Summary

Stacked on #321. Validated against an L40S host running vGPU Manager 580.159.01 on Ubuntu 24.04 / kernel 6.8, where the vendor VFIO framework is active and /sys/class/mdev_bus has no parents.

creatable_vgpu_types is a table with a header row and the type ID first:

ID    : vGPU Name
1145  : NVIDIA L40S-1B
1147  : NVIDIA L40S-1Q
...

The parser expected the name first and the ID last, so every line failed strconv.Atoi on the trailing word:

ListGPUProfilesWithVFs: n=0 err=parse vGPU type ID "Name": strconv.Atoi: parsing "Name": invalid syntax

listProfiles propagates that error, getVGPUStatus swallows it, and the host reports vGPU mode with 64 slots and zero profiles — so no vGPU placement can ever succeed.

Parsing on the : separator and skipping the header row fixes it. On the same host, profile enumeration now returns all 30 L40S types with correct framebuffer sizes.

There is a second case worth locking in: once a profile consumes a GPU's entire framebuffer, the remaining VFs on that GPU print the header and nothing else. That is a legitimately empty list, not a malformed file, so it now yields zero profiles rather than an error — otherwise a single 48Q allocation would take the whole GPU's remaining capacity out of the status response.

Tests

• Updated the fixture to the real sysfs format and added a header-only case.
• go test ./lib/devices/... passes.
• Verified live: 30 profiles enumerated, availability drops from 64 to 32 slots after allocating a 48Q on one GPU, and back to 64 after release.