Re: Status of "Drop nested support for CPUs without NRIPS" patch

[Sean Christopherson <seanjc@google.com>]

On Sun, Dec 14, 2025, Alessandro Ratti wrote:
> Hi,
> 
> I was investigating the TODO in svm_check_intercept() about advertising
> NRIPS unconditionally, and found an old patch by Sean Christopherson
> (with Maciej S. Szmigiero's sign-off) that simply requires NRIPS for
> nested virtualization rather than trying to emulate it.
> 
> Link: https://lore.kernel.org/kvm/f0302382cf45d7a9527b4aebbfe694bbcfa7aff5.1651440202.git.maciej.szmigiero@oracle.com/
> 
> Is there a reason this approach wasn't taken? Was there pushback on
> dropping support for pre-2009 CPUs, or did it just fall through the
> cracks?

Both Maxim[1] and Maciej lightly objected[2], and in the end dropping support
for CPUs without NRIPS barely moves the needle in terms of complexity. 

[1] https://lore.kernel.org/kvm/2327614a18d60a5e1b0d9d3aed754cccebce3117.camel@redhat.com
[2] https://lore.kernel.org/kvm/656aaf33-8c70-8b06-2cdc-fd2685a1348b@maciej.szmigiero.name

> If the approach is still acceptable, I'd be happy to refresh and test
> the patch.

Let's just leave things as-is for now.  Dealing with NRIPS=0 CPUs was annoying
when fixing the soft-int reinjection issues, but it hasn't meaningfully impacted
maintenance in the ~3 years since (which isn't suprising give how little code is
saved by ripping it out).

Thank you for the offer though!