Re: [PATCH 4/4] KVM: arm64: Convert VTCR_EL2 to config-driven sanitisation

[Marc Zyngier <maz@kernel.org>]

On Wed, 03 Dec 2025 16:17:15 +0000,
Joey Gouly <joey.gouly@arm.com> wrote:
> 
> Hi!
> 
> On Sat, Nov 29, 2025 at 02:45:25PM +0000, Marc Zyngier wrote:
> > Describe all the VTCR_EL2 fields and their respective configurations,
> > making sure that we correctly ignore the bits that are not defined
> > for a given guest configuration.
> > 
> > Signed-off-by: Marc Zyngier <maz@kernel.org>
> > ---
> >  arch/arm64/kvm/config.c | 69 +++++++++++++++++++++++++++++++++++++++++
> >  arch/arm64/kvm/nested.c |  3 +-
> >  2 files changed, 70 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/config.c b/arch/arm64/kvm/config.c
> > index a02c28d6a61c9..c36e133c51912 100644
> > --- a/arch/arm64/kvm/config.c
> > +++ b/arch/arm64/kvm/config.c
> > @@ -141,6 +141,7 @@ struct reg_feat_map_desc {
> >  #define FEAT_AA64EL1		ID_AA64PFR0_EL1, EL1, IMP
> >  #define FEAT_AA64EL2		ID_AA64PFR0_EL1, EL2, IMP
> >  #define FEAT_AA64EL3		ID_AA64PFR0_EL1, EL3, IMP
> > +#define FEAT_SEL2		ID_AA64PFR0_EL1, SEL2, IMP
> >  #define FEAT_AIE		ID_AA64MMFR3_EL1, AIE, IMP
> >  #define FEAT_S2POE		ID_AA64MMFR3_EL1, S2POE, IMP
> >  #define FEAT_S1POE		ID_AA64MMFR3_EL1, S1POE, IMP
> > @@ -202,6 +203,8 @@ struct reg_feat_map_desc {
> >  #define FEAT_ASID2		ID_AA64MMFR4_EL1, ASID2, IMP
> >  #define FEAT_MEC		ID_AA64MMFR3_EL1, MEC, IMP
> >  #define FEAT_HAFT		ID_AA64MMFR1_EL1, HAFDBS, HAFT
> > +#define FEAT_HDBSS		ID_AA64MMFR1_EL1, HAFDBS, HDBSS
> > +#define FEAT_HPDS2		ID_AA64MMFR1_EL1, HPDS, HPDS2
> >  #define FEAT_BTI		ID_AA64PFR1_EL1, BT, IMP
> >  #define FEAT_ExS		ID_AA64MMFR0_EL1, EXS, IMP
> >  #define FEAT_IESB		ID_AA64MMFR2_EL1, IESB, IMP
> > @@ -219,6 +222,7 @@ struct reg_feat_map_desc {
> >  #define FEAT_FGT2		ID_AA64MMFR0_EL1, FGT, FGT2
> >  #define FEAT_MTPMU		ID_AA64DFR0_EL1, MTPMU, IMP
> >  #define FEAT_HCX		ID_AA64MMFR1_EL1, HCX, IMP
> > +#define FEAT_S2PIE		ID_AA64MMFR3_EL1, S2PIE, IMP
> >  
> >  static bool not_feat_aa64el3(struct kvm *kvm)
> >  {
> > @@ -362,6 +366,28 @@ static bool feat_pmuv3p9(struct kvm *kvm)
> >  	return check_pmu_revision(kvm, V3P9);
> >  }
> >  
> > +#define has_feat_s2tgran(k, s)						\
> > +  ((kvm_has_feat_enum(kvm, ID_AA64MMFR0_EL1, TGRAN##s##_2, TGRAN##s) && \
> > +    !kvm_has_feat_enum(kvm, ID_AA64MMFR0_EL1, TGRAN##s, NI))	     ||	\
> > +   kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN##s##_2, IMP))
> > +
> > +static bool feat_lpa2(struct kvm *kvm)
> > +{
> > +	return ((kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN4, 52_BIT)    ||
> > +		 !kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN4, IMP))	&&
> > +		(kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN16, 52_BIT)   ||
> > +		 !kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN16, IMP))	&&
> > +		(kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN4_2, 52_BIT)  ||
> > +		 !has_feat_s2tgran(kvm, 4))				&&
> > +		(kvm_has_feat(kvm, ID_AA64MMFR0_EL1, TGRAN16_2, 52_BIT) ||
> > +		 !has_feat_s2tgran(kvm, 16)));
> > +}
> > +
> > +static bool feat_vmid16(struct kvm *kvm)
> > +{
> > +	return kvm_has_feat_enum(kvm, ID_AA64MMFR1_EL1, VMIDBits, 16);
> > +}
> > +
> >  static bool compute_hcr_rw(struct kvm *kvm, u64 *bits)
> >  {
> >  	/* This is purely academic: AArch32 and NV are mutually exclusive */
> > @@ -1168,6 +1194,44 @@ static const struct reg_bits_to_feat_map mdcr_el2_feat_map[] = {
> >  static const DECLARE_FEAT_MAP(mdcr_el2_desc, MDCR_EL2,
> >  			      mdcr_el2_feat_map, FEAT_AA64EL2);
> >  
> > +static const struct reg_bits_to_feat_map vtcr_el2_feat_map[] = {
> > +	NEEDS_FEAT(VTCR_EL2_HDBSS, FEAT_HDBSS),
> > +	NEEDS_FEAT(VTCR_EL2_HAFT, FEAT_HAFT),
> > +	NEEDS_FEAT(VTCR_EL2_TL0		|
> > +		   VTCR_EL2_TL1		|
> > +		   VTCR_EL2_AssuredOnly	|
> > +		   VTCR_EL2_GCSH,
> > +		   FEAT_THE),
> 
> The text for VTCR_EL2.AssuredOnly says:
> 
> 	This field is RES0 when VTCR_EL2.D128 is 1.
> 
> > +	NEEDS_FEAT(VTCR_EL2_D128, FEAT_D128),
> > +	NEEDS_FEAT(VTCR_EL2_S2POE, FEAT_S2POE),
> > +	NEEDS_FEAT(VTCR_EL2_S2PIE, FEAT_S2PIE),
> 
> The text for VTCR_EL2.S2PIE says:
> 
> 	This field is RES1 when VTCR_EL2.D128 is set.
> 
> 
> Are these cases that need to be handled here somehow?

These are not static configurations. They are dynamic behaviours
depending on other control bits.

D128 code, if it ever exists, will have to *interpret* these bits as
RES0 (resp. RES1) when evaluating the page tables.

If you want a similar example in existing code, look at the way we
handle TCR_EL1.HPDn in the S1 PTW. They are treated as RES1 if
TCR2_EL1.PIE is set, as per R_JHSVW.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.