Skip to content

Validate /host-root mount at VfioPciManager startup #1077

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
johnahull wants to merge 1 commit into
base: main
Choose a base branch
from
+8 −2
Open

Validate /host-root mount at VfioPciManager startup #1077

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 8 additions & 2 deletions cmd/gpu-kubelet-plugin/vfio-device.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,12 @@ func NewVfioPciManager(containerDriverRoot string, hostDriverRoot string, nvlib
return nil, fmt.Errorf("IOMMU is not enabled in the kernel")
}

if _, err := os.Stat(hostRoot); os.IsNotExist(err) {

@shivamerla shivamerla Apr 28, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

When do you hit this case? If PassthroughSupport feature gate is enabled, then this host mount should be added through Helm templates. This function itself is only invoked when that feature gate is enabled here:

dra-driver-nvidia-gpu/cmd/gpu-kubelet-plugin/device_state.go

Line 136 in 7f0d03d

if featuregates.Enabled(featuregates.PassthroughSupport) {

@johnahull johnahull Apr 28, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I forked the repo (added some standard attributes to the driver) and i was using a custom-built image, which i deployed via Helm with some custom values. I agree that the stock Helm chart has the mount present. This was just something I hit doing some dev work.

return nil, fmt.Errorf("%s volume mount is missing: the kubelet-plugin container requires a hostPath volume "+
"mounted at %s to check GPU device availability via fuser. Ensure the helm chart has "+
"featureGates.PassthroughSupport=true or add the volume mount manually", hostRoot, hostRoot)
}

vm := &VfioPciManager{
containerDriverRoot: containerDriverRoot,
hostDriverRoot: hostDriverRoot,
Expand Down Expand Up @@ -118,11 +124,11 @@ func (vm *VfioPciManager) WaitForGPUFree(ctx context.Context, info *VfioDeviceIn
if exitErr, ok := cmdErr.(*exec.ExitError); ok && exitErr.ExitCode() == 1 {
return nil
}
err = fmt.Errorf("unexpected error checking if gpu device %q is free: %w", info.PciBusID, cmdErr)
err = fmt.Errorf("error checking if gpu device %q is free (verify %s is mounted and contains fuser): %w", info.PciBusID, hostRoot, cmdErr)
klog.V(6).Infof("[DEBUG] %s", err.Error())
continue
}
err = fmt.Errorf("gpu device %q has open fds by process(es): %q", info.PciBusID, string(out))
err = fmt.Errorf("gpu device %q has open fds by process(es): %q -- external processes (e.g. dcgm-exporter) holding GPU device handles will block VFIO passthrough", info.PciBusID, string(out))
klog.V(6).Infof("[DEBUG] %s", err.Error())
}
}
Expand Down