On 25 September 2018 at 12:47, Damien Hedde <damien.he...@greensocs.com> wrote: > > > On 9/25/18 11:36 AM, Peter Maydell wrote: >> On 17 September 2018 at 09:40, <damien.he...@greensocs.com> wrote: >>> +Outputs must be initialized in the device_reset method to ensure every >>> connected >>> +inputs is updated at machine startup. >> >> device_reset should not set outputs. > > How do the initialization then ? > Only alternative I see is to initialize a clock input when it is > connected to an output. > > If device_reset does not set outputs (which are a direct consequence of > a internal device state being reset). It means calling device_reset at > any other point than simulation startup will break things. Is that the > semantic behind device_reset ?
device_reset's semantics are not great, but basically they are "power on reset", equivalent to a complete simulation restart. If a device has some other kind of warm reset triggered by eg a register write it needs to implement that itself, perhaps sharing code with the device reset method. A warm reset triggered by a register write can obviously assert output signals. I have some thoughts about trying to clean up our reset semantics which I need to write up later this week. thanks -- PMM
