Bug 11531

Summary:	ath5k produces packet loss
Product:	Networking	Reporter:	Patrick Schoenfeld (schoenfeld)
Component:	Wireless	Assignee:	networking_wireless (networking_wireless)
Status:	CLOSED INVALID
Severity:	normal	CC:	mcgrof
Priority:	P1
Hardware:	All
OS:	Linux
Kernel Version:	2.6.26.5	Subsystem:
Regression:	---	Bisected commit-id:
Attachments:	Kernel configuration used to build a kernel where it don't works messages while packet loss dmesg while packet loss

Description Patrick Schoenfeld 2008-09-10 04:36:18 UTC

Latest working kernel version:none
Earliest failing kernel version:2.6.25.14
Distribution:Fedora Core 9
Hardware Environment:
AMD Sempron Processor
Atheros AR2413 wlan chip
According to syslog:
ath5k phy0: Atheros AR2413 chip found (MAC: 0x78, PHY: 0x45)

Software Environment:
wireless-tools version 29
wpa_supplicant v0.6.3
network-manager

Additional information:
Current running kernel is 2.6.26.5 and the problem persists.

Problem Description:
When the ath5k driver is used to connect to a WPA-PSK LAN the connection is very lossy (about 15-25% packet loss) and therefore not really usable. The problem is not related to local infrastructure (reproducible at two totally different environments), not related to system at all (networking via cable works flawless) and not directly related to the used hardware (it works fine with madwifi ath_pci in default settings). It happens with a stock kernel and the distribution kernel, so it does not seem to be affected by patches in the Fedora Kernel. Its untested weither this problem occurs without WPA-PSK encryption, too. I have no infrastructure to test this.

Steps to reproduce:
- Connect to a WLAN which uses WPA-PSK
- Ping a host in the local LAN (e.g. the router)
- Let it ping up to 50-100 times (to get a meaningful value)
- Abort the pinging
- Have a look at the values

Another possible way to test if the problem exists:
After WLAN is established, ssh to a non-local system and try to use it. If it is hanging every now and then, the problem exists. If its working flawless, then probably not. However the ping test above appears to me a bit more scientific ;)

Comment 1 Patrick Schoenfeld 2008-09-10 04:42:42 UTC

Created attachment 17711 [details]
Kernel configuration used to build a kernel where it don't works

I've added a kernel config file which I built for another test, but which were also used to test a vanilla kernel 2.6.26.5.

Comment 2 John W. Linville 2008-09-24 07:28:16 UTC

Can we see the output of dmesg (or the contents of /var/log/message) while this packet loss is occuring?

Comment 3 Patrick Schoenfeld 2008-09-24 10:12:22 UTC

Created attachment 18007 [details]
messages while packet loss

Attached is the wanted messages log. Please note that at the beginning and at the end there are lines includes from madwifi (needed to unload it to test ath5k and load it afterwards to have an internet connection). It seems it does not contain anything interesting. I'll attach dmesg which seems to contain more valueable information.

Comment 4 Patrick Schoenfeld 2008-09-24 10:14:34 UTC

Created attachment 18008 [details]
dmesg while packet loss

The attached file was during the packet loss. Note that it includes loading and unloading of madwifi before ath5k.

Comment 5 Luis Chamberlain 2008-10-04 17:37:56 UTC

Is this still an issue with the latest kernel?

Comment 6 Patrick Schoenfeld 2008-10-06 01:45:20 UTC

This bug was reported against 2.6.26.5 and according to kernel.org
"The latest stable version of the Linux kernel is:  	2.6.26.5"
So: Yes, this is still an issue with the latest kernel. Or are you speaking of a current RC?

Comment 7 John W. Linville 2008-10-06 06:15:22 UTC

I'm fairly certain he was asking about the current HEAD of linux-2.6...

Comment 8 Patrick Schoenfeld 2008-10-06 07:38:47 UTC

Needless to say that beeing more precise in this point would help us both. After all I'm a user and as a user _releases_ are relevant for me, unless someone tells me different. Thats not meant as a offense, but you see that I don't *regulary* compile my kernel with each and every new kernel commit just to test weither the bug is fixed or not. Indeed I do not even follow this. But now that you ask for testing the latest HEAD I'll do so ASAP and report the result.

Comment 9 Patrick Schoenfeld 2008-10-10 03:49:21 UTC

Hi,
I've tested the functionality of the driver with the gurrent git HEAD. Packet loss has not gone away, but the situation is better as it has been. Packet loss has decreased by about 5%. Actually its possible to work with the driver, unless ssh or alike is used.
Regards,
Patrick

Comment 10 Patrick Schoenfeld 2008-10-13 11:11:37 UTC

Hi,

sorry for the noise, but I've noticed that the bug is a hardware issue. It started out with this issues with ath5k, but is now constant with every driver and every operating system (e.g. it happens under Windows). So I close this issue as invalid.

Thanks anyway
Best Regards,
Patrick