Re: [WTF] utterly tasteless ABI in hfi1 (around ->write()/->write_iter())

[Linus Torvalds]

On Wed, Mar 16, 2016 at 8:46 AM, Doug Ledford <dledford@xxxxxxxxxx> wrote:
>
> If we want to maintain back compatibility, then the qib driver has to
> maintain this interface. We could possibly do a new one as well, but we
> can't remove this one.

We've broken more important driver ABI's before - all the nasty X stuff.

Now, the X people did learn their lesson, and it hasn't happened
lately (thank Gods!), but quite frankly, some shit-for-brains
hardware-specific config interface for a rdma device that basically
nobody uses is a _lot_ less important than X ever was.

So I don't care one whit if we break it, and it's not the kind of
backwards compatibility the kernel should worry about. There are
exactly zero regular users of this interface. I assume that people who
use this thing are *so* deeply technical that they can take care of
themselves. And it really is a completely broken interface.

I might be proven wrong, and somebody's dear old grandma ends up
complaining about a new kernel breaking her configuration, and in that
case we'd have to revert anything that causes that breakage. But I
suspect I'm not wrong.

                  Linus