Re: x86: mce: Please revert 22223c9b417be5fd0ab2cf9ad17eb7bd1e19f7b9

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> On Thu, 1 Oct 2009, Borislav Petkov wrote:
> > 
> > Ok, here it is, tested on two Fam10 machines here with injecting 
> > MCEs. The decoding code is now built-in by default (early_initcall 
> > requires !MODULE).
> 
> I don't think it has to require !MODULE. We could do what we do for 
> the other initcalls, ie if MODULE we turn it into just a regular 
> initcall. If that allows something like the EDAC MCE to be built as a 
> module, and people want to, then just go ahead and add the one-liner 
> to <linux/init.h>
> 
> Of course, if it _requires_ being loaded early for some other reason, 
> then that's a different issue. [...]

i think it's borderline.

The one issue that makes it nice to be core code is the fact that many 
hardware problems hit early during bootup, so having the human-readable 
decoder there has practical advantages.

I'd still like to see it nicely abstracted out, and not hardwired into 
lowlevel code. I.e. we should slowly move towards having proper chipset 
drivers in the long run. (the fact that northbridges now sit on the CPU 
die make this easier - there's fewer variations in practice.)

( One detail: if the MODULE=y case is allowed, the unregister side of 
  the callback has to be implemented. That could be .33 material i
  suspect. )

	Ingo