Port DragonFly to Xen platform

Matthew Dillon dillon at apollo.backplane.com
Wed Mar 24 08:41:03 PDT 2010


:I don't see how this has anything to do with a Xen port. Our interrupt
:routing is not broken but rather outdated. We still rely on the mptable
:instead of doing ACPI interrupt routing (and we also lack MSI support), but
:that's a different issue, unless Xen *requires* MSI, which I doubt.

    I think we have MSI support, but its a bad hack.  Actually I'm not
    entirely sure on the MSI[X] front but at least one or two drivers
    uses those APIs.  Whatever it is it isn't a full port.

    Xen doesn't need MSI.

:: Also, I am looking at doing the NUMA project (investigation stage) so
:: that could work with a Xen port nicely.
:: I, myself look forward to a Xen port !
:
:While I see the point of a Xen port, I'm starting to think that something
:along the lines of KVM or hardware virtualization support for vkernels might
:be more interesting and relevant.
:
:Cheers,
:Alex Hornung

    My personal preference would be KVM as well, minus the IOMMU (which I
    think is a huge destabilizing mistake on AMD/Intels part now that
    I've read up on it, and it can't be multi-tasked anyway).

    Xen is interesting but KVM has caught up very quickly and will probably
    completely surpass Xen in the future.  Xen will wind up just being
    another KVM consumer soon IMHO.

    As for the interface, we'd have to implement the linux KVM API for sure
    so things like qemu can run natively.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Kernel mailing list