lkml.org 
[lkml]   [2012]   [Jan]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: RCU qsmask !=0 warnings on large-SMP...
On Wed, Jan 25, 2012 at 10:14:33PM +0100, Steffen Persvold wrote:
> On 1/25/2012 19:14, Paul E. McKenney wrote:
> []
> >
> > So, it would be very interesting to add the values rdp->mynode->gpnum
> > and rdp->mynode->completed to your list, perhaps labeling them something
> > like "rng" and "rnc" respectively.
> >
>
> I added them to the printout. This time I ran with NR_CPUS=512 so there's only two levels but we see more qsmask bits set on the root node :
>
> [ 738.329672] CPU 48, treason uncloaked, rsp @ ffffffff81a1cd80 (rcu_sched), gpnum=10568, completed=10567, n_force_qs=69, n_force_qs_lh=0, n_force_qs_ngp=0, rnp @ ffffffff81a1cd80, qsmask=0x1f
> [ 738.330137] 0 ffff8803f840d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=66 ql=1 qs=..W. b=10 ci=158068 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 12 ffff880bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=1 ri=0 ql=0 qs=.... b=10 ci=715 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 24 ffff8813d040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=0 ql=0 qs=.... b=10 ci=484 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 36 ffff881bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=1 ri=0 ql=0 qs=.... b=10 ci=369 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 48 ffff8823d040d660 c=10567 g=10567 pq=1 pgp=10567 qp=0 of=0 ri=0 ql=28 qs=.RWD b=10 ci=9292 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 60 ffff882bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=1 ql=0 qs=.... b=10 ci=32 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 72 ffff8833d040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=0 ql=0 qs=.... b=10 ci=43 co=0 ca=0 rng=10568 rnc=10567

Yow!!! This is the message from the root rcu_node, right?

One way to check is to print (rnp - &rsp->node[0]), which will be zero
for the root node.

So this is very very strange. This error is happening out of
rcu_start_gp(), which updates rsp->gpnum, and then updates each rnp->gpnum
in rcu_node array order. When you get the error on the root rcu_node
structure, the new ->gpnum value should not have yet propagated to the
leaf rcu_node structures. Or is the error print happening asynchronously?

If it is happening asynchronously, so that the actual printing of the
first set of messages above isn't happening until after initialization
is complete, could you please copy the relevant values out so that the
printout is consistent with the state at the time of the error?

Thanx, Paul

> [ 738.330137] ------------[ cut here ]------------
> [ 738.330137] WARNING: at kernel/rcutree_plugin.h:1011 rcu_preempt_check_blocked_tasks+0x27/0x30()
> [ 738.330137] Hardware name: H8QI6
> [ 738.330137] Modules linked in: rcutorture
> [ 738.330137] Pid: 4611, comm: rcu_torture_rea Not tainted 3.2.1-numaconnect10+ #68
> [ 738.330137] Call Trace:
> [ 738.330137] <IRQ> [<ffffffff810bb327>] ? rcu_preempt_check_blocked_tasks+0x27/0x30
> [ 738.330137] [<ffffffff8106f47b>] warn_slowpath_common+0x8b/0xc0
> [ 738.330137] [<ffffffff8106f4c5>] warn_slowpath_null+0x15/0x20
> [ 738.330137] [<ffffffff810bb327>] rcu_preempt_check_blocked_tasks+0x27/0x30
> [ 738.330137] [<ffffffff810bb43d>] rcu_start_gp+0x10d/0x1b0
> [ 738.330137] [<ffffffff810bc04b>] __rcu_process_callbacks+0x8b/0xd0
> [ 738.330137] [<ffffffff810bc8b0>] rcu_process_callbacks+0x20/0x40
> [ 738.330137] [<ffffffff810757fd>] __do_softirq+0x9d/0x140
> [ 738.330137] [<ffffffffa0001510>] ? rcu_torture_shuffle+0x80/0x80 [rcutorture]
> [ 738.330137] [<ffffffff815d982c>] call_softirq+0x1c/0x30
> [ 738.330137] [<ffffffff8103451a>] do_softirq+0x4a/0x80
> [ 738.330137] [<ffffffff81075b73>] irq_exit+0x43/0x60
> [ 738.330137] [<ffffffff8104aee5>] smp_apic_timer_interrupt+0x45/0x60
> [ 738.330137] [<ffffffffa0001c60>] ? rcu_sync_torture_deferred_free+0xd0/0xd0 [rcutorture]
> [ 738.330137] [<ffffffff815d834b>] apic_timer_interrupt+0x6b/0x70
> [ 738.330137] <EOI> [<ffffffff815d4d79>] ? __schedule+0x349/0x710
> [ 738.330137] [<ffffffff810618d5>] ? update_curr+0x85/0xd0
> [ 738.330137] [<ffffffff8107b816>] ? lock_timer_base+0x36/0x70
> [ 738.330137] [<ffffffff8107ba92>] ? mod_timer+0xf2/0x1d0
> [ 738.330137] [<ffffffffa0001510>] ? rcu_torture_shuffle+0x80/0x80 [rcutorture]
> [ 738.330137] [<ffffffff815d541a>] schedule+0x3a/0x60
> [ 738.330137] [<ffffffffa0001640>] rcu_torture_reader+0x130/0x230 [rcutorture]
> [ 738.330137] [<ffffffffa0001dc0>] ? rcu_torture_writer+0x160/0x160 [rcutorture]
> [ 738.330137] [<ffffffffa0001510>] ? rcu_torture_shuffle+0x80/0x80 [rcutorture]
> [ 738.330137] [<ffffffff8108a716>] kthread+0x96/0xa0
> [ 738.330137] [<ffffffff815d9734>] kernel_thread_helper+0x4/0x10
> [ 738.330137] [<ffffffff8108a680>] ? kthread_stop+0x70/0x70
> [ 738.330137] [<ffffffff815d9730>] ? gs_change+0xb/0xb
> [ 738.330137] ---[ end trace e8e520cce35c7626 ]---
> [ 738.330137] CPU 48, treason uncloaked, rsp @ ffffffff81a1cd80 (rcu_sched), gpnum=10568, completed=10567, n_force_qs=70, n_force_qs_lh=250, n_force_qs_ngp=0, rnp @ ffffffff81a1d180, qsmask=0x1
> [ 738.330137] 0 ffff8803f840d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=67 ql=1 qs=..W. b=10 ci=158068 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 12 ffff880bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=1 ri=1 ql=1 qs=N... b=10 ci=715 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 24 ffff8813d040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=1 ql=1 qs=N... b=10 ci=484 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 36 ffff881bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=1 ri=1 ql=1 qs=N... b=10 ci=369 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 48 ffff8823d040d660 c=10567 g=10567 pq=1 pgp=10567 qp=0 of=0 ri=0 ql=28 qs=.RWD b=10 ci=9292 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 60 ffff882bd040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=1 ql=0 qs=.... b=10 ci=32 co=0 ca=0 rng=10568 rnc=10567
> [ 738.330137] 72 ffff8833d040d660 c=10567 g=10568 pq=1 pgp=10568 qp=0 of=0 ri=0 ql=0 qs=.... b=10 ci=43 co=0 ca=0 rng=10568 rnc=10567
>
> Kind regards,
> --
> Steffen Persvold, Chief Architect NumaChip
> Numascale AS - www.numascale.com
> Tel: +47 92 49 25 54 Skype: spersvold
>



\
 
 \ /
  Last update: 2012-01-25 22:37    [W:0.264 / U:0.176 seconds]
©2003-2014 Jasper Spaans. Advertise on this site