[PATCH v4 3/6] KVM: arm64: Use per guest ID register for ID_AA64PFR0_EL1.[CSV2|CSV3]
Jing Zhang
jingzhangos at google.com
Tue Mar 28 13:01:02 PDT 2023
Hi Faud,
On Tue, Mar 28, 2023 at 5:40 AM Fuad Tabba <tabba at google.com> wrote:
>
> Hi,
>
> On Fri, Mar 17, 2023 at 5:06 AM Jing Zhang <jingzhangos at google.com> wrote:
> >
> > With per guest ID registers, ID_AA64PFR0_EL1.[CSV2|CSV3] settings from
> > userspace can be stored in its corresponding ID register.
> >
> > No functional change intended.
> >
> > Signed-off-by: Jing Zhang <jingzhangos at google.com>
> > ---
> > arch/arm64/include/asm/kvm_host.h | 2 --
> > arch/arm64/kvm/arm.c | 19 +------------------
> > arch/arm64/kvm/hyp/nvhe/sys_regs.c | 7 +++----
> > arch/arm64/kvm/id_regs.c | 30 ++++++++++++++++++++++--------
> > 4 files changed, 26 insertions(+), 32 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> > index fb6b50b1f111..e926ea91a73c 100644
> > --- a/arch/arm64/include/asm/kvm_host.h
> > +++ b/arch/arm64/include/asm/kvm_host.h
> > @@ -230,8 +230,6 @@ struct kvm_arch {
> >
> > cpumask_var_t supported_cpus;
> >
> > - u8 pfr0_csv2;
> > - u8 pfr0_csv3;
> > struct {
> > u8 imp:4;
> > u8 unimp:4;
> > diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> > index 4579c878ab30..c78d68d011cb 100644
> > --- a/arch/arm64/kvm/arm.c
> > +++ b/arch/arm64/kvm/arm.c
> > @@ -104,22 +104,6 @@ static int kvm_arm_default_max_vcpus(void)
> > return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
> > }
> >
> > -static void set_default_spectre(struct kvm *kvm)
> > -{
> > - /*
> > - * The default is to expose CSV2 == 1 if the HW isn't affected.
> > - * Although this is a per-CPU feature, we make it global because
> > - * asymmetric systems are just a nuisance.
> > - *
> > - * Userspace can override this as long as it doesn't promise
> > - * the impossible.
> > - */
> > - if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
> > - kvm->arch.pfr0_csv2 = 1;
> > - if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
> > - kvm->arch.pfr0_csv3 = 1;
> > -}
> > -
> > /**
> > * kvm_arch_init_vm - initializes a VM data structure
> > * @kvm: pointer to the KVM struct
> > @@ -151,9 +135,8 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
> > /* The maximum number of VCPUs is limited by the host's GIC model */
> > kvm->max_vcpus = kvm_arm_default_max_vcpus();
> >
> > - set_default_spectre(kvm);
> > - kvm_arm_init_hypercalls(kvm);
> > kvm_arm_set_default_id_regs(kvm);
> > + kvm_arm_init_hypercalls(kvm);
> >
> > /*
> > * Initialise the default PMUver before there is a chance to
> > diff --git a/arch/arm64/kvm/hyp/nvhe/sys_regs.c b/arch/arm64/kvm/hyp/nvhe/sys_regs.c
> > index 08d2b004f4b7..0e1988740a65 100644
> > --- a/arch/arm64/kvm/hyp/nvhe/sys_regs.c
> > +++ b/arch/arm64/kvm/hyp/nvhe/sys_regs.c
> > @@ -93,10 +93,9 @@ static u64 get_pvm_id_aa64pfr0(const struct kvm_vcpu *vcpu)
> > PVM_ID_AA64PFR0_RESTRICT_UNSIGNED);
> >
> > /* Spectre and Meltdown mitigation in KVM */
> > - set_mask |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
> > - (u64)kvm->arch.pfr0_csv2);
> > - set_mask |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
> > - (u64)kvm->arch.pfr0_csv3);
> > + set_mask |= vcpu->kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] &
> > + (ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
> > + ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
>
> This triggers a compiler warning now since the variable `struct kvm
> *kvm` isn't used anymore, this, however, isn't the main issue.
