Comment 27 for bug 37084

Actually, digging deeper, it looks like this is coming from the handling of scan_mutex, and the early-out mechanism for devices that appear to be off limits because nm_system_device_get_disabled

The mutex is allocated in real_init. But in nm-device.c, the check for whether or not the device is disabled happens before the call to the type-specific init function. That means the cleanup code is getting called for an object that is not fully initialized. The cleanup code tries to free the scan_mutex, which was never allocated, and BOOM!