[LEDE-DEV] Kernel version in next major release

[Hauke Mehrtens]

We had some discussions in IRC about the kernel version which should be
used in the next major release after 17.01.

Probably we would target a release in January 2018, but you know our
track record of hitting such plans. ;-)

We came up with these options:


1. Release with kernel 4.9 only

Most targets are already on kernel 4.9, so this will be the featest
option. Kernel 4.9 will get support from Greg till Jan. 2019 and then
probably Ben Hutchings will take over as this kernel version is used in
Debian and he already maintains all the other kernel version used in Debian.

Kernel 4.9 is already getting old, especially for targets where the main
development happens in mainline, backporting this gets harder. (ARM64
targets, some ARM targets, x86, ...)


2. Release with kernel 4.14 only

This is about one year more recent and will get support from Greg for
the next 6 years, so till end of 2023.
The final kernel 4.14 will probably get released on 12. November 2017.

Currently this is not supported by LEDE and the following task have to
be done:
 * Port the generic patches
 * Make all the out of tree modules work with it
 * Port all targets to this kernel version
  * Some targets are not so well maintained any more and there an
additional delay could be introduced.
 * Find and fix all the new bugs
  * This kernel is probably bigger again and we will have more size
problems on devices with 4 MB flash

This will probably add a delay of 3 months or more compared to using
kernel 4.9.


3. Release with kernel 4.9 and 4.14

Since now OpenWrt and LEDE always used one single kernel version in a
release, but we could also release with two different versions. Then
some targets can be upgraded to kernel 4.14 and some will stay at kernel
4.9.

This will help with targets where most of the development happens in the
upstream kernel as we do not have to backport so much. In the master
branch we use different kernel versions anyway, so in general this is
not a problem.

To do this we have to do the following:
 * Port the generic patches
 * Make all the out of tree modules work with it
 * Port some easy targets to this kernel version
 * Find and fix all the new bugs

This will result in some additional maintenance work, because we would
have to maintain two kernel versions in parallel, but this is mostly
only taking the most recent stable versions anyway.

This will probably add a delay of 1 or 2 month compared to using kernel 4.9.


Kernel 3.18 and 4.4 will not be included in the next major release after
17.01 and all targets which are still on these kernel versions when we
branch of LEDE will not be included in the release.


What is your opinion on this topic? Am I missing some arguments?
Currently I would prefer solution 3 going with kernel 4.9 and 4.14.


Hauke