On 11/2/20 11:38 PM, Rémi Denis-Courmont wrote: > Answering my own patch as I have a policy question here... > > This exposes SEL2 without TTST (small translation tables). On a logical > level, > the two extensions are orthogonal. But per DDI0487, SEL2 implies TTST, so I > am > not sure if this is considered an acceptable deviation in QEMU, or if > implementing TTST is deemed necessary.
We should implement TTST, yes. I don't think we need to be 100% strict on the ordering, so long as they are both done for the next qemu release in the spring. I don't think it should be difficult. FWIW, we left aarch32 fp16 unimplemented for quite some time, even though that was required for full compliance with aarch64 fp16. > Note that there's what seems like an editorial error in the spec: VSTCR > documentation covers the scenario that TTST is not supported by the CPU, even > though then VSTCR should not exist. Indeed. ;-) r~
