Marcel, Brad, Jose,

After a few very busy months at work i finally had the opportunity to go 
back to my bluez flow control patch i left in the middle of nowhere back 
in May. :-)
You will find attached an updated version of the patch against 
2.6.18-mh4, as well as a sample test program that shows the flow control 
stuff is really working.

Major changes compared to previous versions:
 * Packets are queued in the core, and are dequeue from there.
 * Flow control is achieved using high resolution timers. This means the 
code is really HZ independant, and things will even go better when the 
dyntick patch which has been in the air recently will be included in the 
mainline kernel.
 * Fixed a stupid bug that would cause packets to be sent really too 
slowly when HZ was < 1000
 * Should work with HZ = 100, 250, 1000, ... or whatever else. I tested 
with 250 HZ only for now.
 * Wrote a test program, "headsettest". To run this program you will 
need an USB dongle that supports SCO, as well as a
headset. This program records your own voice for 30 seconds, then plays 
what it recorded in your ears for 30 seconds. Beware : Running it 
without the flow control patch is likely to crash your box !!
 * The patch is cleaner. This is mainly due to the fact i am getting to 
a better understanding of Linux socket layer.

Random notes :
I tryed to enable number of completed packets events for synchronous 
connections,
to get rid of all the timer stuff, however it didn't succeed.
 This is achieved using hcitool command while there are no active 
connections :

tannat:/home/fchevalier/tmp/linux-2.6.18-mh4-fch# hcitool cmd 0x003 
0X002F 0x01
< HCI Command: ogf 0x03, ocf 0x002f, plen 1
  01
 > HCI Event: 0x0e plen 4
  01 2F 0C 00
tannat:/home/fchevalier/tmp/linux-2.6.18-mh4-fch# hcitool cmd 0x003 0X002E
< HCI Command: ogf 0x03, ocf 0x002e, plen 0
 > HCI Event: 0x0e plen 5
  01 2E 0C 00 01

I tryed with 3 different USB dongles :
 * one of them did not support this feature (old CSR 1.1 dongle)
 * the two others supported the feature, however did never send any 
completed packets events for SCO packets (Broadcom BCM2035 and ISSC).

Due to this lack of working hardware i took the decision to stick with 
timers, even though it is not the best technical choice :-(

Limitations :
For now the flow control only works with CVSD air encoding
 (ie 16 bit frames). If 8 bit frames are used, data will be sent twice 
too fast.

I am looking forward to your comments,

Cheers,

Fabien