Linux kernel version support policy

[Joseph S. Myers]

On Mon, 27 Jan 2014, Mike Frysinger wrote:

> On Monday, January 27, 2014 23:02:43 Joseph S. Myers wrote:
> > (and more generally, to review
> > the conditionals that would be obsoleted by such a move to make sure the
> > features that would be assumed to be available are in fact available in
> > the chosen versions for relevant architectures).
> 
> i think any bump in min kernel version should be predicated upon this.  if 
> there's no significant code clean up by updating the min version, then we 
> shouldn't be throwing away support for them.

Note that increasing the minimum version makes such a review easier - it 
can start by checking what's in 2.6.32 (for example) without worrying 
about identifying the specific older version where something was added, so 
specific versions need checking only for architectures still missing a 
feature in 2.6.32.

However, when I made 2.6.16 the minimum, I did the cleanup of individual 
__ASSUME_* *after* the general arch_minimum_kernel change.  Given that we 
know about significant mistakes in conditionals for more recent kernels, I 
think the review would need to be done first this time.  Thus, one patch 
would both change the minimum and correct the conditions on individual 
affected macros (possibly to unconditional, possibly not), while 
subsequent patches would then remove macro definitions along with the 
conditionals testing those macros, where a feature is now known to be 
available unconditionally.

-- 
Joseph S. Myers
joseph@codesourcery.com