Re: [PATCH 01/26] KVM: arm64: Check advertised Stage-2 page size capability

[Marc Zyngier <maz@xxxxxxxxxx>]

Hi Suzuki,

On 2020-04-22 14:40, Suzuki K Poulose wrote:
> Hi Marc,
>
> On 04/22/2020 01:00 PM, Marc Zyngier wrote:
> > With ARMv8.5-GTG, the hardware (or more likely a hypervisor) can
> > advertise the supported Stage-2 page sizes.
> >
> > Let's check this at boot time.
> >
> > Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> > ---
> >   arch/arm64/include/asm/kvm_host.h |  2 +-
> >   arch/arm64/include/asm/sysreg.h   |  3 +++
> >   arch/arm64/kernel/cpufeature.c    |  8 +++++++
> >   arch/arm64/kvm/reset.c            | 40 
> > ++++++++++++++++++++++++++++---
> >   virt/kvm/arm/arm.c                |  4 +---
> >   5 files changed, 50 insertions(+), 7 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/kvm_host.h 
> > b/arch/arm64/include/asm/kvm_host.h
> > index 32c8a675e5a4a..7dd8fefa6aecd 100644
> > --- a/arch/arm64/include/asm/kvm_host.h
> > +++ b/arch/arm64/include/asm/kvm_host.h
> > @@ -670,7 +670,7 @@ static inline int kvm_arm_have_ssbd(void)
> >   void kvm_vcpu_load_sysregs(struct kvm_vcpu *vcpu);
> >   void kvm_vcpu_put_sysregs(struct kvm_vcpu *vcpu);
> >   -void kvm_set_ipa_limit(void);
> > +int kvm_set_ipa_limit(void);
> >     #define __KVM_HAVE_ARCH_VM_ALLOC
> >   struct kvm *kvm_arch_alloc_vm(void);
> > diff --git a/arch/arm64/include/asm/sysreg.h 
> > b/arch/arm64/include/asm/sysreg.h
> > index ebc6224328318..5d10c9148e844 100644
> > --- a/arch/arm64/include/asm/sysreg.h
> > +++ b/arch/arm64/include/asm/sysreg.h
> > @@ -686,6 +686,9 @@
> >   #define ID_AA64ZFR0_SVEVER_SVE2		0x1
> >     /* id_aa64mmfr0 */
> > +#define ID_AA64MMFR0_TGRAN4_2_SHIFT	40
> > +#define ID_AA64MMFR0_TGRAN64_2_SHIFT	36
> > +#define ID_AA64MMFR0_TGRAN16_2_SHIFT	32
> >   #define ID_AA64MMFR0_TGRAN4_SHIFT	28
> >   #define ID_AA64MMFR0_TGRAN64_SHIFT	24
> >   #define ID_AA64MMFR0_TGRAN16_SHIFT	20
> > diff --git a/arch/arm64/kernel/cpufeature.c 
> > b/arch/arm64/kernel/cpufeature.c
> > index 9fac745aa7bb2..9892a845d06c9 100644
> > --- a/arch/arm64/kernel/cpufeature.c
> > +++ b/arch/arm64/kernel/cpufeature.c
> > @@ -208,6 +208,14 @@ static const struct arm64_ftr_bits 
> > ftr_id_aa64zfr0[] = {
> >   };
> >     static const struct arm64_ftr_bits ftr_id_aa64mmfr0[] = {
> > +	/*
> > +	 * Page size not being supported at Stage-2 are not fatal. You
> > +	 * just give up KVM if PAGE_SIZE isn't supported there. Go fix
> > +	 * your favourite nesting hypervisor.
> > +	 */
> > +	ARM64_FTR_BITS(FTR_HIDDEN, FTR_NONSTRICT, FTR_EXACT, 
> > ID_AA64MMFR0_TGRAN4_2_SHIFT, 4, 1),
> > +	ARM64_FTR_BITS(FTR_HIDDEN, FTR_NONSTRICT, FTR_EXACT, 
> > ID_AA64MMFR0_TGRAN64_2_SHIFT, 4, 1),
> > +	ARM64_FTR_BITS(FTR_HIDDEN, FTR_NONSTRICT, FTR_EXACT, 
> > ID_AA64MMFR0_TGRAN16_2_SHIFT, 4, 1),
>
> One minor issue with this is, if we get a system with cpus having 
> values
> 0 and 2 (both of which indicates the stage-2 support), we might reset
> the value to 1 for the feature indicating, we don't support and block
> KVM. But, we can blame the nesting hypervisor for not emulating this to
> (2). Do we need a comment to make this explicit here ?

Sure. How about something like:

"There is a small corner case where the hypervisor could explicitly 
advertise
 a given granule size at Stage-2 (value 2) on some vCPUs, and use the 
fallback
 to Stage-1 (value 0) for other vCPUs. Although this is not forbidden by 
the
 architecture, it indicates that the hypervisor is being silly (or 
buggy).
 We make no effort to cope with this and pretend that if these fields 
are
 inconsistent across vCPUs, then it isn't worth trying to bring KVM up."

Thanks,

        M.
--
Jazz is not dead. It just smells funny...