Eric reported that a Seattle system was pretty unhappy about VM
migration, and the trace pointed to a glaring bug in the way the GICv2
emulation code reported the interrupt pending state to userspace for
HW interrupts, specially if the interrupt state is per-CPU, as this is
the case for the timer...
Fixing this actually results in a minor cleanup, followed by a bit of
extra hardening so that we can catch further issues in this area
without completely taking the system down.
Unless someone screams, I plan to take these in as fixes as quickly as
possible, with the first patch being an obvious stable candidate. I'd
appreciate it if people could verify that VM migration still works
correctly for both GICv2 and GICv3.
Thanks,
M.
Marc Zyngier (3):
KVM: arm64: Don't read a HW interrupt pending state in user context
KVM: arm64: Replace vgic_v3_uaccess_read_pending with
vgic_uaccess_read_pending
KVM: arm64: Warn if accessing timer pending state outside of vcpu
context
arch/arm64/kvm/arch_timer.c | 3 +++
arch/arm64/kvm/vgic/vgic-mmio-v2.c | 4 +--
arch/arm64/kvm/vgic/vgic-mmio-v3.c | 40 ++----------------------------
arch/arm64/kvm/vgic/vgic-mmio.c | 19 +++++++++++---
arch/arm64/kvm/vgic/vgic-mmio.h | 3 +++
5 files changed, 26 insertions(+), 43 deletions(-)
--
2.34.1
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
