On Tue, Sep 09, 2025 at 12:30:35PM -0400, Pasha Tatashin wrote:

On Tue, Sep 9, 2025 at 11:54 AM Jason Gunthorpe [off-list ref] wrote:

quoted

On Tue, Sep 09, 2025 at 11:40:18AM -0400, Pasha Tatashin wrote:

quoted

In reality, this is not something that is high priority for cloud
providers, because these kinds of incompatibilities would be found
during qualification; the kernel will fail to update by detecting a
version mismatch during boot instead of during shutdown.

Given I expect CSPs will have to add-in specific version support for
their own special version-pair needs, I think it would be helpful in
the long run to have a tool that reported what versions a kernel build
wrote and parsed. Test-to-learn the same information sounds a bit too
difficult.

Yes, I agree. My point was only about the near term: it's just not a
priority at the moment. This won't block us in the future, as we can
always add a tooling later to inject the required ELF segments for
pre-live update checks.

Yes, but lets design things to have this kind of logical code model
where there are specific serializations, with specific versions that
are at least discoverably by greping for some struct luo_xxx_ops or
whatever.

Let's avoid open coding versioning stuff where it is hard to find and
hard to later make a manifest out of

Jason