intel/intel-technology-enabling-for-openshift

avoid MCO for vfio-pci. Find more elegent way to load vfio-pci by default

Opened this issue · 3 comments

mregmi commented

Issue:
vfio-pci driver is needed by QAT and needs to be loaded before plugin is deployed. For now there is a MCO that loads the driver. We want to move away from the MCO.

Potential solution
We can load the vfio-pci driver using initcontainers.

mythi commented

We want to move away from the MCO.

It's not clear why this is the goal. How do you plan to get rid of setting kernel cmdline args for intel_iommu etc?

uMartinXu commented

We want to move away from the MCO.

It's not clear why this is the goal. How do you plan to get rid of setting kernel cmdline args for intel_iommu etc?

To achieve the final bootless goal.
If we can't make the intel_iommu as the default kernel parameter for RHCOS. We will figure out a way and suggest user use this way to set it as the default parameter on day 1 when the user provisions the cluster.

mythi commented

If we can't make the intel_iommu as the default kernel parameter for RHCOS.

It's probably not going to work out because this may need to be modifiable depending on the features users want.

For DSA/IAA, you are already going to need intel_iommu=on,sm_on but some may also want to turn if off intel_iommu=on,sm_off