Thanks Yangbo,
I'd like to interject some overall comments/questions before digging into
comments on the code.
In my opinion I prefer this v2 (BC) over the previous v1 (TC).
To me, it isn't so much about the name. It is more about aligning the code with
the 802.1AS-2011 requirements ("shall"s). My employer (National Instruments)
has been shipping our own BC-based fork of Linuxptp for years now (e.g. uses
bc_dispatch() and bc_event() as is). That code has passed UNH-IOL conformance
for 802.1AS-2011, performed well in 802.1AS-2011 plugfests (i.e. ICC, Avnu,
ISPCS), and interoperated in customer applications with many other 802.1AS-2011
end-stations and bridges. I realize all of that is anecdotal, in that a
TC-based fork of Linuxptp might have performed well also. Nevertheless, I'm
concerned that a fully TC-based implementation would move the code away from
conformance with 802.1AS-2011, and interoperation would break as a result.
Assuming the group is okay with BC, there is probably one other major question
to answer before we go much further:
Are we okay implementing the 802.1AS-2011 BC specifics in clock.c,
or do we prefer to move most of that to a different file (bridge.c,
gptp_relay.c, or whatever)?
There's a short list of items that make 802.1AS-2011 BC different from
1588-2008 P2P Default BC:
1. Transfer Sync / FollowUp differences from Slave (receive) to Master
(transmit) ports.
2. Support distinct sync intervals on Slave and Master ports (like a typical
BC).
The v2 patch covers item 1 and 2, which are the core differences. Nevertheless,
it might be good to keep other differences in mind (i.e. future patches) before
we decide the preceding question:
3. Transfer Pdelay differences from Slave to Master ports (cumulative rate
ratio).
4. BMCA: Skip over UNCALIBRATED and PRE_MASTER states.
5. BMCA: Remove foreign master qualification (i.e. change threshold from 2 to
1).
6. Check performance requirement in 802.1AS-2011 B.1.1 for neighbor rate ratio
computation.
7. BMCA: FAULTY causes a State Decision Event. 802.1AS-2011's methodology for
faults is to avoid waiting in that FAULTY state in hopes that management will
notice. Instead, 802.1AS moves on to search for a valid non-faulty state (SDE).
If supported, the fault is logged so that management can notice later, but that
logging isn't a requirement.
There's one other item in the upcoming revision (presumably 802.1AS-2020) that
is not explicitly stated in 802.1AS-2011:
8. If-and-only-if the sync interval is the same for a Slave and Master port
pair, "this PTP Port, when operating as a Master port, shall transmit a Sync as
soon as possible after the Slave port received a Sync (ignoring syncInterval)."
To Richard's credit, this is TC-like behavior. In the code, it presumably
represents a call to tc_fwd_sync() as was done in Yangbo's v1 patch. When we
dig into the details, there might be some 802.1AS-specifics, so maybe it will
warrant a new gptp_relay_fwd_sync() function.
Since 802.1AS-2020 isn't published yet, I'd be perfectly happy ignoring item 8
for now. We can always revisit the issue later on.
In the upcoming revision of 1588, all of items 1-8 are supported as options
that a PTP Profile (like 802.1AS) can list as required. For example, the 1588
revision allows UNCALIBRATED and PRE_MASTER to be skipped (whereas 1588-2008
requires those states). Therefore, items 1-8 will be conformant 1588 behavior
that any profile can use. I mention this because if we want to avoid "gptp" or
"802.1AS" in the naming of items 1-8, that sounds fine, because they are not
necessarily exclusive to 802.1AS in the future.
Rodney
> -----Original Message-----
> From: Yangbo Lu <[email protected]>
> Sent: Monday, October 14, 2019 5:27 AM
> To: [email protected]; Rodney Cummings
> <[email protected]>; Richard Cochran <[email protected]>; Erik
> Hons <[email protected]>
> Cc: Yangbo Lu <[email protected]>
> Subject: [EXTERNAL] [RFC V2] Add IEEE 802.1AS-2011 time-aware bridge
> support
>
> This patch is to add IEEE 802.1AS-2011 time-aware bridge support based on
> current BC clock type. It implements only time information relay, and BMCA
> was not touched. To run it, the profile gPTP.cfg could be used with
> multiple interfaces specified using -i option.
