On 17/07/2018 02:21, Jeff Laughton wrote: > > But holding off interrupts is tricky. So far I've not come across any > simple but satisfactory scheme that can ensure a multi-instruction > sequence will remain indivisible. You end up delaying interrupt > recognition (ie, increasing latency) quite a lot. :( That's because > you have to allow for the fact that interrupts aren't always promptly > recognized by the CPU -- see [1]. > Yeah it's suprisingly complicated. You have to leave nmi asserted, because if you deassert it then you'll get another one fired as it can't be masked. But it's hard to know if it's been taken yet. Mapping it to abort gives access to an extra 65816 feature, which might be able to crudely emulate SO & maybe it's useful for something else entirely. It's not perfect, but it's probably better than just not mapping SO at all.Received on 2018-07-17 10:00:44
Archive generated by hypermail 2.2.0.