Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / kernel / irq / spurious.c
diff --git a/kernel/kernel/irq/spurious.c b/kernel/kernel/irq/spurious.c
new file mode 100644 (file)
index 0000000..903a69c
--- /dev/null
@@ -0,0 +1,475 @@
+/*
+ * linux/kernel/irq/spurious.c
+ *
+ * Copyright (C) 1992, 1998-2004 Linus Torvalds, Ingo Molnar
+ *
+ * This file contains spurious interrupt handling.
+ */
+
+#include <linux/jiffies.h>
+#include <linux/irq.h>
+#include <linux/module.h>
+#include <linux/kallsyms.h>
+#include <linux/interrupt.h>
+#include <linux/moduleparam.h>
+#include <linux/timer.h>
+
+#include "internals.h"
+
+static int irqfixup __read_mostly;
+
+#define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10)
+static void poll_spurious_irqs(unsigned long dummy);
+static DEFINE_TIMER(poll_spurious_irq_timer, poll_spurious_irqs, 0, 0);
+static int irq_poll_cpu;
+static atomic_t irq_poll_active;
+
+/*
+ * We wait here for a poller to finish.
+ *
+ * If the poll runs on this CPU, then we yell loudly and return
+ * false. That will leave the interrupt line disabled in the worst
+ * case, but it should never happen.
+ *
+ * We wait until the poller is done and then recheck disabled and
+ * action (about to be disabled). Only if it's still active, we return
+ * true and let the handler run.
+ */
+bool irq_wait_for_poll(struct irq_desc *desc)
+{
+       if (WARN_ONCE(irq_poll_cpu == smp_processor_id(),
+                     "irq poll in progress on cpu %d for irq %d\n",
+                     smp_processor_id(), desc->irq_data.irq))
+               return false;
+
+#ifdef CONFIG_SMP
+       do {
+               raw_spin_unlock(&desc->lock);
+               while (irqd_irq_inprogress(&desc->irq_data))
+                       cpu_relax();
+               raw_spin_lock(&desc->lock);
+       } while (irqd_irq_inprogress(&desc->irq_data));
+       /* Might have been disabled in meantime */
+       return !irqd_irq_disabled(&desc->irq_data) && desc->action;
+#else
+       return false;
+#endif
+}
+
+
+/*
+ * Recovery handler for misrouted interrupts.
+ */
+static int try_one_irq(int irq, struct irq_desc *desc, bool force)
+{
+       irqreturn_t ret = IRQ_NONE;
+       struct irqaction *action;
+
+       raw_spin_lock(&desc->lock);
+
+       /*
+        * PER_CPU, nested thread interrupts and interrupts explicitely
+        * marked polled are excluded from polling.
+        */
+       if (irq_settings_is_per_cpu(desc) ||
+           irq_settings_is_nested_thread(desc) ||
+           irq_settings_is_polled(desc))
+               goto out;
+
+       /*
+        * Do not poll disabled interrupts unless the spurious
+        * disabled poller asks explicitely.
+        */
+       if (irqd_irq_disabled(&desc->irq_data) && !force)
+               goto out;
+
+       /*
+        * All handlers must agree on IRQF_SHARED, so we test just the
+        * first.
