[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZkOEC3PbSmutUdsq@google.com>
Date: Tue, 14 May 2024 08:32:27 -0700
From: Sean Christopherson <seanjc@...gle.com>
To: Xiaoyao Li <xiaoyao.li@...el.com>
Cc: Paolo Bonzini <pbonzini@...hat.com>, linux-kernel@...r.kernel.org, kvm@...r.kernel.org,
Kai Huang <kai.huang@...el.com>, Binbin Wu <binbin.wu@...ux.intel.com>
Subject: Re: [PATCH 04/17] KVM: x86: Move synthetic PFERR_* sanity checks to
SVM's #NPF handler
On Tue, May 14, 2024, Xiaoyao Li wrote:
> On 5/14/2024 1:31 AM, Sean Christopherson wrote:
> > On Mon, May 13, 2024, Xiaoyao Li wrote:
> > > On 5/7/2024 11:58 PM, Paolo Bonzini wrote:
> > > > +#define PFERR_SYNTHETIC_MASK (PFERR_IMPLICIT_ACCESS)
> > > > #define PFERR_NESTED_GUEST_PAGE (PFERR_GUEST_PAGE_MASK | \
> > > > PFERR_WRITE_MASK | \
> > > > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > > > index c72a2033ca96..5562d693880a 100644
> > > > --- a/arch/x86/kvm/mmu/mmu.c
> > > > +++ b/arch/x86/kvm/mmu/mmu.c
> > > > @@ -4502,6 +4502,9 @@ int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
> > > > return -EFAULT;
> > > > #endif
> > > > + /* Ensure the above sanity check also covers KVM-defined flags. */
> > >
> > > 1. There is no sanity check above related to KVM-defined flags yet. It has
> > > to be after Patch 6.
> >
> > Ya, it's not just the comment, the entire changelog expects this patch to land
> > after patch 6.
> > >
> > > 2. I somehow cannot parse the comment properly, though I know it's to ensure
> > > KVM-defined PFERR_SYNTHETIC_MASK not contain any bit below 32-bits.
> >
> > Hmm, how about this?
> >
> > /*
> > * Ensure that the above sanity check on hardware error code bits 63:32
> > * also prevents false positives on KVM-defined flags.
> > */
> >
>
> Maybe it's just myself inability, I still cannot interpret it well.
>
> Can't we put it above the sanity check of error code, and just with a
> comment like
>
> /*
> * Ensure KVM-defined flags not occupied any bits below 32-bits,
> * that are used by hardware.
This is somewhat misleading, as hardware does use bits 63:32 (for #NPF), just not
for #PF error codes. And the reason I'm using rather indirect wording is that
KVM _could_ define synthetic flags in bits 31:0, there's simply a higher probability
of needing to reshuffle bit numbers due to a conflict with a future feature.
Is this better? I think it captures what you're looking for, while hopefully also
capturing that staying out of bits 31:0 isn't a hard requirement.
/*
* Restrict KVM-defined flags to bits 63:32 so that it's impossible for
* them to conflict with #PF error codes, which are limited to 32 bits.
*/
Powered by blists - more mailing lists