On Mon, Sep 04, 2023 at 06:30:06PM +0700, James Clark wrote:
> "chronyc serverstats" looks good: I'm seeing increasing "NTP hardware RX
> timestamps" and
> "NTP hardware TX timestamps".
Is the server running on CM4?
Yes.
> I'm now trying two clients, and I see different results with chronyc
> ntpdata:
>
> - another CM4: here I see RX timestamping : Hardware but TX timestamping :
> Kernel
Can you please enable the measurements log and see if all TX
timestamps are kernel or there is a mix of kernel and hardware?
It's a mixture: 848 "K H"; 22 "K K"; 13 "H H".
The HW
timestamping might be unreliable. You could try increasing hwtstimeout
to 1 to wait even longer for the TX timestamp.
You could also try adding "minpoll 2" to the hwtimestamp directive to
read the clock less frequently to avoid some collisions with the
timestamping access. IIRC the HW or driver might have an issue with
that.
I tried increasing both hwtstimeout and minpoll, but it hasn't fixed the problem.
> - an x86 box with the AQC107: here I see TX timestamping : Hardware and RX
> timestamping : Hardware
> but on the client "chronyc ntpdata" shows Kernel for TX timestamping and RX
> timestamping.
I have no experience with this NIC. What does ethtool -T show?
It shows:
Time stamping parameters for enp1s0:
Capabilities:
hardware-transmit
software-transmit
hardware-receive
software-receive
software-system-clock
hardware-raw-clock
PTP Hardware Clock: 3
Hardware Transmit Timestamp Modes:
off
on
Hardware Receive Filter Modes:
none
ptpv2-l4-event
ptpv2-l2-event
ptpv2-event
The measurements here are consistently K K, apart from the first one which is H H.
James