Re: 6.12 build failure for non-SMP systems

[Greg KH <gregkh@linuxfoundation.org>]

On Sat, Feb 21, 2026 at 02:04:49PM -0500, Nathaniel Filardo wrote:
> Hi stable@
> 
> The commit https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=e61f636cc31042b717a46f6937a2dfaf21f45c91,
> first included in 6.12.64, does not properly guard its new references
> to `stop_sched_class` under `CONFIG_SMP`, leading to build failures
> like the following, even with the significantly newer 6.12.74 release:
> 
> > linux-armv5tel-unknown-linux-gnueabi>   LD      .tmp_vmlinux1
> linux-armv5tel-unknown-linux-gnueabi>
> /nix/store/cn2gkcaa5kfgzjah31c0qlifv9y4yqg5-armv5tel-unknown-linux-gnueabi-binutils-2.44/bin/armv5tel-unknown-linux-gnueabi-ld:
> kernel/sched/build_policy.o: in function `scx_ops_disable_workfn':
> > linux-armv5tel-unknown-linux-gnueabi> /build/linux-6.12.74/build/../kernel/sched/ext.c:4746:(.text+0x704c): undefined reference to `stop_sched_class'
> > linux-armv5tel-unknown-linux-gnueabi> /nix/store/cn2gkcaa5kfgzjah31c0qlifv9y4yqg5-armv5tel-unknown-linux-gnueabi-binutils-2.44/bin/armv5tel-unknown-linux-gnueabi-ld: kernel/sched/build_policy.o: in function `atomic_set':
> > linux-armv5tel-unknown-linux-gnueabi> /build/linux-6.12.74/build/../include/linux/atomic/atomic-instrumented.h:69:(.text+0x11c6c): undefined reference to `stop_sched_class'
> 
> A possible fix is simply to add the requisite guard, thus:
> diff --git a/kernel/sched/ext.c.orig b/kernel/sched/ext.c
> index 9f03255..9a3e5b2 100644
> --- a/kernel/sched/ext.c.orig
> +++ b/kernel/sched/ext.c
> @@ -1059,8 +1059,10 @@ static struct scx_dispatch_q *find_user_dsq(u64 dsq_id)
> 
>  static const struct sched_class *scx_setscheduler_class(struct task_struct *p)
>  {
> +#ifdef CONFIG_SMP
>   if (p->sched_class == &stop_sched_class)
>   return &stop_sched_class;
> +#endif
> 
>   return __setscheduler_class(p->policy, p->prio);
>  }
> 
> But it may also be desirable to backport
> https://github.com/torvalds/linux/commit/cac5cefbade90ff0bb0b393d301fa3b5234cf056
> from mainline, which removes much of the distinction between UP and
> SMP schedulers.

That patch doesn't backport cleanly, so can you just submit this as a
fixup for now in a format that we can apply it in?

thanks,

greg k-h