[Cbe-oss-dev] CONFIG_SCHED_SMT

[Hiroyuki Machida]

Hi Ben,

Thank you for explanation. I think cpu_sibling_map[] will be initialized
according to cpu_spec. It's not platform specific, but common for PPC64.
In case of Cell, SMT flag is set and sibling map is expected to set correctly
in all Cell platform.  So it's better PS3 deconfig follows Cell defconfig.

Regards,
Hiroyuki.


2007/8/12, Benjamin Herrenschmidt <benh at kernel.crashing.org>:
> On Thu, 2007-08-09 at 12:40 +0900, Hiroyuki Machida wrote:
> > Hi Geoff and Arnd.
> >
> > I'm wondering if it is better to turn on CONFIG_SCHED_SMT or not, in
> > Cell B.E. platforms.
> > I realized that it's turned off in ps3defconfig, but turned on in
> > cell_defconfig.
> > I know Cell B.E has SMT but, I'm not sure how CONFIG_SCHED_SMT  works
> > with Cell B.E
> > in details.
>
> The main question is whether the PS3 arch code properly initializes the
> sibling maps ... If yes, then CONFIG_SHED_SMT is a good idea as it will
> make the scheduler avoid the second thread unless the primary is full,
> though I don't expect that to make much difference on a single core
> machine.
>
> Ben
>
>
>