Re: [Patch] sched: new sched domain for representing multi-core

[Andrew Morton <akpm@osdl.org>]

"Siddha, Suresh B" <suresh.b.siddha@intel.com> wrote:
>
> On Tue, Jan 31, 2006 at 05:12:16PM -0800, Andrew Morton wrote:
>  > It's still not clear what's supposed to be happening here.
>  > 
>  > In build_sched_domains() we still have code which does:
>  > 
>  > 
>  > 	for_each_cpu_mask(...) {
>  > 		...
>  > #ifdef CONFIG_SCHED_MC
>  > 		...
>  > #endif
>  > #ifdef CONFIG_SCHED_SMT
>  > 		...
>  > #endif
>  > 		...
>  > 	}
>  > 	...
>  > #ifdef CONFIG_SCHED_SMT
>  > 	...
>  > #endif
>  > 	...
>  > #ifdef CONFIG_SCHED_MC
>  > 	...
>  > #endif
>  > 
>  > So in the first case the SCHED_SMT code will win and in the second case the
>  > SCHED_MC code will win.  I think.  
> 
>  I am not sure what you mean here. At all the above pointed places, both 
>  MC and SMT will win if both are configured.

I was assuming that the code really does something like:

#ifdef CONFIG_SCHED_MC
		some_global_thing = <expr>
#endif
#ifdef CONFIG_SCHED_SMT
		some_global_thing = <expr>
#endif
	}
	...
#ifdef CONFIG_SCHED_SMT
	some_other_global_thing = <expr>
#endif
#ifdef CONFIG_SCHED_MC
	some_other_global_thing = <expr>
#endif

Which, looking a bit closer, was wrong (yes?)

It is a bit irregular that in one place we do the SMT processing first and
in another we do the MC processing first, but I guess it'll work OK.

We do need to be super-careful in the reviewing and testing here.  If we
slip up we won't have a nice crash to tell us.  Instead we'll find that
some machines with some configs will, under some workloads, take a few
percent longer than they should.  We could waste people's time for years
until some developer stumbles across something.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/