On Thu, Dec 05, 2013 at 04:02:38PM +0100, Andreas Eversberg wrote:
> @@ -890,7 +892,18 @@ int rsl_release_request(struct gsm_lchan *lchan, uint8_t
> link_id,
> DEBUGP(DRLL, "%s RSL RLL RELEASE REQ (link_id=0x%02x, reason=%u)\n",
> gsm_lchan_name(lchan), link_id, release_mode);
>
> - return abis_rsl_sendmsg(msg);
> + abis_rsl_sendmsg(msg);
> +
> + /* Do not wait for Nokia BTS to send the confirm. */
> + if (is_nokia_bts(lchan->ts->trx->bts)
> + && release_mode == RSL_REL_LOCAL_END) {
> + DEBUGP(DRLL, "Do not wait for RELease CONFirm, because Nokia"
> + " BTS does not send it.\n");
> + lchan->sapis[link_id & 0x7] = LCHAN_SAPI_UNUSED;
> + rsl_handle_release(lchan);
In addition to what I wrote yesterday. This is also violating the
principle of the least surprise and in fact breaks the error handling
release.
Before:
The "effect" rsl_release_request could be observed after the event loop
has been reached (only then we will parse the result from the BTS).
Your patch:
* RLL indication for the release is missing for the NokiaBTS (that is
an issue and should be resolved by moving the sapi handling code out
of the RSL_MT_REL_IND case into a method)
* You directly call rsl_handle_release and can get ahead of itself. This
is a problem for:
if (deact_sacch == SACCH_DEACTIVATE)
rsl_deact_sacch(lchan);
rsl_release_sapis_from(lchan, 0, RSL_REL_LOCAL_END);
/* TODO: start T3109 now. */
rsl_lchan_set_state(lchan, LCHAN_S_REL_ERR);
So with your change rsl_handle_release will be called when lchan->state
is not set to LCHAN_S_REL_ERR yet.
holger
