt sources:
  69  */
  70 int nr_irqs = NR_IRQS;
  71 EXPORT_SYMBOL_GPL(nr_irqs);
  72
  73 #ifdef CONFIG_SPARSE_IRQ
  74
  75 static struct irq_desc irq_desc_init = {
  76         .irq        = -1,
  77         .status     = IRQ_DISABLED,
  78         .chip       = &no_irq_chip,
  79         .handle_irq = handle_bad_irq,
  80         .depth      = 1,
  81         .lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
  82 };
  83
  84 void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
  85 {
  86         int node;
  87         void *ptr;
  88
  89         node = cpu_to_node(cpu);
  90         ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node);
  91
  92         /*
  93          * don't overwite if can not get new one
  94          * init_copy_kstat_irqs() could still use old one
  95          */
  96         if (ptr) {
  97                 printk(KERN_DEBUG "  alloc kstat_irqs on cpu %d node %d\n",
  98                          cpu, node);
  99                 desc->kstat_irqs = ptr;
 100         }
 101 }
 102
 103 static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu)
 104 {
 105         memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
 106
 107         spin_lock_init(&desc->lock);
 108         desc->irq = irq;
 109 #ifdef CONFIG_SMP
 110         desc->cpu = cpu;
 111 #endif
 112         lockdep_set_class(&desc->lock, &irq_desc_lock_class);
 113         init_kstat_irqs(desc, cpu, nr_cpu_ids);
 114         if (!desc->kstat_irqs) {
 115                 printk(KERN_ERR "can not alloc kstat_irqs\n");
 116                 BUG_ON(1);
 117         }
 118         if (!init_alloc_desc_masks(desc, cpu, false)) {
 119                 printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
 120                 BUG_ON(1);
 121         }
 122         arch_init_chip_data(desc, cpu);
 123 }
 124
 125 /*
 126  * Protect the sparse_irqs:
 127  */
 128 DEFINE_SPINLOCK(sparse_irq_lock);
 129
 130 struct irq_desc **irq_desc_ptrs __read_mostly;
 131
 132 static struct irq_desc irq_desc_legacy[NR_IRQS_LEGACY] __cacheline_aligned_in_smp = {
 133         [0 ... NR_IRQS_LEGACY-1] = {
 134                 .irq        = -1,
 135                 .status     = IRQ_DISABLED,
 136                 .chip       = &no_irq_chip,
 137                 .handle_irq = handle_bad_irq,
 138                 .depth      = 1,
 139                 .lock       = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
 140         }
 141 };
 142
 143 static unsigned int *kstat_irqs_legacy;
 144
 145 int __init early_irq_init(void)
 146 {
 147         struct irq_desc *desc;
 148         int legacy_count;
 149         int i;
 150
 151         init_irq_default_affinity();
 152
 153          /* initialize nr_irqs based on nr_cpu_ids */
 154         arch_probe_nr_irqs();
 155         printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d\n", NR_IRQS, nr_irqs);
 156
 157         desc = irq_desc_legacy;
 158         legacy_count = ARRAY_SIZE(irq_desc_legacy);
 159
 160         /* allocate irq_desc_ptrs array based on nr_irqs */
 161         irq_desc_ptrs = alloc_bootmem(nr_irqs * sizeof(void *));
 162
 163         /* allocate based on nr_cpu_ids */
 164         /* FIXME: invert kstat_irgs, and it'd be a per_cpu_alloc'd thing */
 165         kstat_irqs_legacy = alloc_bootmem(NR_IRQS_LEGACY * nr_cpu_ids *
 166                                           sizeof(int));
 167
 168         for (i = 0; i < legacy_count; i++) {
 169                 desc[i].irq = i;
 170                 desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids;
