Realtek 8169/8110s with IPv6 gives kernel oops

Dear folks,

Hope you'll have the time to read and solve this. I'm running Slackware 10.0
with the 2.4.26 kernel. On my motherboard, I have an integrated NIC: a Realtek
8110s.

When there was some network traffic, the driver "shipped" with the 2.4.26
kernel (drivers/net/r8169.o) used to kernel panic very, very often. So I
downloaded the latest driver from Realtek, compiled it and the problem was
solved.

But, as the Campus internet here is planned to have full IPv6 support and
10Gbit/s in 2005, I thought it was a good idea to load the IPv6 module and
enable IPv6 right now.

But, when I do a "ping6 ::1" (i.e. localhost) the kernel oopses. And again I
heavily suspect the NIC/driver, because my dad also running Slackware 10.0
with kernel 2.4.26 has no problems when doing a "ping6 ::1".

Does someone have the same NIC, can test his results for me or even better,
knows a patch? (Or knows folks at Realtek who can solve this? :-p)

Because I'm not running the target kernel, I can't give you a good ksymoops
report. Here it is anyway, but I guess it's not worth a thing.

Hunderds of errors and warnings like these

Error (pclose_local): read_nm_symbols pclose failed 0x100
Warning (read_object): no symbols in
/lib/modules/2.4.26/kernel/sound/acore/oss/snd-pcm-oss.o.gz

And these

Warning (compare_maps): snd symbol snd_info_create_module_entry not found in
/lib/modules/2.4.26/kernel/sound/acore/snd.o.gz.  Ignoring
/lib/modules/2.4.26/kernel/sound/acore/snd.o.gz entry

The actual report

Nov 23 16:31:03 sjorrit kernel: Unable to handle kernel NULL pointer dereference
at virtual address 0000000c
Nov 23 16:31:03 sjorrit kernel: f8ee90a5
Nov 23 16:31:03 sjorrit kernel: *pde = 00000000
Nov 23 16:31:03 sjorrit kernel: Oops: 0000
Nov 23 16:31:03 sjorrit kernel: CPU:    0
Nov 23 16:31:03 sjorrit kernel: EIP:    0010:[<f8ee90a5>]    Tainted: P 
Using defaults from ksymoops -t elf32-i386 -a i386
Nov 23 16:31:03 sjorrit kernel: EFLAGS: 00010286
Nov 23 16:31:03 sjorrit kernel: eax: 00000000   ebx: ffffffff   ecx: f3051ed0  
edx: 00000001
Nov 23 16:31:03 sjorrit kernel: esi: f3056070   edi: f305b63c   ebp: 00000000  
esp: f3051d74
Nov 23 16:31:03 sjorrit kernel: ds: 0018   es: 0018   ss: 0018
Nov 23 16:31:03 sjorrit kernel: Process ping6 (pid: 1982, stackpage=f3051000)
Nov 23 16:31:03 sjorrit kernel: Stack: f3056070 00000001 00000000 00000000
f305b63c 00000000 00000000 f305b5c8 
Nov 23 16:31:03 sjorrit kernel:        f305b580 00000000 f3051e0c f305b580
00000040 00000000 f8ef93fa f305b580 
Nov 23 16:31:03 sjorrit kernel:        f8ef9070 f3051dec f3051e0c 00000040
00000000 ffffffff 00000800 00000040 
Nov 23 16:31:03 sjorrit kernel: Call Trace:    [<f8ef93fa>] [<f8ef9070>]
[ip_rt_ioctl+307/320] [register_netdevice+108/336] [ethtool_set_eeprom+27/240]
Nov 23 16:31:03 sjorrit kernel: Code: ff 50 0c 85 c0 5a 59 0f 85 2b fa ff ff 8b
54 24 34 31 ed c7 


>>EIP; f8ee90a5 <END_OF_CODE+ad66/????>   <=====

>>ecx; f3051ed0 <_end+32c8dce4/385ace74>
>>esi; f3056070 <_end+32c91e84/385ace74>
>>edi; f305b63c <_end+32c97450/385ace74>
>>esp; f3051d74 <_end+32c8db88/385ace74>

Trace; f8ef93fa <END_OF_CODE+1b0bb/????>
Trace; f8ef9070 <END_OF_CODE+1ad31/????>

