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 (desc->status & IRQ_INPROGRESS)
  49                         cpu_relax();
  50                 raw_spin_lock(&desc->lock);
  51         } while (desc->status & IRQ_INPROGRESS);
  52         /* Might have been disabled in meantime */
  53         return !(desc->status & IRQ_DISABLED) && desc->action;
  54 #else
  55         return false;
  56 #endif
  57 }
  58
  59 /*
  60  * Recovery handler for misrouted interrupts.
  61  */
  62 static int try_one_irq(int irq, struct irq_desc *desc, bool force)
  63 {
  64         struct irqaction *action;
  65         int ok = 0;
  66
  67         raw_spin_lock(&desc->lock);
  68
  69         /* PER_CPU and nested thread interrupts are never polled */
  70         if (desc->status & (IRQ_PER_CPU | IRQ_NESTED_THREAD))
  71                 goto out;
  72
  73         /*
  74          * Do not poll disabled interrupts unless the spurious
  75          * disabled poller asks explicitely.
  76          */
  77         if ((desc->status & IRQ_DISABLED) && !force)
  78                 goto out;
  79
  80         /*
  81          * All handlers must agree on IRQF_SHARED, so we test just the
  82          * first. Check for action->next as well.
  83          */
  84         action = desc->action;
  85         if (!action || !(action->flags & IRQF_SHARED) ||
  86             (action->flags & __IRQF_TIMER) || !action->next)
  87                 goto out;
  88
  89         /* Already running on another processor */
  90         if (desc->status & IRQ_INPROGRESS) {
  91                 /*
  92                  * Already running: If it is shared get the other
  93                  * CPU to go looking for our mystery interrupt too
  94                  */
  95                 desc->status |= IRQ_PENDING;
  96                 goto out;
  97         }
  98
  99         /* Honour the normal IRQ locking and mark it poll in progress */
 100         desc->status |= IRQ_INPROGRESS | IRQ_POLL_INPROGRESS;
 101         do {
 102                 desc->status &= ~IRQ_PENDING;
 103                 raw_spin_unlock(&desc->lock);
 104                 if (handle_IRQ_event(irq, action) != IRQ_NONE)
 105                         ok = 1;
 106                 raw_spin_lock(&desc->lock);
 107                 action = desc->action;
 108         }  while ((desc->status & IRQ_PENDING) && action);
 109
 110         desc->status &= ~(IRQ_INPROGRESS | IRQ_POLL_INPROGRESS);
 111 out:
 112         raw_spin_unlock(&desc->lock);
 113         return ok;
 114 }
 115
 116 static int misrouted_irq(int irq)
 117 {
 118         struct irq_desc *desc;
 119         int i, ok = 0;
 120
 121         if (atomic_inc_return(&irq_poll_active) == 1)
 122                 goto out;
 123
 124         irq_poll_cpu = smp_processor_id();
 125
 126         for_each_irq_desc(i, desc) {
 127                 if (!i)
 128                          continue;
 129
 130                 if (i == irq)   /* Already tried */
 131                         continue;
 132
 133                 if (try_one_irq(i, desc, false))
 134                         ok = 1;
 135         }
 136 out:
 137         atomic_dec(&irq_poll_active);
 138         /* So the caller can adjust the irq error counts */
 139         return ok;
 140 }
 141
 142 static void poll_spurious_irqs(unsigned long dummy)
 143 {
 144         struct irq_desc *desc;
 145         int i;
 146
 147         if (atomic_inc_return(&irq_poll_active) != 1)
 148                 goto out;
 149         irq_poll_cpu = smp_processor_id();
 150
 151         for_each_irq_desc(i, desc) {
 152                 unsigned int status;
 153
 154                 if (!i)
 155                          continue;
 156
 157                 /* Racy but it doesn't matter */
 158                 status = desc->status;
 159                 barrier();
 160                 if (!(status & IRQ_SPURIOUS_DISABLED))
 161                         continue;
 162
 163                 local_irq_disable();
 164                 try_one_irq(i, desc, true);
 165                 local_irq_enable();
 166         }
 167 out:
 168         atomic_dec(&irq_poll_active);
 169         mod_timer(&poll_spurious_irq_timer,
 170                   jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 171 }
 172
 173 /*
 174  * If 99,900 of the previous 100,000 interrupts have not been handled
 175  * then assume that the IRQ is stuck in some manner. Drop a diagnostic
 176  * and try to turn the IRQ off.
 177  *
 178  * (The other 100-of-100,000 interrupts may have been a correctly
 179  *  functioning device sharing an IRQ with the failing one)
 180  */
 181 static void
 182 __report_bad_irq(unsigned int irq, struct irq_desc *desc,
 183                  irqreturn_t action_ret)
 184 {
 185         struct irqaction *action;
 186         unsigned long flags;
 187
 188         if (action_ret != IRQ_HANDLED && action_ret != IRQ_NONE) {
 189                 printk(KERN_ERR "irq event %d: bogus return value %x\n",
 190                                 irq, action_ret);
 191         } else {
 192                 printk(KERN_ERR "irq %d: nobody cared (try booting with "
 193                                 "the \"irqpoll\" option)\n", irq);
 194         }
 195         dump_stack();
 196         printk(KERN_ERR "handlers:\n");
 197
 198         /*
 199          * We need to take desc->lock here. note_interrupt() is called
 200          * w/o desc->lock held, but IRQ_PROGRESS set. We might race
 201          * with something else removing an action. It's ok to take
 202          * desc->lock here. See synchronize_irq().
