[vigh-m]

Description of changes:

• The changes follow the same patterns that were used in Move kmod-6.18 nvidia drivers to use overlayfs at boot #536
• The diff between the 6.12 and 6.18 drivers in minimal

❯ diff packages/kmod-6.12-nvidia-r595 packages/kmod-6.18-nvidia-r595
diff packages/kmod-6.12-nvidia-r595/Cargo.toml packages/kmod-6.18-nvidia-r595/Cargo.toml
2c2
< name = "kmod-6_12-nvidia-r595"
---
> name = "kmod-6_18-nvidia-r595"
12c12
< package-name = "kmod-6.12-nvidia-r595"
---
> package-name = "kmod-6.18-nvidia-r595"
68c68
< kernel-6_12 = { path = "../kernel-6.12" }
---
> kernel-6_18 = { path = "../kernel-6.18" }
Only in packages/kmod-6.12-nvidia-r595: kmod-6.12-nvidia-r595.spec
Only in packages/kmod-6.18-nvidia-r595: kmod-6.18-nvidia-r595.spec


❯ diff packages/kmod-6.12-nvidia-r580 packages/kmod-6.18-nvidia-r580
diff packages/kmod-6.12-nvidia-r580/Cargo.toml packages/kmod-6.18-nvidia-r580/Cargo.toml
2c2
< name = "kmod-6_12-nvidia-r580"
---
> name = "kmod-6_18-nvidia-r580"
12c12
< package-name = "kmod-6.12-nvidia-r580"
---
> package-name = "kmod-6.18-nvidia-r580"
68c68
< kernel-6_12 = { path = "../kernel-6.12" }
---
> kernel-6_18 = { path = "../kernel-6.18" }
Only in packages/kmod-6.12-nvidia-r580: kmod-6.12-nvidia-r580.spec
Only in packages/kmod-6.18-nvidia-r580: kmod-6.18-nvidia-r580.spec

• The diff between the 6.1 and 6.18 drivers is larger since 6.1 doesn't have nvidia-imex support

❯ diff packages/kmod-6.1-nvidia-r595 packages/kmod-6.18-nvidia-r595
diff packages/kmod-6.1-nvidia-r595/Cargo.toml packages/kmod-6.18-nvidia-r595/Cargo.toml
2c2
< name = "kmod-6_1-nvidia-r595"
---
> name = "kmod-6_18-nvidia-r595"
12c12
< package-name = "kmod-6.1-nvidia-r595"
---
> package-name = "kmod-6.18-nvidia-r595"
43a44,53
> url = "https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/x86_64/nvidia-imex-595.91.07-1.amzn2023.x86_64.rpm"
> sha512 = "94db467f6f50cbfdde5e4c4ccc46e576cb9c276aa64bf242356815b1c230869c6de5d98e944e5e18afdb38bcf7687a4eaecd226cc0a813afc95afd39352a5dd2"
> force-upstream = true
>
> [[package.metadata.build-package.external-files]]
> url = "https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/sbsa/nvidia-imex-595.91.07-1.amzn2023.aarch64.rpm"
> sha512 = "7d6233f5fc5b34a6388ac33955a555f44778941d7487381a346761267e37356b3a81a0719699eba13dd2054d409faecda7ff8f7645c39432b68862e97efb46e6"
> force-upstream = true
>
> [[package.metadata.build-package.external-files]]
58c68
< kernel-6_1 = { path = "../kernel-6.1" }
---
> kernel-6_18 = { path = "../kernel-6.18" }
Only in packages/kmod-6.1-nvidia-r595: kmod-6.1-nvidia-r595.spec
Only in packages/kmod-6.18-nvidia-r595: kmod-6.18-nvidia-r595.spec
Only in packages/kmod-6.18-nvidia-r595: nvidia-imex-default-channel.conf
Only in packages/kmod-6.18-nvidia-r595: nvidia-imex-tmpfiles.conf
Only in packages/kmod-6.18-nvidia-r595: nvidia-imex.cfg
Only in packages/kmod-6.18-nvidia-r595: nvidia-imex.service
diff packages/kmod-6.1-nvidia-r595/nvidia-pb-overlay-config.service.in packages/kmod-6.18-nvidia-r595/nvidia-pb-overlay-config.service.in
15a16
>     /etc/nvidia-imex \
17c18,20
<     /etc/.nvidia-config-overlay/work
---
>     /etc/.nvidia-config-overlay/work \
>     /etc/.nvidia-imex-config-overlay/upper \
>     /etc/.nvidia-imex-config-overlay/work
18a22,24
>
> # Writable so downstreams can supply the IMEX node list (nodes_config.cfg).
> ExecStart=/usr/bin/mount -t overlay nvidia-imex-etc --options rw,nosuid,nodev,noexec,noatime,context=system_u:object_r:etc_t:s0,lowerdir=__NVIDIA_ROOT__/etc/nvidia-imex:/etc/nvidia-imex,upperdir=/etc/.nvidia-imex-config-overlay/upper,workdir=/etc/.nvidia-imex-config-overlay/work /etc/nvidia-imex
diff packages/kmod-6.1-nvidia-r595/nvidia-tesla-pb-build-config.toml.in packages/kmod-6.18-nvidia-r595/nvidia-tesla-pb-build-config.toml.in
9c9
< link-objects = ["nvidia.o", "nvidia.mod.o"]
---
> link-objects = ["nvidia.o", "nvidia.mod.o", ".module-common.o"]
15c15
< link-objects = ["nvidia-modeset.o", "nvidia-modeset.mod.o"]
---
> link-objects = ["nvidia-modeset.o", "nvidia-modeset.mod.o", ".module-common.o"]
18c18
< link-objects = ["nvidia-uvm.o", "nvidia-uvm.mod.o"]
---
> link-objects = ["nvidia-uvm.o", "nvidia-uvm.mod.o", ".module-common.o"]


