[dpdk-dev] [PATCH] mbuf: fix tunnel flags check

[Thomas Monjalon]

2015-07-16 00:36, Zhang, Helin:
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > 2015-07-15 23:57, Zhang, Helin:
> > > From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > > > A packet is tunnelled if the tunnel type is identified or if it has an inner part.
> > > >
> > > > Fix also a typo in RTE_PTYPE_INNER_L3_MASK.
> > > >
> > > > Fixes: f295a00a2b44 ("mbuf: add definitions of unified packet
> > > > types")
> > > >
> > > > Signed-off-by: Thomas Monjalon <thomas.monjalon at 6wind.com>
> > |...]
> > > >  /* Check if it is a tunneling packet */  -#define
> > > > RTE_ETH_IS_TUNNEL_PKT(ptype) ((ptype) & RTE_PTYPE_TUNNEL_MASK)
> > > > +#define RTE_ETH_IS_TUNNEL_PKT(ptype) ((ptype) &
> > > > +(RTE_PTYPE_TUNNEL_MASK | \  RTE_PTYPE_INNER_L2_MASK | \
> > > > +RTE_PTYPE_INNER_L3_MASK | \
> > > > + RTE_PTYPE_INNER_L4_MASK))
> > >
> > > Could you help to explain more of why here?
> > > My understanding is that if an inner one can be recognized, there must be a
> > tunnel type there.
> > 
> > Not always.
> > It was my comment in mlx4 patch:
> > 	http://dpdk.org/ml/archives/dev/2015-July/021702.html
> > Currently we can know that mlx4 has detected a tunnel but don't know which
> > one.
> 
> I'd suggest to do more in mlx4 driver, rather than adding more checks in this macro.
> If it can detect the inner l2/l3 type, the driver should tell it is a tunneled packet.
> If it cannot know which tunnel it is, I'd suggest to add one more tunnel type of
> RTE_PTYPE_TUNNEL_UNKNOWN.
> Two reasons:
> - PMD should tell enough info to high level caller or application. It should be clear enough.

RTE_PTYPE_TUNNEL_UNKNOWN doesn't make it clearer, IMHO.

> - Adding more checks in those macro results in more cpu cycles for other NICs to check
> the packet types.

Not sure. It only extends the mask to check. It shouldn't add more cpu cycle.