kernel/irq/spurious.c

   1 /*
   2  * linux/kernel/irq/spurious.c
   3  *
   4  * Copyright (C) 1992, 1998-2004 Linus Torvalds, Ingo Molnar
   5  *
   6  * This file contains spurious interrupt handling.
   7  */
   8
   9 #include <linux/jiffies.h>
  10 #include <linux/irq.h>
  11 #include <linux/module.h>
  12 #include <linux/kallsyms.h>
  13 #include <linux/interrupt.h>
  14 #include <linux/moduleparam.h>
  15 #include <linux/timer.h>
  16
  17 #include "internals.h"
  18
  19 static int irqfixup __read_mostly;
  20
  21 #define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10)
  22 static void poll_spurious_irqs(unsigned long dummy);
  23 static DEFINE_TIMER(poll_spurious_irq_timer, poll_spurious_irqs, 0, 0);
  24 static int irq_poll_cpu;
  25 static atomic_t irq_poll_active;
  26
  27 /*
  28  * We wait here for a poller to finish.
  29  *
  30  * If the poll runs on this CPU, then we yell loudly and return
  31  * false. That will leave the interrupt line disabled in the worst
  32  * case, but it should never happen.
  33  *
  34  * We wait until the poller is done and then recheck disabled and
  35  * action (about to be disabled). Only if it's still active, we return
  36  * true and let the handler run.
  37  */
  38 bool irq_wait_for_poll(struct irq_desc *desc)
  39 {
  40         if (WARN_ONCE(irq_poll_cpu == smp_processor_id(),
  41                       "irq poll in progress on cpu %d for irq %d\n",
  42                       smp_processor_id(), desc->irq_data.irq))
  43                 return false;
  44
  45 #ifdef CONFIG_SMP
  46         do {
  47                 raw_spin_unlock(&desc->lock);
  48                 while (irqd_irq_inprogress(&desc->irq_data))
  49                         cpu_relax();
  50                 raw_spin_lock(&desc->lock);
  51         } while (irqd_irq_inprogress(&desc->irq_data));
  52         /* Might have been disabled in meantime */
  53         return !irqd_irq_disabled(&desc->irq_data) && desc->action;
  54 #else
  55         return false;
  56 #endif
  57 }
  58
  59
  60 /*
  61  * Recovery handler for misrouted interrupts.
  62  */
  63 static int try_one_irq(int irq, struct irq_desc *desc, bool force)
  64 {
  65         irqreturn_t ret = IRQ_NONE;
  66         struct irqaction *action;
  67
  68         raw_spin_lock(&desc->lock);
  69
  70         /*
  71          * PER_CPU, nested thread interrupts and interrupts explicitely
  72          * marked polled are excluded from polling.
  73          */
  74         if (irq_settings_is_per_cpu(desc) ||
  75             irq_settings_is_nested_thread(desc) ||
  76             irq_settings_is_polled(desc))
  77                 goto out;
  78
  79         /*
  80          * Do not poll disabled interrupts unless the spurious
  81          * disabled poller asks explicitely.
  82          */
  83         if (irqd_irq_disabled(&desc->irq_data) && !force)
  84                 goto out;
  85
  86         /*
  87          * All handlers must agree on IRQF_SHARED, so we test just the
  88          * first.
