Add NVIDIA GPU local and remote attestation

[zvonkok]

Using this issue to track the attestation changes for supporting NVIDIA GPUs with our attestation_sdk: https://github.com/NVIDIA/nvtrust, lets start with the obvious:

• Attester
• Verifer

There are more things to come, maybe a KMS backend as well.

[imlk0]

Hi @zvonkok, it's a nice work. I'm curious if this will mean implementing a new Attester in AA crate and a new Verifier in AS crate? Also, I'm wondering if it will use something like nvml-wrapper under the hood or just use ffi to call the attestation_sdk python library?

[zvonkok]

@imlk0 Yes new Attester in AA create and a new Verifier in AS crate. No not using nvml-wrapper this is just a small part of the attestation_sdk. I would need to implement a ton of things to make this work just by using nvml-wrapper.
As a start to see something working maybe just calling an binary (PyInstaller).
We can see from there if we need attestation_sdk updates or FFI lets see but I hope to get enough reviews to move this in the right direction.

[imlk0]

@imlk0 Yes new Attester in AA create and a new Verifier in AS crate. No not using nvml-wrapper this is just a small part of the attestation_sdk. I would need to implement a ton of things to make this work just by using nvml-wrapper. As a start to see something working maybe just calling an binary (PyInstaller). We can see from there if we need attestation_sdk updates or FFI lets see but I hope to get enough reviews to move this in the right direction.

Thanks for your reply.

[1570005763]

Great proposal! I'm curious about the part regarding KMS.
Could you perhaps share the needs for KMS in NVIDIA GPU remote attestation? Currently, CDH has a dedicated KMS plugin interface, which can interface with various KMS. Do you think this feature might be applicable or beneficial in your work?