Hi,

this patch fix some problems:
1. use nonblocking socket on rfcomm connect this allow to wait for more 
than one headset
2. Before seting the SCO filedescriptor to the kernel ir send -1 for 
closing the actual one.
+ If an SCO connection is lost from last time it become close and with 
the next try is available again
3. Move the SCO loopback testroutine to poll informations
4. Handle Headset disconnect (power off)

Cu Thomas

p.s. I hope this tine the patch is in correct format