Code;  f8ee90a5 <END_OF_CODE+ad66/????>
00000000 <_EIP>:
Code;  f8ee90a5 <END_OF_CODE+ad66/????>   <=====
   0:   ff 50 0c                  call   *0xc(%eax)   <=====
Code;  f8ee90a8 <END_OF_CODE+ad69/????>
   3:   85 c0                     test   %eax,%eax
Code;  f8ee90aa <END_OF_CODE+ad6b/????>
   5:   5a                        pop    %edx
Code;  f8ee90ab <END_OF_CODE+ad6c/????>
   6:   59                        pop    %ecx
Code;  f8ee90ac <END_OF_CODE+ad6d/????>
   7:   0f 85 2b fa ff ff         jne    fffffa38 <_EIP+0xfffffa38>
Code;  f8ee90b2 <END_OF_CODE+ad73/????>
   d:   8b 54 24 34               mov    0x34(%esp),%edx
Code;  f8ee90b6 <END_OF_CODE+ad77/????>
  11:   31 ed                     xor    %ebp,%ebp
Code;  f8ee90b8 <END_OF_CODE+ad79/????>
  13:   c7 00 00 00 00 00         movl   $0x0,(%eax)


500 warnings and 31 errors issued.  Results may not be reliable.

Hope some people can help me so IPv6 will run smooth and happy.

Yours,

Jorrit

P.S. Is it bad, when linux crashes and the filesystem check returns
"..inode...dtime of 0.. fixed" kind of stuff? (This is probably answered many
times before...)

Can you reproduce the bug without the tainting module loaded (loaded then unloaded count as loaded) and resend the decoded oops ?

--
Ueimor

I guess I can, but I think it wouldn't make a difference because I've been stupid..

The situation is as follows: because Slackware had no SATA support, I use a precompiled kernel from "bonecrusher". But I have taken the standard *.o modules from the Slackware CD instead of compiling my own kernel and my own modules.

So then I guess the results will always be unreliable. If you think it is still useful to do what you suggested (the tainted module is the nvidia module), then please let me hear.

Till now I haven't had time to compile my own kernel, but maybe I can do that in the Christmas Holidays.

Thanks anyway. From the other topic/post I see you can read the r8169 module source very well. I think it is coded horrible (few comments, bad code style). Technically I don't know how good the module is, but I guess not very good... (seeing all the problems that arise )

Greets,

Jorrit

Subject says it all. It includes link handling changes, ethtool support, SG and Rx/Tx checksums and large frames up to 7200 bytes amongst others.

Patch against 2.4.28:
http://www.fr.zoreil.com/people/fran...9.c-test.patch

For the interested reader, the serie of patch which composes the patch above is available at:
http://www.fr.zoreil.com/linux/kerne.../2.4.28/r8169/

Please test and report status. Cc: on netdev@oss.sgi.com is welcome.

---
Ueimor

I haven't tested the driver you provided yet, because I have some questions:

- Is the driver you provide better than the official Realtek 2.2 driver?
- In what way is your driver different from the official Realtek 2.2 driver?
- Who maintains the 8169/8110 driver in the Linux kernel?

So, the real question is: should I replace my Realtek 2.2 driver by the one you provided in the link?

Thanks again for all help,

Jorrit

> - Is the driver you provide better than the official Realtek 2.2 driver?

Of course )

Realtek merged some of the changes which went into the vanilla version.
They missed some fixes though: check rtl8169_tx_interrupt and ask
yourself what prevents "entry" to overflow. The fix has been in the kernel
version of the driver for ages.

> - In what way is your driver different from the official Realtek 2.2 driver?

- community supported;
- vendor neutral (alas / );
- NAPI, TSO, TX SG and TX csum, VLAN, fancy lockless TX;
- buzzword-compliant (TM);
- audited from time to time;
- evolving through incremental step;
- sensible kernel CodingStyle (for what it means).

Btw, it is not _my_ driver: many people contributed to it.

> - Who maintains the 8169/8110 driver in the Linux kernel?

$ less +/8169 linux-2.6.9/MAINTAINERS
[...]
8169 10/100/1000 GIGABIT ETHERNET DRIVER
P: Francois Romieu
M: romieu@fr.zoreil.com
L: netdev@oss.sgi.com
S: Maintained

> So, the real question is: should I replace my Realtek 2.2 driver by the one you provided in the link ?

Your choice. At least my 2.4.xx testboxes are happy with it.

--
Ueimor