On Thu, 2014-04-03 at 19:55 +0800, Thomas Goirand wrote:
> On 04/03/2014 05:58 PM, Tollef Fog Heen wrote:
> > Am I understanding you correctly that you don't think there are any
> > situations where compiling out features from the kernel can lead to pid1
> > not working would be acceptable?
>
> I'd say the opposite way. Could you please explain in which case you
> find it acceptable to *just crash*, and render the system completely
> unusable, and possibly even not recoverable?
1. If the kernel is configured without a driver for the disk controller,
that happens.
2. If the kernel is configured without the filesystem for the root
partition, that happens.
3. If the kernel is configured without VT or block support, that
probably happens.
4. If the kernel is configured without networking support, the system
may boot but if it's a server it's unusable.
etc.
(I've had boot failures due to several of the above configuration
errors, while doing kernel development and trying to use a minimal
config.)
If cgroups are essential for init, why is this so different from any of
the above?
> Even without cgroup support, the way to handle the situation, IMO, would
> be to at least fall back to the shell with a comprehensive error message.
That would be better, though not much better. You'll need either local
access or a remote serial console either to use the shell or to select a
fallback entry at the GRUB menu.
Ben.
--
Ben Hutchings
The generation of random numbers is too important to be left to chance.
- Robert Coveyou
Attachment:
signature.asc
Description: This is a digitally signed message part