 171                 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
 172                 init_alloc_desc_masks(&desc[i], 0, true);
 173                 irq_desc_ptrs[i] = desc + i;
 174         }
 175
 176         for (i = legacy_count; i < nr_irqs; i++)
 177                 irq_desc_ptrs[i] = NULL;
 178
 179         return arch_early_irq_init();
 180 }
 181
 182 struct irq_desc *irq_to_desc(unsigned int irq)
 183 {
 184         if (irq_desc_ptrs && irq < nr_irqs)
 185                 return irq_desc_ptrs[irq];
 186
 187         return NULL;
 188 }
 189
 190 struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
 191 {
 192         struct irq_desc *desc;
 193         unsigned long flags;
 194         int node;
 195
 196         if (irq >= nr_irqs) {
 197                 WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
 198                         irq, nr_irqs);
 199                 return NULL;
 200         }
 201
 202         desc = irq_desc_ptrs[irq];
 203         if (desc)
 204                 return desc;
 205
 206         spin_lock_irqsave(&sparse_irq_lock, flags);
 207
 208         /* We have to check it to avoid races with another CPU */
 209         desc = irq_desc_ptrs[irq];
 210         if (desc)
 211                 goto out_unlock;
 212
 213         node = cpu_to_node(cpu);
 214         desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
 215         printk(KERN_DEBUG "  alloc irq_desc for %d on cpu %d node %d\n",
 216                  irq, cpu, node);
 217         if (!desc) {
 218                 printk(KERN_ERR "can not alloc irq_desc\n");
 219                 BUG_ON(1);
 220         }
 221         init_one_irq_desc(irq, desc, cpu);
 222
 223         irq_desc_ptrs[irq] = desc;
 224
 225 out_unlock:
 226         spin_unlock_irqrestore(&sparse_irq_lock, flags);
 227
 228         return desc;
 229 }
 230
 231 #else /* !CONFIG_SPARSE_IRQ */
 232
 233 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
 234         [0 ... NR_IRQS-1] = {
 235                 .status = IRQ_DISABLED,
 236                 .chip = &no_irq_chip,
 237                 .handle_irq = handle_bad_irq,
 238                 .depth = 1,
 239                 .lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
 240         }
 241 };
 242
 243 static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
 244 int __init early_irq_init(void)
 245 {
 246         struct irq_desc *desc;
 247         int count;
 248         int i;
 249
 250         init_irq_default_affinity();
 251
 252         printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
 253
 254         desc = irq_desc;
 255         count = ARRAY_SIZE(irq_desc);
 256
 257         for (i = 0; i < count; i++) {
 258                 desc[i].irq = i;
 259                 init_alloc_desc_masks(&desc[i], 0, true);
 260                 desc[i].kstat_irqs = kstat_irqs_all[i];
 261         }
 262         return arch_early_irq_init();
 263 }
 264
 265 struct irq_desc *irq_to_desc(unsigned int irq)
 266 {
 267         return (irq < NR_IRQS) ? irq_desc + irq : NULL;
 268 }
 269
 270 struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
 271 {
 272         return irq_to_desc(irq);
 273 }
 274 #endif /* !CONFIG_SPARSE_IRQ */
 275
 276 void clear_kstat_irqs(struct irq_desc *desc)
 277 {
 278         memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
 279 }
 280
 281 /*
 282  * What should we do if we get a hw irq event on an illegal vector?
 283  * Each architecture has to answer this themself.