>
> The main code changes are,
> - Create syfu_relay_info structure for time information relay.
> - Implement port_syfu_relay_info_insert() to update follow_up (with TLV)
> message with time information for relay.
>
> Signed-off-by: Yangbo Lu <[email protected]>
> ---
> Changes for v2:
> - Implemented based on BC instead of TC.
> ---
> clock.c | 43 ++++++++++++++++++++++++++++++++++++-
> clock.h | 45 +++++++++++++++++++++++++++++++++++++++
> port.c | 75
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++----
> 3 files changed, 158 insertions(+), 5 deletions(-)
>
> diff --git a/clock.c b/clock.c
> index 146576a..1865ecb 100644
> --- a/clock.c
> +++ b/clock.c
> @@ -45,7 +45,6 @@
> #include "util.h"
>
> #define N_CLOCK_PFD (N_POLLFD + 1) /* one extra per port, for the fault
> timer */ -#define POW2_41 ((double)(1ULL << 41))
>
> struct port {
> LIST_ENTRY(port) list;
> @@ -123,6 +122,7 @@ struct clock {
> int stats_interval;
> struct clockcheck *sanity_check;
> struct interface uds_interface;
> + struct syfu_relay_info syfu_relay;
> LIST_HEAD(clock_subscribers_head, clock_subscriber) subscribers; };
>
> @@ -1117,6 +1117,8 @@ struct clock *clock_create(enum clock_type type,
> struct config *config,
> }
> }
>
> + memset(&c->syfu_relay, 0, sizeof(struct syfu_relay_info));
> +
> /* Initialize the parentDS. */
> clock_update_grandmaster(c);
> c->dad.pds.parentStats = 0;
> @@ -1208,6 +1210,15 @@ void clock_follow_up_info(struct clock *c, struct
> follow_up_info_tlv *f)
> sizeof(c->status.lastGmPhaseChange));
> }
>
> +static void clock_get_follow_up_info(struct clock *c, struct
> +follow_up_info_tlv *f) {
> + f->cumulativeScaledRateOffset = c-
> >status.cumulativeScaledRateOffset;
> + f->scaledLastGmPhaseChange = c->status.scaledLastGmPhaseChange;
> + f->gmTimeBaseIndicator = c->status.gmTimeBaseIndicator;
> + memcpy(&f->lastGmPhaseChange, &c->status.lastGmPhaseChange,
> + sizeof(f->lastGmPhaseChange));
> +}
> +
> int clock_free_running(struct clock *c) {
> return c->free_running ? 1 : 0;
> @@ -1583,6 +1594,16 @@ void clock_peer_delay(struct clock *c, tmv_t ppd,
> tmv_t req, tmv_t rx,
> stats_add_value(c->stats.delay, tmv_dbl(ppd)); }
>
> +tmv_t clock_get_path_delay(struct clock *c) {
> + return c->path_delay;
> +}
> +
> +double clock_get_nrr(struct clock *c)
> +{
> + return c->nrr;
> +}
> +
> int clock_slave_only(struct clock *c)
> {
> return c->dds.flags & DDS_SLAVE_ONLY;
> @@ -1776,6 +1797,7 @@ static void handle_state_decision_event(struct clock
> *c)
> c->path_delay = c->initial_delay;
> c->nrr = 1.0;
> fresh_best = 1;
> + clock_disable_syfu_relay(c);
> }
>
> c->best = best;
> @@ -1847,3 +1869,22 @@ enum servo_state clock_servo_state(struct clock *c)
> {
> return c->servo_state;
> }
> +
> +void clock_prepare_syfu_relay(struct clock *c, struct ptp_message *sync,
> + struct ptp_message *fup)
> +{
> + c->syfu_relay.precise_origin_ts = timestamp_to_tmv(fup->ts.pdu);
> + c->syfu_relay.correction = fup->header.correction;
> + clock_get_follow_up_info(c, &c->syfu_relay.fup_info_tlv);
> + c->syfu_relay.avail = 1;
