Re: [PATCH] x86/mce: Don't unregister CPU hotplug notifier in error path

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 06/20/2014 11:58 AM, Borislav Petkov wrote:
> On Fri, Jun 20, 2014 at 11:41:27AM -0400, Boris Ostrovsky wrote:
>> Only in the sense that on Xen misc_register() often fails. But any
>> failure on baremetal will result in the same behavior.
> Ok, thanks for explaining the details.
>
>> Right. And I think we shouldn't because we leave undeleted timers.
> I wonder if we could simply move the oneliner:
>
> 	__register_hotcpu_notifier(&mce_cpu_notifier);
>
> to mcheck_init()? We don't need any locking there because we're pre-SMP
> then and from looking at notifier_chain_register(), it seems there's
> nothing special that needs initializing for the call to be too early -
> we're simply adding the callback to a list...
>
> Hmm, because if that would work, the fix is almost trivial. :)

That's cleaner than what I had in mcheck_cpu_init() (checking for 
boot_cpu_data).

But I think you still need to do the dance in the notifier to make sure 
you are not trying to add/remove device if mcheck_init_device() had 
failed earlier.

-boris