iwl4965 driver bug & a workaround

I've been experiencing random wifi disconnections for a couple of months from now, but assumed these were problems with my AP or with Network Manager. Until now. I've just recently spotted messages indicating firmware/driver problems in dmesg output (I'm running kernel 2.6.30):

Oct  1 16:25:11 pc kernel: wlan0: associated
Oct  1 16:27:21 pc kernel: iwlagn 0000:03:00.0: Microcode SW error detected.  Restarting 0x82000000.
Oct  1 16:27:21 pc kernel: Registered led device: iwl-phy0::radio
Oct  1 16:27:21 pc kernel: Registered led device: iwl-phy0::assoc
Oct  1 16:27:21 pc kernel: Registered led device: iwl-phy0::RX
Oct  1 16:27:21 pc kernel: Registered led device: iwl-phy0::TX
Oct  1 16:28:54 pc kernel: wlan0: disassociating by local choice (reason=3)


Google reveals several bug reports for this error for many popular Linux distributions: Ubuntu, Fedora, Gentoo, Arch...; it was also reported on Intel's bugzilla. The bug has been known for months but it looks like it's not apparent what's its root cause. The problem occurs under high traffic volume or after long inactivity. The connection doesn't work anymore, but it is not reported as such by NetworkManager or iwconfig. To get it working again,you have to reconnect with NetworkManager or reload iwlagn module.

 It seems there are two workarounds, I've tested both of them and both worked for me:

• set "swcrypto=1" option for iwlagn module using modprobe.conf
• use ndiswrapper driver, that is, the wrapper for MS Windows driver.

Of course 1st option is recommended if it works for you. Ndiswrapper works well, except for its ntos process may take as much as 80% CPU under high load (e.g. 200K/s transfer) and around 9% with no traffic (running on 2GHz Core2Duo)... It's not noticable on Core2Duo though.

Judging from the changelong for RC of kernel-2.6.32, there is a buch of fixes related to Intel wireless drivers (and iwl4965 in particular), let's hope the problem will be gone when 2.6.32 is finally released.