Re: [PATCH] x86 NUMA panic compile error

[Andi Kleen <ak@suse.de>]

On Monday 15 May 2006 19:53, Ingo Molnar wrote:
> 
> * Andy Whitcroft <apw@shadowen.org> wrote:
> 
> >  	if (use_cyclone == 0) {
> >  		/* Make sure user sees something */
> > -		static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else."
> > +		static const char s[] __initdata = "Not an IBM x440/NUMAQ. Don't use i386 CONFIG_NUMA anywhere else.";
> >  		early_printk(s);
> >  		panic(s);
> >  	}
> 
> i still strongly oppose the original Andi hack... numerous reasons were 
> given not to apply it (it's nice to simulate/trigger rarer features on 
> mainstream hardware too, and this ability to boot NUMA on my flat x86 
> testbox found at least one other NUMA bug already). Furthermore, the 
> crash i reported was fixed by the NUMA patchset. Andrew, please drop:

The problem is that it's not regularly used on a wide range
of boxes so it will eventually break again. We had this cycle several
times already.

It's also missing a lot of the workarounds for broken SRATs that
are needed for many of the existing NUMA systems.

If there's consensus i386 NUMA is useful I can drop it, but I predict
it will just eventually break again.

>   x86_64-mm-i386-numa-summit-check.patch
> 
> (which has nothing to do with x86_64 anyway)

I have a lot of i386 or combined i386/x86-64 patches in my tree - just Andrew's 
merge script doesn't pick that up.

-Andi