On 02/05/13 11:02, Anup Patel wrote: Hi Anup, > We would like to know KVM ARM/ARM64 plans for handling host implementation > specific registers when underlying host cpu is different from guest cpu. So far, we don't do that. The only case where that could actually be useful is a big.LITTLE-type system. > These host implementation specific registers can be of two types: > 1. Trappable > 2. Non-trappable > > Trappable host implementation specific registers can be easily handled by simply > not allowing guest to access them. > > Non-trappable host implementation specific register are tricky to > handle since, we > cannot trap guest access to these registers. > > One potential solution for this would be to save/restore them in the low-level > context switching routines and KVM VCPU context can have some scratch space > to save/restore host implementation specific registers. You also need to take care of userspace access to these registers (think live migration). > The reason for bringing up this issue is that we have few > implementation specific > registers in APM ARMv8 CPU which cannot be trapped upon guest access. I don't get what this has to do with supporting a guest that is different from your host. You just have an extra set of registers that need to be context-switched in HYP, irrelevant of what the guest thinks it is running on. Am I missing something obvious? M. -- Jazz is not dead. It just smells funny... _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
