On Tue, 23 Dec 2025 12:41:11 +0000 void <[email protected]> wrote: > Hi, > > Question as subject. System is a traditional non-pkgbase. > > A poudriere system, the jail OS version 1500502 has built [0] > nvidia-{driver,kmod}-470. So: > > % pkg search -x nvid > nvidia-driver-470-470.256.02_2 > nvidia-kmod-470-470.256.02.1500502 > > The *host* OS (ie the system that poudriere runs on) is > > % uname -KU > 1500504 1500504 > > I'm presuming this probably won't work and the nvidia kmod > needs to be the *exact* kernel version as the host. > Is this presumption correct? > > The system used to have the following in /etc/src.conf [1] > (I'm thinking about re-adding it and removing nvidia from the > poudriere portslist) > > PORTS_MODULES+=x11/nvidia-driver-470 > > later the line was modified like so: > PORTS_MODULES+=x11/nvidia-driver-470 x11/nvidia-kmod-470 [2] > > [0] accidentally added to regenerated ports list by > `pkg prime-origins | sort | tee portslist.txt` > > [1] `man make.conf` indicates this would go in make.conf but > it works in src.conf too and serves to remind that 'this gets rebuilt > when the kernel gets rebuilt' > > [2] when the port was split into driver plus module > --
Poudriere wants to rebuild ALL speficied (for -a, all installed) ports at least when whichever OSVERSION (comes from __FreeBSD_version in /usr/include/sys/param.h) or REVISION or BRANCH (both defined in /usr/src/sys/conf/newvers.sh and the script runs on base build) is bumped, even if ports tree is NOT AT ALL updated. Safest but annoying (usually too excessive with last 2 digits of OSVERSION in main and stable/* and BRANCH in newvers.sh). OTOH, it's absolutely right for changes for REVISIONS in newvers.sh. And any of graphics/nvidia-drm-*-kmod[-devel] are QUITE STRONGLY tied with LinuxKPI via corresponding graphics/drm-*-kmod sources. Rebuilds are strongly adviced (almost mandatory) if any changes possibly affect video part of it (including parts common with WiFi parts) are made but ports are not updated. For others (x11/nvidia-kmod*), not so tightly tied to kernel upgrades (relatively robust) unless kmod interdaces are changed. Regards. -- Tomoaki AOKI <[email protected]>
