nvidiaPackages.production: libnvidia-pkcs11.so.$DRIVER_VERSION missing

[msanft]

Nixpkgs version

• Unstable (25.05)

Describe the bug

When running a container with pyTorch on version 2.6.0+cu124 with a H100 NVIDIA GPU on the currently-packaged production NVIDIA driver in nixpkgs, PyTorch tries to load the following library:

# excerpt from strace output
openat(AT_FDCWD, "$PATH/libnvidia-pkcs11.so.550.90.07", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)

This file is missing as we delete it in the build of the production driver:

nixpkgs/pkgs/os-specific/linux/nvidia-x11/builder.sh

Lines 81 to 82 in a13cc0b

	# Compatibility with openssl 1.1, unused
	rm -f $i/lib/libnvidia-pkcs11.so*

While we should definitely not include OpenSSL 1.1 by default (e.g. for security reasons), we should still somehow address this incompatibility.

Steps to reproduce

This can be reproduced with the following command:

strace -f -e openat python3 -c 'import torch; torch.cuda.is_available()'

Expected behaviour

PyTorch should load correctly.

Screenshots

No response

Relevant log output

Additional context

No response

System metadata

# uname -a
Linux workload-textgen-0 6.11.0-nvidia-gpu-confidential #1-NixOS SMP Sun Sep 15 14:57:56 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

Not applicable otherwise, as the system does not include nix nor nix-* et al.

Notify maintainers

---

Note for maintainers: Please tag this issue in your pull request description. (i.e. Resolves #ISSUE.)

I assert that this issue is relevant for Nixpkgs

• I assert that this is a bug and not a support request.
• I assert that this is not a duplicate of an existing issue.
• I assert that I have read the NixOS Code of Conduct and agree to abide by it.

Is this issue important to you?

Add a 👍 reaction to issues you find important.