Arch Linux User Repository

For testing users, contributors: check out https://github.com/archlinux-jerry/nvidia-340xx
Warning: force push may happen to match the aur tree

@sebalis I believe user should have the freedom to choose their own C compiler in makepkg.conf. I wouldn't override their preference unless it has to be done. Fixing the warning is a good thing, but we should try fixing the compiler version comparison code instead.
Also structleak_plugin of gcc would have blocked the compilation in case of a compiler version mismatch, so that check seems redundant. Maybe we can remove it entirely.

@JerryXiao When I build the package, I get the following message twice:

warning: the compiler differs from the one used to build the kernel
The kernel was built by: gcc (GCC) 12.1.0
You are using: cc (GCC) 12.1.0

This occurs shortly after the lines

make[1]: Entering directory '/usr/lib/modules/5.18.3-arch1-1/build'

and

make[1]: Entering directory '/usr/lib/modules/5.18.3-arch1-1/build'

The only lines before the warning are echos of commands that are about to be executed.

At one point in the past there was a subsequent message during the installation process that a command had returned a non-zero exit status. I couldn’t reproduce that command when I tried just now.

However, it would still be a good thing in my view to remove this warning as it might cause uncertainty. Also I’m wondering if the issue could be more serious for users that might have an alternative C compilers installed.

@sebalis I don't see any issue with the current PKGBUILD. Tested on real hardware.
@dmanlfc no log, so no solution.
@alexzk ACPI bug? ACPI is working fine on my hardware. TBH how could an acpi bug corrupt your SSD? Do you have a bad sata cable or have you undervolted your CPU?

I installed this package on 5.18.1 and haven’t had issues since. My kernel is now 5.18.3 and I have just successfully built the package again (but didn’t bother to install it). Conincidentally, I did find I still had to add CC=gcc to the two calls to “make” and would appreciate an opinion from the maintainers about their reasons not to include that change. Do they not see this issue?

Doesn't build against the 5.18.2 kernel - is there a compilation patch by any chance?

I installed new SSD into old HP CQ 50 with GTX 8600M card. Installed fresh new arch. Once this driver 340xx boots (gui login) it breaks file system on SSD. Rolled back to nouveau and things work now.

Works like a charm, yes! Including suspend and hibernate (when @Imh69 gave us the first report about a working solution I had this slight worry that ACPI would not be working but that was only the first approach I guess). Many thanks indeed from me too.

I did think that @JerryXiao or @taz-007 might want to use the change I suggested on 2022-05-20 to the two “make” lines in PKGBUILD. As this hasn’t happened I made these changes manually during my update. Therefore I can’t say if the unaltered version would still have warned about a compiler mismatch on my system.

nvidia-340xx-dkms 340.108-30 on 5.18.0-arch1-1 works fine. Thank you very much.

@poluyan I was able to get a clean compilation using the @Viterzgir patch for 5.18 kernel, with 2 modifications. First one to "fix" the error in nv-vm.c

-    return pci_map_sg(dev, sg, 1, PCI_DMA_BIDIRECTIONAL);
+    return dma_map_sg(&dev->dev, sg, 1, DMA_BIDIRECTIONAL);
-    pci_unmap_sg(dev, sg, 1, PCI_DMA_BIDIRECTIONAL);
+    dma_unmap_sg(&dev->dev, sg, 1, DMA_BIDIRECTIONAL);

ERROR: modpost: GPL-incompatible module nvidia.ko uses GPL-only symbol 'acpi_fetch_acpi_dev'

Subject: [PATCH] Buildfix NVIDIA 470.103.01 for Linux 5.18-rc2

acpi_bus_get_device was removed and its trivial replacement acpi_fetch_acpi_dev
is GPL-exported, so just disable ACPI support as I don't want to deal with it.

-#if defined(NV_LINUX_ACPI_EVENTS_SUPPORTED)
+#include <linux/version.h>
+#if defined(NV_LINUX_ACPI_EVENTS_SUPPORTED) && (LINUX_VERSION_CODE < KERNEL_VERSION(5, 18, 0))

I havent check if the module is properly working using this, yet.