High ksoftirqd load on Mellanox ConnectX-4 Lx on Linux

Infrastructure & Networking Software And Drivers Mellanox OFED
1

I’m running xdp service on Mellanox Technologies MT27710 Family [ConnectX-4 Lx]. Just when the service is started and no request, the cpu usage of ksoftirqd processes suddently raises to100% nearly (using top).

Some settings on the system for reference:

kernel: 5.10.0-136.12.0.86.ctl3.x86_64
irqbalance is running.

NIC:
driver: mlx5_core
version: 23.10-2.1.3
firmware-version: 14.32.1010 (ZTE0000000002)
*expansion-rom-version: *
bus-info: 0000:61:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: no
supports-register-dump: no
supports-priv-flags: yes

NIC channel: 2 queue
ethtool -l ens3f0np0
Channel parameters for ens3f0np0:
Pre-set maximums:
RX: n/a
TX: n/a
Other: n/a
Combined: 63
Current hardware settings:
RX: n/a
TX: n/a
Other: n/a
Combined: 2

XDP is attached to the NIC.

image
image953×250 194 KB

I’m wondering why the ksoftiqrd peaks at 100% when the xdp service is started and no incoming request. Any suggestion is appreciated.

Thanks!

2

Hi houguanghua,

Thank you for posting your query on NVIDIA community!

Based on the review of information shared, you are using an OEM Adapter ZTE0000000002 (PSID starting with ZTE) . In such scenarios, we recommend reaching out to the OEM first as the firmware specific to OEM’s can have customizations that may be different than the non-OEM Adapters(starting with PSID “MT_”).

Thanks,
Namrata.