Re: [PATCH 1/3] KVM: VMX: Macrofy GPR swapping in __vmx_vcpu_run()

["Chang S. Bae" <chang.seok.bae@intel.com>]

On 5/14/2026 12:30 PM, Sean Christopherson wrote:
> 
> This is not simpler.  Maybe it ends up being less verbose, but I don't see how
> anyone can claim this is simpler.  E.g. for people like me that aren't already
> familiar with the .ifc directive, the R32_NUM macro is inscrutable.  I can at
> least suss out what e.g. LOAD_REGS and friends are doing, but claiming the code
> is "simpler" is rather ridiculous.


Intended to abstract away the repetitive sequences. But I can read your 
reaction. The compressing macro hurts readability, especially with the 
nested macros and directives.

> For me, this needs to be the focal point of the changelog, and it needs to be
> written with --verbose, because the impact of APX on the VM-Entry/VM-Exit code
> isn't so obvious that a one-line "this makes future life easier" sufficiently
> justifies the macro magic.  And that's coming from someone that generally loves
> macro magic :-)

How about this?

   Convert the repeated register save/restore sequences into macros in
   preparation for  extended GPR support.

   While the resulting macros are more compact, they do not necessarily
   improve readability in fact. They rely nested macros and assembly
   directives, which may not immediately obvious at first glance.

   But upcoming support for additional GPRs (R16-R31) needs to extend the
   VM entry/exit paths. Continuing to grow the existing open-coded
   sequences line-by-line would be increasingly inefficient and tedious.
   Those macros provide a more scalable approach, despite readability
   tradeoff.

Thanks,
Chang