Re: [PATCH] sched_ext: Fix is_bpf_migration_disabled() false negative on non-PREEMPT_RCU

[Kuba Piecuch]

Hi Changwoo,

On Thu Apr 2, 2026 at 2:31 AM UTC, Changwoo Min wrote:
...
>  static bool is_bpf_migration_disabled(const struct task_struct *p)
>  {
> -	if (p->migration_disabled == 1)
> -		return p != current;
> -	else
> -		return p->migration_disabled;
> +	if (p->migration_disabled == 1) {
> +		if (IS_ENABLED(CONFIG_PREEMPT_RCU))
> +			return p != current;
> +		return true;
> +	}
> +	return p->migration_disabled;
>  }

The fix looks correct, but the logic looks somewhat convoluted. How about
something like this:

 static bool is_bpf_migration_disabled(const struct task_struct *p)
 {
-       if (p->migration_disabled == 1)
-               return p != current;
-       else
-               return p->migration_disabled;
+       if (IS_ENABLED(CONFIG_PREEMPT_RCU) &&
+           p == current &&
+           !WARN_ON_ONCE(!p->migration_disabled)) {
+               return p->migration_disabled - 1;
+	}
+       return p->migration_disabled;
 }

My thinking here is: if CONFIG_PREEMPT_RCU is enabled and we're current,
subtract 1 from p->migration_disabled to account for the BPF prologue.
Otherwise just return p->migration_disabled. I've also thrown in a WARN_ON_ONCE
to help catch potential bugs if the assumption about the BPF prologue ever
changes.

Thanks,
Kuba