+        */
+       action = desc->action;
+       if (!action || !(action->flags & IRQF_SHARED) ||
+           (action->flags & __IRQF_TIMER))
+               goto out;
+
+       /* Already running on another processor */
+       if (irqd_irq_inprogress(&desc->irq_data)) {
+               /*
+                * Already running: If it is shared get the other
+                * CPU to go looking for our mystery interrupt too
+                */
+               desc->istate |= IRQS_PENDING;
+               goto out;
+       }
+
+       /* Mark it poll in progress */
+       desc->istate |= IRQS_POLL_INPROGRESS;
+       do {
+               if (handle_irq_event(desc) == IRQ_HANDLED)
+                       ret = IRQ_HANDLED;
+               /* Make sure that there is still a valid action */
+               action = desc->action;
+       } while ((desc->istate & IRQS_PENDING) && action);
+       desc->istate &= ~IRQS_POLL_INPROGRESS;
+out:
+       raw_spin_unlock(&desc->lock);
+       return ret == IRQ_HANDLED;
+}
+
+static int misrouted_irq(int irq)
+{
+       struct irq_desc *desc;
+       int i, ok = 0;
+
+       if (atomic_inc_return(&irq_poll_active) != 1)
+               goto out;
+
+       irq_poll_cpu = smp_processor_id();
+
+       for_each_irq_desc(i, desc) {
+               if (!i)
+                        continue;
+
+               if (i == irq)   /* Already tried */
+                       continue;
+
+               if (try_one_irq(i, desc, false))
+                       ok = 1;
+       }
+out:
+       atomic_dec(&irq_poll_active);
+       /* So the caller can adjust the irq error counts */
+       return ok;
+}
+
+static void poll_spurious_irqs(unsigned long dummy)
+{
+       struct irq_desc *desc;
+       int i;
+
+       if (atomic_inc_return(&irq_poll_active) != 1)
+               goto out;
+       irq_poll_cpu = smp_processor_id();
+
+       for_each_irq_desc(i, desc) {
+               unsigned int state;
+
+               if (!i)
+                        continue;
+
+               /* Racy but it doesn't matter */
+               state = desc->istate;
+               barrier();
+               if (!(state & IRQS_SPURIOUS_DISABLED))
+                       continue;
+
+               local_irq_disable();
+               try_one_irq(i, desc, true);
+               local_irq_enable();
+       }
+out:
+       atomic_dec(&irq_poll_active);
+       mod_timer(&poll_spurious_irq_timer,
+                 jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
+}
+
+static inline int bad_action_ret(irqreturn_t action_ret)
+{
+       if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
+               return 0;
+       return 1;
+}
+
+/*
+ * If 99,900 of the previous 100,000 interrupts have not been handled
+ * then assume that the IRQ is stuck in some manner. Drop a diagnostic
+ * and try to turn the IRQ off.
+ *
+ * (The other 100-of-100,000 interrupts may have been a correctly
+ *  functioning device sharing an IRQ with the failing one)
+ */
+static void
+__report_bad_irq(unsigned int irq, struct irq_desc *desc,
+                irqreturn_t action_ret)
+{
+       struct irqaction *action;
+       unsigned long flags;
+
+       if (bad_action_ret(action_ret)) {
+               printk(KERN_ERR "irq event %d: bogus return value %x\n",
+                               irq, action_ret);
+       } else {
+               printk(KERN_ERR "irq %d: nobody cared (try booting with "
+                               "the \"irqpoll\" option)\n", irq);
+       }
+       dump_stack();
+       printk(KERN_ERR "handlers:\n");
+
+       /*
+        * We need to take desc->lock here. note_interrupt() is called
+        * w/o desc->lock held, but IRQ_PROGRESS set. We might race
+        * with something else removing an action. It's ok to take
+        * desc->lock here. See synchronize_irq().
+        */
+       raw_spin_lock_irqsave(&desc->lock, flags);
+       action = desc->action;
+       while (action) {
+               printk(KERN_ERR "[<%p>] %pf", action->handler, action->handler);
+               if (action->thread_fn)
+                       printk(KERN_CONT " threaded [<%p>] %pf",
+                                       action->thread_fn, action->thread_fn);
+               printk(KERN_CONT "\n");
+               action = action->next;
+       }
+       raw_spin_unlock_irqrestore(&desc->lock, flags);
+}
+
+static void
+report_bad_irq(unsigned int irq, struct irq_desc *desc, irqreturn_t action_ret)
+{
+       static int count = 100;
+
+       if (count > 0) {
+               count--;
+               __report_bad_irq(irq, desc, action_ret);
+       }
+}
+
+static inline int
+try_misrouted_irq(unsigned int irq, struct irq_desc *desc,
+                 irqreturn_t action_ret)
+{
+       struct irqaction *action;
+
+       if (!irqfixup)
+               return 0;
+
+       /* We didn't actually handle the IRQ - see if it was misrouted? */
+       if (action_ret == IRQ_NONE)
+               return 1;
+
+       /*
+        * But for 'irqfixup == 2' we also do it for handled interrupts if
+        * they are marked as IRQF_IRQPOLL (or for irq zero, which is the
+        * traditional PC timer interrupt.. Legacy)
+        */
+       if (irqfixup < 2)
+               return 0;
+
+       if (!irq)
+               return 1;
+
+       /*
+        * Since we don't get the descriptor lock, "action" can
+        * change under us.  We don't really care, but we don't
+        * want to follow a NULL pointer. So tell the compiler to
+        * just load it once by using a barrier.
