Re: problems with 2.5.61-mm1

[Martin J. Bligh]

>> No, that's a kirq broke no_irq_balance thing (I presume this is NUMA-Q?).
> 
> Nope, it's an 8-way Summit box.
> 
> I just booted 2.5.61, and the problem still happens there, so it not
> surprisingly isn't just -mm.

Ah, OK. Sorry, "assumptions > coffee" error.
 
>> There's a bootflag option to disable it as well, but that's broken too. I
>> can't fix do it right now, but someone needs to go through and fix all
>> the disable bits so they work.
> 
> Disabling it is easy.  Any idea what might be wrong.

Yup, lots of the code assumes things are in flat logical mode, and/or
that you can target arbitrary bitmasks of CPUs ... see the fix I sent out
yesterday for smp_affinity, for instance.

M.