Message-ID: <20230116130243.ghtythag3og6745y@skbuf>
Date:   Mon, 16 Jan 2023 15:02:43 +0200
From:   Vladimir Oltean <olteanv@...il.com>
To:     nikhil.gupta@....com
Cc:     linux-arm-kernel@...ts.infradead.org,
        Yangbo Lu <yangbo.lu@....com>, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org, vakul.garg@....com,
        rajan.gupta@....com, richardcochran@...il.com
Subject: Re: [PATCH v1] ptp_qoriq: fix latency in ptp_qoriq_adjtime()
 operation.

Hi Nikhil,

On Mon, Jan 16, 2023 at 03:54:40PM +0530, nikhil.gupta@....com wrote:
> From: Nikhil Gupta <nikhil.gupta@....com>
> 
> 1588 driver loses about 1us in adjtime operation at PTP slave.
> This is because adjtime operation uses a slow non-atomic tmr_cnt_read()
> followed by tmr_cnt_write() operation.
> In the above sequence, since the timer counter operation keeps
> incrementing, it leads to latency.
> The tmr_offset register (which is added to TMR_CNT_H/L register
> gives the current time) must be programmed with the delta
> nanoseconds.
> 
> Signed-off-by: Nikhil Gupta <nikhil.gupta@....com>
> Reviewed-by: Yangbo Lu <yangbo.lu@....com>
> ---

Your patch breaks eTSEC1 [ and eTSEC2 ] on LS1021A.

Before:

root@...ck:~# ip link set eth1 up
root@...ck:~# [   54.321664] fsl-gianfar soc:ethernet@...0000 eth1: Link is Up - 1Gbps/Full - flow control off
[   54.331331] IPv6: ADDRCONF(NETDEV_CHANGE): eth1: link becomes ready

root@...ck:~# ptp4l -i eth1 -2 -P -m
ptp4l[57.231]: selected /dev/ptp0 as PTP clock
ptp4l[57.351]: port 1: INITIALIZING to LISTENING on INIT_COMPLETE
ptp4l[57.353]: port 0: INITIALIZING to LISTENING on INIT_COMPLETE
ptp4l[57.355]: port 1: link down
ptp4l[57.356]: port 1: LISTENING to FAULTY on FAULT_DETECTED (FT_UNSPECIFIED)
ptp4l[57.455]: selected local clock 001f7b.fffe.630228 as best master
ptp4l[57.458]: port 1: assuming the grand master role
ptp4l[57.464]: selected local clock 001f7b.fffe.630228 as best master
ptp4l[57.466]: port 1: assuming the grand master role
[   60.086366] fsl-gianfar soc:ethernet@...0000 eth1: Link is Up - 1Gbps/Full - flow control off
ptp4l[60.089]: port 1: link up
ptp4l[60.183]: port 1: FAULTY to LISTENING on INIT_COMPLETE
ptp4l[66.713]: port 1: LISTENING to MASTER on ANNOUNCE_RECEIPT_TIMEOUT_EXPIRES
ptp4l[66.714]: selected local clock 001f7b.fffe.630228 as best master
ptp4l[66.715]: port 1: assuming the grand master role
ptp4l[102.753]: port 1: new foreign master 00049f.fffe.05f4ad-1
ptp4l[106.757]: selected best master clock 00049f.fffe.05f4ad
ptp4l[106.758]: port 1: MASTER to UNCALIBRATED on RS_SLAVE
ptp4l[107.763]: master offset -363290000824 s0 freq      +0 path delay       735
ptp4l[108.764]: master offset -363289992530 s1 freq   +8285 path delay       736
ptp4l[109.767]: master offset      -1648 s2 freq   +6637 path delay       736
ptp4l[109.768]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
ptp4l[110.770]: master offset         -6 s2 freq   +7784 path delay       736
ptp4l[111.772]: master offset        491 s2 freq   +8280 path delay       735
ptp4l[112.773]: master offset        495 s2 freq   +8431 path delay       735
ptp4l[113.776]: master offset        342 s2 freq   +8426 path delay       731
ptp4l[114.778]: master offset        198 s2 freq   +8385 path delay       724
ptp4l[115.781]: master offset        100 s2 freq   +8346 path delay       722

After:

root@...ck:~# ip link set eth1 up
root@...ck:~# [  311.001924] fsl-gianfar soc:ethernet@...0000 eth1: Link is Up - 1Gbps/Full - flow control off
[  311.013236] IPv6: ADDRCONF(NETDEV_CHANGE): eth1: link becomes ready

root@...ck:~# ptp4l -i eth1 -2 -P -m
ptp4l[333.797]: selected /dev/ptp0 as PTP clock
ptp4l[333.916]: port 1: INITIALIZING to LISTENING on INIT_COMPLETE
ptp4l[333.918]: port 0: INITIALIZING to LISTENING on INIT_COMPLETE
ptp4l[333.919]: port 1: link down
ptp4l[333.919]: port 1: LISTENING to FAULTY on FAULT_DETECTED (FT_UNSPECIFIED)
ptp4l[334.012]: selected local clock 001f7b.fffe.630228 as best master
ptp4l[334.013]: port 1: assuming the grand master role
[  336.626213] fsl-gianfar soc:ethernet@...0000 eth1: Link is Up - 1Gbps/Full - flow control off
ptp4l[336.631]: port 1: link up
ptp4l[336.713]: port 1: FAULTY to LISTENING on INIT_COMPLETE
ptp4l[343.832]: port 1: LISTENING to MASTER on ANNOUNCE_RECEIPT_TIMEOUT_EXPIRES
ptp4l[343.833]: selected local clock 001f7b.fffe.630228 as best master
ptp4l[343.834]: port 1: assuming the grand master role
ptp4l[344.873]: port 1: new foreign master 00049f.fffe.05f4ad-1
ptp4l[348.879]: selected best master clock 00049f.fffe.05f4ad
ptp4l[348.879]: port 1: MASTER to UNCALIBRATED on RS_SLAVE
ptp4l[349.888]: master offset 215392521359 s0 freq      +0 path delay       737
ptp4l[350.890]: master offset 215392529808 s1 freq   +8433 path delay       738
ptp4l[351.896]: master offset 215392529845 s2 freq +32767999 path delay       735
ptp4l[351.896]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
ptp4l[352.897]: master offset 215359819000 s2 freq +32767999 path delay       731
ptp4l[353.899]: master offset 215326980590 s2 freq +32767999 path delay       731
ptp4l[354.900]: master offset 215294167912 s2 freq +32767999 path delay       729
ptp4l[355.902]: master offset 215261357366 s2 freq +32767999 path delay       725
ptp4l[356.905]: master offset 215228532618 s2 freq +32767999 path delay    -24741
ptp4l[357.907]: master offset 215195717627 s2 freq +32767999 path delay    -51970
ptp4l[358.909]: master offset 215162898062 s2 freq +32767999 path delay    -57014
ptp4l[359.912]: master offset 215130057602 s2 freq +32767999 path delay    -59714
ptp4l[360.915]: master offset 215097182978 s2 freq +32767999 path delay    -60737
ptp4l[361.917]: master offset 215064357344 s2 freq +32767999 path delay    -60783

I have forwarded you an internal email which gives more details about
how the 1588 block was integrated on the 3 eTSEC ports on LS1021A.
The general takeaway is: please don't break that SoC, and NACK for this
patch.

Powered by blists - more mailing lists