Oh...forgot to mention where this line was at :-)  rtl8169_start_xmit is the function...

On Tue, 25 May 2004 23:08:31 -0500
Ray Cole [off-list ref] wrote:

I found the problem - I think.  At least I'll say I found a line of code that I modified and it started working :-)

Here is the offending line:

		tp->TxDescArray[entry].status = cpu_to_le32(OWNbit | FSbit |
			LSbit | len | (EORbit * !((entry + 1) % NUM_TX_DESC)));

Here is what I changed it into (based on the 2.6.5 flavor of r8169.c):

		if (entry != (NUM_TX_DESC - 1))
			tp->TxDescArray[entry].status =
			    (OWNbit | FSbit | LSbit) | len;
		else
			tp->TxDescArray[entry].status =
			    (OWNbit | EORbit | FSbit | LSbit) | len;

Once I changed this it started working great.  I picked up on this when I noticed it was only sending 64 packets then failing.  I looked for differences in the way 2.6.5 and 2.6.6 handled a full transmit ring and this line looked suspicious...I figure it is perhaps a difference in the compiler I'm using (2.95.3) vs what others using this module are using...perhaps?

-- Ray		


On Tue, 25 May 2004 09:19:15 +0200
Francois Romieu [off-list ref] wrote:

quoted

Ray Cole [off-list ref] :
[...]

quoted

I was wondering if you were aware of such issues in the 2.6.6 flavor of
r8169.  If not, I'm more than willing to provide any information you need.

Nope, you are the first one. If time permits today, I'll check if the mobo
appears on the usual suspects list.

quoted

 Either way I'm more than willing to try any patches for it.

Nice :o)

quoted

Attach is my config for the kernel in case that helps.  Oh, BTW, I don't
actually have a gigbit network - running a 100M network.  I don't know if
that makes any difference or not.  I also don't see any error messages when

It may make things simpler.

quoted

the network stops working (aside from my application, which simply reports
a timeout error communicating with the POP email server :-)

Some more information may help. Typically:
- lspci -vx output
- dmesg after boot and insertion of the module (usually a combination of
  /var/log/dmesg and 'dmesg' output as this one may be truncated)
- lsmod output
- /proc/interrupts contents before/after foobar
- which (if any) was the latest known working kernel, be it 2.4.x or 2.6.x
  (2.5.5 apparently) ?
- can you enumerate all the dysfunctionnal kernel ?
- which distribution/compiler ?
- does it still fail if you do a simple 'ping' for an extended period of
  time instead of pop (same thing with lengthy ftp/http transfer, say a file
  of a few megs) ?

A few things you may want to try:
- disable ACPI (acpi=none on the boot cmd line, do not confuse it with APIC)
- PREEMPT may change a few things though it is less on the radar than ACPI
- http://www.fr.zoreil.com/people/francois/misc/20040525-2.6.7-rc1-r8169.c-test.patch applied on 2.6.6 (some link related things may come into play but
  I am a bit sceptical)
- as the system does not hang (only the network trafic), can you see if
  it makes a difference if you
  - ifconfig down/up the network card ?
  - same thing + removal/insertion of the driver ?
  - unplug/plug again the cable (wait for 10~15 seconds then) ?
  -> dmesg output welcome.

Please Cc: future traffic to netdev@oss.sgi.com as well (if you can tell
sylphid to cut the lines at 72/80 chars, I will not complain either :o) ).

Do not hesitate to ping me if you feel like I forget your problem (I try
to avoid it but it may happen).

--
Ueimor