On 5/10/19 9:20 AM, Peter Maydell wrote: > On Fri, 10 May 2019 at 17:17, Richard Henderson > <richard.hender...@linaro.org> wrote: >> >> On 5/10/19 9:01 AM, Peter Maydell wrote: >>> Don't these need to be marked ARM_CP_IO for the benefit >>> of -icount ? >> >> I don't think so. There's no lock taken, as for mmio devices. It's not not >> related to time, virtual or otherwise. There are no possible exceptions. I >> can't think of anything that would make icount care. >> >> Have I missed something? > > If icount decides it needs to replay execution of the TB > which the RNDR access is in, won't it get a different > number back the second time it executes ?
Yes, it will. I forgot about replay. r~