Broadcom BCM57608 Ethernet 100Gb 2-port QSFP112 Adapter - PCI-PT and SRIOV on VM Perf Degrade in 25.11 and 26.03
http://bugs.dpdk.org/show_bug.cgi?id=1931
Bug ID: 1931
Summary: Broadcom BCM57608 Ethernet 100Gb 2-port QSFP112
Adapter - PCI-PT and SRIOV on VM Perf Degrade in 25.11
and 26.03
Product: DPDK
Version: 25.11
Hardware: All
OS: Linux
Status: UNCONFIRMED
Severity: normal
Priority: Normal
Component: core
Assignee: [email protected]
Reporter: [email protected]Target Milestone: ---
Created attachment 347
--> http://bugs.dpdk.org/attachment.cgi?id=347&action=edit
git bisect search offending commit perf test results
[Steps to Reproduce]
1. Install RHEL 10 - 6.12.0-55.43.1.el10_0.x86_64
2. Tested with following latest BCM57608 driver and firmware:
Driver:1.10.3-236.1.155.0 || firmware:235.1.164.6/pkg 235.1.164.14
Driver:1.10.3-237.1.131.0 || firmware:237.1.131.0/pkg 237.1.132.03. Install DPDK - Tested with 25.11 and 26.03
4. Start testpmd; command: with tuning params build/app/dpdk-testpmd -l 2-4 --main-lcore 2 -v -a 0000:07:00.0 -a
0000:08:00.0 --socket-mem=2048 -- -i --socket-num=0 --burst=64 --txd=2048 --rxd=2048 --mbcache=512 --rxq=1 --txq=1 --nb-cores=2 --port-topology=paired --forward-mode=io '--port-numa-config=(0,0),(1,0)' --ring-numa-config=0,0 --ring-numa-config=1,0 -a
5. Start RFC2544 Throughput Test - I used Hardware Traffic Gen [Xena]
6. Result from my lab test:
---------------------------------------------------------------------------------------------
Frame Size 25.11 - BAD 26.03 - BAD 64 26.012 25.838 128 45.186 45.383 256 71.955 81.777 512 100.000 100.000 1024 100.000 100.000 1280 100.000 100.000 1518 100.000 100.000
---------------------------------------------------------------------------------------------
Test Setup Info:
HPE ProLiant DL345 Gen12
AMD EPYC 9845 160-Core Processor 2100 MHz
Broadcom BCM57608 Ethernet 100Gb 2-port QSFP112 Adapter - PCIe Slot 1
[Additional Info]
This issue is not seen with DPDK 25.07.
DPDK 25.07 Result:
Frame Size 25.07 - GOOD 64 45.635 128 79.789 256 100.000 512 100.000 1024 100.000 1280 100.000 1518 100.000
Test results of git bisect search for the offending commit: Ref attachment
It looks like the performance degradation is mostly due to the multiple commits from both Intel and Broadcom between September and October 2025 around the 25.11-rc0-25.11-rc2 phase.
After this commit, the performance has remained BAD even with DPDK 26.03.
-------------------------------------------------------------------------------------------------------
Date DPDK Version Commit ID Driver Description
Result
Oct 30, 2025 v25.11-rc2 ef86bb9c55 bnxt Add feature bit Rx miss
handling BAD
-------------------------------------------------------------------------------------------------------
[Expected result]
Result must be similar or better than 25.07 result.
[Actual result]
64b/128b/256b frame size performance degradation with 25.11 and 26.03.
Signature · 2 lines
You are receiving this mail because: You are the assignee for the bug.