Chained interrupt handlers dont have an irqaction and hence are not
handled during migrating interrupts when some cores go offline.
Handle this by introducing a irq_is_chained() check which is based
on the IRQ_NOREQUEST flag being set for such interrupts. fixup_irq()
can then handle such interrupts and not skip them over.
Signed-off-by: Sundar Iyer <[email protected]>
---
v1: use accessors instead of new variables to identify chained irqs
arch/x86/kernel/irq.c | 6 ++++--
include/linux/irqdesc.h | 10 ++++++++++
2 files changed, 14 insertions(+), 2 deletions(-)
diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c
index 6c0802e..acd7949 100644
--- a/arch/x86/kernel/irq.c
+++ b/arch/x86/kernel/irq.c
@@ -249,8 +249,10 @@ void fixup_irqs(void)
data = irq_desc_get_irq_data(desc);
affinity = data->affinity;
- if (!irq_has_action(irq) || irqd_is_per_cpu(data) ||
- cpumask_subset(affinity, cpu_online_mask)) {
+ /* include IRQs who have no action, but are chained */
+ if ((!irq_has_action(irq) && irq_is_chained(irq)) ||
+ irqd_is_per_cpu(data) ||
+ cpumask_subset(affinity, cpu_online_mask)) {
raw_spin_unlock(&desc->lock);
continue;
}
diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index 2d921b3..7880722 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -120,6 +120,16 @@ static inline int irq_has_action(unsigned int irq)
return desc->action != NULL;
}
+/*
+ * Test to see if the IRQ is chained; it would not be requested and hence
+ * _IRQ_NOREQUEST would be set
+ */
+static inline int irq_is_chained(unsigned int irq)
+{
+ struct irq_desc *desc = irq_to_desc(irq);
+ return !(desc->status_use_accessors & IRQ_NOREQUEST);
+}
+
/* caller has locked the irq_desc and both params are valid */
static inline void __irq_set_handler_locked(unsigned int irq,
irq_flow_handler_t handler)
--
1.7.1
On Tue, May 29, 2012 at 07:10:08PM +0530, Sundar Iyer wrote:
> diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
> index 2d921b3..7880722 100644
> --- a/include/linux/irqdesc.h
> +++ b/include/linux/irqdesc.h
> @@ -120,6 +120,16 @@ static inline int irq_has_action(unsigned int irq)
> return desc->action != NULL;
> }
>
> +/*
> + * Test to see if the IRQ is chained; it would not be requested and hence
> + * _IRQ_NOREQUEST would be set
> + */
> +static inline int irq_is_chained(unsigned int irq)
> +{
> + struct irq_desc *desc = irq_to_desc(irq);
> + return !(desc->status_use_accessors & IRQ_NOREQUEST);
> +}
> +
This approach looks highly suspect. There are many non-chained cases that
are also NOREQUEST. Chained IRQs are at the very least NOREQUEST,
NOPROBE, and NOTHREAD. You could test a mask of those and at least have a
vague change of not catching other unrelated IRQs. It would still be a
hack though, given that none of those status flags in and of themselves
require an IRQ to be chained.
On Tue, 29 May 2012, Sundar Iyer wrote:
> Chained interrupt handlers dont have an irqaction and hence are not
> handled during migrating interrupts when some cores go offline.
>
> Handle this by introducing a irq_is_chained() check which is based
> on the IRQ_NOREQUEST flag being set for such interrupts. fixup_irq()
> can then handle such interrupts and not skip them over.
No, we need a separate flag for those. IRQ_NORREQUEST is not a
reliable indicator.
Thanks,
tglx
>-----Original Message-----
>From: Thomas Gleixner [mailto:[email protected]]
>Sent: Thursday, May 31, 2012 1:01 PM
>>
>> Handle this by introducing a irq_is_chained() check which is based on
>> the IRQ_NOREQUEST flag being set for such interrupts. fixup_irq() can
>> then handle such interrupts and not skip them over.
>
>No, we need a separate flag for those. IRQ_NORREQUEST is not a reliable indicator.
I posted a v2 for this patch with all the flags; else will it be okay me adding
a new flag called IRQ_CHAINED and a new function similar to set_noprobe* helpers?
Cheers!