>
> The main issue is that `struct kvm` here (vcpu->kvm) is the
> hypervisor's version for protected vms, and not the host's. Therefore,
> reading that value is wrong. That said, this is an existing bug in
> pKVM since kvm->arch.pfr0_csv2 and kvm->arch.pfr0_csv3 are not
> initialized.
>
> The solution would be to track the spectre/meltown state at hyp and
> use that. I'll submit a patch that does that. In the meantime, I think
> that it would be better not to set the CSV bits for protected VMs,
> which is the current behavior in practice.
>
> Non-protected VMs in protected mode go back to the host on id register
> traps, and use the host's `struct kvm`, so they should behave as
> expected.
You mean just remove these two lines:
/* Spectre and Meltdown mitigation in KVM */
set_mask |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
(u64)kvm->arch.pfr0_csv2);
set_mask |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
(u64)kvm->arch.pfr0_csv3);
Will it cause any problem for pKVM without your incoming patch?
>
> Thanks,
> /fuad
>
>
> >
> > return (id_aa64pfr0_el1_sys_val & allow_mask) | set_mask;
> > }
> > diff --git a/arch/arm64/kvm/id_regs.c b/arch/arm64/kvm/id_regs.c
> > index e393b5730557..b60ca1058301 100644
> > --- a/arch/arm64/kvm/id_regs.c
> > +++ b/arch/arm64/kvm/id_regs.c
> > @@ -61,12 +61,6 @@ u64 kvm_arm_read_id_reg(const struct kvm_vcpu *vcpu, u32 id)
> > if (!vcpu_has_sve(vcpu))
> > val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_SVE);
> > val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_AMU);
> > - val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2);
> > - val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
> > - (u64)vcpu->kvm->arch.pfr0_csv2);
> > - val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3);
> > - val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
> > - (u64)vcpu->kvm->arch.pfr0_csv3);
> > if (kvm_vgic_global_state.type == VGIC_V3) {
> > val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_GIC);
> > val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_GIC), 1);
> > @@ -201,6 +195,7 @@ static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
> > u64 val)
> > {
> > u8 csv2, csv3;
> > + u64 sval = val;
> >
> > /*
> > * Allow AA64PFR0_EL1.CSV2 to be set from userspace as long as
> > @@ -225,8 +220,7 @@ static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
> > if (val)
> > return -EINVAL;
> >
> > - vcpu->kvm->arch.pfr0_csv2 = csv2;
> > - vcpu->kvm->arch.pfr0_csv3 = csv3;
> > + vcpu->kvm->arch.id_regs[IDREG_IDX(reg_to_encoding(rd))] = sval;
> >
> > return 0;
> > }
> > @@ -529,4 +523,24 @@ void kvm_arm_set_default_id_regs(struct kvm *kvm)
> > val = read_sanitised_ftr_reg(id);
> > kvm->arch.id_regs[IDREG_IDX(id)] = val;
> > }
> > + /*
> > + * The default is to expose CSV2 == 1 if the HW isn't affected.
> > + * Although this is a per-CPU feature, we make it global because
> > + * asymmetric systems are just a nuisance.
> > + *
> > + * Userspace can override this as long as it doesn't promise
> > + * the impossible.
> > + */
> > + val = kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)];
> > +
> > + if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
> > + val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2);
> > + val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2), 1);
> > + }
> > + if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
> > + val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3);
> > + val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3), 1);
> > + }
> > +
> > + kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] = val;
> > }
> > --
> > 2.40.0.rc1.284.g88254d51c5-goog
> >
Thanks,
Jing
More information about the linux-arm-kernel
mailing list