diff options
-rw-r--r-- | doc/Makefile.am | 12 | ||||
-rw-r--r-- | doc/assignment.msc | 181 | ||||
-rw-r--r-- | doc/handover.msc | 164 |
3 files changed, 357 insertions, 0 deletions
diff --git a/doc/Makefile.am b/doc/Makefile.am index 5a231074c..cc2e277b5 100644 --- a/doc/Makefile.am +++ b/doc/Makefile.am @@ -1,3 +1,15 @@ SUBDIRS = \ examples \ $(NULL) + +msc: \ + $(builddir)/handover.png \ + $(builddir)/assignment.png \ + $(NULL) + +$(builddir)/%.png: $(srcdir)/%.msc + mscgen -T png -o $@ $< + +.PHONY: poll +poll: + while true; do $(MAKE) msc; sleep 1; done diff --git a/doc/assignment.msc b/doc/assignment.msc new file mode 100644 index 000000000..e2bdf845d --- /dev/null +++ b/doc/assignment.msc @@ -0,0 +1,181 @@ +msc { + hscale=3; + ms [label="MS"], bts [label="BTS"], bsc[label="BSC"], bsc_gscon[label="BSC conn FSM"], bsc_mgcp[label="BSC mgcp FSM"], mgw[label="MGW"], msc_[label="MSC"]; + + ms note msc_ [label="lchan allocation sequence for BSSMAP Assignment Request"]; + + bsc <= msc_ [label="BSSMAP Assignment Request"]; + bsc box bsc [label="bssmap_handle_assignm_req()"]; + bsc -> bsc_gscon [label="GSCON_EV_A_ASSIGNMENT_CMD"]; + + --- [label="is the chan_mode a speech mode?"]; + + bsc_gscon abox bsc_gscon [label="ST_WAIT_CRCX_BTS (MGCP_MGW_TIMEOUT = 4s)"]; + bsc_gscon -> bsc_mgcp [label="mgcp_conn_create()"]; + bsc_mgcp => mgw [label="CRCX (for BTS)"]; + bsc_mgcp abox bsc_mgcp [label="ST_CRCX (MGCP_MGW_TIMEOUT = 4s)"]; + bsc_gscon note bsc_mgcp [label="two timeouts running in parallel"]; + bsc_gscon note bsc_mgcp [label="note: #define MGCP_MGW_TIMEOUT exists twice, + once in libosmo-mgcp-client, + once in bsc_subscr_conn_fsm.c"]; + bsc_mgcp -> bsc_gscon [label="mgcp_conn_create() exits"]; + bsc_gscon -> bsc [label="bssmap_handle_assignm_req() exits"]; + ...; + --- [label="On Timeout"]; + bsc_gscon note bsc_gscon [label="The conn FSM likely timeouts first"]; + bsc_gscon => msc_ [label="BSSMAP Assignment Failure"]; + bsc_gscon abox bsc_gscon [label="ST_ACTIVE"]; + bsc_mgcp note bsc_mgcp [label="The MGCP FSM will timeout right after that, and terminate itself, + emitting the parent_term event set upon mgcp_conn_create():"]; + bsc_mgcp -> bsc_gscon [label="GSCON_EV_MGW_FAIL_BTS"]; + bsc_gscon note bsc_gscon [label="GSCON_EV_MGW_FAIL_BTS is handled by the conn FSM allstate + handler. It sets conn->user_plane.fi_bts = NULL. There is code + that would emit a BSSMAP Assignment Failure, but not in + ST_ACTIVE"]; + --- [label="end: 'On Timeout'"]; + ...; + + bsc_mgcp <= mgw [label="CRCX OK (for BTS)"]; + bsc_mgcp box bsc_mgcp [label="libosmo-mgcp-client fsm_crcx_resp_cb()"]; + bsc_mgcp -> bsc_gscon [label="GSCON_EV_MGW_CRCX_RESP_BTS"]; + --- [label="end: 'is the chan_mode a speech mode?'"]; + + bsc_gscon note bsc_gscon [label="for mode=sign, we're still handling GSCON_EV_A_ASSIGNMENT_CMD; + for speech mode, we're handling GSCON_EV_MGW_CRCX_RESP_BTS"]; + bsc <- bsc_gscon [label="gsm0808_assign_req()"]; + + bsc box bsc [label="lchan_alloc(): pick available lchan"]; + bsc box bsc [label="rsl_chan_activate_lchan()"]; + + --- [label="is the chosen lchan on dynamic timeslot that is currently used as PDCH?"]; + bts <= bsc [label="i) RSL RF Chan Release of PDCH (Osmocom dyn TS)"]; + bts <= bsc [label="OR ii) RSL PDCH Deact (ip.access dyn TS)"]; + bsc -> bsc_gscon [label="gsm0808_assign_req() returns early"]; + bsc_gscon