> +}
> +
> +void clock_disable_syfu_relay(struct clock *c) {
> + c->syfu_relay.avail = 0;
> +}
> +
> +struct syfu_relay_info *clock_get_syfu_relay(struct clock *c) {
> + return &c->syfu_relay;
> +}
> diff --git a/clock.h b/clock.h
> index 9d3133a..8ff1181 100644
> --- a/clock.h
> +++ b/clock.h
> @@ -29,8 +29,18 @@
> #include "tmv.h"
> #include "transport.h"
>
> +#define POW2_41 ((double)(1ULL << 41))
> +
> struct ptp_message; /*forward declaration*/
>
> +struct syfu_relay_info {
> + tmv_t precise_origin_ts;
> + Integer64 correction;
> + struct follow_up_info_tlv fup_info_tlv;
> + /* Auxiliary info */
> + int avail;
> +};
> +
> /** Opaque type. */
> struct clock;
>
> @@ -240,6 +250,20 @@ void clock_peer_delay(struct clock *c, tmv_t ppd,
> tmv_t req, tmv_t rx,
> double nrr);
>
> /**
> + * Get the path delay as measured on a slave port.
> + * @param c The clock instance.
> + * @return The path delay as measured on a slave port.
> + */
> +tmv_t clock_get_path_delay(struct clock *c);
> +
> +/**
> + * Get the neighbor rate ratio as measured on a slave port.
> + * @param c The clock instance.
> + * @return The neighbor rate ratio as measured on a slave
> port.
> + */
> +double clock_get_nrr(struct clock *c);
> +
> +/**
> * Set clock sde
> * @param c A pointer to a clock instance obtained with
> clock_create().
> * @param sde Pass one (1) if need a decision event and zero if not.
> @@ -359,4 +383,25 @@ void clock_check_ts(struct clock *c, uint64_t ts);
> */
> double clock_rate_ratio(struct clock *c);
>
> +/**
> + * Prepare sync/follow_up relay.
> + * @param c The clock instance.
> + * @param sync The sync message.
> + * @param fup The follow_up message.
> + */
> +void clock_prepare_syfu_relay(struct clock *c, struct ptp_message *sync,
> + struct ptp_message *fup);
> +
> +/**
> + * Disable sync/follow_up relay.
> + * @param c The clock instance.
> + */
> +void clock_disable_syfu_relay(struct clock *c);
> +
> +/**
> + * Get sync/follow_up relay information.
> + * @param c The clock instance.
> + * @return The sync/follow_up relay information.
> + */
> +struct syfu_relay_info *clock_get_syfu_relay(struct clock *c);
> #endif
> diff --git a/port.c b/port.c
> index 07ad3f0..073e71a 100644
> --- a/port.c
> +++ b/port.c
> @@ -1241,6 +1241,10 @@ static void port_syfufsm(struct port *p, enum
> syfu_event event,
> break;
> case FUP_MATCH:
> syn = p->last_syncfup;
> + if (p->follow_up_info)
> + clock_prepare_syfu_relay(p->clock, syn, m);
> + else
> + clock_disable_syfu_relay(p->clock);
> port_synchronize(p, syn->hwts.ts, m->ts.pdu,
> syn->header.correction,
> m->header.correction,
> @@ -1261,6 +1265,10 @@ static void port_syfufsm(struct port *p, enum
> syfu_event event,
> break;
> case SYNC_MATCH:
> fup = p->last_syncfup;
> + if (p->follow_up_info)
> + clock_prepare_syfu_relay(p->clock, m, fup);
> + else
> + clock_disable_syfu_relay(p->clock);
> port_synchronize(p, m->hwts.ts, fup->ts.pdu,
> m->header.correction,
> fup->header.correction,
> @@ -1450,6 +1458,60 @@ int port_tx_announce(struct port *p, struct address
