Hi,
I would like to get some suggestions to implement the interrupt clearing
methods in kdump kernel.
After a panic situation, first kernel goes through each irq_desc
structure and if any interrupt is in progress, it calls eoi handler for
that interrupt. eoi handler is called from the crashing cpu context.
for_each_irq(irq) {
struct irq_desc *desc = irq_desc + irq;
if (desc->status & IRQ_INPROGRESS)
desc->chip->eoi(irq);
if (!(desc->status & IRQ_DISABLED))
desc->chip->disable(irq);
}
But initially some another cpu might have acknowledged the interrupt and
started processing the interrupt. But eoi handler may not be called by
the cpu which acknowledged the interrupt.
Is there any problem with the above approach? Is there any restriction
such that a cpu which acknowledged the interrupt only should send the eoi
signal?
Regards,
Mohan.
Is there any problem with the above approach? Is there any restriction
such that a cpu which acknowledged the interrupt only should send the eoi
signal?
While I don't think there -should- be any restriction, I also seem to
remember that the xics "emulation" done by the firmware on things like
JS20 requires the eoi to be done from the CPU where the interrupt
originated.
Ben.
On Tue, Mar 06, 2007 at 10:04:47PM +0100, Benjamin Herrenschmidt wrote:
quoted
Is there any problem with the above approach? Is there any restriction
such that a cpu which acknowledged the interrupt only should send the eoi
signal?
While I don't think there -should- be any restriction, I also seem to
remember that the xics "emulation" done by the firmware on things like
JS20 requires the eoi to be done from the CPU where the interrupt
originated.
Thanks Ben.
If thats the case, is it okay to call eoi handler from all cpus so that
above case will be satisfied?
Regards,
Mohan.
On Wed, 2007-03-07 at 10:38 +0530, Mohan Kumar M wrote:
On Tue, Mar 06, 2007 at 10:04:47PM +0100, Benjamin Herrenschmidt wrote:
quoted
quoted
Is there any problem with the above approach? Is there any restriction
such that a cpu which acknowledged the interrupt only should send the eoi
signal?
While I don't think there -should- be any restriction, I also seem to
remember that the xics "emulation" done by the firmware on things like
JS20 requires the eoi to be done from the CPU where the interrupt
originated.
Thanks Ben.
If thats the case, is it okay to call eoi handler from all cpus so that
above case will be satisfied?
I suppose so.
Ben.