On Thu, Apr 28, 2016 at 06:49:23PM +0200, Arnd Bergmann wrote: > On Thursday 28 April 2016 17:39:20 Mark Brown wrote: > > It's not just platforms that use these things though - there's things > > like the SolarFlare NICs where the firmware update mechanism essentially > > involves exposing a SPI flash as part of a PCI device and we just merged > > an ASoC driver for a video card which was reusing some existing IPs and > > chips. > That's of course fine: you essentially have a discoverable bus there, > and if we need something like that, we can always add it later to > any subsystem. No, the issue with those cases is that there are devices on discoverable buses that instantiate non-discoverable buses as part of the discoverable device so the implementation of the discoverable device is in part essentially a board file. > In contrast, the interface in the proposed slimbus subsystem seems > designed for board files, and is in the best case just dead code > that can be removed, or has a risk of being misused e.g. if some > device manufacturer decides to use a board file for this instead > of describing the slimbus slaves in DT. Which can be more of a pain than it should be if the code has been written to assume that all the world is DT. That's the most problematic bit of this meme - I've seen people doing things like duplicating IP drivers inside larger devices because they don't think they're allowed to reuse the IP driver. We can fix this sort of thing up when we notice it but we should also try to avoid people making the mistakes in the first place since it saves everyone a lot of effort. With subsystems and to a lesser extent off SoC drivers we need to be careful about just blindly removing non-DT code and writing things so they're hard to use without DT. Like I say in this case I think the board file stuff is probably never going to get used so it's probably sensible to remove that code but it does need to be thought about.
Attachment:
signature.asc
Description: PGP signature