nanoBTS - First Call Ok, but not others -> I have an ugly hack to fix it

[Harald Welte]

On Thu, Aug 27, 2009 at 01:16:12AM +0200, Sylvain Munaut wrote:

> I finally found that the chan_rate field of the channel_mode must be 0
> for that request to work. And forcing it to 0 in
> rsl_chan_activate_lchan, like this :

thanks, nice spot!

> --- SNIP ---
> 
>         rc = channel_mode_from_lchan(&cm, lchan);
>         if (rc < 0)
>                 return rc;
>         cm.chan_rate = 0; // Hack !
> 
> -----------
> 
> makes the thing work.
> 
> But that's obviously not the root cause. I thing the lchan tch_mode
> should be GSM48_CMODE_SIGN, so that channel_mode_from_lchan would set
> the proper value directly.
> But I haven't traced yet why tch_mode is not at the good value for the
> second call ...

the reason is that we only set it in two places:

1) when we initialize / memset our data structures at program startup
2) in gsm48_tx_chan_mode_modify() when we typically switch from signalling
   to voice mode.

So what is missing is re-setting this before rsl_chan_activate_lchan is called.
Right now the architecture of the code is not good enough to have a "right
spot" for it, since we only activate a channel from the code where we process
channel required messages.  So probably there would
be the right location to do that. (Around line 1087 of abis_rsl.c before
rsl_chan_activate_lchan() is being called).

Testing + patch submission is very welcome.

Regards,
-- 
- Harald Welte <laforge at gnumonks.org>           http://laforge.gnumonks.org/
============================================================================
"Privacy in residential applications is a desirable marketing option."
                                                  (ETSI EN 300 175-7 Ch. A6)