 203          */
 204         raw_spin_lock_irqsave(&desc->lock, flags);
 205         action = desc->action;
 206         while (action) {
 207                 printk(KERN_ERR "[<%p>]", action->handler);
 208                 print_symbol(" (%s)",
 209                         (unsigned long)action->handler);
 210                 printk("\n");
 211                 action = action->next;
 212         }
 213         raw_spin_unlock_irqrestore(&desc->lock, flags);
 214 }
 215
 216 static void
 217 report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
 218 {
 219         static int count = 100;
 220
 221         if (count > 0) {
 222                 count--;
 223                 __report_bad_irq(irq, desc, action_ret);
 224         }
 225 }
 226
 227 static inline int
 228 try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
 229                   irqreturn_t action_ret)
 230 {
 231         struct irqaction *action;
 232
 233         if (!irqfixup)
 234                 return 0;
 235
 236         /* We didn't actually handle the IRQ - see if it was misrouted? */
 237         if (action_ret == IRQ_NONE)
 238                 return 1;
 239
 240         /*
 241          * But for 'irqfixup == 2' we also do it for handled interrupts if
 242          * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
 243          * traditional PC timer interrupt.. Legacy)
 244          */
 245         if (irqfixup < 2)
 246                 return 0;
 247
 248         if (!irq)
 249                 return 1;
 250
 251         /*
 252          * Since we don't get the descriptor lock, "action" can
 253          * change under us.  We don't really care, but we don't
 254          * want to follow a NULL pointer. So tell the compiler to
 255          * just load it once by using a barrier.
 256          */
 257         action = desc->action;
 258         barrier();
 259         return action && (action->flags & IRQF_IRQPOLL);
 260 }
 261
 262 void note_interrupt(unsigned int irq, struct irq_desc *desc,
 263                     irqreturn_t action_ret)
 264 {
 265         if (desc->status & IRQ_POLL_INPROGRESS)
 266                 return;
 267
 268         if (unlikely(action_ret != IRQ_HANDLED)) {
 269                 /*
 270                  * If we are seeing only the odd spurious IRQ caused by
 271                  * bus asynchronicity then don't eventually trigger an error,
 272                  * otherwise the counter becomes a doomsday timer for otherwise
 273                  * working systems
 274                  */
 275                 if (time_after(jiffies, desc->last_unhandled + HZ/10))
 276                         desc->irqs_unhandled = 1;
 277                 else
 278                         desc->irqs_unhandled++;
 279                 desc->last_unhandled = jiffies;
 280                 if (unlikely(action_ret != IRQ_NONE))
 281                         report_bad_irq(irq, desc, action_ret);
 282         }
 283
 284         if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
 285                 int ok = misrouted_irq(irq);
 286                 if (action_ret == IRQ_NONE)
 287                         desc->irqs_unhandled -= ok;
 288         }
 289
 290         desc->irq_count++;
 291         if (likely(desc->irq_count < 100000))
 292                 return;
 293
 294         desc->irq_count = 0;
 295         if (unlikely(desc->irqs_unhandled > 99900)) {
 296                 /*
 297                  * The interrupt is stuck
 298                  */
 299                 __report_bad_irq(irq, desc, action_ret);
 300                 /*
 301                  * Now kill the IRQ
 302                  */
 303                 printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
 304                 desc->status |= IRQ_DISABLED | IRQ_SPURIOUS_DISABLED;
 305                 desc->depth++;
 306                 irq_disable(desc);
 307
 308                 mod_timer(&poll_spurious_irq_timer,
 309                           jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
 310         }
 311         desc->irqs_unhandled = 0;
 312 }
 313
 314 int noirqdebug __read_mostly;
 315
 316 int noirqdebug_setup(char *str)
 317 {
 318         noirqdebug = 1;
 319         printk(KERN_INFO "IRQ lockup detection disabled\n");
 320
 321         return 1;
 322 }
 323
 324 __setup("noirqdebug", noirqdebug_setup);
 325 module_param(noirqdebug, bool, 0644);
 326 MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
 327
 328 static int __init irqfixup_setup(char *str)
 329 {
 330         irqfixup = 1;
 331         printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
 332         printk(KERN_WARNING "This may impact system performance.\n");
 333
 334         return 1;
 335 }
 336
 337 __setup("irqfixup", irqfixup_setup);
 338 module_param(irqfixup, int, 0644);
 339
 340 static int __init irqpoll_setup(char *str)
 341 {
 342         irqfixup = 2;
 343         printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
 344                                 "enabled\n");
 345         printk(KERN_WARNING "This may significantly impact system "
 346                                 "performance\n");
 347         return 1;
 348 }
 349
 350 __setup("irqpoll", irqpoll_setup);