There are a number of cases where we evaluate PSTATE early on guest
exit. Nothing wrong with that. Except that we actually synchronise
KVM's view of PSTATE pretty late, way after we needed it. Oopsie boo.
Thankfully, there are only two paths that require it: GICv3 emulation
for 32bit guests, and trap handling of 32bit guests in protected
mode. There is no known need of the former (though you could enable it
on the command line), and the latter is still a work in progress. In
any case, this needs fixing.
Funnily enough, this is something that I had already solved on NV, so
the solution isn't that different from what I have there.
Unless someone shouts, I intend to merge these as fixes.
Marc Zyngier (2):
KVM: arm64: Save PSTATE early on exit
KVM: arm64: Move pkvm's special 32bit handling into a generic
infrastructure
arch/arm64/kvm/hyp/include/hyp/switch.h | 14 ++++++++++++++
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h | 7 ++++++-
arch/arm64/kvm/hyp/nvhe/switch.c | 8 +-------
arch/arm64/kvm/hyp/vhe/switch.c | 4 ++++
4 files changed, 25 insertions(+), 8 deletions(-)
--
2.30.2
