KVM: arm64: Expose ID_AA64PFR2_EL1 to userspace and guests
authorMarc Zyngier <maz@kernel.org>
Tue, 20 Aug 2024 13:18:02 +0000 (14:18 +0100)
committerMarc Zyngier <maz@kernel.org>
Tue, 27 Aug 2024 06:59:27 +0000 (07:59 +0100)
Everything is now in place for a guest to "enjoy" FP8 support.
Expose ID_AA64PFR2_EL1 to both userspace and guests, with the
explicit restriction of only being able to clear FPMR.

All other features (MTE* at the time of writing) are hidden
and not writable.

Reviewed-by: Mark Brown <broonie@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20240820131802.3547589-9-maz@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
arch/arm64/kvm/sys_regs.c

index 51627ad..2d1e451 100644 (file)
@@ -1539,6 +1539,10 @@ static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
 
                val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_SME);
                break;
+       case SYS_ID_AA64PFR2_EL1:
+               /* We only expose FPMR */
+               val &= ID_AA64PFR2_EL1_FPMR;
+               break;
        case SYS_ID_AA64ISAR1_EL1:
                if (!vcpu_has_ptrauth(vcpu))
                        val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_APA) |
@@ -2381,7 +2385,7 @@ static const struct sys_reg_desc sys_reg_descs[] = {
                   ID_AA64PFR0_EL1_AdvSIMD |
                   ID_AA64PFR0_EL1_FP), },
        ID_SANITISED(ID_AA64PFR1_EL1),
-       ID_UNALLOCATED(4,2),
+       ID_WRITABLE(ID_AA64PFR2_EL1, ID_AA64PFR2_EL1_FPMR),
        ID_UNALLOCATED(4,3),
        ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
        ID_HIDDEN(ID_AA64SMFR0_EL1),