Re: 2.6.8.1-mm4 - more cpu hotplug breakage

[Nathan Lynch <nathanl@austin.ibm.com>]

On Mon, 2004-08-23 at 17:18, Nathan Lynch wrote: 
> Hi-
> 
> In both 2.6.8.1-mm3 and -mm4, a task which tries to offline a cpu
> becomes permanently hung, seemingly because the high-priority per-cpu
> kernel threads which __stop_machine_run starts never get to run.  I did
> a bisection search on the 2.6.8.1-mm4 patch series and narrowed it down
> to one of these three patches, listed below along with their behavior
> when trying to offline a cpu:
> 
> 191: sched-smt-fixes.patch     - panic
> 192: sched-smt-fixes-fix.patch - panic
> 193: nicksched.patch           - process hangs

In case anyone cares, commenting out the sys_sched_setscheduler() call
in stop_machine() allows the kstopmachine threads to run and the cpu
goes offline.  However, there's still something wrong as
__stop_machine_run never returns from wait_for_completion.

Any ideas here?  Zwane and I are kinda stuck using 2.6.8.1-mm2 for cpu
hotplug work until we get this sorted out.

Nathan