]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
KVM: arm64: Allow userspace to change ID_AA64PFR1_EL1
authorShaoqin Huang <shahuang@redhat.com>
Tue, 23 Jul 2024 07:20:02 +0000 (03:20 -0400)
committerMarc Zyngier <maz@kernel.org>
Sun, 25 Aug 2024 16:48:44 +0000 (17:48 +0100)
commit78c4446b5f957fb14737582e503b1b25f66edc45
treec4554cd7d2da606570010384e9bb7e42269376c0
parente8d164974cfa46fe5ec87869c8a7113641f322d5
KVM: arm64: Allow userspace to change ID_AA64PFR1_EL1

Allow userspace to change the guest-visible value of the register with
different way of handling:

  - Since the RAS and MPAM is not writable in the ID_AA64PFR0_EL1
    register, RAS_frac and MPAM_frac are also not writable in the
    ID_AA64PFR1_EL1 register.

  - The MTE is controlled by a separate UAPI (KVM_CAP_ARM_MTE) with an
    internal flag (KVM_ARCH_FLAG_MTE_ENABLED).
    So it's not writable.

  - For those fields which KVM doesn't know how to handle, they are not
    exposed to the guest (being disabled in the register read accessor),
    those fields value will always be 0.
    Those fields don't have a known behavior now, so don't advertise
    them to the userspace. Thus still not writable.
    Those fields include SME, RNDR_trap, NMI, GCS, THE, DF2, PFAR,
    MTE_frac, MTEX.

  - The BT, SSBS, CSV2_frac don't introduce any new registers which KVM
    doesn't know how to handle, they can be written without ill effect.
    So let them writable.

Besides, we don't do the crosscheck in KVM about the CSV2_frac even if
it depends on the value of CSV2, it should be made sure by the VMM
instead of KVM.

Signed-off-by: Shaoqin Huang <shahuang@redhat.com>
Link: https://lore.kernel.org/r/20240723072004.1470688-4-shahuang@redhat.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
arch/arm64/kvm/sys_regs.c