1f75e1a665
Adds patches 0081-0082 on top of the unicast client FSM series
(0070-0080) to fix state machine and BMC bugs specific to the
two-port inhibit_announce boundary clock topology used in StarlingX.
These patches complete the inhibit_announce support introduced in
0078-0079. The earlier patches fixed the unicast client FSM grant
negotiation and multi-master failover paths. This submission fixes
the BMC state decision logic and foreign master record handling that
interact with inhibit_announce to cause stuck or incorrect states.
Patch 0081 — bmc, clock: two-port unicast BC fixes
- Restructure bmc_state_decision() M1/M2/M3 paths so GRAND_MASTER
and MASTER are only reachable for non-inhibit_announce ports.
Prevents split-brain where both upstream and downstream ports
claim GM simultaneously during holdover (clockClass 135 vs 248).
- Guard clock_update_grandmaster() in PS_GRAND_MASTER handler with
if (!c->best) to prevent overwriting parentPortIdentity that
clock_update_slave() set from the upstream master's announce.
Fixes port stuck in UNCALIBRATED when inhibit_announce port is
listed before the downstream port in config.
- Add unconditional clock_update_grandmaster() when best is NULL
before the per-port loop, fixing stale gm.ClockClass in
PARENT_DATA_SET after GRANDMASTER_SETTINGS_NP changes.
- Cache has_inhibit_port flag in clock_add_port() to replace O(n)
LIST_FOREACH walk on every state decision event.
- Add [BMC], [GM_UPDATE], [UC_ORDER] debug instrumentation.
Patch 0082 — port, unicast_client: Fix stuck state in unicast BC
- Skip fc_prune() on inhibit_announce ports. These ports cannot
re-announce, so pruned foreign master records cannot be rebuilt,
causing a ~750ms BMCA fault loop after restart.
- In UNCALIBRATED announce receipt timeout, only call fc_clear()
when p->best->n_messages == 0. Prevents clearing a live foreign
master record while announces are still arriving.
- Set ANNOUNCE bit in ucma->granted on forced UC_WAIT -> UC_NEED_SYDY
transition. Without it, sydymsk (0xa01) is never satisfiable and
the FSM is permanently stuck even after SYNC and DELAY_RESP grants
arrive.
- Add [BMCA] fc_clear debug logging on announce/sync timeout.
Test plan:
Pass: Build and deploy linuxptp_3.1.1-4.stx.16 to two-node BC chain
(BC1 controller-0, BC2 controller-1) with dual upstream GMs.
Pass: GM1 stop/start: clean failover to holdover, clean
recovery. gm.ClockClass tracks 6 -> 7 -> 6 correctly.
Pass: BC1 ptp4l stop/start: BC2 enters holdover, BC1
re-locks to GM in <1s, full chain restored.
Pass: GM2 stop/start: no impact (backup)
Pass: clockClass degradation flows: class 6->7, 6->135,
135->6, 6->165 transitions. BMCA tiebreak correct (lower
clockID wins). No stale gm.ClockClass in PARENT_DATA_SET
after GRANDMASTER_SETTINGS_NP changes.
Pass: GM failure during degraded state: clean failover to
BC1 holdover
Pass: ptp-instance-apply: all instances restart and
re-converge to GM in ~3s.
Pass: controller lock/unlock: BC2 retains holdover during
lock, full chain restored after unlock.
Pass: Group H — controller-0 eno1 link down/up: port on eno1 goes
FAULTY, recovers cleanly on link up. Full chain restored in
~3s.
Story: 2011536
Task: 53990
Change-Id: Iff70be62c4cc6be98dd64cc1e6b48aba5da0ba04
Signed-off-by: Cole Walker <cole.walker@windriver.com>
integ
StarlingX Integration