[PATCH] IRQ: prevent enabling of previously disabled interrupt
authorBryan Holty <lgeek@frontiernet.net>
Sat, 25 Mar 2006 11:07:37 +0000 (03:07 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Sat, 25 Mar 2006 16:22:55 +0000 (08:22 -0800)
This fix prevents re-disabling and enabling of a previously disabled
interrupt.  On an SMP system with irq balancing enabled; If an interrupt is
disabled from within its own interrupt context with disable_irq_nosync and is
also earmarked for processor migration, the interrupt is blindly moved to the
other processor and enabled without regard for its current "enabled" state.
If there is an interrupt pending, it will unexpectedly invoke the irq handler
on the new irq owning processor (even though the irq was previously disabled)

The more intuitive fix would be to invoke disable_irq_nosync and
enable_irq, but since we already have the desc->lock from __do_IRQ, we
cannot call them directly.  Instead we can use the same logic to disable
and enable found in disable_irq_nosync and enable_irq, with regards to the
desc->depth.

This now prevents a disabled interrupt from being re-disabled, and more
importantly prevents a disabled interrupt from being incorrectly enabled on
a different processor.

Signed-off-by: Bryan Holty <lgeek@frontiernet.net>
Cc: Andi Kleen <ak@muc.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
kernel/irq/migration.c

index 6bdd03c..52a8655 100644 (file)
@@ -18,9 +18,17 @@ void move_native_irq(int irq)
        cpumask_t tmp;
        irq_desc_t *desc = irq_descp(irq);
 
-       if (likely (!desc->move_irq))
+       if (likely(!desc->move_irq))
                return;
 
+       /*
+        * Paranoia: cpu-local interrupts shouldn't be calling in here anyway.
+        */
+       if (CHECK_IRQ_PER_CPU(desc->status)) {
+               WARN_ON(1);
+               return;
+       }
+
        desc->move_irq = 0;
 
        if (likely(cpus_empty(pending_irq_cpumask[irq])))
@@ -29,7 +37,8 @@ void move_native_irq(int irq)
        if (!desc->handler->set_affinity)
                return;
 
-       /* note - we hold the desc->lock */
+       assert_spin_locked(&desc->lock);
+
        cpus_and(tmp, pending_irq_cpumask[irq], cpu_online_map);
 
        /*
@@ -42,9 +51,13 @@ void move_native_irq(int irq)
         * Being paranoid i guess!
         */
        if (unlikely(!cpus_empty(tmp))) {
-               desc->handler->disable(irq);
+               if (likely(!(desc->status & IRQ_DISABLED)))
+                       desc->handler->disable(irq);
+
                desc->handler->set_affinity(irq,tmp);
-               desc->handler->enable(irq);
+
+               if (likely(!(desc->status & IRQ_DISABLED)))
+                       desc->handler->enable(irq);
        }
        cpus_clear(pending_irq_cpumask[irq]);
 }