[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [RFC PATCH v1 10/10] xen/arm: GICv3 device tree parsing

On 03/25/2014 11:04 AM, Stefano Stabellini wrote:
> On Mon, 24 Mar 2014, Julien Grall wrote:
>> Hi Stefano,
>> On 24/03/14 17:34, Stefano Stabellini wrote:
>>> I think that for Dom0 we have to use vgic_v3, because it doesn't only
>>> give you support for more vcpus but also MSI and MSI-X delivery.
>> Right.
>>> For DomUs it might be important to support vgic_v2 on gicv3 hardware,
>>> however I wouldn't want to defer the decision to the user (i.e.
>>> introduce yet another VM config option), if not for debugging.
>>> Would it be possible to advirtise both gicv2 and gicv3 on device tree?
>>> What would the guest kernel do in that case?
>> Linux will try to load both GICv2 and GICv3 drivers. In any case I don't 
>> think
>> it's a solution because:
>>      - you don't know how the kernel will react
>>         - how will you choose which backend to use?
>>> Otherwise we could default to vgic_v2 if vcpus <= 8 and vgic_v3 if
>>> vcpus > 8.
>> What about kernel which only support GICv3 and have less than 8 VCPUS?
>> What about of device (MSI, ...) passthrough with this kind of kernel?
>> I think we need to have a VM config option in this specific case.
> Having a VM config option is OK, but it should be the last resort for
> people doing something very uncommon.
> GICv3 support in Linux and distros is probably going to be widespread
> soon, so we might not have to worry about it though.

I don't think GICv3 will supported in 3.15. The current series doesn't
support GICv3 on 32 bits

But, I'm more worry about other OS than Linux (e.g FreeRTOS, *BSD,...)
which may not support GICv2 before a while.


Julien Grall

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.