Hi James, On 2017/10/23 23:26, James Morse wrote: > If you're suggesting Qemu should set a default 'unknown' ESR for use when KVM > doesn't know what to do, and SError is pretty much its only option: > > Why would Qemu set anything other than impdef:all-zeros? > > The only use would be to fake this back into a RAS error. Qemu isn't involved > here so this can't be used to emulate NOTIFY_SEI without the kernel support. We > don't have support for emulating the RAS ERR registers either, so this can't be > used to emulate kernel first. > > > If you're suggesting Qemu should specify a set of ESR values for the different > cases where KVM doesn't know what do: this would be an early shortcut that burns > us in the future, these things are going to be changed. This is too specific to > KVMs internals: > > When we add NOTIFY_SEI support, we will call out of this KVM code and APEI > should 'claim' any SError that arrived with CPER records. > > I expect this code to change dramatically if/when we add kernel first support. > > > The KVM patches on the end of this series are just the minimum needed to enable > IESB and keep the status quo for all other behaviour. And this is just so we can > tackle the other bits of support that depends on the cpufeature without > reposting all the same code. Thanks very much for your explanation, understand it now. _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
