On 09/11/2015 12:09, Peter Maydell wrote: >>> >> I could have handled that by making the CPU init code always >>> >> register two ASes (using the same one twice if the board code >>> >> only passes one or using system_address_space twice if the >>> >> board code doesn't pass one at all), but that seemed a bit wasteful. >> > >> > I think it's simpler though. Complicating the init code is better than >> > handling the condition throughout all the helpers... > It was the overhead of having an extra AddressSpace that concerned > me (plus it shows up in things like 'info mtree' somewhat confusingly > if you didn't expect your board to really have 2 ASes).
I don't think it shows up twice with address_space_init_shareable, does it? Paolo > But I don't > feel very strongly about it (or have anything solid to base an > argument either way on).