[ath9k-devel] ath9k buggy in 2.6.31-rc7

[Thomas Backlund <tmb@mandriva.org>]

Hi,

I'm tracking a bug in ath9k in 2.6.31-rc7 series, and have been 
cherrypicking patches from wireless-testing to try to get a stable ath9k...

The problem is that if the user configures the interface it works, but 
have a weak signal and tends to drop the connection.

If he reboots, the signal is even weaker, and also drops the connection
faster. He has to reconfigure the interface to get it back up.

So here is what I have applied:

net-wireless-ath9k-downgrade-ASSERT-in-ath_clone_txbuf.patch
net-wireless-ath9k-Manipulate-and-report-the-correct-RSSI.patch
net-wireless-ath9k-RX-stucks-during-heavy-traffic-in-HT40-mode.patch
net-wireless-ath9k-Handle-tx-desc-shortage-more-appropriately.patch
net-wireless-ath9k-Trivial-fix-in-Kconfig.patch
net-wireless-ath9k-Update-beacon-RSSI.patch
net-wireless-ath9k-Fix-bug-in-PCI-resume.patch
net-wireless-ath9k-Set-HW-state-properly.patch
net-wireless-ath9k-Fix-bug-in-retrieving-average-beacon-rssi.patch


Now with theese patches the signal strength is correct, and the 
connection survives the reboot, but does disconnect when transfer
load increases.

The good part of the above patches is that only a simple network restart
is needed to get the connection back...
(In some cases the connection resets itself and starts working again)

The fix for the connection dropping is to apply the patch:

net-wireless-ath9k-Fix-TX-hang-issue-with-Atheros-chipsets.patch

with this patch applied the link is stable, and keeps working no
matter how much data is pushed through it ...

But the patch has bad side-effects....
The transfers are slower, and loading webpages in firefox takes longer...
It also makes cpu usage higher...

So... What am I missing to get this stable for 2.6.31 ?
(going full wireless-testing is not an option, as this intended for
  main distribution kernel.)

And to help debug the connection dropping issue, attached is a kernel 
2.6.31-rc7 log with "debug=0xffffffff" (and without the last patch)

seems that AR_INTR_SYNC_LOCAL_TIMEOUT happends just before the link
dies...

I seem to recall a fix for that, but cant find it now...

--
Thomas
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ath9k.messages.bug
Url: http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20090823/9893fe48/attachment.txt