Re: [PATCH, RFC] A development process document

[Roland Dreier <rdreier@cisco.com>]

 > > On the other hand, the drivers for completely new hardware (that has
 > > been totally unsupported by the kernel previously, so there is no
 > > possibility for introducing regressions) are still allowed even after
 > > -rc1, right? 
 > 
 > Clearly it's been known to happen (can you say GRU?).  But I don't know
 > that has ever really been adopted as an "official" policy.  My
 > preference would be to lay down the merge window rules as a fairly firm
 > thing rather than suggest to people that there might be ways to slip
 > around them sometimes.  But, certainly, I'm not tied to that if we want
 > to send a different message.

I think Linus has said pretty explicitly that he is willing to merge new
low-level hardware drivers at pretty much any stage, as long as they are
obviously self-contained and can't break anything.  As an example of
this, the UVC webcam driver was one of the very last things merged just
before the final 2.6.26 release, very late in the release cycle.

Obviously there is some judgement involved in what qualifies as new
hardware drivers, but Linus also often comes out in favor of thinking
about things rather than trying to apply mechanical rules anyway.