Re: Emulator not unregistered while sending ATH0

[Guillaume Zajac <guillaume.zajac@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 1985 bytes --]

Hi again Denis,

I have found the issue I will submit a fix.
See explanations below.

On 19/05/2011 14:12, Guillaume Zajac wrote:
> Hi Denis,
>
> I have mixed some stuff concerning GIOChannels...
>
> On 19/05/2011 12:13, Guillaume Zajac wrote:
>> Hi Denis,
>>
>> I noticed that when I send +++ -> ATH0 during a PPP session the ppp 
>> server is well unregistered.
>> However when the physical layer is removed (bluetooht or TCP), the 
>> associated emulator is not unregistered.
>> As we used in gatutil.c set_close_on_unref(TRUE), it means there is 
>> still on reference on the GIOChannel.
>
> The set_close_on_unref() from gatutil.c is done for ppp_net interface 
> GIOChannel.
> It has nothing to do with the physical layer.
>
> However, we do also:
>     - set_close_on_unref(TRUE) into example/emulator.c
>     - set_close_on_unref(FALSE) into plugins/dun_gw.c
>
> So after the client send ATH0, the connection is removed and emulator 
> should be unregistered and removed but it is not the case.
>
>> In the case the PPP session is ended normally, we call 
>> ppp_ipcp_down_notify() to free the ppp_net interface.
>> Then the physical layer is removed and we got the emulator well 
>> unregistered.
>
> That's why I assume unregistration of the emulator is bound to the way 
> PPP session is ended e.g.
> normally ---> ppp_ipcpdown_notify()
> forced ---> ATH0
>
>> Do you have any idea about what is happening there?
>> Maybe I forgot to do something into dun_ath_cb().

While doing g_at_ppp_unref(), g_at_io_set_disconnect_function() is 
called and clear the io_disconnect function.
Thus, when connection with client is removed, io_disconnect() from 
GAtServer is no more called.

We have to do in dun_ath_cb() this sequence:

g_at_server_suspend(em->server);

g_at_ppp_unref(em->ppp);  /* Clear io_disconnect */
em->ppp = NULL;

g_at_server_resume(em->server); /* Set io_disconnect */


Kind regards,
Guillaume