Re: ia64 SN2 - migration costs: 1) nearly zero 2) BUG 3) repeated

[Ingo Molnar <mingo@elte.hu>]

* Paul Jackson <pj@sgi.com> wrote:

> I turned on migration_debug=3 at boot, enabled CONFIG_PREEMPT and some 
> other CONFIG DEBUG options, and captured the output from the tty 
> console instead of via the /var/log/messages file.  This resulted in 
> the following.  Now I don't see the BUG, and the cost matrix makes 
> more sense (not all 0's), and the matrix output is not repeated.  I 
> don't understand why.  Feel free to ask me to try various 
> combinations, Ingo, if this doesn't make sense to you either.

hm, these calibration results looks quite sane. Could you try 
migration_debug=3 with the .config that failed before? (if it doesnt 
fail, could you re-check it still fails with migration_debug=2?) Zero 
results in the cost matrix usually are a sign of time resolution 
problems - but that shouldnt be an issue on ia64 which should have a 
globally synced TSC, right?

until these problems are solved, you should be able to work around it by 
booting with the migration_cost=0,7512036,45850376 boot option.

you can work around the warning message via changing the:

	int cpu1 = -1, cpu2 = -1, cpu, orig_cpu = smp_processor_id();

line to:

	int cpu1 = -1, cpu2 = -1, cpu, orig_cpu = raw_smp_processor_id();

i'll do the correct fix.

	Ingo