[MeeGo-dev] [Connman] & [N900]: Very flaky connman behaviour regarding wifi, debug log linked
holtmann at linux.intel.com
Wed Jul 14 02:12:17 PDT 2010
> We are experiencing very flaky wifi on the N900 and wlan0 seems to
> work fine with standard wireless-tools and wpa_supplicant etc. But not
> with connman (0.54 from trunk + fix for BMC#3807, which made connman
> extremely flaky with SIGSEGVs.)
> I have implemented a testcase which runs at startup:
> * In rc.sysinit, rm -rf /var/lib/connman/default.profile to ensure a
> clean slate. Patch for BMC#3807 is included in connman binary.
> * In rc.sysinit, run connmand -n -d '*' &> /connman.debug
> * In rc.sysinit, disable ofonod, bluetoothd and Nokia N900 usb
> networking as to not interfere
> * After 60 seconds, run /usr/lib/connman/test/test-manager
> * Check output for /org/moblin/connman/technology/wifi which would
> indicate wifi technology being available, indicate date/time +
> PASS/FAIL in log
> * Dump /var/log/messages, /var/log/debug and /connman.debug into log
> * Reboot, try again.
> In one run, wifi only appeared 6 out of 16 times which is considered
> slightly flaky in my books. I haven't verified if the same issue
> happens on other hardware configurations, but it might.
> At http://stskeeps.subnetmask.net/meego/connman-14jul2010.log you can
> download my test log, there's a good number of FAIL cases and some
> PASS cases to look at.
> It would be great if you connman guys (or anyone else) could come with
> theories on why this is so flaky, it would be fantastic as this is
> keeping us from a usable image on N900.
my experience with TI WiFi hardware is actually limited, but one clear
thing I see here is this one:
connmand: Failed to open RFKILL control device
connmand: Enabling udev based RFKILL processing
I have no idea why the kernel is compiled without CONFIG_RFKILL support
or why udev might be not creating /dev/rfkill properly.
The udev based RFKILL processing is racy and we did actually remove it
lately. And a 2.6.35-rc4 clearly has /dev/rfkill support.
A MeeGo kernel should always have CONFIG_RFKILL enabled. Maybe that
More information about the MeeGo-dev