❯ diff packages/kmod-6.1-nvidia-r580 packages/kmod-6.18-nvidia-r580
diff packages/kmod-6.1-nvidia-r580/Cargo.toml packages/kmod-6.18-nvidia-r580/Cargo.toml
2c2
< name = "kmod-6_1-nvidia-r580"
---
> name = "kmod-6_18-nvidia-r580"
12c12
< package-name = "kmod-6.1-nvidia-r580"
---
> package-name = "kmod-6.18-nvidia-r580"
43a44,53
> url = "https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/x86_64/nvidia-imex-580.178.04-1.amzn2023.x86_64.rpm"
> sha512 = "89f4b38004f5d621da2b79e15f10ff46365544d418fdd1349dfa5598dd980738cf35f223953b7ce8fe061c29e64025c604696613985e72b08fef52816a3f286f"
> force-upstream = true
>
> [[package.metadata.build-package.external-files]]
> url = "https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/sbsa/nvidia-imex-580.178.04-1.amzn2023.aarch64.rpm"
> sha512 = "39e50f9bbf7d65c7926305757d0b71fec1f8529690d6f42dc66029bc05fd7db7f18027f4b2284ea00905313d32e908f0c53462edd0e3dfecc0609191a718594b"
> force-upstream = true
>
> [[package.metadata.build-package.external-files]]
58c68
< kernel-6_1 = { path = "../kernel-6.1" }
---
> kernel-6_18 = { path = "../kernel-6.18" }
Only in packages/kmod-6.1-nvidia-r580: kmod-6.1-nvidia-r580.spec
Only in packages/kmod-6.18-nvidia-r580: kmod-6.18-nvidia-r580.spec
Only in packages/kmod-6.18-nvidia-r580: nvidia-imex-default-channel.conf
Only in packages/kmod-6.18-nvidia-r580: nvidia-imex-tmpfiles.conf
Only in packages/kmod-6.18-nvidia-r580: nvidia-imex.cfg
Only in packages/kmod-6.18-nvidia-r580: nvidia-imex.service
diff packages/kmod-6.1-nvidia-r580/nvidia-lts-overlay-config.service.in packages/kmod-6.18-nvidia-r580/nvidia-lts-overlay-config.service.in
15a16
>     /etc/nvidia-imex \
17c18,20
<     /etc/.nvidia-config-overlay/work
---
>     /etc/.nvidia-config-overlay/work \
>     /etc/.nvidia-imex-config-overlay/upper \
>     /etc/.nvidia-imex-config-overlay/work
18a22,24
>
> # Writable so downstreams can supply the IMEX node list (nodes_config.cfg).
> ExecStart=/usr/bin/mount -t overlay nvidia-imex-etc --options rw,nosuid,nodev,noexec,noatime,context=system_u:object_r:etc_t:s0,lowerdir=__NVIDIA_ROOT__/etc/nvidia-imex:/etc/nvidia-imex,upperdir=/etc/.nvidia-imex-config-overlay/upper,workdir=/etc/.nvidia-imex-config-overlay/work /etc/nvidia-imex
diff packages/kmod-6.1-nvidia-r580/nvidia-tesla-lts-build-config.toml.in packages/kmod-6.18-nvidia-r580/nvidia-tesla-lts-build-config.toml.in
9c9
< link-objects = ["nvidia.o", "nvidia.mod.o"]
---
> link-objects = ["nvidia.o", "nvidia.mod.o", ".module-common.o"]
15c15
< link-objects = ["nvidia-modeset.o", "nvidia-modeset.mod.o"]
---
> link-objects = ["nvidia-modeset.o", "nvidia-modeset.mod.o", ".module-common.o"]
18c18
< link-objects = ["nvidia-uvm.o", "nvidia-uvm.mod.o"]
---
> link-objects = ["nvidia-uvm.o", "nvidia-uvm.mod.o", ".module-common.o"]

Testing done:

Tested with a custom AMI per kernel version for the following instance families: g4dn, g5, g6, g7

• Both driver trees, /usr/share/.nvidia/lts and /usr/share/.nvidia/pb, are in the image.
• Branch selection: g4dn, g5, and g6 select lts. g7 selects pb. Each node had exactly one /run/nvidia/branch-<branch> marker, and the loaded driver matched it.
• Units: nvidia-select-branch and the selected branch's overlay-driver, overlay-config, and ldconfig units were active.
• systemctl --failed was empty.
• the host ldconfig -p lists the NVIDIA libraries, and libnvidia-ml.so.1 is at /usr/lib64
• IMEX: kernel 6.1 has no nvidia-imex.service on either branch. Kernel 6.12 has the unit on both branches.

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.

[arnaldo2792]

The ideal fix would be to have a capabilitiy that provides the kernel and a conflict with itself, so that we don't have to add this line per kernel that we ship.

[vigh-m]

⬆️ Updates following comments