Re: [PATCH] selftests/sched_ext: Fix the incorrect logic validation

[Feng Yang]

On Mon, 2 Mar 2026 08:47:17 +0100, Andrea Righi wrote:
> On Mon, Mar 02, 2026 at 02:48:51PM +0800, Feng Yang wrote:
> > From: Feng Yang <yangfeng@xxxxxxxxxx>
> > 
> > When __COMPAT_scx_bpf_pick_idle_cpu_node selects an idle CPU,
> > it reports that the CPU should be marked as busy.
> > 
> > Fixes: 5ae5161820e5 ("selftests/sched_ext: Add NUMA-aware scheduler test")
> > Signed-off-by: Feng Yang <yangfeng@xxxxxxxxxx>
> > ---
> >  tools/testing/selftests/sched_ext/numa.bpf.c | 8 ++++----
> >  1 file changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/tools/testing/selftests/sched_ext/numa.bpf.c b/tools/testing/selftests/sched_ext/numa.bpf.c
> > index a79d86ed54a1..98423628b05c 100644
> > --- a/tools/testing/selftests/sched_ext/numa.bpf.c
> > +++ b/tools/testing/selftests/sched_ext/numa.bpf.c
> > @@ -44,12 +44,12 @@ s32 BPF_STRUCT_OPS(numa_select_cpu,
> >  	 */
> >  	cpu = __COMPAT_scx_bpf_pick_idle_cpu_node(p->cpus_ptr, node,
> >  					__COMPAT_SCX_PICK_IDLE_IN_NODE);
> > -	if (cpu < 0)
> > +	if (cpu < 0) {
> >  		cpu = __COMPAT_scx_bpf_pick_any_cpu_node(p->cpus_ptr, node,
> >  						__COMPAT_SCX_PICK_IDLE_IN_NODE);
> > -
> > -	if (is_cpu_idle(cpu, node))
> > -		scx_bpf_error("CPU %d should be marked as busy", cpu);
> > +		if (is_cpu_idle(cpu, node))
> > +			scx_bpf_error("CPU %d should be marked as busy", cpu);
> > +	}
> 
> No, this is not correct. The CPU returned by scx_bpf_pick_idle_cpu_node()
> should be marked as busy at this point (bit is set in the idle cpumask),
> essentially it has been reserved/allocated by the caller.

Okay, I misunderstood.
The CI test of this patch (https://patchwork.kernel.org/project/netdevbpf/list/?series=1058866&state=*) 
resulted in the error "CPU 0 should be marked as busy"(https://github.com/kernel-patches/bpf/actions/runs/22548310848/job/65314767856),
but I don't see the connection between them that would cause this error. Could you explain why this happens?

Thanks,
Feng

> Maybe we can add a comment if we think it's not clear enough, but the
> original code looks correct.
> 
> >  
> >  	if (__COMPAT_scx_bpf_cpu_node(cpu) != node)
> >  		scx_bpf_error("CPU %d should be in node %d", cpu, node);
> > -- 
> > 2.43.0
> > 
> 
> Thanks,
> -Andrea