Nvidia Legacy driver (340.106) will not compile under RH 7.5 kernel

Graphics / Linux Linux
1

I’m upgrading some old linux machines from 7.4 to 7.5 with legacy graphics card. 340.106 works fine under the 3.10.0-693.21.1 kernel , but when I upgrade to 3.10.0-862 (or 861) it fails to install. The error messages seem to be related to get_user_pages function. However that function prototype doesn’t seem to have changed between those two versions.

–D

In file included from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv.c:13:0:
   /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv-linux.h:2126:13: error: too few arguments to function ‘get_user_pages’
                return get_user_pages(start, nr_pages, flags, pages, vmas);
                ^
   In file included from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv-linux.h:68:0,
                    from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv.c:13:
   /usr/src/kernels/3.10.0-861.el7.x86_64/include/linux/mm.h:1381:6: note: declared here
    long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,



  In file included from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv.c:13:0:
   /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv-linux.h:2153:45: error: too few arguments to function ‘get_user_pages_remote’
                                                pages, vmas);
                                                ^
   In file included from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv-linux.h:68:0,
                    from /tmp/selfgz23204/NVIDIA-Linux-x86_64-340.106/kernel/nv.c:13:

nvidia-compile.txt (16.2 KB)

2

RH did a lot of backporting. Attached are two patches. One changes the get_user_pages wrapper to the current version, and the other one injects the proper defines for 7.5 (unconditionally). Works for me.

As an aside: RH did so much backporting, you now even see something like this:

nvidia 0000:01:00.0: irq 31 for MSI/MSI-X
resource sanity check: requesting [mem 0x000c0000-0x000fffff], which spans more than PCI Bus 0000:00 [mem 0x000d0000-0x000dffff window]
caller os_map_kernel_space+0xad/0xe0 [nvidia] mapping multiple BARs

→ https://devtalk.nvidia.com/default/topic/1026975/nvidia-related-warning-on-booting-linux-4-14/

Verdammte Schlamperei!

.
fix_config.txt (929 Bytes)
fix_wrapper.txt (4.69 KB)

3

Chemal,

I tried your patching suggestion and it now builds successfully!
Thank you very much for posting this answer!

4

Hi, how exactly do you apply those patches?
I’m stuck trying to install CUDA 8.0 on Ubuntu 18.04.

5

This thread is obsolete. Install 340.107 instead. And 340 is only good for cuda up to 6.5.

6

For Ubuntu, the patches are included in the repo drivers

  • Don’t use the .run installers, use --uninstall to uninstall them
  • purge anthing nvidia/cuda
  • install the driver from repo (sudo apt install nvidia-driver-340)
  • download the cuda .deb
  • add the repo to your system (first three steps from install instructions on download page)
  • don’t install cuda
  • instead, run sudo apt install cuda-toolkit-8-0