Hi Chuanyu,
> -----Original Message-----
> From: Chuanyu Xue <chuanyu....@uconn.edu>
> Sent: Saturday, December 30, 2023 5:30 AM
> To: Su, Simei <simei...@intel.com>
> Cc: Xing, Beilei <beilei.x...@intel.com>; chuanyu....@uconn.edu;
> dev@dpdk.org; Zhang, Qi Z <qi.z.zh...@intel.com>; Lu, Wenzhuo
> <wenzhuo...@intel.com>
> Subject: RE: [PATCH] net/e1000: support launchtime feature
>
> >>
> >> >> +static int
> >> >> +eth_igb_read_clock(__rte_unused struct rte_eth_dev *dev, uint64_t
> >> >> +*clock) {
> >> >> + uint64_t systime_cycles;
> >> >> + struct e1000_adapter *adapter = dev->data->dev_private;
> >> >> +
> >> >> + systime_cycles = igb_read_systime_cyclecounter(dev);
> >> >> + uint64_t ns = rte_timecounter_update(&adapter->systime_tc,
> >> >> systime_cycles);
> >> >
> >> >Do you also run "ptp timesync" when testing this launchtime feature?
> >> >
> >>
> >> I used `rte_eth_timesync_enable` function during the test. I am not
> >> familiar with the `ptp timesync` in DPDK. If you are referring to
> >> something else, could you please guide me on how to test it?
> >
> >Do you use your own application or DPDK application to test this
> >launchtime feature, for example, dpdk testpmd?
>
> Yes, I used my own application to test it. The benefit of launch time feature
> in
> boundable delay and jitter is significant compared with when it is disabled.
>
> Specifically, my app periodically calls `rte_eth_tx_burst` with
> `rte_dynfield_timestamp` field on talker, and compares whether the receiving
> time in NIC hardware timestamp on listener is as expected. Talker and listener
> are directly connected by a RJ45 cable, both installed with i210 NIC. The
> feature works perfect in my test.
OK, it sounds good.
>
> I also tested it with testpmd with `txtimes` config. However it seems there
> is an
> issue in testpmd. Specifically the tx_only mode sends packets as fast as
> possible, results in an increasing gap between the current time and the
> scheduled transmission time.
> Based on i210 NIC sheet Sec 7.2.2.2.3, the launch time should be within
> (current_time, current time + 0.5 Sec), thus most of tx packets are not
> scheduled.
> I got the similar test results with dpdk igc driver which already implemeted
> launch time feature.
>
> Following is how I try to test with testpmd. Please let me know if I did
> something wrong.
>
> sudo ./dpdk-testpmd -- -i --forward-mode=txonly
>
> testpmd> port stop 0
> testpmd> set burst 1
> testpmd> set txtimes 100000000,0
> testpmd> port config 0 tx_offload send_on_timestamp on
> testpmd> port start 0
> testpmd> start
When testing launch time feature with igc driver, firstly, some code change
made in txonly.c:
pkt->ol_flags |= RTE_MBUF_F_TX_IEEE1588_TMST; (this flag should be added to
forward PTP packet with hardware Tx timestamp)
# ./build/app/dpdk-testpmd -a 0000:81:00.0 -c f -n 4 -- -i
--tx-offloads=0x200000
testpmd> set burst 1
testpmd> set fwd txonly
testpmd> set txtimes 1000000,0
testpmd> start
On receiver side (with tcpdump):
# tcpdump -Q in -ttt -ni ens25f3 --time-stamp-precision=nano -j
adapter_unsynced -c 32
Thanks,
Simei
>
> >
> >> +-----------+---------------+---------------+---------------+---------------+---------------+
> >> | 1G | 16880 | 16880 | 16880 |
> 16880
> >> | 16880 |
> >> +-----------+---------------+---------------+---------------+---------------+---------------+
> >>
> >> Any suggestions here? Is it supposed to be embedded directly here or
> >> left to the application level to compensate? I can fix it accordingly.
> >
> >I think it can be put here directly just as you do.
>
> Got it. Will keep this delay compensiation here and revise it in the next
> batch
> version.
