Re: [PATCH net-next RFC v4 4/5] net: Let the active time stamping layer be selectable.

["Köry Maincent" <kory.maincent@bootlin.com>]

On Thu, 11 May 2023 16:35:47 -0700
Jakub Kicinski <kuba@kernel.org> wrote:

> On Fri, 12 May 2023 02:18:03 +0300 Vladimir Oltean wrote:
> > > Why can't we treat ndo_hwtstamp_set() == -EOPNOTSUPP as a signal 
> > > to call the PHY? ndo_hwtstamp_set() does not exist, we can give
> > > it whatever semantics we want.    
> > 
> > Hmm, because if we do that, bridged DSA switch ports without hardware
> > timestamping support and without logic to trap PTP to the CPU will just
> > spew those PTP frames with PHY hardware timestamps everywhere, instead
> > of just telling the user hey, the configuration isn't supported?  
> 
> I see, so there is a legit reason to abort. 
> 
> We could use one of the high error codes, then, to signal 
> the "I didn't care, please carry on to the PHY" condition?
> -ENOTSUPP?
> 
> I guess we can add a separate "please configure traps for PTP/NTP" 
> NDO, if you prefer. Mostly an implementation detail.

I am not as expert as you on the network stack therefore I am trying to follow
and understand all the remarks. Please correct me if I say something wrong. It
is interesting to understand all the complications that these changes bring.

To summary, what do you think is preferable for this patch series?
- New ops for TS as suggested by Russell.

- Continue on this implementation and check that Vladimir A,B and C cases are
  handled. Which imply, if I understand well, find a good way to deal with PTP
  change trap (bit or new ndo ops), convert most drivers from IOCTL to NDO
  beforehand. 

- Add MAC-DMA TS? It think it is needed as MAC-DMA TS seems already used and
  different from simple MAC TS in term of quality, as described by Jakub.