[SDFAB-1097] Do not generate INT report for pruned packets #502
Conversation
| @@ -297,7 +297,7 @@ control IntEgress (inout v1model_header_t hdr_v1model, | |||
| fabric_md.int_report_md.mirror_type = FabricMirrorType_t.INT_REPORT; | |||
| fabric_md.int_report_md.report_type = fabric_md.bridged.int_bmd.report_type; | |||
| fabric_md.int_report_md.ig_port = fabric_md.bridged.base.ig_port; | |||
| fabric_md.int_report_md.eg_port = (PortId_t)fabric_v1model.recirc_preserved_egress_port; | |||
| fabric_md.int_report_md.eg_port = (PortId_t)standard_md.egress_port; | |||
There was a problem hiding this comment.
based on the logic from fabric-tna, we use the egress port number from intrinsic metadata, which is the standard metadata of v1model
There was a problem hiding this comment.
@EmanueleGallone 👆does this change make sense?
There was a problem hiding this comment.
Yes it does, because init_int_metadata is performed BEFORE cloning/recirculating the packet. the standard_md.egress_port information is lost AFTER cloning/recirculating the packet.
This is also confirmed by the CI, otherwise I would've expected some INT tests to fail when asserting the original packet's egress port. BTW, Thanks for catching this.
Codecov Report
@@ Coverage Diff @@
## main #502 +/- ##
=========================================
Coverage 69.85% 69.85%
+ Complexity 730 729 -1
=========================================
Files 63 63
Lines 4771 4771
Branches 528 528
=========================================
Hits 3333 3333
Misses 1161 1161
Partials 277 277 Continue to review full report at Codecov.
|
Yi-Tseng
changed the title
| @@ -2149,8 +2148,12 @@ def runIPv4MulticastTest(self, pkt, in_port, out_ports, in_vlan, out_vlan): | |||
| # Send packets and verify | |||
| self.send_packet(in_port, pkt) | |||
| for out_port in out_ports: | |||
| if out_port == in_port: | |||
| # Packet will be dropped by egress pipeline. | |||
| continue | |||
There was a problem hiding this comment.
Does that mean before we were expecting to receive a packet on in_port? If yes, since the pipeline performs inport pruning, how is it possible that the multicast test was passing before this change?
There was a problem hiding this comment.
For IPv4 multicast test, we didn't include ingress port to the egress port list, which means the clone group won't clone the packet to the ingress port
https://github.com/stratum/fabric-tna/blob/main/ptf/tests/unary/test.py#L118-L119
When sending multicast/broadcast traffic, we always drop packets if the egress port is equal to the ingress port.
This caused the INT pipeline to generate unnecessary INT drop reports.
To avoid this, we can set the report type to NO_REPORT for this case.