]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
KVM: arm64: Ensure a VMID is allocated before programming VTTBR_EL2
authorOliver Upton <oliver.upton@linux.dev>
Wed, 19 Feb 2025 22:07:37 +0000 (14:07 -0800)
committerMarc Zyngier <maz@kernel.org>
Thu, 20 Feb 2025 16:29:28 +0000 (16:29 +0000)
commitfa808ed4e199ed17d878eb75b110bda30dd52434
tree22e36cfb8bb22d22e8b5d387583b6a63d8ee18ca
parent102c51c50db88aedd00a318b7708ad60dbec2e95
KVM: arm64: Ensure a VMID is allocated before programming VTTBR_EL2

Vladimir reports that a race condition to attach a VMID to a stage-2 MMU
sometimes results in a vCPU entering the guest with a VMID of 0:

| CPU1                                            |   CPU2
|                                                 |
|                                                 | kvm_arch_vcpu_ioctl_run
|                                                 |   vcpu_load             <= load VTTBR_EL2
|                                                 |                            kvm_vmid->id = 0
|                                                 |
| kvm_arch_vcpu_ioctl_run                         |
|   vcpu_load             <= load VTTBR_EL2       |
|                            with kvm_vmid->id = 0|
|   kvm_arm_vmid_update   <= allocates fresh      |
|                            kvm_vmid->id and     |
|                            reload VTTBR_EL2     |
|                                                 |
|                                                 |   kvm_arm_vmid_update <= observes that kvm_vmid->id
|                                                 |                          already allocated,
|                                                 |                          skips reload VTTBR_EL2

Oh yeah, it's as bad as it looks. Remember that VHE loads the stage-2
MMU eagerly but a VMID only gets attached to the MMU later on in the
KVM_RUN loop.

Even in the "best case" where VTTBR_EL2 correctly gets reprogrammed
before entering the EL1&0 regime, there is a period of time where
hardware is configured with VMID 0. That's completely insane. So, rather
than decorating the 'late' binding with another hack, just allocate the
damn thing up front.

Attaching a VMID from vcpu_load() is still rollover safe since
(surprise!) it'll always get called after a vCPU was preempted.

Excuse me while I go find a brown paper bag.

Cc: stable@vger.kernel.org
Fixes: 934bf871f011 ("KVM: arm64: Load the stage-2 MMU context in kvm_vcpu_load_vhe()")
Reported-by: Vladimir Murzin <vladimir.murzin@arm.com>
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Link: https://lore.kernel.org/r/20250219220737.130842-1-oliver.upton@linux.dev
Signed-off-by: Marc Zyngier <maz@kernel.org>
arch/arm64/include/asm/kvm_host.h
arch/arm64/kvm/arm.c
arch/arm64/kvm/vmid.c