Re: Kernel modules not loading on 15-prerelease

On Tue, Sep 02, 2025 at 11:45:35AM -0700, John Kennedy wrote:
> On Mon, Sep 01, 2025 at 06:03:21PM -0700, bob prohaska wrote:
> > H Pi2 self-hosting with 
> > bob@pelorus:~ % uname -a
> > FreeBSD pelorus.zefox.org 15.0-PRERELEASE FreeBSD 15.0-PRERELEASE #63 main-n279991-21628c79d541: Mon Sep  1 12:06:31 PDT 2025     bob@pelorus.zefox.org:/usr/obj/usr/src/arm.armv7/sys/GENERIC arm
> > bob@pelorus:~ % uname -KU
> > 1500063 1500061
> >         ^^^^^^^
> > 	[other hosts are at ...63]     
> >  ...
> > The simplest problem seems to be the failure to update the userland
> > version. Git seems to run and reports no errors, git reset --hard
> > didn't change anything. ...
> 
>   No way to be sure where you ended up, but it superficially looks like you've
> done a kernel install without a world install.  I know you're an old hat at
> this and you've got others at 63, so you're probably looking for some weird
> fluke on that one box (I suspect).
> 
Yes, it's just this one box. A second has kept world and kernel in sync
and loads both filemon and uftdi normally

>   On the thing 15 has done is default to a "no clean" scenario, and there were
> some "META" mode changes around early-August that might mean things are in
> a slightly odd state.  I'd suggest you clean out your object files and
> recompile everything (which I think REALLY hurts time-wise for your RPI2?).
> 
>   My META issues usually just resulted in compile errors that would go away
> when I blew away the object tree.  I presume you're trying to do NO_CLEAN
> builds just to speed things up on RPI and maybe some dependency isn't getting
> evaluated right, or clock was iffy for some timestamp, or whatever.
>

For a long time (months?) -DWITH_META_MODE has been part of the build
commands. On the troublesome host I've eliminated that and used
make -j3   buildworld > buildworld.log && make -j4  buildkernel   > buildkernel.log && etcupdate -p > etcupdate.log
but it hasn't changed the kernel/world/modules mismatch.

>   You look like you're running with WITHOUT_REPRODUCIBLE_BUILD set, but I've
> also taken to nuking /usr/obj/usr/src/amd64.amd64/sys/GENERIC/vers.[co] on
> rebuild to make sure that current version strings land up in the binary.

I didn't know about WITHOUT_REPRODUCIBLE_BUILD and have not set it intentionally.
What's the purpose of the variable? I don't find it in man build...

Thanks for writing!

bob prohaska