abox bsc_gscon [label="ST_WAIT_ASS_COMPL (GSM0808_T10_VALUE=6s)"]; + ...; + bts note bsc_gscon [linecolor="red", + label="There seems to be no timer watching over Chan Release nor dyn TS switchover!"]; + ...; + bts => bsc [label="i) RSL RF Chan Release ACK (Osmocom dyn TS)"]; + bts => bsc [label="OR ii) RSL PDCH Deact ACK (ip.access dyn TS)"]; + bsc box bsc [label="rsl_chan_activate_lchan() re-invoked"]; + bts <= bsc [label="RSL Chan Activ"]; + --- [label="else (no dyn TS switchover)"]; + + bts <= bsc [label="RSL Chan Activ"]; + bsc -> bsc_gscon [label="gsm0808_assign_req() returns"]; + bsc_gscon abox bsc_gscon [label="ST_WAIT_ASS_COMPL (GSM0808_T10_VALUE=6s)"]; + ---; + + ...; + --- [label="On Timeout"]; + bsc_gscon => msc_ [label="BSSMAP Assignment Failure"]; + bsc_gscon abox bsc_gscon [label="ST_ACTIVE"]; + bsc_gscon note bsc_mgcp [linecolor="red", + label="The mgcp FSM from CRCX above apparently lacks a cleanup action for this case. + It should be cleaned up eventually when the conn is torn down, but we should + release RTP endpoints as soon as possible."]; + --- [label="end: 'On Timeout'"]; + ...; + + bts => bsc [label="RSL Chan Activ ACK"]; + bsc box bsc [label="bsc_api.c handle_chan_ack()"]; + ms <= bsc [label="RR Assignment Command"]; + + ...; + ms note bsc_gscon [label="We rely on the overall conn FSM ST_WAIT_ASS_COMPL timeout."]; + ...; + + ms => bsc [label="RR Assignment Complete"]; + bsc box bsc [label="handle_ass_compl()"]; + --- [label="Release old lchan"]; + bsc box bsc [label="_lchan_handle_release(sacch_deact=0)"]; + bsc box bsc [label="rsl_release_sapis_from(start=1)"]; + bts <= bsc [label="RSL Release Request (Local End)..."]; + bts <= bsc [label="...for each SAPI except link_id=0"]; + bsc box bsc [label="rsl_release_request(link_id=0)"]; + bts <= bsc [label="RSL Release Request (Local End) for link_id=0"]; + bsc box bsc [label="_lchan_handle_release() returns here, the remaining release is asynchronous; + see `End: 'Release old lchan'` below."]; + ...; + bts note bsc_gscon [linecolor="red", + label="There seems to be no timer watching over RSL Release Request!"]; + ...; + bts => bsc [label="RSL Release Confirm..."]; + bts => bsc [label="...for each SAPI and link_id=0"]; + bsc abox bsc [label="start T3111"]; + ...; + bsc box bsc [label="T3111 expires"]; + bsc abox bsc [label="Start lchan->act_timer with lchan_deact_tmr_cb"]; + bts <= bsc [label="RSL RF Channel Release"]; + ...; + --- [label="On timeout"]; + bsc box bsc [label="lchan_deact_tmr_cb()"]; + bsc box bsc [label="rsl_lchan_mark_broken(): state=LCHAN_S_BROKEN"]; + bsc box bsc [label="lchan_free()"]; + bsc -> bsc [label="S_LCHAN_UNEXPECTED_RELEASE"]; + bsc box bsc [label="bsc_api.c handle_release()"]; + bsc box bsc [label="bsc->assign_fail()"]; + bsc -> bsc_gscon [label="GSCON_EV_RR_ASS_FAIL"]; + bsc note bsc_gscon [linecolor="orange", + label="The name 'RR_ASS_FAIL' might suggest the event means an actual RR Assignment + Failure message being received. Maybe this should be called GSCON_EV_ASSIGNMENT_ERROR."]; + ...; + bsc box bsc [label="bsc->clear_request()"]; + bsc box bsc [label="bsc_clear_request encodes a BSSMAP Clear Request message and passes it on + to the conn FSM as data argument via:"]; + bsc -> bsc_gscon [label="GSCON_EV_TX_SCCP"]; + bsc_gscon => msc_ [label="BSSMAP Clear Request"]; + bsc note bsc_gscon [linecolor="red", + label="Instead of sending an arbitrary message, the conn FSM should + be explicitly instructed to clear the connection, to be able + to notice if the MSC failed to respond to the Clear Request. + Currently, this relies on the MSC responding with a Clear + Command, hopefully, some time later."]; + --- [label="End: 'On timeout'"]; + ...; + bts => bsc [label="RSL RF Channel Release Ack"]; + --- [label="End: 'Release old lchan'"]; + bsc box bsc [label="still in handle_ass_compl()"]; + bsc note bsc [label="officially take over new lchan: conn->lchan = conn->secondary_lchan"]; + --- [label="is BTS using IPA Abis? (osmo-bts, ip.access)"]; + bts <= bsc [label="IPACC CRCX"]; + ---; + bsc -> bsc [label="handle_ass_compl() calls bsc_api->assign_compl()"]; + --- [label="is BTS using IPA Abis? (osmo-bts, ip.access) && conn->user_plane.rtp_ip"]; + bsc box bsc [label="bsc_assign_compl()"]; + bsc note bsc [label="set ass_compl.valid = true, + postponing GSCON_EV_RR_ASS_COMPL until after the + IPACC MDCX ACK received in osmo_bsc_audio.c"]; + bsc box bsc [label="exit early: bsc_assign_compl()"]; + bsc box bsc [label="exit early: handle_ass_compl()"]; + bsc box bsc [label="osmo_bsc_audio.c"]; + bts => bsc [label="IPACC CRCX ACK"]; + bts <= bsc [label="IPACC MDCX"]; + bts => bsc [label="IPACC MDCX ACK"]; + bsc box bsc [label="handle_abisip_signal()"]; + bsc -> bsc_gscon [label="GSCON_EV_RR_ASS_COMPL"]; + --- [label="else"]; + bsc box bsc [label="bsc_assign_compl()"]; + bsc -> bsc_gscon [label="GSCON_EV_RR_ASS_COMPL"]; + --- ; + + --- [label="is chan_mode a speech mode?"]; + bsc_gscon abox bsc_gscon [label="ST_WAIT_MDCX_BTS"]; + bsc_gscon -> bsc_mgcp [label="mgcp_conn_modify()"]; + bsc_mgcp note bsc_mgcp [label="same mgcp FSM as above, for BTS side"]; + bsc_mgcp => mgw [label="MDCX (for BTS)"]; + bsc_mgcp <= mgw [label="MDCX OK"]; + bsc_mgcp -> bsc_gscon [label="GSCON_EV_MGW_MDCX_RESP_BTS"]; + bsc_gscon abox bsc_gscon [label="ST_WAIT_CRCX_MSC"]; + bsc_gscon -> bsc_mgcp [label="mgcp_conn_create()"]; + bsc_mgcp note bsc_mgcp [label="second mgcp FSM for MSC side"]; + bsc_mgcp => mgw [label="CRCX (for MSC)"]; + bsc_mgcp <= mgw [label="CRCX OK (for MSC)"]; + bsc_gscon <- bsc_mgcp [label="GSCON_EV_MGW_CRCX_RESP_MSC"]; + ---; + + bsc_gscon => msc_ [label="BSSMAP Assignment Complete"]; + + bsc_gscon abox bsc_gscon [label="ST_ACTIVE"]; +} diff --git a/doc/handover.msc b/doc/handover.msc new file mode 100644 index 000000000..e5e787c0e --- /dev/null +++ b/doc/handover.msc @@ -0,0 +1,164 @@ +# Handover between cells, intra-BSC +msc { + hscale=3; + ms [label="MS"], bts [label="BTS"], bsc[label="BSC"], bsc_gscon[label="BSC conn FSM"], bsc_mgcp[label="BSC mgcp FSM"], mgw[label="MGW"]; + + ms note mgw [label="intra-BSC Handover sequence"]; + + bsc_gscon abox bsc_gscon [label="ST_ACTIVE"]; + bsc box bsc [label="bsc_handover_start(): init conn->ho"]; + bsc -> bsc_gscon [label="GSCON_EV_HO_START"]; + bsc <- bsc_gscon [label="bsc_handover_start_gscon()"]; + + bsc box bsc [label="lchan_alloc(): pick available lchan"]; + bsc box bsc [label="rsl_chan_activate_lchan()"]; + + --- [label="is the chosen lchan on dynamic timeslot that is currently used as PDCH?"]; + bts <= bsc [label="i) RSL RF Chan Release of PDCH (Osmocom dyn TS)"]; + bts <= bsc [label="OR ii) RSL PDCH Deact (ip.access dyn TS)"]; + bsc -> bsc_gscon [label="bsc_handover_start_gscon() returns early"]; + bsc_gscon abox bsc_gscon [label="ST_WAIT_HO_COMPL (no timeout, relies on T3103 below)"]; + ...; + bts note bsc_gscon [linecolor="red", + label="There seems to be no timer watching over Chan Release nor dyn TS switchover!"]; + ...; + bts => bsc [label="i) RSL RF Chan Release ACK (Osmocom dyn TS)"]; + bts => bsc [label="OR ii) RSL PDCH Deact ACK (ip.access dyn TS)"]; + bsc box bsc [label="rsl_chan_activate_lchan() re-invoked"]; + bts <= bsc [label="RSL Chan Activ"]; + --- [label="else (no dyn TS switchover)"]; + + bts <= bsc [label="RSL Chan Activ"]; + bsc -> bsc_gscon [label="bsc_handover_start_gscon() returns"]; + ---; + bsc_gscon abox bsc_gscon [label="ST_WAIT_HO_COMPL (no timeout, relies on T3103 below)"]; + + ...; + bts note bsc_gscon [linecolor="red", + label="There seems to be no timer watching out for RSL Chan Activ ACK/NACK!"]; + ...; + bts => bsc [label="RSL Chan Activ ACK"]; + bsc -> bsc [label="S_LCHAN_ACTIVATE_ACK"]; + bsc box bsc [label="handover_logic.c ho_logic_sig_cb()"]; + bsc box bsc [label="ho_chan_activ_ack()"]; + bsc note bsc [label="gsm48_send_ho_cmd()"]; + ms <= bsc [label="RR Handover Command"]; + bsc abox bsc [label="start T3103"]; + --- [label="is BTS using IPA Abis? (osmo-bts, ip.access)"]; + bts <= bsc [label="IPACC CRCX"]; + bsc -> bsc [label="ho_chan_activ_ack() returns"]; + bts note bsc [linecolor="red", + label="There seems to be no timer watching over IPACC CRCX ACK/NACK! + If no response is received, we simply ignore that fact and carry on as if + everything was fine."]; + ...; + bts note bsc [label="The IPACC CRCX and MDCX ACKs may come back at any time: + before or after the Handover Detect, before or after Handover Complete."]; + bts note bsc_mgcp [linecolor="red", + label="The CRCX ACK contains vital information for routing the RTP stream. + If the CRCX ACK were very slow, we would not know which RTP/RTPC ports + to point the MGW at, below at mgcp_conn_modify()! + Even though this being unrealistic, we must make sure to receive a CRCX ACK."]; + ...; + bsc box bsc [label="osmo_bsc_audio.c"]; + bts => bsc [label="IPACC CRCX ACK"]; + bts <= bsc [label="IPACC MDCX"]; + ...; + bts note bsc [linecolor="red", + label="There seems to be no timer watching over IPACC MDCX ACK/NACK! + If no response is received, we simply ignore that fact and carry on as if + everything was fine."]; + ...; + bts => bsc [label="IPACC MDCX ACK"]; + bts note bsc [label="IPACC MDCX ACK triggers no events or actions"]; + ---; + + ...; + ms => bsc [label="RR Handover Detect"]; + bsc -> bsc [label="S_LCHAN_HANDOVER_DETECT"]; + bsc box bsc [label="ho_rsl_detect(): no action, only logging"]; + bsc note bsc_gscon [label="Handover Detect triggers no events or actions"]; + bsc note bsc_gscon [linecolor="red", + label="upon Handover Detect, we should already start re-routing the RTP! + Instead we wait for Handover Complete."]; + + ...; + ms => bsc [label="RR Handover Complete"]; + bsc -> bsc [label="S_LCHAN_HANDOVER_COMPL"]; + bsc box bsc [label="handover_logic.c ho_logic_sig_cb()"]; + bsc box bsc [label="ho_gsm48_ho_compl()"]; + bsc box bsc [label="stop T3103"]; + bts note bsc_gscon [label="If anything goes wrong from this point on, we will not move back + to the old lchan: would be pointless after Handover Complete."]; + bsc note bsc [label="officially take