The ID_AA64PFR1_EL1.MTE_frac field is currently hidden from KVM. However, when ID_AA64PFR1_EL1.MTE==2, ID_AA64PFR1_EL1.MTE_frac==0 indicates that MTE_ASYNC is supported. On a host with ID_AA64PFR1_EL1.MTE==2 but without MTE_ASYNC support a guest with the MTE capability enabled will incorrectly see MTE_ASYNC advertised as supported. This series fixes that.
This was found by inspection and the current behaviour is not known to break anything. Linux doesn't check MTE_frac, and wrongly, assumes MTE async faults can be generated whenever MTE is supported. This is a separate problem and not addressed here.
I am looking for feedback on whether this change is valuable or otherwise.
Changes since v1: Only pass MTE_Frac hw value to the guest when it is the exact failure case. Changed base commit to v6.15-rc5 but still applies on v6.16-rc2 as well.
Ben Horgan (3): arm64/sysreg: Expose MTE_frac so that it is visible to KVM KVM: arm64: Make MTE_frac masking conditional on MTE capability KVM: selftests: Confirm exposing MTE_frac does not break migration
arch/arm64/kernel/cpufeature.c | 1 + arch/arm64/kvm/sys_regs.c | 28 ++++++- .../testing/selftests/kvm/arm64/set_id_regs.c | 77 ++++++++++++++++++- 3 files changed, 103 insertions(+), 3 deletions(-)
base-commit: 92a09c47464d040866cf2b4cd052bc60555185fb