On Thu, 2024-01-11 at 15:41 +0000, Mark Brown wrote: > On Thu, Jan 11, 2024 at 03:11:32PM +0100, Nuno Sá wrote: > > On Thu, 2024-01-11 at 13:33 +0000, Mark Brown wrote: > > > > I tend to agree that we shouldn't be exposing this to SPI device drivers > > > however we will want to keep track of if the unit is busy, and designing > > > it to cope with multiple offloads does seem like sensible future > > > proofing. There's also the possibility that one engine might be able to > > > Fair enough. But wouldn't a simple DT integer property (handled by the spi core) > > to identify the offload index be easier for SPI device drivers? We could still > > have dedicated interfaces for checking if the unit is busy or not... The point > > is that we would not need an explicit get() from SPI drivers. > > It feels like we'd need a get/release operation of some kind for mutual > exclusion, it's not just the discovery it's also figuring out if the > hardware is in use at a given moment. > Hmm did not thought about the busy case. Still, I could see this being easily done on the controller driver (at least until we have a clear idea if this is useful or if it will attract more users) or even at the spi core on the prepare + unprepare interfaces. A flag could be enough to return EBUSY if someone is already in the process of preparing + enabling the engine. Bah, anyways, it's just I'm really not thrilled about that kind of interface in here but yeah, as long as you think it's worth it... > - Nuno Sá