  89          */
  90         action = desc->action;
  91         if (!action || !(action->flags & IRQF_SHARED) ||
  92             (action->flags & __IRQF_TIMER))
  93                 goto out;
  94
  95         /* Already running on another processor */
  96         if (irqd_irq_inprogress(&desc->irq_data)) {
  97                 /*
  98                  * Already running: If it is shared get the other
  99                  * CPU to go looking for our mystery interrupt too
 100                  */
 101                 desc->istate |= IRQS_PENDING;
 102                 goto out;
 103         }
 104
 105         /* Mark it poll in progress */
 106         desc->istate |= IRQS_POLL_INPROGRESS;
 107         do {
 108                 if (handle_irq_event(desc) == IRQ_HANDLED)
 109                         ret = IRQ_HANDLED;
 110                 /* Make sure that there is still a valid action */
 111                 action = desc->action;
 112         } while ((desc->istate & IRQS_PENDING) && action);
 113         desc->istate &= ~IRQS_POLL_INPROGRESS;
 114 out:
 115         raw_spin_unlock(&desc->lock);
 116         return ret == IRQ_HANDLED;
 117 }
 118
 119 static int misrouted_irq(int irq)
 120 {
 121         struct irq_desc *desc;
 122         int i, ok = 0;
 123
 124         if (atomic_inc_return(&irq_poll_active) != 1)
 125                 goto out;
 126
 127         irq_poll_cpu = smp_processor_id();
 128
 129         for_each_irq_desc(i, desc) {
 130                 if (!i)
 131                          continue;
 132
 133                 if (i == irq)   /* Already tried */
 134                         continue;
 135
 136                 if (try_one_irq(i, desc, false))
 137                         ok = 1;
 138         }
 139 out:
 140         atomic_dec(&irq_poll_active);
 141         /* So the caller can adjust the irq error counts */
 142         return ok;
 143 }
 144
 145 static void poll_spurious_irqs(unsigned long dummy)
 146 {
 147         struct irq_desc *desc;
 148         int i;
 149
 150         if (atomic_inc_return(&irq_poll_active) != 1)
 151                 goto out;
 152         irq_poll_cpu = smp_processor_id();
 153
 154         for_each_irq_desc(i, desc) {
 155                 unsigned int state;
 156
 157                 if (!i)
 158                          continue;
 159
 160                 /* Racy but it doesn't matter */
 161                 state = desc->istate;
 162                 barrier();
 163                 if (!(state & IRQS_SPURIOUS_DISABLED))
 164                         continue;
 165
 166                 local_irq_disable();
 167                 try_one_irq(i, desc, true);
 168                 local_irq_enable();
 169         }
 170 out:
 171         atomic_dec(&irq_poll_active);
 172         mod_timer(&poll_spurious_irq_timer,
 173                   jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 174 }
 175
 176 static inline int bad_action_ret(irqreturn_t action_ret)
 177 {
 178         if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
 179                 return 0;
 180         return 1;
 181 }
 182
 183 /*
 184  * If 99,900 of the previous 100,000 interrupts have not been handled
 185  * then assume that the IRQ is stuck in some manner. Drop a diagnostic
 186  * and try to turn the IRQ off.
 187  *
 188  * (The other 100-of-100,000 interrupts may have been a correctly
 189  *  functioning device sharing an IRQ with the failing one)
 190  */
 191 static void
 192 __report_bad_irq(unsigned int irq, struct irq_desc *desc,
 193                  irqreturn_t action_ret)
 194 {
 195         struct irqaction *action;
 196         unsigned long flags;
 197
 198         if (bad_action_ret(action_ret)) {
 199                 printk(KERN_ERR "irq event %d: bogus return value %x\n",
 200                                 irq, action_ret);
 201         } else {
 202                 printk(KERN_ERR "irq %d: nobody cared (try booting with "
 203                                 "the \"irqpoll\" option)\n", irq);
 204         }
 205         dump_stack();
 206         printk(KERN_ERR "handlers:\n");
 207
 208         /*
 209          * We need to take desc->lock here. note_interrupt() is called
 210          * w/o desc->lock held, but IRQ_PROGRESS set. We might race
 211          * with something else removing an action. It's ok to take
 212          * desc->lock here. See synchronize_irq().
