[vfio-users] Host can't wake from suspend after running a passthrough VM

Ryan A Young rayoung at utexas.edu
Sun Oct 23 21:46:21 UTC 2016

Well, I am not ready to declare my issue "solved" just yet, but so
far, sleep's been working as expected with vanilla 4.8.3.

I have a similar setup - I often put the system to sleep with the VFIO
VM still running by mistake, so I wrote a shell script that runs on
the sleep trigger and blocks until the VM is powered down. Hasn't been
working lately, probably some systemd change that I missed...

On Sun, Oct 23, 2016 at 4:08 PM, Nicolas Roy-Renaud
<nicolas.roy-renaud.1 at ens.etsmtl.ca> wrote:
> I've had this problem for a while, and I just figured it was due to VFIO GPU
> passthroughs being generally incompatible with S3. I currently just set a
> sleep inhibitor while my VM is running so I don't accidentally put the host
> into sleep while it's still running. I've been trying to add it to my
> libvirt hooks, but libvirt seems to detect when a process is forking and
> hold on to the subprocess regardless, so no luck there.
> - Nicolas
> On 10/21/16 20:09, Ryan A Young wrote:
>> I will give 4.8 a try, thanks for the heads up!
>> On Fri, Oct 21, 2016 at 6:58 PM, Alex Williamson
>> <alex.williamson at redhat.com> wrote:
>>> On Thu, 20 Oct 2016 20:40:24 -0500
>>> Ryan A Young <rayoung at utexas.edu> wrote:
>>>> After I start and then shutdown my passthrough VM and then suspend the
>>>> host, it won't wake from suspend anymore. As in, once in S3, pressing
>>>> the power button simply has no effect. I have to press the reset
>>>> button (and sometimes that doesn't work either) and restart the
>>>> machine. Suspend works correctly if I never start the VM.
>>>> Has anyone seen this issue before?
>>>> Hardware: i5-4690, ASRock H97, GTX 1060.
>>>> Software: Fedora 24, kernel 4.7.7, qemu 2.6.2, GTX 1060 video and
>>>> audio taken by vfio-pci, OVMF VM with Windows 10 managed by libvirt.
>>> Try a v4.8 kernel, I was doing some testing on a Fedora v4.7 system and
>>> seeing a number of strange things after running a VM with device
>>> assignment.  Disabling KVM resolved the problem, but is clearly not a
>>> useful workaround.  I wasn't able to reproduce on v4.8, which makes me
>>> think KVM in v4.7 might have some issues.

- Ryan

More information about the vfio-users mailing list