On Thu, 30 Dec 2021 10:53:30 +0000, Nikita Yushchenko <nikita.yoush@xxxxxxxxxxxxxxxxxx> wrote: > > >>>>> The right way to do it is to use the existing API by exposing the > >>>>> inverter (there are existing examples in the tree, using the > >>>>> hierarchical model)... > >>> > >>> A much simpler version can be written in a few minutes, see below... > >> > >> Can something like that be used if the parent domain is not > >> hierarchical (i.e. does not provide alloc(), but provides map() > >> instead)? > > > > No. This definitely relies on the parent being hierarchical, as that's > > exactly what it was designed for the first place. > > Is supporting hierarchical API now mandatory for kernel irqchips? It isn't. But you are definitely giving me some ideas now. > > If yes, then perhaps you can at least document it somewhere? > E.g. declare irq_domain.map() as deprecated? > > If no, then I'd like to discuss a solution for irq_inverter that can > work for non-hierarchical case. You are still missing the point. Any active element on the interrupt path that changes the signal without multiplexing must be described in the hierarchical model. Anything else is wrong, and I am not interested in reinventing that particular wheel (it was painful enough to kill all these hacks years ago, and I'm not doing it again). Moving these GPIO chips into the hierarchical model isn't rocket science, as there is plenty of support for that already, and is the right thing to do. M. -- Without deviation from the norm, progress is not possible.
