Re: [dpdk-dev] [PATCH 2/2] doc: announce new mbuf field for LRO

[Olivier Matz <olivier.matz@6wind.com>]

On Mon, Jul 27, 2020 at 08:41:47AM +0000, Matan Azrad wrote:
> 
> Hi Olivier
> 
> From: Olivier Matz:
> > Hi,
> > 
> > On Tue, Jun 02, 2020 at 06:49:01AM +0000, Matan Azrad wrote:
> > > Hi
> > >
> > > From: Thomas Monjalon
> > > > 10/08/2019 23:31, Thomas Monjalon:
> > > > > 06/08/2019 20:17, Andrew Rybchenko:
> > > > > > On 8/6/19 5:56 PM, Matan Azrad wrote:
> > > > > > > The API breakage is because the ``tso_segsz`` field was
> > > > > > > documented for LRO.
> > > > > > >
> > > > > > > The ``tso_segsz`` field in mbuf indicates the size of each
> > > > > > > segment in the LRO packet in Rx path and should be provided by
> > > > > > > the LRO packet port.
> > > > > > >
> > > > > > > While the generic LRO packet may aggregate different segments
> > > > > > > sizes in one packet, it is impossible to expose this
> > > > > > > information for each segment by one field and it doesn't make
> > > > > > > sense to expose all the segments sizes in the mbuf.
> > > > > > >
> > > > > > > A new field may be added as union with the above field to
> > > > > > > expose the number of segments aggregated in the LRO packet.
> > > > > > >
> > > > > > > Signed-off-by: Matan Azrad <matan@mellanox.com>
> > > > > > > ---
> > > > > > > --- a/doc/guides/rel_notes/deprecation.rst
> > > > > > > +++ b/doc/guides/rel_notes/deprecation.rst
> > > > > > > +* mbuf: Remove ``tso_segsz`` mbuf field providing for LRO
> > > > > > > +support. Use union
> > > > > > > +  block for the field memory to be shared with a new field
> > > > > > > +``lro_segs_n``
> > > > > > > +  indicates the number of segments aggregated in the LRO packet.
> > > > > >
> > > > > > I think that the number of segments is more logical in the case of LRO.
> > > > > > The question (already asked by Konstantin) is why it is needed
> > > > > > at all (statistics?). If so, it still makes sense.
> > > > > >
> > > > > > Segment size is misleading here, since not all segments could be
> > > > > > the same size. So,
> > > > > >
> > > > > > Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
> > > > > >
> > > > > > As far as I can see bnxt and qede do not fill it in.
> > > > > > mlx5 and vmxnet3 have the number of segments (vmxnet3 has
> > > > > > segment size sometimes and sometimes use a function to guess the
> > value).
> > > > > > So both will win from the change.
> > > > > > It looks like virtio does not have number of segments. CC Maxime
> > > > > > to comment.
> > > > >
> > > > > I support improving the API for LRO.
> > > > > Unfortunately, the consensus is not strong enough at the moment.
> > > >
> > > > We had no progress about LRO field in mbuf.
> > > > Is it a change we would like to have in 20.11?
> > > >
> > > +1 to make the change.
> > 
> > Thinking about this, having the segment size for LRO is useful: it is expected
> > to give the size of the segments, except the last one. The advantage of this is
> > to be able to resegment exactly at the same MSS on Tx (so it does not break
> > PMTU). This is needed if you want to do a bridge or a router with LRO
> > enabled.
> 
> Yes, you right it may be useful.
> 
> I don't familiar with other vendors, but mlx5 HWs supply the number of segments aggregated by the HW and doesn't assume all the head segments are in the same size.
> 
> So, we just put in the current field packet_size/num of segments.
> 
> So, maybe we need the 2 options as uinion or as 2 separated fields to be more accurate.

Yes.

Having ethdev capas would help to know what the PMD can do (basically,
reassembly of same segment sizes or not).

Renaming PKT_RX_LRO to PKT_RX_LRO_SEGSIZE could make sense, and the new
one PKT_RX_LRO_PKTNUM could be added in case the packet number is
provided. I have no strong opinion about union vs 2 separate fields.

If a new field is added, I think we should prefer a dynamic mbuf field.


> 
> > What is described above is more GRO than LRO. But today, it is possible to do
> > this with the virtio driver, so removing the segment size would break this use
> > case.
> > 
> > Regards,
> > Olivier