Issue building nvidia-kmod for 7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64

I have issues building the NVIDIA driver 615.71.09-1 nvidia-kmod on Rawhide for latest kernel version 7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64, while it builds correctly for the earlier kernel version 7.3.0-0.rc0.260819gbd5f485f3f02.5.fc46.x86_64:

root@BigComputer:~# akmods --verbose --rebuild --force --kernels $(uname -r)
Checking kmods exist for 7.3.0-0.rc0.260819gbd5f485f3f02.5.fc46.x86_64 [  OK  ]
Building and installing nvidia-kmod [  OK  ]
root@BigComputer:~# akmods --verbose --rebuild --force --kernels 7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64
Checking kmods exist for 7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64 [  OK  ]
Building and installing nvidia-kmod [FAILED]
Building rpms failed; see /var/cache/akmods/nvidia/615.71.09-1-for-7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64.failed.log for details

Hint: Some kmods were ignored or failed to build or install.
You can try to rebuild and install them by by calling
'/usr/sbin/akmods --force' as root.

Looking at the akmods log file it shows the following errors:

2026/09/17 06:59:01 akmodsbuild: nvidia/os-interface.c: In function ‘os_dmem_cgroup_register_region’:
2026/09/17 06:59:01 akmodsbuild: nvidia/os-interface.c:3031:48: error: passing argument 1 of ‘dmem_cgroup_register_region’ makes pointer from integer without a cast [-Wint-conversion]
2026/09/17 06:59:01 akmodsbuild:  3031 |     void *region = dmem_cgroup_register_region(size, name);
2026/09/17 06:59:01 akmodsbuild:       |                                                ^~~~
2026/09/17 06:59:01 akmodsbuild:       |                                                |
2026/09/17 06:59:01 akmodsbuild:       |                                                NvU64 {aka long long unsigned int}
2026/09/17 06:59:01 akmodsbuild: In file included from nvidia/os-interface.c:50:
2026/09/17 06:59:01 akmodsbuild: /usr/src/kernels/7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64/include/linux/cgroup_dmem.h:49:60: note: expected ‘const struct dmem_cgroup_init *’ but argument is of type ‘NvU64’ {aka ‘long long unsigned int’}
2026/09/17 06:59:01 akmodsbuild:    49 | dmem_cgroup_register_region(const struct dmem_cgroup_init *init,
2026/09/17 06:59:01 akmodsbuild:       |                             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
2026/09/17 06:59:01 akmodsbuild: make[4]: *** [/usr/src/kernels/7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64/scripts/Makefile.build:290: nvidia/os-interface.o] Error 1
2026/09/17 06:59:01 akmodsbuild: make[4]: *** Waiting for unfinished jobs....

Which causes the build to fail:

2026/09/17 06:59:01 akmodsbuild: make[3]: *** [/usr/src/kernels/7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64/Makefile:2265: .] Error 2
2026/09/17 06:59:01 akmodsbuild: make[2]: *** [/usr/src/kernels/7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64/Makefile:248: __sub-make] Error 2
2026/09/17 06:59:01 akmodsbuild: make[1]: *** [Makefile:248: __sub-make] Error 2
2026/09/17 06:59:01 akmodsbuild: make: *** [Makefile:115: modules] Error 2
2026/09/17 06:59:01 akmodsbuild: error: Bad exit status from /var/tmp/rpm-tmp.Zx7QQs (%build)
2026/09/17 06:59:01 akmodsbuild: 
2026/09/17 06:59:01 akmodsbuild: RPM build errors:
2026/09/17 06:59:01 akmodsbuild:     Unable to open current directory: Permission denied
2026/09/17 06:59:01 akmodsbuild:     Bad exit status from /var/tmp/rpm-tmp.Zx7QQs (%build)
2026/09/17 06:59:01 akmodsbuild: 
2026/09/17 06:59:06 akmods: Building rpms failed; see /var/cache/akmods/nvidia/615.71.09-1-for-7.3.0-0.rc3.260914g704340f1cd0d.32.fc46.x86_64.failed.log for details

If it builds correctly with 7.3.0-0.rc0 but it fails with 7.3.0-0.rc3, this is probably a regression, am I right? I tried looking for an alternative such as the open driver, but the instructions on the RPMFusion wiki seem to be slightly outdated, and the akmods-nvidia-open package in the nonfree-tainted repo is a few versions behind the rest. All my nvidia packages are on version 615.71.09-1, installationwise it looks good.

My knowledge to correct this error in the source code, unfortunately, is insufficient, so I guess I’ll be stuck on 7.3.0-0.rc0 for a while. Not a huge problem, but I thought it would be useful to share the situation. And I wonder: am I missing some important detail?

Greetings!

did you look at the log

Don’t use it, akmod-nvidia auto-detects if open or prop driver is required and builds it.

I have no intention to fix the compile issue with the 7.3rc driver until it’s a stable release.

No problem, and thanks. I will use nvidia-580xx for now, as building its kmod doesn’t show the same issue with these newer-than-7.3rc0 kernels.

It does feel a bit funny to use a stable nvidia driver on an otherwise unstable system. But it works! :grinning_face:

As far as I know, the only reason to remain on the 580xx version of the driver is if you have one of the older (pascal or maxwell chipset) GPUs. Those would probably be the 9xx or 10xx cards which are still supported by the 580xx driver and not supported by the newer 595 and newer drivers.

All cards 16xx and newer are properly supported by the current nvidia drivers. The newer drivers provide additional features as well as better support especially with the later card versions.

Well sure, I agree. It makes sense that newer cards are better supported by the current driver. And I would like to use the current driver for my RTX 5070, not this older one.

But when akmods-nvidia offered with the current (615.71.09) nvidia driver from the rpmfusion-nonfree-rawhide repo fails to build the kmod with a kernel newer than 7.3rc0 (see earlier in the topic), what other driver would you suggest I use with current kernel 7.3rc4?

Until the compilation issue that arises with kernels newer than 7.3rc0 is fixed, I find this a workable temporary solution.

DIY

https://patch-diff.githubusercontent.com/raw/NVIDIA/open-gpu-kernel-modules/pull/1330.patch