Mellanox ConnectX-3 Pro (MT27520) not creating network interface on Proxmox VE 9 (6.14.8-2-pve)

jdan

New Member
Aug 6, 2025
3
0
1
Hi everyone,

I'm trying to get a Mellanox ConnectX-3 Pro (MT27520) network card to work on Proxmox VE 9, but I'm encountering an issue. The kernel detects the card and loads the drivers, but it doesn't create a network interface that is visible in ip a or /proc/net/dev.

I've tried a few things, but the problem persists. Here are the key details and logs.

System Information:

# Proxmox VE 9, based on Debian 13 "Trixie"
root@# uname -a
Linux XXX 6.14.8-2-pve
1 SMP PREEMPT_DYNAMIC PMX 6.14.8-2 (2025-07-22T10:04Z) x86_64 GNU/Linux

Mellanox Card Information:
root@:~# lspci | grep Mellanox
81:00.0 Network controller: Mellanox Technologies MT27520 Family [ConnectX-3 Pro]
root@:# lspci -vv -s 81:00.0 | grep "Part number" -A 3
[PN] Part number: 764284-B21
[EC] Engineering changes: A2
[SN] Serial number: IL262700ML
[V0] Vendor specific: PCIe FDR x8 13W
Loaded Kernel Modules:

I've loaded the mlx4_en module in an attempt to get an Ethernet interface, but as you can see, both mlx4_ib (for InfiniBand) and mlx4_en (for Ethernet) are active. This is unusual, and despite both modules being loaded, no interface is created.

root@:# lsmod | grep mlx4
mlx4_ib 241664 0
ib_uverbs 184320 1 mlx4_ib
mlx4_en 159744 0
mlx4_core 417792 2 mlx4_ib,mlx4_en
ib_core 487424 2 mlx4_ib,ib_uverbs
Kernel Logs (dmesg):

The logs show that the kernel correctly initializes the card. There are no errors visible in dmesg related to the mlx4 drivers.
root@# dmesg | grep -i mell
[ 1.914473] mlx4_core: Mellanox ConnectX core driver v4.0-0
[ 10.407016] <mlx4_ib> mlx4_ib_probe: mlx4_ib: Mellanox ConnectX InfiniBand driver v4.0-0
Problem Summary:

Despite the mlx4_en and mlx4_ib modules being loaded, no network interface (e.g., p1p1 or ethX) appears in ip a or /proc/net/dev. I also tried to use mstflint and mstregdump from the Debian repositories, but they both result in a "Segmentation fault" error, suggesting a deeper incompatibility with the Proxmox kernel.

Could you please provide some guidance on what might be causing this issue and how to resolve it? Is there a known problem with the mlx4 drivers on the Proxmox 6.14 kernel?

Thank you for your help.
 
Following up on my previous post. I have tried several of the suggested solutions, but unfortunately, the Mellanox ConnectX-3 Pro card is still not creating a network interface. The issue seems to be more complex.

Here is a summary of the actions I have taken and the results:

1. GRUB Configuration (iommu=pt, pci=realloc=off)
I modified the /etc/default/grub file to include iommu=pt and pci=realloc=off.
GRUB_CMDLINE_LINUX_DEFAULT="quiet iommu=pt pci=realloc=off"

After running update-grub and rebooting, there was no change. The network interface still did not appear.

2. Blacklisting mlx4_ib t
T
o force the system to use the Ethernet driver (mlx4_en), I blacklisted the InfiniBand driver. I created the file /etc/modprobe.d/blacklist-mellanox.conf with the following content:

blacklist mlx4_ib
install mlx4_ib /bin/false
After running update-initramfs -u and rebooting, the mlx4_ib module was successfully blacklisted and did not load automatically. This was a positive step.

3. Manual Loading of mlx4_en
With the mlx4_ib module blocked, I manually loaded the mlx4_en module. The lsmod output confirms this:
# lsmod |grep mlx
mlx4_en 159744 0
mlx4_core 417792 1 mlx4_en
The mlx4_en module is loaded and is properly linked to mlx4_core.

Current Status:
Despite all these steps, the network interface is still not listed in ip a or /proc/net/dev. The dmesg output shows that the kernel initializes the card, but no further action seems to occur. The logs are very brief:

# dmesg |grep -i mlx
[ 1.908995] mlx4_core: Mellanox ConnectX core driver v4.0-0
[ 1.909018] mlx4_core: Initializing 0000:81:00.0
[ 10.102887] mlx4_core 0000:81:00.0: DMFS high rate steer mode is: disabled performance optimized steering
[ 10.103140] mlx4_core 0000:81:00.0: 63.008 Gb/s available PCIe bandwidth (8.0 GT/s PCIe x8 link)

The kernel seems to stop after detecting the PCIe bandwidth, without creating a network device.

Remaining Problem:
The kernel successfully loads the mlx4_en driver.
The dmesg logs show no errors, but no network interface is created.
The mstflint tool from Debian repos still fails with a "Segmentation fault" error, making it impossible to check or update the card's firmware.

It seems there might be a fundamental compatibility issue between the Proxmox 6.14 kernel and the ConnectX-3 Pro card, possibly related to the driver implementation or interaction with the card's firmware.

Does anyone have further ideas on what might be the problem or what else I can try?
 
I'm closing this thread to report the resolution. Today, while performing server maintenance in the data center, I simply swapped the cable and transceiver for the Mellanox card. This seemingly minor change turned out to be the key.

The server's BIOS/UEFI detected a new part, applied configuration changes, and upon booting into the OS, the network interface enp129s0 finally appeared. I did not have to change any kernel or module settings. It seems a simple hardware compatibility issue was the culprit all along. I was convinced the previous transceiver was detected at boot, but it looks like Murphy's Law prevailed once again.

Thank you for your help and suggestions regarding the kernel module settings. Your guidance was crucial in confirming that the software configuration was likely correct, allowing me to focus on the hardware aspect of the problem.
 
I have another problem with the Mellanox ConnectX-3 Pro (MT27520) and Proxmox VE 9 (6.14.8-2-pve):

When I backup VMs via CIFs on a Windows computer, the backup runs to 100% and then nothing happens.
Sometimes it works, but mostly not. -

A tmp folder and a .dat file remain on the target computer:

INFO: 98% (129.5 GiB of 132.0 GiB) in 3m 28s, read: 166.2 MiB/s, write: 162.6 MiB/s
INFO: 99% (130.7 GiB of 132.0 GiB) in 3m 39s, read: 115.0 MiB/s, write: 113.5 MiB/s
INFO: 100% (132.0 GiB of 132.0 GiB) in 3m 49s, read: 128.6 MiB/s, write: 125.7 MiB/s
INFO: backup is sparse: 119.22 GiB (90%) total zero data
INFO: transferred 132.00 GiB in 229 seconds (590.3 MiB/s)

Then nothing happens...
Then I have to restart Proxmox to stop it.
When shutting down, Proxmox stops at that point because the CIFS connection cannot be unmounted.
Then the only option is a hard reset.

Switching to a different PCIe port didn't help either.
If I back up using a 2.5G card (RTL8125b), the backup completes without any problems.
If I swap the Mellanox MCX311 card for a 10G Intel 82599, I don't have these problems.
Does Debian 13 (Trixie) have a driver problem with the Mellanox card?

greetings ms1958
 
Last edited: