Failed to build 3.07 with Mellanox

[Haklat haklat]

Hi,
is Mellanox supported for the new 3.07 release?
When I tried compiling it could not configure before build:

cd ../linux_dpdk/
./b configure --new-memory --with-bird

This step failed because of line 661,662 list in 

1. trex-core
2. /linux_dpdk

/ws_main.py

  is only is only 3 long:

         [ 'HAVE_MLX5_IBV_IMPORT_CTX_PD_AND_MR', 'infiniband/verbs.h',
            'ibv_import_device' ],

Just to make configure pass, I changed this to:
         [ 'HAVE_MLX5_IBV_IMPORT_CTX_PD_AND_MR', 'infiniband/verbs.h',
            type, 'struct ibv_import_device' ],

This made it pass. But I could not start TRex on CX6 NIC with the resulting build.
Has TRex version 3.07  been verified working with Mellanox NICs?

BR//Håkan

[Shlomo Yaakov]

Hi Haklat,

Thanks for reporting this.
We have mlx5 included in our regression suite, and it was successfully tested with the following setup:

• OS: Ubuntu 22.04

• OFED: 25.07-0.9.7

• Firmware: 16.35.4554

The issue you encountered happens only when configuring the build in an environment that includes mlx5 headers. Our build server doesn’t have mlx5 installed, which is why this problem wasn’t triggered there. However, the compiled executables are still tested on setups with Mellanox NICs, so functionality is verified.
We’ll address this issue in the next update to ensure the configuration step works properly in mlx5 environments.

You can try our compiled package here:
https://trex-tgn.cisco.com/trex/release/v3.07.tar.gz

[Haklat haklat]

Thanks,
yes the pre-compiled package works (except few things in my case and not on all our servers (most important I need it to support DPDK LCORE more than the TRex 128 (120) default limit)). 

I am running on kubernetes cluster and need to apply DPDK LCORE  and some other patches to the code to make it work in all our environments.

Did not know I could compile for Nvidia NICs without OFED installed. I did this now and think I got it working OK for now even if OFED is not fully inline with the K8s worker nodes.

Thanks//Håkan