[Ksummit-2013-discuss] [ATTEND] Interface stability guarantees

[Greg KH]

On Thu, Jul 25, 2013 at 10:43:38AM +0300, Dan Carpenter wrote:
> I've seen a couple times where we end up deliberately breaking
> support for some closed source firmware debugging tool.  For example
> there was a wireless driver that writes custom debug firmware to the
> device by catting it to a debugfs file.  The firmware comes with a
> close source tool to use it.

What is wrong with moving that to the proper firmware interface?  I see
this a lot, people abuse sysfs in their out-of-tree drivers, and then
get upset when we change them when the code is accepted.  It's been
happening for 15+ years, and it's part of a company's evolution to
working with the community and accepting feedback on their code.

> Here is a recent one:
> https://lkml.org/lkml/2013/7/25/41
> 
> It has custom ioctls for their optional SDK.  We would like to
> delete the ioctls or change them to sysfs files and netlink etc.

As this driver _just_ went into staging, and hasn't hit Linus's tree, I
don't think there are any interface guarantee yet for it, so that might
be a bad example :)

Also, the developer in that example said it was ok to remove the
offending ioctl, as they would change their userspace tool, so all is
good...

thanks,

greg k-h