> *dst)
> return err;
> }
>
> +static void port_syfu_relay_info_insert(struct port *p,
> + struct ptp_message *sync,
> + struct ptp_message *fup)
> +{
> + struct syfu_relay_info *syfu_relay = clock_get_syfu_relay(p->clock);
> + struct follow_up_info_tlv *fui_relay = &syfu_relay->fup_info_tlv;
> + struct follow_up_info_tlv *fui = follow_up_info_extract(fup);
> + tmv_t ingress, egress, residence, path_delay;
> + double gm_rr, nrr, rr;
> + struct timestamp ts;
> +
> + if (syfu_relay->avail == 0)
> + return;
> +
> + fup->follow_up.preciseOriginTimestamp =
> + tmv_to_Timestamp(syfu_relay->precise_origin_ts);
> + fup->header.correction = syfu_relay->correction;
> +
> + /* Calculate residence time. */
> + ingress = clock_ingress_time(p->clock);
> + egress = sync->hwts.ts;
> + residence = tmv_sub(egress, ingress);
> + rr = clock_rate_ratio(p->clock);
> + if (rr != 1.0) {
> + residence = dbl_tmv(tmv_dbl(residence) * rr);
> + }
> +
> + gm_rr = 1.0 + (fui_relay->cumulativeScaledRateOffset + 0.0) /
> POW2_41;
> + nrr = clock_get_nrr(p->clock);
> +
> + /* Add corrected residence time into correction. */
> + fup->header.correction += tmv_to_TimeInterval(residence) * gm_rr *
> +nrr;
> +
> + /* Add corrected path delay into correction. */
> + path_delay = clock_get_path_delay(p->clock);
> + fup->header.correction += tmv_to_TimeInterval(path_delay) * gm_rr;
> +
> + /* Update follow_up TLV */
> + gm_rr *= nrr;
> + fui->cumulativeScaledRateOffset = gm_rr * POW2_41 - POW2_41;
> + fui->scaledLastGmPhaseChange = fui_relay->scaledLastGmPhaseChange;
> + fui->gmTimeBaseIndicator = fui_relay->gmTimeBaseIndicator;
> + memcpy(&fui->lastGmPhaseChange, &fui_relay->lastGmPhaseChange,
> + sizeof(fui->lastGmPhaseChange));
> +
> + ts.sec = fup->follow_up.preciseOriginTimestamp.seconds_msb;
> + ts.sec = ts.sec << 32 | fup-
> >follow_up.preciseOriginTimestamp.seconds_lsb;
> + ts.nsec = fup->follow_up.preciseOriginTimestamp.nanoseconds;
> + pr_debug("port %hu: syfu_relay info:", portnum(p));
> + pr_debug("port %hu: precise_origin_ts %" PRIu64 ".%u", portnum(p),
> ts.sec, ts.nsec);
> + pr_debug("port %hu: correction %" PRId64, portnum(p), fup-
> >header.correction >> 16);
> + pr_debug("port %hu: fup_info %.9f", portnum(p), gm_rr);
> +}
> +
> int port_tx_sync(struct port *p, struct address *dst) {
> struct ptp_message *msg, *fup;
> @@ -1543,10 +1605,15 @@ int port_tx_sync(struct port *p, struct address
> *dst)
> fup->address = *dst;
> fup->header.flagField[0] |= UNICAST;
> }
> - if (p->follow_up_info && follow_up_info_append(fup)) {
> - pr_err("port %hu: append fup info failed", portnum(p));
> - err = -1;
> - goto out;
> +
> + if (p->follow_up_info) {
> + if (follow_up_info_append(fup)) {
> + pr_err("port %hu: append fup info failed", portnum(p));
> + err = -1;
> + goto out;
> + }
> +
> + port_syfu_relay_info_insert(p, msg, fup);
> }
>
> err = port_prepare_and_send(p, fup, TRANS_GENERAL);
> --
> 2.7.4
_______________________________________________
Linuxptp-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
