Description of problem: I am using ath5k on a Thinkpad T41p. Nearly all network traffic seems to be fine, but when I rsync or git pull, some initial control traffic gets through and then it hangs. I must ctrl-c the program. Setting mtu to 1498 fixes the problem for rsync. ath5k_pci 0000:02:02.0: Atheros AR5213 chip found (MAC: 0x56, PHY: 0x41) ath5k_pci 0000:02:02.0: RF5111 5GHz radio found (0x17) ath5k_pci 0000:02:02.0: RF2111 2GHz radio found (0x23) Version-Release number of selected component (if applicable): kernel-2.6.23.8-63.fc8
I doubt if it makes a difference, but please try at least -80.fc8 (the last round of wireless updates for .fc8): http://koji.fedoraproject.org/koji/buildinfo?buildID=26862
Hmm, interesting. Are you forcing the driver to 11M ? Right now the driver will get you large packet loss at higher rates. I'm curious if your issues 'also' go away at when you force driver to 11Mbps: iwconfig wlan0 rate 11M
Nope, neither 11M setting nor -80.fc8 make any difference. But mtu 1498 continues to work perfectly.
kernel -85.fc8 still has the same problem.
Could there be an ethernet LAN using VLAN tagging somewhere between you and the Internet?
I don't think so. The wireless machines are directly on the Internet. We can ping other local addresses with another machine with this: ping -M do -s 1472 ... With IP + ICMP + payload = 20 + 8 + 1472 = 1500. That works fine using ipw2200. On the ath5k machine, if mtu is 1500, then a ping payload of 1472 or 1471 results in no remote response. Ping payload of 1470 works.
Do you see the >1470 payload ICMP frames at the other host? I would be curious to know if the frames are getting dropped on Tx or Rx.
I have tried ping both ways. The frames are dropped only on Tx. Rx works fine.
Seems kernel-2.6.24.2-7.fc8 fixes this problem! I can ping and rsync works with MTU 1500 (the default). I did get a panic at one point, though. Couldn't see the output since X was running, and it hasn't happened since.
Wow, you are quick! :-) FWIW, I think it this patch fixes the problem: ath5k: correct padding in tx descriptors I was going to ask you to test it, but you beat me to it. Thanks for the report!