(AMD EPYC 9224+H100) Stuck in booting

Accelerated Computing Confidential Computing
,
1

Motherboard: gigabyte G493-ZB2 (G493-ZB2 (rev. AAP1) | GPU Servers - GIGABYTE Korea)
OS: Ubuntu 22.04 server
CPU: AMD EPYC 9224
RAM: 64G / SSD: 2T

Hi there,
I was trying to do confidential computing with H100.

But, based on the Confidential Computing Deployment Guide(https://docs.nvidia.com/confidential-computing-deployment-guide.pdf), I succssed until installing the Host OS (i.e., 5.19 snp-aware kernel).

But, I was stuck on the booting after installing the 5.19 kernel.

So, I was trting to figure out what’s problem in, and then found out that when I changed SMEE mode in BIOS back to ‘Auto’ (default value), it works.

My gigabyte BIOS has following options about SEV-SNP,

SMEE (default Auto)
SEV-ES ASID Space Limit (default 1)
SNP Memory Coveragy (default Auto)

IOMMU (default Enabled)
SEV-SNP support (default Auto)

How to fix it?
Thanks in advance.

stuck on screenshot

2

The log shows SEV-SNP: failed to INIT error 0x3.

Could you boot the VM without H100 using the BIOS configuration listed in the guide (turn to “on”)?

3

Hi @Yifan-Tan ,

This is the host machine, not VM.

In that case, you also say that removing H100 from the machine physically?

4

Oh, my mistake.

It is probably the issue of AMD SEV. Issues · AMDESE/AMDSEV · GitHub has plenty of similar issues for error 0x3.

I guess it is not the issue of H100. You could exclude H100’s problem by not installing (or uninstalling) nvidia driver on the host.

5

Per the Deployment Guide, we want to have these options configured specifically, not on “Auto” (as I see you have set):

  • SEV ASID Count → 509 ASIDs
  • SEV-ES ASID space Limit Control → Manual
  • SEV-ES ASID space limit → 100
  • SNP Memory Coverage → Enabled
  • SMEE → Enabled
  • SEV-SNP Support → Enabled
  • IOMMU → Enabled