Here's a bunch of vgic-v3 fixes I have been sitting on for some
time. None of them are critical, though some are rather entertaining.
The first one is a leftover from the initial Apple-M1 enablement,
which doesn't advertise the GIC support via ID_AA64PFR0_EL1 (which is
expected, as it only has half a GIC...). We address it by forcefully
advertising the feature if the guest has a GICv3.
The second patch is really fun, and shows how things can go wrong when
they are badly specified. The gist of it is that on systems that
advertise ICH_VTR_EL2.SEIS, we need to fallback to the full GICv3
cpuif emulation. The third patch is a direct consequence of the
previous one.
The last two patches are more of a harmless oddity: virtual LPIs
happen to have an active state buried into the pseudocode (and only
there). Fun! Nothing goes wrong with that, but we can perform a minor
optimisation, and we need to align the emulation to match the
pseudocode.
All of this is only targeting 5.16, and I don't plan to backport any
of it.
Marc Zyngier (5):
KVM: arm64: Force ID_AA64PFR0_EL1.GIC=1 when exposing a virtual GICv3
KVM: arm64: Work around GICv3 locally generated SErrors
KVM: arm64: vgic-v3: Don't advertise ICC_CTLR_EL1.SEIS
KVM: arm64: vgic-v3: Don't propagate LPI active state from LRs into
the distributor
KVM: arm64: vgic-v3: Align emulated cpuif LPI state machine with the
pseudocode
arch/arm64/kvm/hyp/vgic-v3-sr.c | 22 ++++++++--------------
arch/arm64/kvm/sys_regs.c | 5 +++++
arch/arm64/kvm/vgic/vgic-v3.c | 11 ++++++++++-
3 files changed, 23 insertions(+), 15 deletions(-)
--
2.30.2
