On Tue, 11 Jul 2023 at 17:28, Vladimir Yakovlev via Linuxptp-devel <
linuxptp-devel@lists.sourceforge.net> wrote:
> In current version, when port fault occurs, unicast service in G.8275.2
> mode will lock up.
>
> When i disconnect cable between master and slave, and reconnect it back,
> master stops sending messages.
>
> Series of events occurs:
> 1. unicast_service_rearm_timer rearms timer, but it can skip execution if
> port fault occurs.
> 2. next time ptp client connects, unicast_service_add will check client
> list and find it in itmp->clients list, so it does not arm timer.
> 3. unicast_service_timer will not fire
>
> Master will receive and process unicast_service_extend messages, but does
> not send announce and sync.
>
That looks like an ugly workaround.
1. The clear seems to apply to slaves as well. Is that necessary?
2. Why is that a problem at all, what is the root problem?
Do you have other traffic? Are you using any link detection protocols
like LLDP?
https://en.wikipedia.org/wiki/Link_Layer_Discovery_Protocol
3. PS_FAULTY, is a detection for several errors.
If you wish to detect disconnected cable, you should add a new handler
that captures the *RTMGRP_LINK *events and issues a link down event.
https://man7.org/linux/man-pages/man7/netlink.7.html
The netlink link up/down events are a bit problematic as there might be
false events that require a more delicate handling to filter them.
You might need a link detection protocol like LLDP.
Erez
> I used this patch to solve a problem. It clears the client list, if port
> goes to faulty state.
>
That sounds like a daemon restart.
>
> Index: linuxptp-2.0/port.c
> ===
> --- linuxptp-2.0.orig/port.c
> +++ linuxptp-2.0/port.c
> @@ -2976,6 +2976,7 @@ int port_state_update(struct port *p, en
> if (PS_FAULTY == next) {
> struct fault_interval i;
> fault_interval(p, last_fault_type(p), );
> + unicast_service_clear_clients(p);
> if (clear_fault_asap()) {
> pr_notice("port %hu: clearing fault immediately", portnum(p));
> next = p->state_machine(next, EV_FAULT_CLEARED, 0);
> Index: linuxptp-2.0/unicast_service.c
> ===
> --- linuxptp-2.0.orig/unicast_service.c
> +++ linuxptp-2.0/unicast_service.c
> @@ -378,6 +378,23 @@ void unicast_service_cleanup(struct port
> free(p->unicast_service);
> }
>
> +void unicast_service_clear_clients(struct port *p)
> +{
> + struct unicast_service_interval *itmp, *inext;
> + struct unicast_client_address *ctmp, *cnext;
> +
> + if (!p->unicast_service) {
> + return;
> + }
> + LIST_FOREACH_SAFE(itmp, >unicast_service->intervals, list, inext) {
> + LIST_FOREACH_SAFE(ctmp, >clients, list, cnext) {
> + LIST_REMOVE(ctmp, list);
> + free(ctmp);
> + }
> + LIST_REMOVE(itmp, list);
> + }
> +}
> +
> int unicast_service_deny(struct port *p, struct ptp_message *m,
> struct tlv_extra *extra)
> {
> Index: linuxptp-2.0/unicast_service.h
> ===
> --- linuxptp-2.0.orig/unicast_service.h
> +++ linuxptp-2.0/unicast_service.h
> @@ -87,4 +87,6 @@ void unicast_service_remove(struct port
> */
> int unicast_service_timer(struct port *p);
>
> +void unicast_service_clear_clients(struct port *p);
> +
> #endif
> ___
> Linuxptp-devel mailing list
> Linuxptp-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
>
___
Linuxptp-devel mailing list
Linuxptp-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
