[ubuntu] Support non-standard kernel version naming

[jfernandez]

Query apt-cache for linux-headers package info once and attempt to parse the version using the existing major.minor.patch-revision regex. When the regex does not match (e.g. custom/distro kernels with non-standard naming), keep KERNEL_VERSION unchanged instead of failing. This preserves the existing resolution logic for standard Ubuntu kernels while allowing custom kernels to work without modification.

Applied to all Ubuntu/Debian variants and vgpu-manager Ubuntu variants.

Closes: #601

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[jfernandez]

@shivakunv ready for another review.

[shivakunv]

@jfernandez Please share the test logs for both types of kernels, if available.

@rajathagasthya LGTM. PTAL

[jfernandez]

@shivakunv I will verify the latest version with our infra this week. I should get you test logs before the end of the week.

[shivakunv]

@jfernandez : Have you had a chance to test it?

[jfernandez]

@shivakunv Testing surfaced a gap in the initial patch: the regex ([0-9]+).* was accidentally matching our kernel's package version (6.x.y-00004-g00fabdb6286a-1), extracting 6.x.y-00004 and producing a mangled KERNEL_VERSION whose headers package doesn't exist, so the fallback never triggered. I pushed a fix changing .* to \..* to require a dot after the revision, which is always present in standard Ubuntu versions but not in our custom format.

Logic is working as expected after this change, here are the logs for both our custom kernel and standard Ubuntu:

Custom kernel (Ubuntu 22.04, 6.x.ynetflix-00004-g00fabdb6286a):

Resolving Linux kernel version...
Proceeding with Linux kernel version 6.x.ynetflix-00004-g00fabdb6286a

Standard Ubuntu kernel (Ubuntu 22.04 VM, 5.15.0-173-generic):

Resolving Linux kernel version...
Proceeding with Linux kernel version 5.15.0-173-generic

[shivakunv]

/ok-to-test bbf2bf3

[shivakunv]

LGTM .
PTAL @rahulait @rajathagasthya

[shivakunv]

@shivakunv Testing surfaced a gap in the initial patch: the regex ([0-9]+).* was accidentally matching our kernel's package version (6.x.y-00004-g00fabdb6286a-1), extracting 6.x.y-00004 and producing a mangled KERNEL_VERSION whose headers package doesn't exist, so the fallback never triggered. I pushed a fix changing .* to \..* to require a dot after the revision, which is always present in standard Ubuntu versions but not in our custom format.

Logic is working as expected after this change, here are the logs for both our custom kernel and standard Ubuntu:

Custom kernel (Ubuntu 22.04, 6.x.ynetflix-00004-g00fabdb6286a):

Resolving Linux kernel version...
Proceeding with Linux kernel version 6.x.ynetflix-00004-g00fabdb6286a

Standard Ubuntu kernel (Ubuntu 22.04 VM, 5.15.0-173-generic):

Resolving Linux kernel version...
Proceeding with Linux kernel version 5.15.0-173-generic

Thanks @jfernandez .

Could you please provide the log files for both kerbel version ?

[rahulait]

Overall, LGTM. We only support ubuntu22.04 and higher, so changes to versions lower than that will not be used.

[rahulait]

Merging this in. These changes will be available once we release driver versions 535.309.01, 580.159.03 and 595.71.05 and onwards.