Add the memory addressing mode as an attribute for each device

[shivamerla]

On coherent-memory systems, the possible modes are:

• HMM - (Hardware Memory Management mode)
• ATS - (Address Translation Service mode)
• None - (Supported by the platform but currently inactive)

On others, its explicitly set to empty to indicate that it is not supported by the platform

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[shivamerla]

Is blocked by NVIDIA/go-nvlib#75

[shivamerla]

@klueska go-nvlib is now updated to fetch the AddressingMode attribute. PTAL.

$ kubectl get resourceslice -o json | jq .items[1].spec.devices[0].attributes
{
  "addressingMode": {
    "string": "HMM"
  },
  "architecture": {
    "string": "Ampere"
  },
  "brand": {
    "string": "Nvidia"
  },
  "cudaComputeCapability": {
    "version": "8.0.0"
  },
  "cudaDriverVersion": {
    "version": "13.0.0"
  },
  "driverVersion": {
    "version": "580.95.5"
  },
  "pcieBusID": {
    "string": "0000:3b:00.0"
  },
  "productName": {
    "string": "NVIDIA A100-PCIE-40GB"
  },
  "resource.kubernetes.io/pcieRoot": {
    "string": "pci0000:3a"
  },
  "type": {
    "string": "gpu"
  },
  "uuid": {
    "string": "GPU-c021ba0a-76fd-9f12-7e4a-a8c3fce1c92d"
  }
}

[klueska]

Looks good. Can you please squash into a single commit before merging.

[klueska]

I still see 4 commits in this PR and on main I see a single commit without a merge commit. How did you merge this?

[shivamerla]

I still see 4 commits in this PR and on main I see a single commit without a merge commit. How did you merge this?

With the Squash and Merge option which doesn't create the merge commit.

[klueska]

In the future, please squash locally and repush so that all merges happen with a merge commit.