Main Linux kernel vs Ubuntu kernel
Thadeu Lima de Souza Cascardo
cascardo at canonical.com
Fri Jun 15 12:58:45 UTC 2018
On Fri, Jun 15, 2018 at 07:43:55AM -0500, Seth Forshee wrote:
> On Fri, Jun 15, 2018 at 04:55:38PM +0530, Subhashini Rao Beerisetty wrote:
[...]
> > Basically i’m looking for ubuntu disto with N years of security
> > support\updates along with supporting\developing our out-of-tree module for
> > new kernel versions. What would be the best strategy?
>
> The LTS releases are your best bet, each of those comes with 5 years of
> support. 16.04 will be supported until April 2021 and 18.04 until April
> 2023. As long as you don't install any hwe kernel packages your system
> should remain with the GA kernel version.
>
[...]
> > > In broad strokes yes, as there is a pretty strict policy upstream about
> > > not introducing regressions to userspace. However sometimes issues do
> > > slip in, and you might also see problems if you try to install kernel
> > > packages from a newer Ubuntu release into an older release. The only
> > > supported way to run newer kernels is through the hwe packages, and
> > > those are only provided for LTS releases.
>
> Just to clarify this -- the "no regressions" policy applies to
> userspace, not to kernel modules. There's no guarantee about internal
> kernel interfaces remaining backwards compatible.
>
> Seth
And they also break between stable (4.4.y) releases, and we also apply fixes
that break this internal ABI. In fact, we do not go through the effort of
trying not to break them, and require all DKMS modules to rebuild after each
Ubuntu release. As kernels go into -proposed, we have ourselves the chance to
fix any DKMS module that breaks before those go into -updates.
But there is no guarantee that an out-of-tree module won't need any updates
even if they stick to the LTS security updates. Still, there should be much
less changes between an Ubuntu-4.4.0-90 and Ubuntu-4.4.0-120 kernel, then say,
between an Ubuntu-4.4.0-90 and Ubuntu-4.13.0-30 kernel. So, sticking to the LTS
kernels is still the best strategy here.
Better yet, you should try to follow the proposed kernels on developer/testing
systems. That gives you the chance to fix anything that breaks before deploying
those.
Cascardo.
More information about the kernel-team
mailing list