Re: [Linuxptp-users] Synchronization fault with stmmac

[Mohamed B. <mbe...@ad...>]

Richard,

Could you please give us your insight on this matter?
I would appreciate any lead to pinpoint the cause :).

Thanks & Best Regards,
Mohamed

> On Thu, 2014-03-20 at 18:03 +0100, Mohamed Belaouad wrote:
> > Jake,
> > 
> > > Out of curiosity where are you printing the value ppb?
> > 
> > I am printing right at the beginning of stmmac_adjust_freq function
> > (drivers/net/ethernet/stmicro/stmmac/stmmac_ptp.c).
> > So the value should be clean.
> > 
> > > > 
> > > > The driver implementation is not checking the max frequency
> > > > adjustment...
> > > > I still did some tests and strange enough there is a value change to
> > > > -32,768,000 when passing ppb greater than 32,768,000. Also, the value
> > > > saturates to -32,768,000 with ppb lesser than -32,768,000.
> > > > So I am forced to use values between [-32,768,000 | 32,768,000]. I
> > > > guess
> > > > your test suggestion will pass once I get through this.
> > > > 
> > > > Could you please shed some light on that? Is it something related with
> > > > the
> > > > kernel ptp clock driver? I did not find
> > > > 
> > > 
> > > Your driver does have to fill in the maximum value to register to the
> > > ptp core, and ptp core handles checking, but maybe there is a bug
> > > regarding that in your kernel... It should be keeping the sign the same
> > > when it clips the maximum value, however.
> > > 
> > > It should be changing it to the max adjust, non negative.
> > > 
> > > Look at your driver, where you set the max_adj field of ptp_caps.
> > > 
> > 
> > It is correctly filled with 62,500,000 and I checked with testptp. So there
> > is still some margin.
> > I don't know if I am looking at the right location but I did not find any
> > check on max_adj of ppb in drivers/ptp.
> > 
> > 
> > Thanks,
> > Mohamed
> 
> That's interesting.. It's possible you have a bug in the kernel you are
> using.. That'd be my only suggestion.. Maybe Richard who wrote the ptp
> core module knows more about this..
> 
> Thanks,
> Jake
>