 284  */
 285 static void ack_bad(unsigned int irq)
 286 {
 287         struct irq_desc *desc = irq_to_desc(irq);
 288
 289         print_irq_desc(irq, desc);
 290         ack_bad_irq(irq);
 291 }
 292
 293 /*
 294  * NOP functions
 295  */
 296 static void noop(unsigned int irq)
 297 {
 298 }
 299
 300 static unsigned int noop_ret(unsigned int irq)
 301 {
 302         return 0;
 303 }
 304
 305 /*
 306  * Generic no controller implementation
 307  */
 308 struct irq_chip no_irq_chip = {
 309         .name           = "none",
 310         .startup        = noop_ret,
 311         .shutdown       = noop,
 312         .enable         = noop,
 313         .disable        = noop,
 314         .ack            = ack_bad,
 315         .end            = noop,
 316 };
 317
 318 /*
 319  * Generic dummy implementation which can be used for
 320  * real dumb interrupt sources
 321  */
 322 struct irq_chip dummy_irq_chip = {
 323         .name           = "dummy",
 324         .startup        = noop_ret,
 325         .shutdown       = noop,
 326         .enable         = noop,
 327         .disable        = noop,
 328         .ack            = noop,
 329         .mask           = noop,
 330         .unmask         = noop,
 331         .end            = noop,
 332 };
 333
 334 /*
 335  * Special, empty irq handler:
 336  */
 337 irqreturn_t no_action(int cpl, void *dev_id)
 338 {
 339         return IRQ_NONE;
 340 }
 341
 342 static void warn_no_thread(unsigned int irq, struct irqaction *action)
 343 {
 344         if (test_and_set_bit(IRQTF_WARNED, &action->thread_flags))
 345                 return;
 346
 347         printk(KERN_WARNING "IRQ %d device %s returned IRQ_WAKE_THREAD "
 348                "but no thread function available.", irq, action->name);
 349 }
 350
 351 DEFINE_TRACE(irq_handler_entry);
 352 DEFINE_TRACE(irq_handler_exit);
 353
 354 /**
 355  * handle_IRQ_event - irq action chain handler
 356  * @irq:        the interrupt number
 357  * @action:     the interrupt action chain for this irq
 358  *
 359  * Handles the action chain of an irq event
 360  */
 361 irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
 362 {
 363         irqreturn_t ret, retval = IRQ_NONE;
 364         unsigned int status = 0;
 365
 366         WARN_ONCE(!in_irq(), "BUG: IRQ handler called from non-hardirq context!");
 367
 368         if (!(action->flags & IRQF_DISABLED))
 369                 local_irq_enable_in_hardirq();
 370
 371         do {
 372                 trace_irq_handler_entry(irq, action);
 373                 ret = action->handler(irq, action->dev_id);
 374                 trace_irq_handler_exit(irq, action, ret);
 375
 376                 switch (ret) {
 377                 case IRQ_WAKE_THREAD:
 378                         /*
 379                          * Set result to handled so the spurious check
 380                          * does not trigger.
 381                          */
 382                         ret = IRQ_HANDLED;
 383
 384                         /*
 385                          * Catch drivers which return WAKE_THREAD but
 386                          * did not set up a thread function
 387                          */
 388                         if (unlikely(!action->thread_fn)) {
 389                                 warn_no_thread(irq, action);
 390                                 break;
 391                         }
 392
 393                         /*
 394                          * Wake up the handler thread for this
 395                          * action. In case the thread crashed and was
 396                          * killed we just pretend that we handled the
 397                          * interrupt. The hardirq handler above has
 398                          * disabled the device interrupt, so no irq
 399                          * storm is lurking.
 400                          */
 401                         if (likely(!test_bit(IRQTF_DIED,
 402                                              &action->thread_flags))) {
 403                                 set_bit(IRQTF_RUNTHREAD, &action->thread_flags);
 404                                 wake_up_process(action->thread);
 405                         }
 406
 407                         /* Fall through to add to randomness */
 408                 case IRQ_HANDLED:
 409                         status |= action->flags;
 410                         break;
 411
 412                 default:
 413                         break;
 414                 }
 415
 416                 retval |= ret;
 417                 action = action->next;
 418         } while (action);
 419
 420         if (status & IRQF_SAMPLE_RANDOM)
 421                 add_interrupt_randomness(irq);
 422         local_irq_disable();
 423
 424         return retval;
 425 }
 426
 427 #ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ
 428
 429 #ifdef CONFIG_ENABLE_WARN_DEPRECATED
 430 # warning __do_IRQ is deprecated. Please convert to proper flow handlers
 431 #endif
 432
 433 /**
 434  * __do_IRQ - original all in one highlevel IRQ handler
 435  * @irq:        the interrupt number
 436  *
 437  * __do_IRQ handles all normal device IRQ's (the special
 438  * SMP cross-CPU interrupts have their own specific
 439  * handlers).
 440  *
 441  * This is the original x86 implementation which is used for every
 442  * interrupt type.
