On Fri, 28 Apr 2006, Chandra Seetharaman wrote:
> - if we are ok with a loss of a kbyte or two, 2.6.17 is fine as is
> (with my incorrect patches in).
> - if we want to save that memory, we can revert the two patches and fix
> xfs to make the register calls only when hotplug cpu is defined. This
> change is also minimal. It is a step in the right direction.
> Only downside i can see in reverting my patch is that if there is any
> other modules that are doing the same as what xfs was doing, we might
> trip in a similar oops.
Once register_cpu_notifier is placed in an init section, everything should
be okay. If some other module does _exactly_ what xfs did, it won't oops
-- instead the module will get an unresolved symbol error whenever someone
tries to insmod it, because the register_cpu_notifier symbol won't be
defined. I think this is an appropriate kind of failure mode.
However, it wouldn't hurt to add some comments to the definition and
declaration of register_cpu_notifier, explaining the circumstances in
which it should be used.