Hello! > That's what I see as correct way to go > and along the way teach KVM to not derive mpidr encoding from vcpuid > and use QEMU provided mpidr value, that way mpidr would be consistent.
Yes, and this would make modelling of real hardware working much better. However, unfortunately, i have just tested it, and it does not work. MPIDR can be set for the vCPU, and i believe it will even give back the value when read. But PSCI completely ignores this setting. > As it currently stands QEMU notion of mpidr and KVM's will diverge > onece CPU count goes above 16 CPUs. Yes, but fortunately the code which cares about it (PSCI and Shlomo's GICv3 software emulation) is not used with KVM. So i think we have no other choice if we want to be compatible with current KVM APIs. Kind regards, Pavel Fedin Expert Engineer Samsung Electronics Research center Russia
