Re: [PATCH 1/3] KVM: x86: Add CR3 to guest debug info

[Sean Christopherson <seanjc@google.com>]

On Fri, Nov 21, 2025, Yosry Ahmed wrote:
> On Fri, Nov 21, 2025 at 01:01:40PM -0800, Sean Christopherson wrote:
> > On Fri, Nov 21, 2025, Yosry Ahmed wrote:
> > KVM already provides kvm_run.kvm_valid_regs to let userspace grab register state
> > on exit to userspace.  If userspace is debugging, why not simply save all regs on
> > exit?
> > 
> > If the answer is "because it slows down all other exits", then I would much rather
> > give userspace the ability to conditionally save registers based on the exit reason,
> > e.g. something like this (completely untested, no CAP, etc.)
> 
> I like this approach conceptually, but I think it's an overkill for this
> use case tbh. Especially the memory usage, that's 1K per vCPU for the
> bitmap. I know it can be smaller, but probably not small either because
> it will be a problem if we run out of bits.
> > diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> > index 52f6000ab020..452805c1337b 100644
> > --- a/include/uapi/linux/kvm.h
> > +++ b/include/uapi/linux/kvm.h
> > @@ -494,8 +494,12 @@ struct kvm_run {
> >                 struct kvm_sync_regs regs;
> >                 char padding[SYNC_REGS_SIZE_BYTES];
> >         } s;
> > +
> > +       __u64 kvm_save_regs_on_exit[16];

Heh, check your math.  It's 1024 bits, 128 bytes.  Reserving space for 1024 exits
is likely extreme overkill given that KVM is sitting at 40 exits after ~18 years,
so as you say we could cut that down significantly depending on how willing we are
to risk having to add kvm_save_regs_on_exit2 in the future.