Re: [RFC][PATCH 1/3] net: per skb control messages

[Stephen Hemminger <shemminger@vyatta.com>]

On Thu, 24 Jul 2008 14:56:00 -0700 (PDT)
David Miller <davem@davemloft.net> wrote:

> From: Octavian Purdila <opurdila@ixiacom.com>
> Date: Fri, 25 Jul 2008 00:49:46 +0300
> 
> > On Thursday 24 July 2008, David Miller wrote:
> > 
> > > Adding new fields to struct sk_buff that take up space is generally
> > > not allowed unless the new field adds substantially to the benefit of
> > > a large group of users of Linus.
> > >
> > > I don't think that applied here for this hw-tstamps stuff.
> > 
> > What about the approach proposed in the patch? Is it ok to add a pointer which 
> > may resolve other future similar issues?
> 
> Traversing the list and maintaining the reference counting and sharing
> issues is expensive and error prone.
> 
> This is what OpenBSD uses for their IPSEC state attached to MBUFs and
> it's a nightmare.
> 
> We have a timestamp in the SKB already, why don't you simply override
> it when your feature is enable and set a single flag bit that
> indicates you used a HW timestamp to set that timestamp?

How hard would it be to add PLL support to use same kind of timestamp.
Enlist the help of some NTP/clock experts to help.