[PATCH 0 of 7] pci passthrough with stubdoms

[PATCH 0 of 7] pci passthrough with stubdoms

[Stefano Stabellini]

Hi all,
this patch series enables pci passthrough with stubdoms.
Few things still don't work: MSI-X, iomul, and more importantly pci
coldplug.
This means that you still cannot write pci=['bdf'] in your VM config
file, but you can use xm pci-attach to hotplug a device into a VM
without any trouble.
This is due to a circular dependency introduced by "xend: hot-plug PCI
devices at boot-time": during initialization xend waits for the device
model to complete pci hotplug while the device model (the stubdom) waits
for xend to finish its own initialization.

The patch series consists of the following patches:

[PATCH 1 of 7] pcifront: translate physical into virtual addresses
[PATCH 2 of 7] pcilib: define endianness for minios
[PATCH 3 of 7] xen: allows more hypercalls from stubdoms
[PATCH 4 of 7] minios: fix minios console
[PATCH 5 of 7] xend: allow a device to be assigned to a guest and its stubdom
[PATCH 6 of 7] xend: call pci_device_configure on the stubdom
[PATCH 7 of 7] qemu: passthrough support in stubdoms

the first 6 patches are for the xen repository while the last one is for
qemu.

Cheers,

Stefano

Re: [PATCH 0 of 7] pci passthrough with stubdoms

[Simon Horman]

On Mon, Oct 12, 2009 at 06:19:29PM +0100, Stefano Stabellini wrote:
> Hi all,
> this patch series enables pci passthrough with stubdoms.
> Few things still don't work: MSI-X, iomul, and more importantly pci
> coldplug.
> This means that you still cannot write pci=['bdf'] in your VM config
> file, but you can use xm pci-attach to hotplug a device into a VM
> without any trouble.
> This is due to a circular dependency introduced by "xend: hot-plug PCI
> devices at boot-time": during initialization xend waits for the device
> model to complete pci hotplug while the device model (the stubdom) waits
> for xend to finish its own initialization.

Is a possible solution for xend not to wait on qemu in this case?

Re: [PATCH 0 of 7] pci passthrough with stubdoms

[Stefano Stabellini]

On Mon, 12 Oct 2009, Simon Horman wrote:
> On Mon, Oct 12, 2009 at 06:19:29PM +0100, Stefano Stabellini wrote:
> > Hi all,
> > this patch series enables pci passthrough with stubdoms.
> > Few things still don't work: MSI-X, iomul, and more importantly pci
> > coldplug.
> > This means that you still cannot write pci=['bdf'] in your VM config
> > file, but you can use xm pci-attach to hotplug a device into a VM
> > without any trouble.
> > This is due to a circular dependency introduced by "xend: hot-plug PCI
> > devices at boot-time": during initialization xend waits for the device
> > model to complete pci hotplug while the device model (the stubdom) waits
> > for xend to finish its own initialization.
> 
> Is a possible solution for xend not to wait on qemu in this case?
>

Yes, I think that would solve the problem.