Re: [2.6.14-rc1/sparc54]: BUG: soft lockup detected on CPU#0!

[Tomasz Kłoczko]

On Thu, 15 Sep 2005, David S. Miller wrote:

> From: Tomasz Kłoczko <kloczek@xxxxxxxxxxxxxxxxxx>
> Date: Thu, 15 Sep 2005 19:40:27 +0200 (CEST)
>
> > I'm just catch series of kernel messages with soft lockup detected reports
> > (in attachment).
> > It occures during store big amout of data on NFS volume.
> >
> > As NIC I use now Sun Swift gigabit eth (cassini driver). Probably this is
> > NFS related because I'm just browse yesterday logs and simillar was also
> > on Sun Happy Meal.
>
> Interesting.  Can you reproduce this with SLAB poisioning disabled?
> That debugging feature is extremely expensive, although it shouldn't
> make the CPU stop scheduling processes for more than 10 seconds.
>
> I wonder if the NFS daemon code needs to have some limits put on
> how much cpu it consumes handling requests before it gives up the
> cpu.  Perhaps, it has such throttling already, I don't know.

But this not case NFS server but NSF client. During this lookups I observe 
rpciod takes 90-99% time of single processor. Load is between 10 and 20.

> I'll also try to see if there can be some kind of sparc64 specific
> issue which would cause this.
>
> Where did you get that Cassini driver btw?  It's not upstream,
> although if it exists it should be.

It is avalaible on Sun pages (Copyright by Sun by it is GPLed):

http://www.sun.com/download/index.jsp?cat=Hardware%20Drivers&tab=3

Few days ago I'm talk about this driver with spot on #aurora channel and 
it was included during prepare next kernel package.

But continue ..

Yesterday before testing NFS I was trying utilize NIC is using only 
ftp/http/rcync/scp and all works correctly (without reporting lookups). 
~23:00 CET after reported series of lookups system hangs completly. After 
restart to now I see in dmesg some known messages:

[root@boss]# dmesg | sort | uniq -c | sort -n | tail -n 2
     87 svc: bad direction 268435456, dropping request   <==========
    285 hw tcp v4 csum failed

Also I have question about second (hw tcp v4 csum failed):

[root@boss]# ifconfig eth0
eth0      Link encap:Ethernet  HWaddr 00:03:BA:18:41:F9
          inet addr:153.19.33.230  Bcast:153.19.33.255  Mask:255.255.255.0
          inet6 addr: fe80::203:baff:fe18:41f9/64 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:2361199 errors:0 dropped:0 overruns:0 frame:0
          TX packets:4060978 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:352428791 (336.1 Mb)  TX bytes:4884521951 (4658.2 Mb)
          Interrupt:192

As you see in both errors is "0". Is this correct reporting broken packets 
in kernel messages instead in RX errors ?

kloczek
--
-----------------------------------------------------------
*Ludzie nie mają problemów, tylko sobie sami je stwarzają*
-----------------------------------------------------------
Tomasz Kłoczko, sys adm @zie.pg.gda.pl|*e-mail: kloczek@xxxxxxxxxxxxxxxxxx*