Created attachment 1274519 [details] Logs Description of problem: VF leakage on start VM with VFIO host device. When starting a VM with host device which is a VF and then shutting it down, the VF is leaked and didn't released on refresh caps. He considered taken by the engine and can't be used by other VMs. Version-Release number of selected component (if applicable): 4.1.2-0.1.el7 vdsm-4.19.11-1.el7ev.x86_64 How reproducible: 100% Steps to Reproduce: 1. Enable 1 VF on a capable sr-iov host 2. Add the VF to the VM as a host device 3. Run VM (VF is taken from host on refresh caps) 4. Shutdown the VM Actual results: VF has leaked Expected results: VF should released on VM shutdown
VF released only on host reboot.
Please clarify what do you mean by "leaked" and "released". If you mean that the hsot device is detached from the host and not reattached on VM shutdown then this is by design
I mean that the host device is not reattached on VM shutdown as it designed with sr-iov 'passthorugh' vNICs. And can't be used by other VMs. If this is the design, Ok(although we not sure it's the best design), why it's not reattached back to the host on host device remove from the VM?(it's pretty bad). It should at least reattached back to the host, if the host device removed from the VM. And it's not get reattached on VM remove as well..(which is pretty bad as well). Should be at least reattached back on host device remove/VM remove.
(In reply to Michael Burman from comment #3) > I mean that the host device is not reattached on VM shutdown as it designed > with sr-iov 'passthorugh' vNICs. And can't be used by other VMs. > > If this is the design, Ok(although we not sure it's the best design), why > it's not reattached back to the host on host device remove from the VM?(it's > pretty bad). It should at least reattached back to the host, if the host > device removed from the VM. > And it's not get reattached on VM remove as well..(which is pretty bad as > well). > Should be at least reattached back on host device remove/VM remove. it is not reattached to the host since it is a risky operation. You are not able to use this device on the host, but you should still be able to use it in an another VM. Is this the case? Are you able to attach the device to an another VM?
Now that we no longer reattach devices to host, we should at least provide a simple tool to let a local admin (that knows his hardware and drivers) to explicitly request that. something like vdsm-tool reattach-detached-devices
Dan, so is the state it is still unavailable to other VMs? How is that a RFE?
The devices are available for VMs, but not for the host itself.
This will not make it in a reasonable time. Please re-open if you still feel this should be fixed
*** Bug 1672252 has been marked as a duplicate of this bug. ***