Bug 296450 – "Kernel unaligned access at TPC[address]" message when accessing CIFS mounted share

• Product: Gentoo Linux
• Status: RESOLVED UPSTREAM

[Description noybman 2009-12-10 20:13:56 UTC]

When a read/write request is issued to a mounted cifs share, the following error block is seen (in groups of 4):
Kernel unaligned access at TPC[6b17d4] tcp_packet+0xcd4/0xd00
where the above address block tcp/id varies.



Reproducible: Always

Steps to Reproduce:
1. Boot a Sparc64 system with a Gentoo minimal install build, (happens with current nightly build 12072009).
2. mkdir /mnt/win
3. mount -t cifs -o username=someuser //ip.ip.ip.ip/shar-folder-name /mnt/win
4. enter password when prompted.
5. The message is seen upon mount operation
5. do a:
   ls -al of /mnt/win   (or)
   mkdir, or touch, etc... any read/write based operation to the /mnt/win 

If (as in my case) you are doing a dd operation, or any constant read.write operation, the error message is seen in groups of 4, every second or so, until the operation is done.

Actual Results:  
You receive the described error. (repetively).

Expected Results:  
Not displayed said error message. (the actual disk operations both read and write seem not to be affected by said error, however, it does cause the user severe console issues as the message is constant.)

A google search revealed the following (debian based info), which seems to be identical, and perhaps will allow the following patch to be applied promptly:

commit cbb1dbb63271d976f82819fb5cba88f47ac90616
Author: Mark H. Weaver <[e-mail]>
Date:   Mon Mar 23 13:46:12 2009 +0100
    netfilter: nf_conntrack_tcp: fix unaligned memory access in tcp_sack
    
    [ Upstream commit 534f81a5068799799e264fd162e9488a129f98d4 ]
    
    This patch fixes an unaligned memory access in tcp_sack while reading
    sequence numbers from TCP selective acknowledgement options.  Prior to
    applying this patch, upstream linux-2.6.27.20 was occasionally
    generating messages like this on my sparc64 system:
    
      [54678.532071] Kernel unaligned access at TPC[6b17d4] tcp_packet+0xcd4/0xd00
    
    Acked-by: David S. Miller <[e-mail]>
    Signed-off-by: Patrick McHardy <[e-mail]>
    Signed-off-by: Chris Wright <[e-mail]>
 
 
patch 05/88 - netfilter: nf_conntrack_tcp: fix unaligned memory access in tcp_sack by Greg KH on 2009-04-30T17:09:40+00:00 
2.6.28-stable review patch.  If anyone has any objections, please let us know.
  [54678.532071] Kernel unaligned access at TPC[6b17d4] tcp-packet+0xcd4/0xd00
Acked-by: David S. Miller <[e-mail]>
Signed-off-by: Patrick McHardy <[e-mail]>
Signed-off-by: Greg Kroah-Hartman <[e-mail]>
 
 #include <net/tcp.h>
 
@@ -466,7 +467,7 @@ static void tcp-sack(const struct sk-buf
     for (i = 0;
          i < (opsize - TCPOLEN-SACK-BASE);
          i += TCPOLEN-SACK-PERBLOCK) {
-     tmp = ntohl(*((
 

http://linux.derkeiler.com/Mailing-Lists/Kernel/2009-02/msg01340.html

[Comment 1 Mike Pagano 2009-12-30 01:17:05 UTC]

kernel version, emerge --info

[Comment 2 George Kadianakis (RETIRED) 2009-12-30 15:40:33 UTC]

Did you try the patch you mentioned? Did it work?

[Comment 3 noybman 2010-01-20 18:37:09 UTC]

(In reply to comment #2)
> Did you try the patch you mentioned? Did it work?
> 

It is a Debian based patch so no I did not.
Also, I only have easy access to download the nightly iso's. I do not have source nor do I have a good environment to build it in.

I was hoping by providing the details of the debian patch perhaps it would illude to the location of the issue. The steps I provided should cause this issue to present itself to whomever is willing to resolve it and apply the changes to cvs. (I unfortunately can not).

[Comment 4 George Kadianakis (RETIRED) 2010-01-21 12:09:33 UTC]

Can you execute: `uname -r` in the Live CD environment?
If the kernel's version is above 2.6.28 then the patch you suggested is already applied in and you should rule it out.

[Comment 5 noybman 2010-01-21 15:01:54 UTC]

Thanks for the reply.
It is 2.6.30-gentoo-r8, from Live cd 20091207.

Can someone duplicate the steps outlined above? I have been able to consistently reproduce this error on every machine I have used.

(In reply to comment #4)
> Can you execute: `uname -r` in the Live CD environment?
> If the kernel's version is above 2.6.28 then the patch you suggested is already
> applied in and you should rule it out.

[Comment 6 noybman 2010-01-21 15:04:52 UTC]

Quick note as to not lead anyone astray, (in case it matters)...
I am bringing the machine online in an environment without DHCP, therefore I issue ifconfig eth0 ip and then ping test, and then perform the mount.

[Comment 7 Mike Pagano 2010-02-15 21:03:43 UTC]

Have you tested with 2.6.31 or 2.6.32 ?

[Comment 8 noybman 2010-03-04 14:21:33 UTC]

Yes I have. I tested it with the latest nightly build 20100301 which is kernel version 2.6.31.

(I don't know who marked this as resolved, but there was no comment so I'm reopening it until I can test a build that shows it is resolved).

(In reply to comment #7)
> Have you tested with 2.6.31 or 2.6.32 ?

[Comment 9 Mike Pagano 2010-03-04 16:23:50 UTC]

User is not using gentoo-sources. Release team, any thoughts?

[Comment 10 Raúl Porcel (RETIRED) 2010-04-16 20:46:18 UTC]

This is gentoo-sources, since he is using the installcd.

On the other hand, we can't do nothing about unaligned accesses...this should be reported to kernel upstream. 

There's nothing we can't do here.