Re: [PATCH] for account_group_exec_runtime(), make sure ->signal can't be freed under rq->lock

[Frank Mayhar <fmayhar@google.com>]

On Tue, 2008-11-11 at 18:16 +0100, Ingo Molnar wrote:
> * Frank Mayhar <fmayhar@google.com> wrote:
> 
> > On Tue, 2008-11-11 at 11:35 +0100, Ingo Molnar wrote:
> > > * Oleg Nesterov <oleg@redhat.com> wrote:
> > > > The patch is ugly, but I don't see the better fix for now. Needs the 
> > > > review from Peter/Ingo.
> > > this is indeed too ugly, and if we do it we'll get both this ugliness 
> > > and the CPU loop upstream forever. Frank, if you dont have time to fix 
> > > this code, then i guess the best thing is to do the full revert that 
> > > Peter sent.
> > 
> > Well, at the moment I'm up to my armpits in alligators.  That said, 
> > we're going to have to pull in this code regardless, ugliness and 
> > all, since we're guaranteed to run into the soft lockup bug 
> > otherwise.  This means that I'll have strong incentive to come back 
> > and readdress the fix to remove the ugliness and address Peter's 
> > concerns.  I have no idea when that will be, however.
> 
> well, we wont leave buggy code in there for .28 - it could trigger 
> anytime on any SMP box, no matter how narrow the race is.

Sorry, by "we" I meant where I work, not the Linux kernel folks; I guess
it's as true for you guys as it is for us but I'm certainly not in a
position to speak for you.

> I've picked up the spin-wait fix from Oleg, because it fixes the bug. 
> But we should really fix the fundamental issues here too.

Agreed, in spades.  In fact I plan to track this internally so it
doesn't fall off my radar.
-- 
Frank Mayhar <fmayhar@google.com>
Google, Inc.