+        */
+       action = desc->action;
+       barrier();
+       return action && (action->flags & IRQF_IRQPOLL);
+}
+
+#define SPURIOUS_DEFERRED      0x80000000
+
+void note_interrupt(unsigned int irq, struct irq_desc *desc,
+                   irqreturn_t action_ret)
+{
+       if (desc->istate & IRQS_POLL_INPROGRESS ||
+           irq_settings_is_polled(desc))
+               return;
+
+       if (bad_action_ret(action_ret)) {
+               report_bad_irq(irq, desc, action_ret);
+               return;
+       }
+
+       /*
+        * We cannot call note_interrupt from the threaded handler
+        * because we need to look at the compound of all handlers
+        * (primary and threaded). Aside of that in the threaded
+        * shared case we have no serialization against an incoming
+        * hardware interrupt while we are dealing with a threaded
+        * result.
+        *
+        * So in case a thread is woken, we just note the fact and
+        * defer the analysis to the next hardware interrupt.
+        *
+        * The threaded handlers store whether they sucessfully
+        * handled an interrupt and we check whether that number
+        * changed versus the last invocation.
+        *
+        * We could handle all interrupts with the delayed by one
+        * mechanism, but for the non forced threaded case we'd just
+        * add pointless overhead to the straight hardirq interrupts
+        * for the sake of a few lines less code.
+        */
+       if (action_ret & IRQ_WAKE_THREAD) {
+               /*
+                * There is a thread woken. Check whether one of the
+                * shared primary handlers returned IRQ_HANDLED. If
+                * not we defer the spurious detection to the next
+                * interrupt.
+                */
+               if (action_ret == IRQ_WAKE_THREAD) {
+                       int handled;
+                       /*
+                        * We use bit 31 of thread_handled_last to
+                        * denote the deferred spurious detection
+                        * active. No locking necessary as
+                        * thread_handled_last is only accessed here
+                        * and we have the guarantee that hard
+                        * interrupts are not reentrant.
+                        */
+                       if (!(desc->threads_handled_last & SPURIOUS_DEFERRED)) {
+                               desc->threads_handled_last |= SPURIOUS_DEFERRED;
+                               return;
+                       }
+                       /*
+                        * Check whether one of the threaded handlers
+                        * returned IRQ_HANDLED since the last
+                        * interrupt happened.
+                        *
+                        * For simplicity we just set bit 31, as it is
+                        * set in threads_handled_last as well. So we
+                        * avoid extra masking. And we really do not
+                        * care about the high bits of the handled
+                        * count. We just care about the count being
+                        * different than the one we saw before.
+                        */
+                       handled = atomic_read(&desc->threads_handled);
+                       handled |= SPURIOUS_DEFERRED;
+                       if (handled != desc->threads_handled_last) {
+                               action_ret = IRQ_HANDLED;
+                               /*
+                                * Note: We keep the SPURIOUS_DEFERRED
+                                * bit set. We are handling the
+                                * previous invocation right now.
+                                * Keep it for the current one, so the
+                                * next hardware interrupt will
+                                * account for it.
+                                */
+                               desc->threads_handled_last = handled;
+                       } else {
+                               /*
+                                * None of the threaded handlers felt
+                                * responsible for the last interrupt
+                                *
+                                * We keep the SPURIOUS_DEFERRED bit
+                                * set in threads_handled_last as we
+                                * need to account for the current
+                                * interrupt as well.
