netdev
[Top] [All Lists]

Re: Net module Realtek 8110S/8169S

To: "markes.albert@xxxxxxxxx" <markes.albert@xxxxxxxxx>
Subject: Re: Net module Realtek 8110S/8169S
From: Francois Romieu <romieu@xxxxxxxxxxxxx>
Date: Thu, 25 Mar 2004 00:04:31 +0100
Cc: netdev <netdev@xxxxxxxxxxx>
In-reply-to: <HV3PKU$46A35E9C06E9DB9727EB43EC9861B7A1@xxxxxxxxx>; from markes.albert@xxxxxxxxx on Wed, Mar 24, 2004 at 11:10:06PM +0100
References: <HV3PKU$46A35E9C06E9DB9727EB43EC9861B7A1@xxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mutt/1.2.5.1i
markes.albert@xxxxxxxxx <markes.albert@xxxxxxxxx> :
[...]
>       Fisrt, excuse for my bad english. I have big problem with module r8169
> shipped with kernel 2.6.4. My system is a Pentium4HT, mainborad is a MSI 865PE
> Neo2-P with integrated Fast Ethernet Realtek 8110s/8169S (as I see in bios
> message). When I transfer a big file (size > 5Mb) from my system to another 
> PC,
> it freeze without any message. The

Please describe the freeze in a few words:
- transfer dead but computer is still fine;
- keyboard stuck;
- leds do not change any more.

> only messages that I see in the log are some line with text "eth0: Too much 
> work
> at interrupt!", not at same time with hang. I think is NIC module problem

Good to know.

> because  only operation I do is transfer file (with samba client) in tty (not 
> in
> X). With kernel 2.6.3 the problem isn't show.

Can you issue a 'dd if=/dev/my_hard_drive of=/dev/null bs=1024k' and
see if it performs well for, say 2 or 3 minutes ? If it works dd your file
to some other file so that it gets cached in memory (if possible) and try
the copy again.

[...]
>      With Kernel 2.6.3 I noticed only a thing, if I unplugged the cable and
> re-plug the network not respond (not link on NIC). Only if I stop and start
> then daemon, network return-up. 

It is known but your report is still welcome.

[...]
>      For nmi watchdog I don't see the option in kernel menu. If you tell me
> how... I'll try...

Processor type and features  --->
[...]
  x x[ ] Symmetric multi-processing support                               x x   
  x x[ ] Preemptible Kernel                                               x x   
  x x[*] Local APIC support on uniprocessors                              x x   

Enable as well:

Kernel hacking  --->
[...]
  x x            [*] Magic SysRq key                                      x x   

Pass "nmi_watchdog=1" through the kernel boot command line, verify that you
have "kernel.sysrq = 1" in your /etc/sysctl.conf and if something goes wrong
try to issue an alt+sysrq+t (sysrq appears as the "print" key amongst the
three keys left to the leds). Verify that alt+sysrq+t is fine _before_ the
problem appears. If you do not understand, ask again.

>      I tryed vanilla-kernel and Andrew Morton's kernel.
> Any Idea? 

1 - Let's rule out the disk first;
2 - then if kernel gets stuck and leds refuse to change state, r8169 may be
    spinning in irq context, either because there is still a off by one error
    or because something corrupted the Rx/Tx descriptors. Try again -mm with
    attached patch and see if messages pop-up (easy to apply by hand if it
    needs to).

Please send lspci, dmesg and friends off-list (tarball for instance).

--
Ueimor

Attachment: r8169-break-irq-loop.patch
Description: Text document

<Prev in Thread] Current Thread [Next in Thread>