From Newsgroup: comp.os.linux.advocacy

$ tail -n 3 linux-6.18-rc4/*.out

linux-6.18-rc4/nohup.out <==

real 424.85
user 20312.86
sys 3881.19

linux-6.18-rc4/uname.out <==

Linux lm 6.18.0-rc4 #1 SMP PREEMPT Tue Nov 4 20:22:14 PST 2025 x86_64
x86_64 x86_64 GNU/Linux

I had to make a fiddling patch to the graphics driver to get the unified
memory module to compile:

% - cut here - %<-

$ diff -u NVIDIA-Linux-x86_64-580.105.08/kernel-open/nvidia-uvm/uvm_va_range_device_p2p.c~ NVIDIA-Linux-x86_64-580.105.08/kernel-open/nvidia-uvm/uvm_va_range_device_p2p.c
--- NVIDIA-Linux-x86_64-580.105.08/kernel-open/nvidia-uvm/uvm_va_range_device_p2p.c~ 2025-10-29 15:30:21.000000000 -0700
+++ NVIDIA-Linux-x86_64-580.105.08/kernel-open/nvidia-uvm/uvm_va_range_device_p2p.c 2025-11-08 00:33:09.491801970 -0800
@@ -360,7 +360,7 @@
// a reference to them, so take one now if using DEVICE_COHERENT pages.
if (gpu->parent->cdmm_enabled) {
get_page(page);
- get_dev_pagemap(page_to_pfn(page), NULL);
+ get_dev_pagemap(page_to_pfn(page));
}
#else
// CDMM P2PDMA will never be enabled for this case

% - cut here - %<-

gcc made this easy -- the output showed the prototype that it was
matching against (which the nvidia installer puts in /var/log/nvidia-installer.log).

(One could build without nvidia-uvm, but then CUDA wouldn't work, and I actually use CUDA.)

TL;DR: I can test a release candidate kernel on my daily driver.

First stop: Intensive graphical space simulation (aka, Elite Dangerous Odyssey).
--
-v System76 Thelio Mega v1.1 x86_64 NVIDIA RTX 3090Ti 24G
OS: Linux 6.18.0-rc4 D: Mint 22.2 DE: Xfce 4.18
NVIDIA: 580.105.08 Mem: 258G
"You, you, and you, panic. Everyone else, come with me."
--- Synchronet 3.21a-Linux NewsLink 1.2