[Slackbuilds-users] nvidia-driver CUDA glitch

[B Watson]

On 7/13/14, King Beowulf <kingbeowulf at gmail.com> wrote:
> 1. put this in doinst.sh
>     mknod -m 660 /dev/nvidia-uvm c 249 0
>     chgrp video /dev/nvidia-uvm

Will the device node persist across a reboot? Last time I tried something
like this (a while back), the device node had to be created in something
like rc.local (due to /dev being virtual).

You could create the node in /lib/udev/devices (which is used to
populate /dev on every boot, and contains nodes like null, zero,
urandom). doinst.sh should probably create both /dev/nvidia-uvm (for
use now) and /lib/udev/devices/nvidia-uvm (to be used on next reboot).

I've seen shutdown scripts for other Linux distros that look for
new device nodes in /dev and store them elsewhere, so they can be
re-created on the next boot. I don't see anything like this in either
rc.0 or rc.udev on Slack 14, but I only took a minute to look at them
(I might have missed it).

> 2. udev rule?
>     KERNEL=="nvidia_uvm", RUN+="/usr/bin/bash -c '/usr/bin/mknod -m 666
> /dev/nvidia-uvm c $(grep nvidia-uvm /proc/devices | cut -d \ -f 1) 0;
> /usr/bin/chgrp video /dev/nvidia-uvm'"

This seems like the "proper" solution, assuming it actually works
(haven't tested, I'm currently using nouveau, not nvidia).

> 3. Compile nvidia-modprobe by default (optional now) so that X userland
> can make the node, or invoke via doinst.sh if not.

No clue about this one, so no comment.