[dpdk-users] mlx5 driver didn't set corresponding packet_type flag in TCP_ACK packet case

[Yongseok Koh]

On Tue, Mar 27, 2018 at 09:14:34AM +0200, Nélio Laranjeiro wrote:
> Hi Bin,
> 
> On Mon, Mar 26, 2018 at 06:09:35PM +0800, Bin Huang wrote:
> > Greetings all,
> > 
> > 
> > I have run l3fwd example program using two mlx5 NIC(
> > MLNX_OFED_LINUX-4.3-1.0.1.0 ) base on DPDK 17.08 and use pktgen as
> > flowgen.
> > 
> > This demo provided by DPDK works fine in UDP case, but in pktgen TCP
> > case, l3fwd can't forward packet to nexthop port.
> > 
> > 
> > After look into function rxq_cq_to_pkt_type(), I found while handling
> > IPv4 TCP ACK packets:
> > 
> > > l4_hdr_type bits[2:0] in CQE is (100)
> > 
> > > l3_hdr_type bits[1:0] in CQE is (10)
> > 
> > 
> > It looks fine according to CQE format in
> > <Ethernet_Adapters_Programming_Manual> doc:
> > 
> > > l4_hdr_type:
> >     0 - None
> >     1 - TCP header was present in the packet
> >     2 - UDP header was present in the packet
> >     3 - TCP header was present in the packet with Empty TCP ACK
> > indication. (TCP packet <ACK> flag is set, and packet carries no data)
> >     4 - TCP header was present in the packet with TCP ACK indication.
> > (TCP packet <ACK> flag is set, and packet carries data).
> > 
> > > l3_hdr_type:
> > 
> >     00 - None
> >     01 - IPv6
> >     10 - IPv4
> > 
> > 
> > But combined l4_hdr_type and l3_hdr_type into idx, the idx would be
> > 0x12, which was not reserved in mlx5_set_ptype_type().
> > 
> > Then RTE_PTYPE_UNKNOWN would be return to caller which caused
> > sub-sequence procedure going wrong.
> > 
> > 
> > Did I omit any possible configuration to make ptype flag work for TCP
> > ACK packets, or should I add (*p)[0x12] in mlx5_set_ptype_type() to
> > make this work?
> 
> According to the mlx5_set_ptype_table(), the index of the array is a
> little more complex:
> 
>  /*
>   * The index to the array should have:
>   * bit[1:0] = l3_hdr_type
>   * bit[4:2] = l4_hdr_type
>   * bit[5] = ip_frag
>   * bit[6] = tunneled
>   * bit[7] = outer_l3_type
>   */
> 
> If in the CQE the other three fields are 0, your hypothesis is correct
> and index 0x12 should be filled.
> 
> Can you also check it?

When I changed it to the current array-based conversion, I followed the use of 
MLX5_CQE_RX_TCP_PACKET [1], but it looks Bin is right. A packet should be
identified as TCP packet if l4_hdr_type is 1, 3 or 4.

Bin, will you send out a patch for it or shall I?
I'd love to see a patch from you. :-)


[1] http://dpdk.org/browse/dpdk/commit/?id=ea16068c00647fb6c7fe8704d8ad2adff6bf378f

Thanks,
Yongseok