* Re: + sched-force-sbin-init-off-isolated-cpus.patch added to -mm tree
[not found] <200607280451.k6S4pEY9027994@shell0.pdx.osdl.net>
@ 2006-07-28 5:19 ` Paul Jackson
0 siblings, 0 replies; only message in thread
From: Paul Jackson @ 2006-07-28 5:19 UTC (permalink / raw)
To: linux-kernel; +Cc: akpm, mm-commits, npiggin, mingo, sivanich
Nick wrote:
> Force /sbin/init off isolated cpus (unless every CPU is specified as an
> isolcpu).
For our big honkin NUMA iron SGI customers, who are serious about this
sort of thing, we actually already do this, entirely in user space. I
use the init= kernel command line boot option to run a pre-init program,
that sets up a configurable (by a config file in /etc) cpuset, then
exec's the real init in that cpuset. This feature is known to SGI
customers as the "bootcpuset."
But Nick's patch looks ok to me. It seems harmless enough for our NUMA
power users - as the bootcpuset they configured will just override what
Nick's code did a few milliseconds (to read the pre-init executable off
the disk) earlier, harmlessly. Meanwhile, the remaining (and likely
vast majority of) isolcpu users will get the behaviour they prefer.
Acked-by: Paul Jackson <pj@sgi.com>
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <pj@sgi.com> 1.925.600.0401
^ permalink raw reply [flat|nested] only message in thread