Hi Harald,
On Sun, Feb 10, 2013 at 07:46:10AM +0100, Harald Welte wrote:
Hi Pablo,
I am currently experimenting with porting osmo-bts over to libosmo-abis for the OML/RSL links. However, it seems that libosmo-abis might not be as complete as I was assuming. Maybe you have some ideas how to address the issue.
As you probably know, upon start of the BTS only the OML link is established to the BSC. It is then via OML that the BSC instructs the BTS to establish a RSL link to a specified address. In the case of current OpenBSC, those IP addressee for OML and RSL are identical, but we cannot make that an assumption.
Furthermore, in the case of a multi-TRX BTS, there will be one OML connection but multiplt RSL connections (one for each TRX). Those RSL connections will originate from different BTS hardware units.
So far I've been using e1inp_ipa_bts_test.c as a template for the BTS side, but it seems that it
- establishes OML and RSL at the same time, without waiting for the OML command to establish the RSL link
Just sent you a patch for this. What I'm proposing is that you call e1inp_line_update twice, once to get the OML link up and another to get the RSL link up. So you control when the RSL becomes up.
The patch has been tested with the example templates.
- uses the same destination IP for OML and RSL without option for configuring this.
Hm, let me check this, will try to get back to you with anything asap.
Do you have any ideas how to best implement this? There is of course the option to bypass the entire e1inp_line / e1inp_sign_link abstraction and to use the ipa.c code directly (ipa_client_conn_create() and friends). But then I get only the multiplex without the andling of the ID_GET/ID_RESP/ID_ACK contained in ipaccess.c.
Oh, I would be happy if you use that abstraction, let me try to fix it. If I fail to provide anything that you like, you can default on that, of course.
Regards.