 213          */
 214         raw_spin_lock_irqsave(&desc->lock, flags);
 215         action = desc->action;
 216         while (action) {
 217                 printk(KERN_ERR "[<%p>] %pf", action->handler, action->handler);
 218                 if (action->thread_fn)
 219                         printk(KERN_CONT " threaded [<%p>] %pf",
 220                                         action->thread_fn, action->thread_fn);
 221                 printk(KERN_CONT "\n");
 222                 action = action->next;
 223         }
 224         raw_spin_unlock_irqrestore(&desc->lock, flags);
 225 }
 226
 227 static void
 228 report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
 229 {
 230         static int count = 100;
 231
 232         if (count > 0) {
 233                 count--;
 234                 __report_bad_irq(irq, desc, action_ret);
 235         }
 236 }
 237
 238 static inline int
 239 try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
 240                   irqreturn_t action_ret)
 241 {
 242         struct irqaction *action;
 243
 244         if (!irqfixup)
 245                 return 0;
 246
 247         /* We didn't actually handle the IRQ - see if it was misrouted? */
 248         if (action_ret == IRQ_NONE)
 249                 return 1;
 250
 251         /*
 252          * But for 'irqfixup == 2' we also do it for handled interrupts if
 253          * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
 254          * traditional PC timer interrupt.. Legacy)
 255          */
 256         if (irqfixup < 2)
 257                 return 0;
 258
 259         if (!irq)
 260                 return 1;
 261
 262         /*
 263          * Since we don't get the descriptor lock, "action" can
 264          * change under us.  We don't really care, but we don't
 265          * want to follow a NULL pointer. So tell the compiler to
 266          * just load it once by using a barrier.
 267          */
 268         action = desc->action;
 269         barrier();
 270         return action && (action->flags & IRQF_IRQPOLL);
 271 }
 272
 273 #define SPURIOUS_DEFERRED       0x80000000
 274
 275 void note_interrupt(unsigned int irq, struct irq_desc *desc,
 276                     irqreturn_t action_ret)
 277 {
 278         if (desc->istate & IRQS_POLL_INPROGRESS ||
 279             irq_settings_is_polled(desc))
 280                 return;
 281
 282         if (bad_action_ret(action_ret)) {
 283                 report_bad_irq(irq, desc, action_ret);
 284                 return;
 285         }
 286
 287         /*
 288          * We cannot call note_interrupt from the threaded handler
 289          * because we need to look at the compound of all handlers
 290          * (primary and threaded). Aside of that in the threaded
 291          * shared case we have no serialization against an incoming
 292          * hardware interrupt while we are dealing with a threaded
 293          * result.
 294          *
 295          * So in case a thread is woken, we just note the fact and
 296          * defer the analysis to the next hardware interrupt.
 297          *
 298          * The threaded handlers store whether they sucessfully
 299          * handled an interrupt and we check whether that number
 300          * changed versus the last invocation.
 301          *
 302          * We could handle all interrupts with the delayed by one
 303          * mechanism, but for the non forced threaded case we'd just
 304          * add pointless overhead to the straight hardirq interrupts
 305          * for the sake of a few lines less code.
 306          */
 307         if (action_ret & IRQ_WAKE_THREAD) {
 308                 /*
 309                  * There is a thread woken. Check whether one of the
 310                  * shared primary handlers returned IRQ_HANDLED. If
 311                  * not we defer the spurious detection to the next
 312                  * interrupt.
 313                  */
 314                 if (action_ret == IRQ_WAKE_THREAD) {
 315                         int handled;
 316                         /*
 317                          * We use bit 31 of thread_handled_last to
 318                          * denote the deferred spurious detection
 319                          * active. No locking necessary as
 320                          * thread_handled_last is only accessed here
 321                          * and we have the guarantee that hard
 322                          * interrupts are not reentrant.
 323                          */
 324                         if (!(desc->threads_handled_last & SPURIOUS_DEFERRED)) {
 325                                 desc->threads_handled_last |= SPURIOUS_DEFERRED;
 326                                 return;
 327                         }
 328                         /*
 329                          * Check whether one of the threaded handlers
 330                          * returned IRQ_HANDLED since the last
 331                          * interrupt happened.
 332                          *
 333                          * For simplicity we just set bit 31, as it is
 334                          * set in threads_handled_last as well. So we
 335                          * avoid extra masking. And we really do not
 336                          * care about the high bits of the handled
 337                          * count. We just care about the count being
 338                          * different than the one we saw before.
 339                          */
 340                         handled = atomic_read(&desc->threads_handled);
 341                         handled |= SPURIOUS_DEFERRED;
 342                         if (handled != desc->threads_handled_last) {
 343                                 action_ret = IRQ_HANDLED;
 344                                 /*
 345                                  * Note: We keep the SPURIOUS_DEFERRED
 346                                  * bit set. We are handling the
 347                                  * previous invocation right now.
