Re: [Bug #10815] 2.6.26-rc4: RIP find_pid_ns+0x6b/0xa0

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Sun, Jun 15, 2008 at 11:17:10AM -0700, Paul E. McKenney wrote:
> On Sun, Jun 15, 2008 at 08:21:50PM +0400, Alexey Dobriyan wrote:
> > On Sat, Jun 14, 2008 at 08:30:01PM -0700, Paul E. McKenney wrote:
> > > On Sat, Jun 14, 2008 at 11:43:38PM +0400, Alexey Dobriyan wrote:
> > > > On Sat, Jun 14, 2008 at 11:12:12AM -0700, Paul E. McKenney wrote:
> > > > > On Sat, Jun 14, 2008 at 06:58:39PM +0400, Oleg Nesterov wrote:
> > > > > > On 06/14, Paul E. McKenney wrote:
> > > > > > >
> > > > > > > On Fri, Jun 13, 2008 at 04:52:55PM +0300, Adrian Bunk wrote:
> > > > > > > > On Sat, Jun 07, 2008 at 10:42:57PM +0200, Rafael J. Wysocki wrote:
> > > > > > > > > This message has been generated automatically as a part of a report
> > > > > > > > > of recent regressions.
> > > > > > > > > 
> > > > > > > > > The following bug entry is on the current list of known regressions
> > > > > > > > > from 2.6.25.  Please verify if it still should be listed.
> > > > > > > > > 
> > > > > > > > > 
> > > > > > > > > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=10815
> > > > > > > > > Subject		: 2.6.26-rc4: RIP find_pid_ns+0x6b/0xa0
> > > > > > > > > Submitter	: Alexey Dobriyan <adobriyan@gmail.com>
> > > > > > > > > Date		: 2008-05-27 09:23 (12 days old)
> > > > > > > > > References	: http://lkml.org/lkml/2008/5/27/9
> > > > > > > > > Handled-By	: Oleg Nesterov <oleg@tv-sign.ru>
> > > > > > > > > 		  Linus Torvalds <torvalds@linux-foundation.org>
> > > > > > > > > 		  Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > > > > > > > Patch		: http://lkml.org/lkml/2008/5/28/16
> > > > > > > > 
> > > > > > > > What happened with this issue?
> > > > > > > 
> > > > > > > The patch listed above works for me, passes rcutorture, &c.  However,
> > > > > > > I never have been able to reproduce the original problem, so cannot say
> > > > > > > whether it qualifies as a fix.
> > > > > > 
> > > > > > I doubt very much RCU was the reason of this problem.
> > > > > 
> > > > > Although I very much appreciate your confidence in my code, it is new
> > > > > code, so therefore under suspicion.
> > > > > 
> > > > > > Alexey, how did you trigger this problem?
> > > > > 
> > > > > One of them involved running LTP while doing 170 kernel builds in
> > > > > parallel.
> > > > 
> > > > My gut feeling is that find_pid_ns oops, __d_lookup oops and
> > > > __call_for_each_cic oops are the same bug.
> > > > 
> > > > And rcutorture failures I've mentioned to Paul privately.
> > > 
> > > Yep, running rcutorture in parallel with LTP, which didn't reproduce
> > > for me either.
> > > 
> > > Did the patch at http://lkml.org/lkml/2008/5/28/16 help?
> > > 
> > > > Oleg, debugging you've posted never triggered.
> > > > 
> > > > kerneloops suggests that I'm alone. :-(
> > > 
> > > Assuming that the above patch didn't help...  As a desperation measure,
> > > I could suggest the following patch.
> > 
> > > --- linux-2.6.26-rc4/kernel/rcupreempt.c
> > > +++ linux-2.6.26-rc4-alexey/kernel/rcupreempt.c
> > > @@ -77,7 +77,7 @@
> > >   *
> > >   * GP in GP_STAGES stands for Grace Period ;)
> > >   */
> > > -#define GP_STAGES    2
> > > +#define GP_STAGES    3
> > >  struct rcu_data {
> > >  	spinlock_t	lock;		/* Protect rcu_data fields. */
> > >  	long		completed;	/* Number of last completed batch. */
> > 
> > Both patches (independently) do not help with rcutortures failures:
> > 
> > [   58.968404] rcu-torture:--- Start of test: nreaders=4 nfakewriters=4 stat_interval=0 verbose=0 test_no_idle_hz=0 shuffle_interval = 5
> > [  159.044524] rcu-torture: rtc: 0000000000000000 ver: 53859 tfle: 0 rta: 53859 rtaf: 18 rtf: 53797 rtmbe: 0
> > [  159.044527] rcu-torture: !!! Reader Pipe:  65565142 4275 1 0 0 0 0 0 0 0 0
> > [  159.044529] rcu-torture: Reader Batch:  65564196 5207 7 3 1 1 0 1 1 0 1
> > [  159.044530] rcu-torture: Free-Block Circulation:  53858 53853 53846 53843 53834 53825 53816 53808 53803 53797 0
> > [  159.044976] rcu-torture:--- End of test: FAILURE: nreaders=4 nfakewriters=4 stat_interval=0 verbose=0 test_no_idle_hz=0 shuffle_interval = 5
> 
> And the repeat-by is simply running LTP in parallel with rcutorture?
> This is a one-hour run of rcutorture or thereabouts?

I also tried running LTP in parallel with rcutorture on POWER, and I
cannot reproduce on that platform either.

							Thanx, Paul