Re: [PATCH 22/28] KVM: x86/mmu: introduce cpu_role bit for availability of PFEC.I/D

[mlevitsk@redhat.com]

On Tue, 2026-05-05 at 21:52 +0200, Paolo Bonzini wrote:
> While GMET looks a lot like SMEP, it has several annoying differences.
> The main one is that the availability of the I/D bit in the page fault
> error code still depends on the host CR4.SMEP and EFER.NXE bits.  If the

Hi!

What do you think if we reword this like this:

"still depends on either host's CR4.SMEP or host's EFER.NXE being set"

I initially thought that we have to have both of these settings enabled and it
confused me.

> base.cr4_smep bit of the cpu_role is (ab)used to enable GMET, there needs
> to be another place where the host CR4.SMEP is read from; just merge it
> with EFER.NXE into a new cpu_role bit that tells paging_tmpl.h whether
> to set the I/D bit at all.


I am thinking:

For KVM point of view, has_pferr_fetch will always be true on NPT, because the kernel itself always enables EFER.NX,
if it can (for reference, the code is in head_64.S), and there seems to be no override for that.

And then, KVM refuses to load when EFER.NX is not supported by the CPU 
So host's EFER.NX will always be set for KVM (we can add an assert somewhere to be 100% sure)

In fact if this is the case, we can simplifiy things by assuming that we always have PFERR_FETCH for NPT.
What do you think?


> 
> Tested-by: David Riley <d.riley@proxmox.com>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  arch/x86/include/asm/kvm_host.h | 7 +++++++
>  arch/x86/kvm/mmu/mmu.c          | 8 ++++++++
>  arch/x86/kvm/mmu/paging_tmpl.h  | 2 +-
>  3 files changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 23a7ac8d7fbe..7dde4ca87752 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -414,6 +414,13 @@ union kvm_mmu_extended_role {
>   unsigned int cr4_smap:1;
>   unsigned int cr4_la57:1;
>   unsigned int efer_lma:1;
> +
> + /*
> + * True if either CR4.SMEP or EFER.NXE are set.  For AMD NPT
> + * this is the "real" host CR4.SMEP whereas cr4_smep is
> + * actually GMET.
> + */


If we adopt my suggestion of using 'role.has_user_exec_permission',
then I guess we won't need this patch?
What do you think? 


Side question: Is it true that CR4.SMEP *on the host* is not needed for NPT/GMET?
(I wasn't able to find anything in the manual)


Best regards,
	Maxim Levitsky


> + unsigned int has_pferr_fetch:1;
>   };
>  };
>  
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 156bab8afbc6..912c8e97ef61 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -234,6 +234,11 @@ BUILD_MMU_ROLE_ACCESSOR(ext,  cr4, la57);
>  BUILD_MMU_ROLE_ACCESSOR(base, efer, nx);
>  BUILD_MMU_ROLE_ACCESSOR(ext,  efer, lma);
>  
> +static inline bool has_pferr_fetch(struct kvm_mmu *mmu)
> +{
> + return mmu->cpu_role.ext.has_pferr_fetch;
> +}
> +
>  static inline bool is_cr0_pg(struct kvm_mmu *mmu)
>  {
>          return mmu->cpu_role.base.level > 0;
> @@ -5793,6 +5798,8 @@ static union kvm_cpu_role kvm_calc_cpu_role(struct kvm_vcpu *vcpu,
>   role.ext.cr4_pke = ____is_efer_lma(regs) && ____is_cr4_pke(regs);
>   role.ext.cr4_la57 = ____is_efer_lma(regs) && ____is_cr4_la57(regs);
>   role.ext.efer_lma = ____is_efer_lma(regs);
> +
> + role.ext.has_pferr_fetch = role.base.efer_nx | role.base.cr4_smep;
>   return role;
>  }
>  
> @@ -5946,6 +5953,7 @@ void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
>  
>   /* NPT requires CR0.PG=1. */
>   WARN_ON_ONCE(cpu_role.base.direct || !cpu_role.base.guest_mode);
> + cpu_role.base.cr4_smep = false;
>  
>   root_role = cpu_role.base;
>   root_role.level = kvm_mmu_get_tdp_level(vcpu);
> diff --git a/arch/x86/kvm/mmu/paging_tmpl.h b/arch/x86/kvm/mmu/paging_tmpl.h
> index 047400af924d..07100bbfc270 100644
> --- a/arch/x86/kvm/mmu/paging_tmpl.h
> +++ b/arch/x86/kvm/mmu/paging_tmpl.h
> @@ -489,7 +489,7 @@ static int FNAME(walk_addr_generic)(struct guest_walker *walker,
>  
>  error:
>   errcode |= write_fault | user_fault;
> - if (fetch_fault && (is_efer_nx(mmu) || is_cr4_smep(mmu)))
> + if (fetch_fault && has_pferr_fetch(mmu))
>   errcode |= PFERR_FETCH_MASK;
>  
>   walker->fault.vector = PF_VECTOR;