On Tue, 2021-02-02 at 16:13 +1100, Herbert Xu wrote: > The issue is that we always require a software implementation for > any given hardware algorithm. As otherwise kernel users cannot > rely on the algorithm to work.
I understand. This sounds very reasonable to me. > Of course we don't want to add every single algorithm out there > to the kernel so that's why require there to be an actual in-kernel > user before adding a given algorithm. I see. Just to clarify: does the in-kernel user requirement also apply to the case when the author of a device driver also provides the software implementation for the new algorithms supported by device driver / HW?