Re: [RFC PATCH bpf-next v3 1/2] net: Rename mono_delivery_time to tstamp_type for scalabilty

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

> >>>>  static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
> >>>> -					 bool mono)
> >>>> +					  u8 tstamp_type)
> >>>>  {
> >>>>  	skb->tstamp = kt;
> >>>> -	skb->mono_delivery_time = kt && mono;
> >>>> +
> >>>> +	switch (tstamp_type) {
> >>>> +	case CLOCK_REAL:
> >>>> +		skb->tstamp_type = CLOCK_REAL;
> >>>> +		break;
> >>>> +	case CLOCK_MONO:
> >>>> +		skb->tstamp_type = kt && tstamp_type;
> >>>> +		break;
> >>>> +	}
> >>>
> >>> Technically this leaves the tstamp_type undefined if (skb, 0, CLOCK_REAL)
> >> Do you think i should be checking for valid value of tstamp before setting the tstamp_type ? Only then set it. 
> > 
> > A kt of 0 is interpreted as resetting the type. That should probably
> > be maintained.
> > 
> > For SO_TIMESTAMPING, a mono delivery time of 0 does have some meaning.
> > In __sock_recv_timestamp:
> > 
> >         /* Race occurred between timestamp enabling and packet
> >            receiving.  Fill in the current time for now. */
> >         if (need_software_tstamp && skb->tstamp == 0) {
> >                 __net_timestamp(skb);
> >                 false_tstamp = 1;
> >         }
> 
> Well in that case the above logic still resets the tstamp and sets the tstamp_type to CLOCK_REAL(value 0). 
> Anyway the tstamp_type will be 0 to begin with. 
> The logic is still inline with previous implementation, because previously if kt was 0 then kt && mono sets the tstamp_type (previously called as mono_delivery_time) to 0 (i.e SKB_CLOCK_REAL). 

Sorry, I got my defaults confused. If we maintain that a zero tstamp
resets the type, then here should be no case with skb->tstamp 0 and
skb->tstamp_type SKB_CLOCK_REAL (or SKB_CLOCK_TAI or whatever). I
think it's preferable to make that obvious in the
skb_set_delivery_time implementation, rather than depend on knowledge
of its callers.