Re: TCP stuck (RH 2.2.12-20 <-> 2.2.13)

[Jeff Garzik]

David Woodhouse wrote:
> I left my stuck connexion overnight, and it came back to life almost exactly
> two hours after I last sent it data from my end.
> 
> The offending machine is running the RH 2.2.12-20 UP kernel on a single
> Pentium, and the interface it's using for this traffic is
> eth1: RealTek RTL8139 Fast Ethernet at 0x6600, IRQ 11, 00:e0:7d:02:bf:f2.
> 
> I'm not sure exactly what board it is:
> 00:11.0 Ethernet controller: Realtek Semiconductor Co., Ltd. 8139 (rev 10)
>         Subsystem: Unknown device 10ec:8139

I cannot claim any knowledge of the TCP code (and your tcpdump), but
with regards to this net driver, there is a known problem where it stops
receiving packets.
	ifdown eth0 ; ifup eth0
seems to cure this problem if you don't want to wait 2 hours :)

Does this occur in the latest 2.3.x 8139too driver?
Does this occur in Becker's latest test rtl8139 driver?

This problem seems easily reproduced by ping-flooding the machine in
question.

If you do test with the 8139too driver, sending me 'working' and 'dead'
snapshots of the 8139 registers would be awesome.  You can get these
from

	rtl8139-diag -mmmaaavvvf

-- 
Jeff Garzik         | The ultimate Windows 98 keyboard:
Building 1024       | Ctrl, Alt + Del on one big key!
MandrakeSoft, Inc.  | 	- another cute fortune