• Vladimir Oltean's avatar
    net: dsa: sja1105: implement tc-gate using time-triggered virtual links · 834f8933
    Vladimir Oltean authored
    Restrict the TTEthernet hardware support on this switch to operate as
    closely as possible to IEEE 802.1Qci as possible. This means that it can
    perform PTP-time-based ingress admission control on streams identified
    by {DMAC, VID, PCP}, which is useful when trying to ensure the
    determinism of traffic scheduled via IEEE 802.1Qbv.
    
    The oddity comes from the fact that in hardware (and in TTEthernet at
    large), virtual links always need a full-blown action, including not
    only the type of policing, but also the list of destination ports. So in
    practice, a single tc-gate action will result in all packets getting
    dropped. Additional actions (either "trap" or "redirect") need to be
    specified in the same filter rule such that the conforming packets are
    actually forwarded somewhere.
    
    Apart from the VL Lookup, Policing and Forwarding tables which need to
    be programmed for each flow (virtual link), the Schedule engine also
    needs to be told to open/close the admission gates for each individual
    virtual link. A fairly accurate (and detailed) description of how that
    works is already present in sja1105_tas.c, since it is already used to
    trigger the egress gates for the tc-taprio offload (IEEE 802.1Qbv). Key
    point here, we remember that the schedule engine supports 8
    "subschedules" (execution threads that iterate through the global
    schedule in parallel, and that no 2 hardware threads must execute a
    schedule entry at the same time). For tc-taprio, each egress port used
    one of these 8 subschedules, leaving a total of 4 subschedules unused.
    In principle we could have allocated 1 subschedule for the tc-gate
    offload of each ingress port, but actually the schedules of all virtual
    links installed on each ingress port would have needed to be merged
    together, before they could have been programmed to hardware. So
    simplify our life and just merge the entire tc-gate configuration, for
    all virtual links on all ingress ports, into a single subschedule. Be
    sure to check that against the usual hardware scheduling conflicts, and
    program it to hardware alongside any tc-taprio subschedule that may be
    present.
    
    The following scenarios were tested:
    
    1. Quantitative testing:
    
       tc qdisc add dev swp2 clsact
       tc filter add dev swp2 ingress flower skip_sw \
               dst_mac 42:be:24:9b:76:20 \
               action gate index 1 base-time 0 \
               sched-entry OPEN 1200 -1 -1 \
               sched-entry CLOSE 1200 -1 -1 \
               action trap
    
       ping 192.168.1.2 -f
       PING 192.168.1.2 (192.168.1.2) 56(84) bytes of data.
       .............................
       --- 192.168.1.2 ping statistics ---
       948 packets transmitted, 467 received, 50.7384% packet loss, time 9671ms
    
    2. Qualitative testing (with a phase-aligned schedule - the clocks are
       synchronized by ptp4l, not shown here):
    
       Receiver (sja1105):
    
       tc qdisc add dev swp2 clsact
       now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
               sec=$(echo $now | awk -F. '{print $1}') && \
               base_time="$(((sec + 2) * 1000000000))" && \
               echo "base time ${base_time}"
       tc filter add dev swp2 ingress flower skip_sw \
               dst_mac 42:be:24:9b:76:20 \
               action gate base-time ${base_time} \
               sched-entry OPEN  60000 -1 -1 \
               sched-entry CLOSE 40000 -1 -1 \
               action trap
    
       Sender (enetc):
       now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
               sec=$(echo $now | awk -F. '{print $1}') && \
               base_time="$(((sec + 2) * 1000000000))" && \
               echo "base time ${base_time}"
       tc qdisc add dev eno0 parent root taprio \
               num_tc 8 \
               map 0 1 2 3 4 5 6 7 \
               queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
               base-time ${base_time} \
               sched-entry S 01  50000 \
               sched-entry S 00  50000 \
               flags 2
    
       ping -A 192.168.1.1
       PING 192.168.1.1 (192.168.1.1): 56 data bytes
       ...
       ^C
       --- 192.168.1.1 ping statistics ---
       1425 packets transmitted, 1424 packets received, 0% packet loss
       round-trip min/avg/max = 0.322/0.361/0.990 ms
    
       And just for comparison, with the tc-taprio schedule deleted:
    
       ping -A 192.168.1.1
       PING 192.168.1.1 (192.168.1.1): 56 data bytes
       ...
       ^C
       --- 192.168.1.1 ping statistics ---
       33 packets transmitted, 19 packets received, 42% packet loss
       round-trip min/avg/max = 0.336/0.464/0.597 ms
    Signed-off-by: default avatarVladimir Oltean <vladimir.oltean@nxp.com>
    Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
    834f8933
sja1105_flower.c 12.5 KB