I have this same issue, it’s actually only happening on a testing machine where the number of total available vcpus is less than 16. I also know kernel-ml isn’t supported but was wanting to do some
testing with nested virt live migration which apparently is available with the kernel 5.x line.
Does anyone know how I can manually adjust the maxcpus=16 parameter that is passed to qemu-kvm to start VMs on this one machine?
Cheers,
Chris.
From: Jillian Morgan <jillian.morgan@primordial.ca>
Sent: Wednesday, March 23, 2022 5:09 PM
To: users@ovirt.org
Subject: [ovirt-users] 4.4.9 -> 4.4.10 Cannot start or migrate any VM (hotpluggable cpus requested exceeds the maximum cpus supported by KVM)
After upgrading the engine from 4.4.9 to 4.4.10, and then upgrading one host, any attempt to migrate a VM to that host or start a VM on that host results in the following error:
Number of hotpluggable cpus requested (16) exceeds the maximum cpus supported by KVM (8)
While the version of qemu is the same across hosts, (qemu-kvm-6.0.0-33.el8s.x86_64), I traced the difference to the upgraded kernel on the new host. I have always
run elrepo's kernel-ml on these hosts to support bcache which RHEL's kernel doesn't support. The working hosts still run kernel-ml-5.15.12. The upgraded host ran kernel-ml-5.17.0.
In case anyone else runs kernel-ml, have you run into this issue?
Does anyone know why KVM's KVM_CAP_MAX_VCPUS value is lowered on the new kernel?
Does anyone know how to query the KVM capabilities from userspace without writing a program leveraging kvm_ioctl()'s?
Related to this, it seems that ovirt and/or libvirtd always runs qmu-kvm with an -smp argument of "maxcpus=16". This causes qemu's built-in check to fail on the new kernel which is supporting max_vpus of 8.
Why does ovirt always request maxcpus=16?
And yes, before you say it, I know you're going to say that running kernel-ml isn't supported.
--
Jillian Morgan (she/her) 🏳️⚧️
Systems & Networking Specialist
Primordial Software Group & I.T. Consultancy
https://www.primordial.ca