Re: [PATCH nf-next v2] netfilter: conntrack: collect start time as early as possible

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Wed, Nov 06, 2024 at 09:26:44AM +0100, Florian Westphal wrote:
> Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > On Tue, Nov 05, 2024 at 06:32:47PM +0100, Florian Westphal wrote:
> > > Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > > > Thanks, I'd rather convince you this is the way to go, if after
> > > > quickly sketching a patchset you think it is not worth for more
> > > > reasons, we can revisit.
> > > 
> > > Untested.  I'm not sure about skb_tstamp() usage.
> > > As-is CTA_EVENT_TIMESTAMP in the NEW event would be before
> > > the start time reported as the start time by the timestamp extension.
> > 
> > Is there any chance this timestamp can be enabled via toggle?
> 
> Can you clarify?  Do you mean skb_tstamp() vs ktime_get_real_ns()
> or tstamp sampling in general?

I am referring to ktime_get_real_ns(), I remember to have measured
25%-30% performance drop when this is used, but I have not refreshed
those numbers for long time.

As for skb_tstamp(), I have to dig in the cost of it.

> > > +	CTA_EVENT_TIMESTAMP,
> > 
> >         CTA_TIMESTAMP_EVENT
> > 
> > for consistency with CTA_TIMESTAMP_{START,...}
> 
> Sure, updated.