Re: [Linuxptp-devel] [PATCHv5 00/12] Add linuxptp bond failover support

[Miroslav L. <mli...@re...>]

On Sat, Sep 30, 2017 at 04:25:08PM +0800, Hangbin Liu wrote:
> This patch set is to add linuxptp bond fail over support.

> v4 -> v5: Based on Miroslav's tests and advises.

Thanks for the fixes and improvements. In my testing it all seems to
be working as expected, with both HW and SW timestamping.

I was testing the failover in this setup:

CLOCK_REALTIME -> phc2sys -a -r -r -> bond(I210, I350) -> ptp4l
-> LAN ->
ptp4l -> bond(I350, 82579LM) -> phc2sys -a -r -> CLOCK_REALTIME

I suspect the default fault reset interval (16 seconds) might be a bit
too long for people interested in this failover, but it is
configurable.

> 1. For [PATCHv4 08/10] ptp4l: use ts label to get ts info
>   - In function port_link_status() we should only switch phc with HW time
>     stamping mode.
> 2. For [PATCHv4 10/10] phc2sys: update clock clkid and phc_index if device changed
>   - Split this patch into three small patches to make the logic more clear.
>     So now we have two parts for this patchset. Patches 1-9 are to add ptp4l
>     bond failover support. And patches 10-12 are to add phc2sys failover
>     support.
>   - Split servo_add from function clock_add for later use.
>   - In reconfigure() we only do clock_reinit() when new state is PS_MASTER
>     or PS_SLAVE. But when get failover the state changed to PS_FAULTY first,
>     during this time we still choose the old phc device instead of new phc.
>     Fix this by call clock_reinit() whenever we have a new state.
>   - In clock_reinit(), the new phc device may have different maximum
>     adjustment, then it will try to make a larger adjustment than the HW can
>     handle and breaks the synchronization. So we need to destroy the old servo
>     and create a new one.

-- 
Miroslav Lichvar