io-apic interrupt migration on bhyve
Ashutosh Kumar
mrashutosh at hotmail.com
Fri Oct 17 15:25:46 UTC 2014
Hi,
On more analysis I found that there is a race condition in Guest OS such that before complete initialization happens for APs the interrupt migration takes place and hence incorrect cpu is written to io-apic's rte for that particular pin. The debug prints either in Guest OS or vmm just kills this race condition.
Thanks.
RegardsAshutosh
> From: mrashutosh at hotmail.com
> To: freebsd-virtualization at freebsd.org
> Subject: io-apic interrupt migration on bhyve
> Date: Fri, 17 Oct 2014 13:24:34 +0000
>
>
>
> Hi,
>
>
>
> I am witnessing some strange behavior w.r.t io-apic
> interrupt migration on bhyve.
>
> I have below assignment for a level triggered
> interrupt:
>
>
>
> io-apic pin - p
>
> cpu vector - v
>
> cpu id - 0
>
>
>
> when this interrupt is migrated to cpu 1 then
> the assignment is as below
>
>
>
> io-apic pin - p
>
> cpu vector - v1
>
> cpu id - 1
>
>
>
> When VM boots up then OS is getting interrupts for
>
>
>
> io-apic pin - p
>
> cpu vector - v1
>
> cpu id - 0
>
>
>
> Added debug prints in vioapic_write function
> to check the values written to io-apic rte for this pin and then VM gets
> correct interrupts i.e (vector v, cpu 0 ) before the interrupts are migrated and
> (vector v1, cpu 1) after the migration When prints are removed then the
> problem happens and we see interrupts for (vector v1 and cpu 0).
> RegardsAshutosh
> _______________________________________________
> freebsd-virtualization at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-virtualization
> To unsubscribe, send any mail to "freebsd-virtualization-unsubscribe at freebsd.org"
More information about the freebsd-virtualization
mailing list