Re: [PATCH 08/16] KVM: x86/mmu: WARN and skip MMIO cache on private, reserved page faults

[Sean Christopherson]

On Wed, Mar 06, 2024, Kai Huang wrote:
> On 5/03/2024 4:51 am, Sean Christopherson wrote:
> > In other words, KVM_EXIT_MEMORY_FAULT essentially communicates to userspace that
> > (a) userspace can likely fix whatever badness triggered the -EFAULT, and (b) that
> > KVM is in a state where fixing the underlying problem and resuming the guest is
> > safe, e.g. won't corrupt the guest (because KVM is in a half-baked state).
> > 
> 
> Sure.  One small issue might be that, in a later code check, we actually
> return KVM_EXIT_MEMORY_FAULT when private fault hits RET_PF_EMULATION -- see
> your patch:
> 
> [PATCH 01/16] KVM: x86/mmu: Exit to userspace with -EFAULT if private fault
> hits emulation
> 
> So here if we just return -EFAULT w/o reporting KVM_EXIT_MEMORY_FAULT when
> private+reserved is hit, it seems there's a little bit inconsistency here.

It's intentionally inconsistent.  -EFAULT without KVM_EXIT_MEMORY_FAULT is
essentially KVM saying "something bad happened, and it can't be fixed", whereas
exiting with KVM_EXIT_MEMORY_FAULT says "there's an issue, but you may be able
to resolve it".

The ABI is a bit messy, e.g. in some ways it would be cleaner if KVM returned '0'.
But doing that in a backwards compatible way would have required a rather ugly
opt-in, and it would also make it more tedious to extend KVM_EXIT_MEMORY_FAULT,
e.g. pairing it with -EHWPOISON didn't require any new flags.