GPU Direct Storage Not Patching NVME

[schmaustech]

What happened:

In the nic cluster policy there are two env variables that should enable GPUDirect Storage

ENABLE_NFSRDMA which should enable GDS for nvme or NFS

In some cases though there are only nvme drives in a system and the customer only wants NFS so this option was created

ENABLE_NFSRDMA_NO_NVME which enabled NFS only

What we are seeing in 25.7 is that when we use ENABLE_NFSRDMA it does not patch the nvme driver and GDS does not work for nvme.

What you expected to happen:

If nvme is patches properly we should see the following:

cat /proc/kallsyms | grep nvfs_dma
0000000000000000 t nvme_v1_unregister_nvfs_dma_ops [nvme]
0000000000000000 t nvme_v1_register_nvfs_dma_ops [nvme]

How to reproduce it (as minimally and precisely as possible):

Install 25.7 in OpenShift and set ENABLE_NFSRDMA

Anything else we need to know?:

Logs:
Will provide if needed.

Environment:
Dell R760xa
Mellanox CX7 or BF3
OpenShift 4.19