Re: WARNING: CPU: 12 PID: 44394 at arch/powerpc/kernel/process.c:437 giveup_all+0x90/0x1d0 (v6.14.8, Talos II)

[Erhard Furtner <erhard_f@mailbox.org>]

On Thu, 29 May 2025 10:41:22 +0530
Srikar Dronamraju <srikar@linux.ibm.com> wrote:

> * Erhard Furtner <erhard_f@mailbox.org> [2025-05-27 20:47:28]:
> 
> > On Tue, 27 May 2025 14:19:13 +0530
> > Srikar Dronamraju <srikar@linux.ibm.com> wrote:
> >   
> > > > Getting these from time to time when I run the machine for longer periods building:
> > > > 
> > > > [...]
> > > > ------------[ cut here ]------------
> > > > WARNING: CPU: 12 PID: 44394 at arch/powerpc/kernel/process.c:437 giveup_all+0x90/0x1d0    
> > > 
> > > Its a warning with says that the task's userspace MSR register has
> > > VSX set but not the VEC or FP.
> > > 
> > > Were you building the kernel or were you building something else.
> > > Can you paste the command you attempted?  
> > 
> > I was doing an 'emerge -e @world' in Gentoo Linux and probably building a kernel in another shell. Noticed it only in the logs on the next day so I can't tell exactly which command triggered the issue.
> > 
> > But I will pay closer attention from now on.
> >   
> > > If this is happening with only the recent kernel and working with a previous
> > > kernel, do you happen to know the previous kernel version.
> > > In which case, is there a possibility of doing a git bisect?  
> > 
> > I think I have not seen this in v6.13, certainly not in v6.12.
> >   
> 
> I did go through the git log from v6.12 till v6.15-rc7 and I dont see any
> direct changes that could affect VSX VEC or FP changes in kernel.
> 
> On a side note, Was there any significant userspace updates like library,
> compiler updates from when you saw it working to now?

Looks like these warnings easily happen when I build stuff with GCC 15, currently on my Gentoo Linux musl ppc partition. No kernel builds, just emerging regular Gentoo ebuilds.

I'll sort out whether I see this on GCC 14 too and/or glibc partitions too. I'll report back when I know a bit more and maybe found a good starting point for a bisect.

Regards,
Erhard