KVM: arm64: Take a protected vCPU's TWI and TWE only at vCPU load

Commit 06499de8f998 ("KVM: arm64: Skip fixed-feature state flush for
protected vCPUs") moved the copy of the host's TWI and TWE in
flush_hyp_vcpu() into the non-protected branch: a protected vCPU takes
them in __pkvm_vcpu_load() instead, and only VSE comes from the host on
each entry.

Commit 04447b95c62b ("KVM: arm64: Use the host's HCR_EL2 for
non-protected VMs in pKVM") was written against the old code, where
both VM types took TWI, TWE and VSE on each entry, and kept that for a
protected VM as PKVM_HCR_EL2_HOST_PVM. The merge of the two applies
PKVM_HCR_EL2_HOST_PVM in the common path, so a protected vCPU takes TWI
and TWE from the host's vCPU on every entry again.

That doesn't change what a protected guest runs with today: once the
vCPU is initialised, the host only changes TWI and TWE in
vcpu_set_wfx_traps(), which pkvm_vcpu_load() calls right before
__pkvm_vcpu_load(). But it undoes part of 06499de8f998, and has EL2
read a protected vCPU's trap configuration from host memory on every
entry, which nothing needs.

Make PKVM_HCR_EL2_HOST_PVM VSE alone and add TWI and TWE to
PKVM_HCR_EL2_HOST_NPVM, so that each mask again lists what EL2 takes
from the host on each entry for its VM type. A non-protected VM takes
the same bits as before. Say in the comments that a protected vCPU's
TWI and TWE come at vCPU load.

llminus, an LLM-assisted merge conflict resolution tool, found the
difference while resolving the same conflict in an integration tree.

Fixes: 439d8f9cb01c ("Merge tag 'kvmarm-7.4' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm into HEAD")
Assisted-by: LLM
Signed-off-by: Sasha Levin <sashal@kernel.org>
Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev>
Tested-by: Fuad Tabba <fuad.tabba@linux.dev>
Link: https://patch.msgid.link/20261017165204.3343275-1-sashal@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>

0 files changed, 0 insertions, 0 deletions