over new lchan: conn->lchan = ho->new_lchan"]; + + --- [label="Release old lchan"]; + bsc box bsc [label="_lchan_handle_release(sacch_deact=0)"]; + bsc box bsc [label="rsl_release_sapis_from(start=1)"]; + bts <= bsc [label="RSL Release Request (Local End)..."]; + bts <= bsc [label="...for each SAPI except link_id=0"]; + bsc box bsc [label="rsl_release_request(link_id=0)"]; + bts <= bsc [label="RSL Release Request (Local End) for link_id=0"]; + bsc box bsc [label="_lchan_handle_release() returns here, the remaining release is asynchronous; + see `End: 'Release old lchan'` below."]; + ...; + bts note bsc_gscon [linecolor="red", + label="There seems to be no timer watching over RSL Release Request!"]; + ...; + bts => bsc [label="RSL Release Confirm..."]; + bts => bsc [label="...for each SAPI and link_id=0"]; + bsc abox bsc [label="start T3111"]; + ...; + bsc box bsc [label="T3111 expires"]; + bsc abox bsc [label="Start lchan->act_timer with lchan_deact_tmr_cb"]; + bts <= bsc [label="RSL RF Channel Release"]; + ...; + --- [label="On timeout"]; + bsc box bsc [label="lchan_deact_tmr_cb()"]; + bsc box bsc [label="rsl_lchan_mark_broken(): state=LCHAN_S_BROKEN"]; + bsc box bsc [label="lchan_free()"]; + bsc -> bsc [label="S_LCHAN_UNEXPECTED_RELEASE"]; + bsc box bsc [label="bsc_api.c handle_release()"]; + bsc box bsc [label="bsc->clear_request()"]; + bsc box bsc [label="bsc_clear_request encodes a BSSMAP Clear Request message and passes it on + to the conn FSM as data argument via:"]; + bsc -> bsc_gscon [label="GSCON_EV_TX_SCCP"]; + bsc_gscon rbox bsc_gscon [label="BSSMAP Clear Request to MSC"]; + bsc note bsc_gscon [linecolor="red", + label="During Handover, we actually release the entire conn just because we failed to + gracefully release the old lchan. That is obviously nonsense."]; + bsc note bsc [label="Stop T3101 (but was not active in this code path)"]; + bsc -> bsc [label="S_CHALLOC_FREED"]; + --- [label="End: 'On timeout'"]; + ...; + bts => bsc [label="RSL RF Channel Release Ack"]; + ---; + + bsc box bsc [label="still in ho_gsm48_ho_compl()"]; + bsc note bsc [label="handover_free(), conn->ho = NULL"]; + bsc -> bsc_gscon [label="GSCON_EV_HO_COMPL"]; + bsc note bsc_gscon [linecolor="orange", + label="Handover information is cleared before signalling the conn FSM. + That means the conn FSM cannot possibly log sensible information about exactly + which Handover has just completed."]; + + bsc_gscon abox bsc_gscon [label="ST_WAIT_MDCX_BTS_HO + (MGCP_MGW_TIMEOUT=4s with MGCP_MGW_HO_TIMEOUT_TIMER_NR)"]; + + bsc_gscon -> bsc_mgcp [label="mgcp_conn_modify()"]; + bsc_mgcp note bsc_mgcp [label="mgcp FSM that was established for old lchan, for BTS side"]; + bsc_mgcp => mgw [label="MDCX (for BTS)"]; + ...; + bsc_gscon note mgw [ + label="If we get no MDCX ACK, the MGCP FSM terminates, and emits GSCON_EV_MGW_FAIL_BTS. + Besides invalidating the MGCP FSM pointer, this event has no + effect in ST_WAIT_MDCX_BTS_HO, and we rely on above conn FSM + timeout instead."]; + bsc_gscon note bsc_gscon [linecolor="red", + label="A timeout of ST_WAIT_MDCX_BTS_HO simply transitions back to ST_ACTIVE! + Even though the MGW failed, we carry on as if everything were fine."]; + ...; + bsc_mgcp <= mgw [label="MDCX OK"]; + bsc_mgcp -> bsc_gscon [label="GSCON_EV_MGW_MDCX_RESP_BTS"]; + bsc_gscon abox bsc_gscon [label="ST_ACTIVE"]; +} |