On 7/18/19 12:29 PM, Laine Stump wrote:

On 7/18/19 10:29 AM, Daniel Henrique Barboza wrote:

Hi,

I have a PoC that enables partial coldplug assignment of multifunction
PCI devices with managed mode. At this moment, Libvirt can't handle
this scenario - the code will detach only the hostdevs from the XML,
when in fact the whole IOMMU needs to be detached. This can be
verified by the fact that Libvirt handles the unmanaged scenario
well, as long as the user detaches the whole IOMMU beforehand.

I have played with 2 approaches. The one I am planning to contribute
back is a change inside virHostdevGetPCIHostDeviceList(), that
adds the extra PCI devices for detach/re-attach in case a PCI
Multifunction device in managed mode is presented in the XML.

If you're thinking of doing that automatically, then I should warn you that we had discussed that a long time ago, and decided that it was a bad idea to do it because it was likely someone would, e.g. try to assign an audio device to their guest that happened to be one function on a multifunction device that also contained a disk controller (or some other device) that the host needed for proper operation.

Let's say that I have a Multi PCI card with 4 functions, and I want a guest to use
only the function 0 of that card. At this moment, I'm only able to do that if I
manually execute nodedev-detach on all 4 functions beforehand and use function
0 as a hostdev with managed=false.

What I've implemented is a way of doing the detach/re-attach of the whole IOMMU
for the user, if the hostdev is set with managed=true (and perhaps I should also
consider verifying the 'multifunction=yes' attribute as well, for more clarity).
I am not trying to assign all the IOMMU devices to the guest - not sure if that's
what you were talking about up there, but I'm happy to emphasize that's not
the case.

Now, yes, if the user is unaware of the consequences of detaching all devices
of the IOMMU from the host, bad things can happen. If that's what you're saying,
fair enough. I can make an argument about how we can't shield the user from
his/her own 'unawareness' forever, but in the end it's better to be on the safe
side.

It may be that in *your* particular case, you understand that the functions you don't want to assign to the guest are not otherwise used, and it's not dangerous to suddenly detach them from their host driver. But you can't assume that will always be the case.

If you *really* can't accept just assigning all the devices in that IOMMU group to the guest (thus making them all explicitly listed in the config, and obvious to the administrator that they won't be available on the host) and simply not using them, then you either need to separately detach those particular functions from the host, or come up with a way of having the domain config explicitly list them as "detached from the host but not actually attached to the guest".

I can live with that - it will automate the detach/re-attach process, which is
my goal here, and it force the user to know exactly what is going to be detached
from the host, minimizing errors. If no one is against adding an extra
parameter 'unassigned=true' to the hostdev in these cases, I can make this
happen.

Thanks,

DHB

Now, there's a catch. Inside both virHostdevPreparePCIDevices()
and virHostdevReAttachPCIDevices() there are code to save/restore
the network configuration for SR-IOV devices. These functions iterates
in the hostdevs list, instead of the pcidevs list I'm changing. The final
result, given that the current conditions used for SR-IOV matches the
conditions for multifunction PCI devices as well, is that not all virtual
functions will get their network configuration saved/restored.

If you're not going to use a device (which is implied by the fact that it's not in the hostdevs list) then nothing about its network config will change, so there is no reason to save/restore it.

For example, a guest that uses 3 of 4 functions of a PCI MultiFunction
card, let's say functions 0,1 and 3. The code will handle the detach
of all the IOMMU, including the function 2 that isn't declared in the
XML.

Again, the above sentence implies that you're wanting to make this completely automatic, which we previously decided was something we didn't want to do.

However, since function 2 isn't a hostdev, its network config
will not be restored after the VM shutdown.

You're talking about something that will never occur - on every SRIOV network card I've ever seen each VF is in its own IOMMU group, and can be assigned to a guest independent of what's done with any other VF. I've never seen a case (except maybe once with a newly released motherboard that had broken IOMMU firmware(?)) where a VF was in the same IOMMU group as any other device.

Now comes the question: how much effort should be spent into making
the network config of all the functions be restored? Is this a blocker
for the whole code to be accepted or, given it is proper documented
somewhere, it can be done later on?