VMs fail to boot on RHV 4.4

Description

On RHV 4.4, VMs fail to boot with the error Host doesn't support passthru of host PCI device. This issue affects GPU pass through deployments with all supported GPUs and NVIDIA vGPU deployments with GPUs based on the NVIDIA Ampere architecture. This issue occurs because the intel_iommu parameter and the nouveau.modeset parameter are not set correctly.

Version

This issue affects RHV 4.4.

Workaround

Perform this workaround on the hypervisor host. This workaround requires root user privileges on the hypervisor host.

1. In a plain-text editor, edit the file /boot/loader/entries/rhvh-4.4.1.1-0.20200722.0+1-4.18.0-193.13.2.el8_2.x86_64.conf to add the following options to the boot options.

   • nouveau.modeset=0
   • intel_iommu=on
   Note:

   Line breaks have been added to this example to enhance readability.

   Copy

   Copied!

               
   
               
   title rhvh-4.4.1.1-0.20200722.0 (4.18.0-193.13.2.el8_2.x86_64)
   version 4.18.0-193.13.2.el8_2.x86_64
   linux //rhvh-4.4.1.1-0.20200722.0+1/vmlinuz-4.18.0-193.13.2.el8_2.x86_64
   initrd //rhvh-4.4.1.1-0.20200722.0+1/initramfs-4.18.0-193.13.2.el8_2.x86_64.img
   options crashkernel=auto resume=/dev/mapper/rhvh00-swap \
   rd.lvm.lv=rhvh00/rhvh-4.4.1.1-0.20200722.0+1 rd.lvm.lv=rhvh00/swap \
   root=/dev/rhvh00/rhvh-4.4.1.1-0.20200722.0+1 \
   boot=UUID=38ff2175-b761-403d-8a91-d7ec9f7ec2f7 rootflags=discard \
   img.bootid=rhvh-4.4.1.1-0.20200722.0+1 intel_iommu=on nouveau.modeset=0
   id rhel-20200825140238-4.18.0-193.13.2.el8_2.x86_64
   grub_users $grub_users
   grub_arg --unrestricted
   grub_class kernel
           

2. Reboot the hypervisor host machine.

Status

Not an NVIDIA bug

Ref. #

200653675