Hi Bruce, This patch attempts to resolve the SCTP mismatching on 82599ES. A sample flow rule is as follows: flow create 0 ingress pattern eth / ipv4 dst is 245.194.135.241 src is 215.201.218.98 / sctp / end actions drop / end
NIC will pass the packet to host , however we expect the packet to be dropped. The reason for this is that ixgbe_parse_fdir_filter_normal() does not change the SCTP mask, which has a default value of 0. This causes HW to ignore L4 protocol type because the L4P register is set. (The default value is set by the following patch, called patch_df. https://patches.dpdk.org/project/dpdk/patch/20250418074309.705337-1-yuanx.w...@intel.com/) However, This patch seems to need to be discard due to another issue. This scenario is based on patch_df and create 2 rules on 82599ES. flow create 0 ingress pattern fuzzy thresh is 6 / ipv4 dst is 102.23.130.154 src is 70.247.152.105 / end actions queue index 6 / end flow create 0 ingress pattern fuzzy thresh is 4 / ipv4 dst is 193.23.234.17 src is 59.247.66.16 / udp dst is 57827 src is 23877 / end actions queue index 11 / end The second rule creation fails (ixgbe_flow.c line 3168) because the port mask is different from the first one. The first is 0x0 and the other is 0xffff. Because of this scenario, I think it would be better to retore the default value of port mask to 0xffff. Therefore, we should reconsider the patch_df solution. If all mask are formatted as 0xffff, raw IP packet will not match on E610, I think it is not appropriate to use port masks as L4P condition. I am considering using IXGBE_ATR_L4TYPE_MASK. if (info->mask.dst_port_mask == 0 && info->mask.src_port_mask == 0) fdirm |= IXGBE_FDIRM_L4P; Do you have any suggestions? Thanks, Yuan > -----Original Message----- > From: Richardson, Bruce <bruce.richard...@intel.com> > Sent: Monday, June 30, 2025 5:11 PM > To: Wang, YuanX <yuanx.w...@intel.com> > Cc: Burakov, Anatoly <anatoly.bura...@intel.com>; Medvedkin, Vladimir > <vladimir.medved...@intel.com>; dev@dpdk.org; sta...@dpdk.org > Subject: Re: [PATCH] net/ixgbe: fix sctp mask in flow director > > On Fri, Jun 27, 2025 at 04:59:53PM +0800, Yuan Wang wrote: > > Since the default value of the port mask is set to 0, the port mask > > does not change in some cases when creating SCTP flow rules, which > > results in incorrect L4P register configuration. > > > > This patch fixes this issue by setting the mask to 0xffff in these cases. > > > > Fixes: c81daae2383a (net/ixgbe: fix port mask default value in filter) > > Cc: sta...@dpdk.org > > > > Signed-off-by: Yuan Wang <yuanx.w...@intel.com> > > --- > > drivers/net/intel/ixgbe/ixgbe_flow.c | 2 ++ > > 1 file changed, 2 insertions(+) > > > > diff --git a/drivers/net/intel/ixgbe/ixgbe_flow.c > > b/drivers/net/intel/ixgbe/ixgbe_flow.c > > index 6278646720..9f2e470ad9 100644 > > --- a/drivers/net/intel/ixgbe/ixgbe_flow.c > > +++ b/drivers/net/intel/ixgbe/ixgbe_flow.c > > @@ -2161,6 +2161,8 @@ ixgbe_parse_fdir_filter_normal(struct > rte_eth_dev *dev, > > item, "Not supported by fdir filter"); > > return -rte_errno; > > } > > + rule->mask.src_port_mask = 0xffff; > > + rule->mask.dst_port_mask = 0xffff; > > } > > > > Hi, > > can you give a quick example of how to demonstrate the issue here, so I can > test the patch? Presumably without this patch some packets are incorrectly > classified/filtered based on the rte_flow rules? > > Thanks, > /Bruce