+                                */
+                               action_ret = IRQ_NONE;
+                       }
+               } else {
+                       /*
+                        * One of the primary handlers returned
+                        * IRQ_HANDLED. So we don't care about the
+                        * threaded handlers on the same line. Clear
+                        * the deferred detection bit.
+                        *
+                        * In theory we could/should check whether the
+                        * deferred bit is set and take the result of
+                        * the previous run into account here as
+                        * well. But it's really not worth the
+                        * trouble. If every other interrupt is
+                        * handled we never trigger the spurious
+                        * detector. And if this is just the one out
+                        * of 100k unhandled ones which is handled
+                        * then we merily delay the spurious detection
+                        * by one hard interrupt. Not a real problem.
+                        */
+                       desc->threads_handled_last &= ~SPURIOUS_DEFERRED;
+               }
+       }
+
+       if (unlikely(action_ret == IRQ_NONE)) {
+               /*
+                * If we are seeing only the odd spurious IRQ caused by
+                * bus asynchronicity then don't eventually trigger an error,
+                * otherwise the counter becomes a doomsday timer for otherwise
+                * working systems
+                */
+               if (time_after(jiffies, desc->last_unhandled + HZ/10))
+                       desc->irqs_unhandled = 1;
+               else
+                       desc->irqs_unhandled++;
+               desc->last_unhandled = jiffies;
+       }
+
+       if (unlikely(try_misrouted_irq(irq, desc, action_ret))) {
+               int ok = misrouted_irq(irq);
+               if (action_ret == IRQ_NONE)
+                       desc->irqs_unhandled -= ok;
+       }
+
+       desc->irq_count++;
+       if (likely(desc->irq_count < 100000))
+               return;
+
+       desc->irq_count = 0;
+       if (unlikely(desc->irqs_unhandled > 99900)) {
+               /*
+                * The interrupt is stuck
+                */
+               __report_bad_irq(irq, desc, action_ret);
+               /*
+                * Now kill the IRQ
+                */
+               printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
+               desc->istate |= IRQS_SPURIOUS_DISABLED;
+               desc->depth++;
+               irq_disable(desc);
+
+               mod_timer(&poll_spurious_irq_timer,
+                         jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
+       }
+       desc->irqs_unhandled = 0;
+}
+
+bool noirqdebug __read_mostly;
+
+int noirqdebug_setup(char *str)
+{
+       noirqdebug = 1;
+       printk(KERN_INFO "IRQ lockup detection disabled\n");
+
+       return 1;
+}
+
+__setup("noirqdebug", noirqdebug_setup);
+module_param(noirqdebug, bool, 0644);
+MODULE_PARM_DESC(noirqdebug, "Disable irq lockup detection when true");
+
+static int __init irqfixup_setup(char *str)
+{
+#ifdef CONFIG_PREEMPT_RT_BASE
+       pr_warn("irqfixup boot option not supported w/ CONFIG_PREEMPT_RT_BASE\n");
+       return 1;
+#endif
+       irqfixup = 1;
+       printk(KERN_WARNING "Misrouted IRQ fixup support enabled.\n");
+       printk(KERN_WARNING "This may impact system performance.\n");
+
+       return 1;
+}
+
+__setup("irqfixup", irqfixup_setup);
+module_param(irqfixup, int, 0644);
+
+static int __init irqpoll_setup(char *str)
+{
+#ifdef CONFIG_PREEMPT_RT_BASE
+       pr_warn("irqpoll boot option not supported w/ CONFIG_PREEMPT_RT_BASE\n");
+       return 1;
+#endif
+       irqfixup = 2;
+       printk(KERN_WARNING "Misrouted IRQ fixup and polling support "
+                               "enabled\n");
+       printk(KERN_WARNING "This may significantly impact system "
+                               "performance\n");
+       return 1;
+}
+
+__setup("irqpoll", irqpoll_setup);