 443  */
 444 unsigned int __do_IRQ(unsigned int irq)
 445 {
 446         struct irq_desc *desc = irq_to_desc(irq);
 447         struct irqaction *action;
 448         unsigned int status;
 449
 450         kstat_incr_irqs_this_cpu(irq, desc);
 451
 452         if (CHECK_IRQ_PER_CPU(desc->status)) {
 453                 irqreturn_t action_ret;
 454
 455                 /*
 456                  * No locking required for CPU-local interrupts:
 457                  */
 458                 if (desc->chip->ack) {
 459                         desc->chip->ack(irq);
 460                         /* get new one */
 461                         desc = irq_remap_to_desc(irq, desc);
 462                 }
 463                 if (likely(!(desc->status & IRQ_DISABLED))) {
 464                         action_ret = handle_IRQ_event(irq, desc->action);
 465                         if (!noirqdebug)
 466                                 note_interrupt(irq, desc, action_ret);
 467                 }
 468                 desc->chip->end(irq);
 469                 return 1;
 470         }
 471
 472         spin_lock(&desc->lock);
 473         if (desc->chip->ack) {
 474                 desc->chip->ack(irq);
 475                 desc = irq_remap_to_desc(irq, desc);
 476         }
 477         /*
 478          * REPLAY is when Linux resends an IRQ that was dropped earlier
 479          * WAITING is used by probe to mark irqs that are being tested
 480          */
 481         status = desc->status & ~(IRQ_REPLAY | IRQ_WAITING);
 482         status |= IRQ_PENDING; /* we _want_ to handle it */
 483
 484         /*
 485          * If the IRQ is disabled for whatever reason, we cannot
 486          * use the action we have.
 487          */
 488         action = NULL;
 489         if (likely(!(status & (IRQ_DISABLED | IRQ_INPROGRESS)))) {
 490                 action = desc->action;
 491                 status &= ~IRQ_PENDING; /* we commit to handling */
 492                 status |= IRQ_INPROGRESS; /* we are handling it */
 493         }
 494         desc->status = status;
 495
 496         /*
 497          * If there is no IRQ handler or it was disabled, exit early.
 498          * Since we set PENDING, if another processor is handling
 499          * a different instance of this same irq, the other processor
 500          * will take care of it.
 501          */
 502         if (unlikely(!action))
 503                 goto out;
 504
 505         /*
 506          * Edge triggered interrupts need to remember
 507          * pending events.
 508          * This applies to any hw interrupts that allow a second
 509          * instance of the same irq to arrive while we are in do_IRQ
 510          * or in the handler. But the code here only handles the _second_
 511          * instance of the irq, not the third or fourth. So it is mostly
 512          * useful for irq hardware that does not mask cleanly in an
 513          * SMP environment.
 514          */
 515         for (;;) {
 516                 irqreturn_t action_ret;
 517
 518                 spin_unlock(&desc->lock);
 519
 520                 action_ret = handle_IRQ_event(irq, action);
 521                 if (!noirqdebug)
 522                         note_interrupt(irq, desc, action_ret);
 523
 524                 spin_lock(&desc->lock);
 525                 if (likely(!(desc->status & IRQ_PENDING)))
 526                         break;
 527                 desc->status &= ~IRQ_PENDING;
 528         }
 529         desc->status &= ~IRQ_INPROGRESS;
 530
 531 out:
 532         /*
 533          * The ->end() handler has to deal with interrupts which got
 534          * disabled while the handler was running.
 535          */
 536         desc->chip->end(irq);
 537         spin_unlock(&desc->lock);
 538
 539         return 1;
 540 }
 541 #endif
 542
 543 void early_init_irq_lock_class(void)
 544 {
 545         struct irq_desc *desc;
 546         int i;
 547
 548         for_each_irq_desc(i, desc) {
 549                 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
 550         }
 551 }
 552
 553 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
 554 {
 555         struct irq_desc *desc = irq_to_desc(irq);
 556         return desc ? desc->kstat_irqs[cpu] : 0;
 557 }
 558 EXPORT_SYMBOL(kstat_irqs_cpu);
 559