Re: sections mismatches: How to mark new false positives?

[Sam Ravnborg <sam@ravnborg.org>]

On Tue, Jan 29, 2008 at 07:02:55PM +0200, Adrian Bunk wrote:
> I'm getting the following in the latest -git:
> 
> <--  snip  -->
> 
> ...
> WARNING: arch/x86/kernel/built-in.o(.exit.text+0x1db): Section mismatch in reference from the function msr_exit() to the variable .cpuinit.data:msr_class_cpu_notifier
> ...
> 
> <--  snip  -->
> 
> That's obviously a false positive (unregister_hotcpu_notifier() is a 
> noop if CONFIG_HOTPLUG_CPU=n), but how can I silence the warning?

What is the purpose of __cpuinit?
It seems to be used for two purposes:

To annotate code that is used to initialize cpu's
  if CONFIG_HOTPLUG_CPU=n then discard it after
                               init has completed
  if CONFIG_HOTPLUG_CPU=y then keep it

To annotate all 'core' cpu hotplug related code
  if CONFIG_HOTPLUG_CPU=n then it is not used and
                               can safely be discarded
  if CONFIG_HOTPLUG_CPU=y then keep it

And the variable msr_class_cpu_notifier belongs to the last category.
So the root cause of all the __cpu* related section mismatch
warnings are the misuse of __cpuinit to mark all core functions.

How are we going to fix this?

I see a couple of possibilities:

1) annotate like hell to hide the misuse of __cpuinit
2) introduce __cpu to make cpu hotplug 'core' stuff
3) drop section mismatch checks for __cpu stuff


	Sam