[PATCH] libxl: enable Xen's e820_host setting

[Jim Fehlig]

On 4/14/20 4:07 PM, Marek Marczykowski-Górecki wrote:
> On Tue, Apr 14, 2020 at 03:56:47PM -0600, Jim Fehlig wrote:
>> On 4/13/20 1:17 PM, Marek Marczykowski-Górecki wrote:
>>> FWIW, in Qubes we have a patches adding e820_host setting here:
>>> https://github.com/QubesOS/qubes-core-libvirt/
>>> (patches 8-11)
>>> Not submitted before, exactly to avoid adding temporary options. But
>>> since 8+ years later it is still there, I think it's safe to assume it
>>> will be there for some more. Or at least it's worth to unbreak some
>>> configurations in the meantime.
>>
>> BTW, the other piece of the puzzle for PCI passthrough starting with xen
>> 4.13 is the xl.cfg 'passthrough' option. See xen commit babde47a3fe. Without
>> that enabled, hotplugging a PCI device to an HVM or PV domain is not
>> possible, with or without e820_host. AFAICT there are no patches in your
>> queue for the 'passthrough' option. Perhaps you haven't rebased to xen 4.13.
> 
> TBH, I don't do PCI hotplugging.

Lucky you :-).

>> Do you have any thoughts on modeling that option?
>>
>> My thoughts are adding another xen hypervisor feature (similar to e820_host)
>> or map it to the new IOMMU device in libvirt [1]. I suppose it depends on
>> the need to support the {sync,share}_pt values vs a simple enable/disable.
>> If enabling and disabling passthrough is enough, I'd lean towards another
>> xen hypervisor feature. Otherwise we'll need to explore the IOMMU device?
> 
> Isn't IOMMU device about allowing the guest to control (v)IOMMU? I've seen
> some discussion about it in context of Xen, so IMO we should not use it
> now for something else.
> What about adding <controller type='pci'/> ?
> I'm not sure about {sync,share}_pt, but probably could be squeezed there
> if needed.

I suppose adding a PCI controller is one way of configuring the guest to later 
hotplug PCI devices, although I'm not sure if (sync,share}_pt could be 
considered attributes of a PCI controller. Also, xen has no notion of a PCI 
controller that I'm aware of, so it would only exist in libvirt. It seems like a 
heavy-weight approach for a setting that essentially enables or disables PCI 
passthrough.

Daniel, do you have any ideas about modeling the xl.cfg(5) 'passthrough' option? 
For convenience, see 'passthrough' under "Other Options"

https://xenbits.xen.org/docs/unstable/man/xl.cfg.5.html#Other-Options

Regards,
Jim