On Tue, Jul 17, 2012 at 10:22:55AM -0700, Stefan Bader wrote:
> On 07/03/2012 08:17 AM, Herton Ronaldo Krzesinski wrote:
> >On Tue, Jul 03, 2012 at 03:01:58PM +0200, Stefan Bader wrote:
> >>Though reproducing this somehow required (or at least made much
> >>more likely) a certain setup (two CPUs and the described loop of
> >>a certain command), the result (crash) is probably bad enough to
> >>warrant applying it before it appears upstream. Peter queued it
> >>up but needs Ingo to push it onwards.
> >>Also this may cause other subtle to find issues as the current
> >>code calls task_group() four times to assign values to cfs and
> >>rt schdeduler elements. So while the observed crash happened when
> >>cfs elements were inconsistent, there could be other problems when
> >>the inconsistency hits cfs/rt or rt only.
> >>Quantal still has the same race but at least the code was changed
> >>to call task_group only once. So the value could be old but at least
> >>it is always old (or the new one).
> >>This backport should apply to Natty till Precise (maybe a bit of
> >>fuzz in Natty). But later that 3.3 all the files moved around.
> The patch at least made it into linux-next as
> commit 7fac251a97f36c9aef31b08ce7ad6ef8f06e54d1
> Author: Peter Zijlstra <email@example.com>
> Date: Fri Jun 22 13:36:05 2012 +0200
> sched: Fix race in task_group()
> I would like to see the backport I submitted to go into our kernels
> as soon as possible. Upstream unlikely is in a hurry (so I would
> think it gets there not before 3.6) because the issue is made
> covered up there.
Yes, looks sensible to get this now.
kernel-team mailing list