Simon Kagstrom wrote: > Well isn't the accelerometer driver a pretty good example?
That's what I meant :-) But in order to make the case for a lean and fast async interface, we'd have to have some code that proves that it's actually better. Hence my suggestion for the DMA-based approach, which should be fairly close to the optimum one can do on the kernel side. (Beware that SPI DMA isn't trivial, though. E.g., there are some quirks with the first and last bytes.) - Werner