 348                                  * Keep it for the current one, so the
 349                                  * next hardware interrupt will
 350                                  * account for it.
 351                                  */
 352                                 desc->threads_handled_last = handled;
 353                         } else {
 354                                 /*
 355                                  * None of the threaded handlers felt
 356                                  * responsible for the last interrupt
 357                                  *
 358                                  * We keep the SPURIOUS_DEFERRED bit
 359                                  * set in threads_handled_last as we
 360                                  * need to account for the current
 361                                  * interrupt as well.
 362                                  */
 363                                 action_ret = IRQ_NONE;
 364                         }
 365                 } else {
 366                         /*
 367                          * One of the primary handlers returned
 368                          * IRQ_HANDLED. So we don't care about the
 369                          * threaded handlers on the same line. Clear
 370                          * the deferred detection bit.
 371                          *
 372                          * In theory we could/should check whether the
 373                          * deferred bit is set and take the result of
 374                          * the previous run into account here as
 375                          * well. But it's really not worth the
 376                          * trouble. If every other interrupt is
 377                          * handled we never trigger the spurious
 378                          * detector. And if this is just the one out
 379                          * of 100k unhandled ones which is handled
 380                          * then we merily delay the spurious detection
 381                          * by one hard interrupt. Not a real problem.
 382                          */
 383                         desc->threads_handled_last &= ~SPURIOUS_DEFERRED;
 384                 }
 385         }
 386
 387         if (unlikely(action_ret == IRQ_NONE)) {
 388                 /*
 389                  * If we are seeing only the odd spurious IRQ caused by
 390                  * bus asynchronicity then don't eventually trigger an error,
 391                  * otherwise the counter becomes a doomsday timer for otherwise
 392                  * working systems
 393                  */
 394                 if (time_after(jiffies, desc->last_unhandled + HZ/10))
 395                         desc->irqs_unhandled = 1;
 396                 else
 397                         desc->irqs_unhandled++;
 398                 desc->last_unhandled = jiffies;
 399         }
 400
 401         if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
 402                 int ok = misrouted_irq(irq);
 403                 if (action_ret == IRQ_NONE)
 404                         desc->irqs_unhandled -= ok;
 405         }
 406
 407         desc->irq_count++;
 408         if (likely(desc->irq_count < 100000))
 409                 return;
 410
 411         desc->irq_count = 0;
 412         if (unlikely(desc->irqs_unhandled > 99900)) {
 413                 /*
 414                  * The interrupt is stuck
 415                  */
 416                 __report_bad_irq(irq, desc, action_ret);
 417                 /*
 418                  * Now kill the IRQ
 419                  */
 420                 printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
 421                 desc->istate |= IRQS_SPURIOUS_DISABLED;
 422                 desc->depth++;
 423                 irq_disable(desc);
 424
 425                 mod_timer(&poll_spurious_irq_timer,
 426                           jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 427         }
 428         desc->irqs_unhandled = 0;
 429 }
 430
 431 bool noirqdebug __read_mostly;
 432
 433 int noirqdebug_setup(char *str)
 434 {
 435         noirqdebug = 1;
 436         printk(KERN_INFO "IRQ lockup detection disabled\n");
 437
 438         return 1;
 439 }
 440
 441 __setup("noirqdebug", noirqdebug_setup);
 442 module_param(noirqdebug, bool, 0644);
 443 MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
 444
 445 static int __init irqfixup_setup(char *str)
 446 {
 447         irqfixup = 1;
 448         printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
 449         printk(KERN_WARNING "This may impact system performance.\n");
 450
 451         return 1;
 452 }
 453
 454 __setup("irqfixup", irqfixup_setup);
 455 module_param(irqfixup, int, 0644);
 456
 457 static int __init irqpoll_setup(char *str)
 458 {
 459         irqfixup = 2;
 460         printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
 461                                 "enabled\n");
 462         printk(KERN_WARNING "This may significantly impact system "
 463                                 "performance\n");
 464         return 1;
 465 }
 466
 467 __setup("irqpoll", irqpoll_setup);