On Wed, 11 Feb 2009, Connie Sieh wrote:

> There are lots of reports of issues like this for tg3 but most reports end in 
> "did not hear from requester so closing" or "Fedora Core X is not supported 
> any more so closing".  Does anyone know of a real solution to this problem. 
> ...
> We have been experiencing intermittent network failures on systems running 
> SLF47 2.6.9-78.0.1 kernel/tg3 3.86 driver .  They waiting for the tcp to 
> finish, which never happens.  The failures are load-related.
> 01:05.1 Ethernet controller: Broadcom Corporation NetXtreme BCM5704 Gigabit 
> Ethernet (rev 10)

We have only 3 systems in the Tier-1 centre requiring the tg3 driver;
our card model is:
   Broadcom Corporation NetXtreme BCM5721 Gigabit Ethernet PCI Express (rev 11).
We have never seen a problem with it (running SL 4.7  2.6.9-78.0.8.ELsmp)

> ...
> The hints we have got thus far is that the new driver v3.86 is
> trying to implement one of the tcp offload features on the NIC
> which in the case of the BCM5704 is broken.

I came across this link:
http://www.linuxquestions.org/questions/linux-networking-3/any-type-of-bittorrent-downloads-kills-linux-network-476368/?s=924f03df77c95c1f9d769fef0253f55a

  - where they suggested turning off TSO (assuming it is turned on).  But I do
not think it is turned on by default; at least on our 3 boxes it is off by
default (but I suppose the driver turns on/off features depending on the model):

# ethtool -k eth0
Offload parameters for eth0:
Cannot get device udp large send offload settings: Operation not supported
Cannot get device generic segmentation offload settings: Operation not supported
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp segmentation offload: off
udp fragmentation offload: off
generic segmentation offload: off

# grep eth0 /var/log/dmesg
divert: allocating divert_blk for eth0
eth0: Tigon3 [partno(BCM95721) rev 4101 PHY(5750)] (PCI Express) 10/100/1000Base-T Ethernet 00:14:22:7b:7c:34
eth0: RXcsums[1] LinkChgREG[0] MIirq[0] ASF[1] WireSpeed[1] TSOcap[1]
eth0: dma_rwctrl[76180000] dma_mask[64-bit]

If any tcp offload features are suspected, have you tried turning them off?
i.e.
   ethtool -K eth0 sg off rx off tx off

cheers, etc.
-- 
deatrich @ triumf.ca, Science/Atlas         PH: +1 604-222-7665
<*> This moment's fortune cookie:
The amount of time between slipping on the peel and landing on the
pavement is precisely 1 bananosecond.