Re: [kvm-unit-tests PATCH 00/17] x86/vmx: align with Linux kernel VMX definitions

[Sean Christopherson <seanjc@google.com>]

On Fri, Nov 14, 2025, Jon Kohler wrote:
> > On Nov 12, 2025, at 2:02 PM, Sean Christopherson <seanjc@google.com> wrote:
> > On Tue, Sep 16, 2025, Jon Kohler wrote:
> >> This series modernizes VMX definitions to align with the canonical ones
> >> within Linux kernel source. Currently, kvm-unit-tests uses custom VMX
> >> constant definitions that have grown organically and have diverged from
> >> the kernel, increasing the overhead to grok from one code base to
> >> another.
> >> 
> >> This alignment provides several benefits:
> >> - Reduces maintenance overhead by using authoritative definitions
> >> - Eliminates potential bugs from definition mismatches
> >> - Makes the test suite more consistent with kernel code
> >> - Simplifies future updates when new VMX features are added
> >> 
> >> Given the lines touched, I've broken this up into two groups within the
> >> series:
> >> 
> >> Group 1: Import various headers from Linux kernel 6.16 (P01-04)
> > 
> > Hrm.  I'm definitely in favor of aligning names, and not opposed to pulling
> > information from the kernel, but I don't think I like the idea of doing a straight
> > copy+paste.  The arch/x86/include/asm/vmxfeatures.h insanity in particular is pure
> > overhead/noise in KUT.  E.g. the layer of indirection to find out the bit number is
> > _really_ annoying, and the shifting done for VMFUNC is downright gross, but at
> > least in the kernel we get pretty printing in /proc/cpuinfo.
> > 
> > Similarly, I don't want to pull in trapnr.h verbatim, because KVM already provides
> > <nr>_VECTOR in a uapi header, and I strongly prefer the <nr>_VECTOR macros
> > ("trap" is very misleading when considering fault-like vs. trap-like exceptions).
> > 
> > This is also a good opportunity to align the third player: KVM selftests.  Which
> > kinda sorta copy the kernel headers, but with stale and annoying differences.
> > 
> > Lastly, if we're going to pull from the kernel, ideally we would have a script to
> > semi-automate updating the KUT side of things.
> > 
> > So, I think/hope we can kill a bunch of birds at once by creating a script to
> > parse the kernel's vmxfeatures.h, vmx.h, trapnr.h, msr-index.h (to replace lib/x86/msr.h),
> > and generate the pieces we want.  And if we do that for KVM selftests, then we
> > can commit the script to the kernel repo, i.e. we can make it the kernel's
> > responsibility to keep the script up-to-date, e.g. if there's a big rename or
> > something.
> 
> Thanks, Sean - Happy to take a swing at if you don’t already have something
> cooked up to magic that into existence. Any chance any other subsystems do
> something similar? Want to make sure we don’t re-invent the wheel if so.

AFAIK, there's no prior art.  :-/

People do have scripts to manage headers, but they're for simple use cases of
copying kernel headers elsewhere.