Re: [PATCH v3 3/3] sched_ext: Guarantee rq lock on scx_bpf_cpu_rq()

[Christian Loehle <christian.loehle@arm.com>]

On 8/10/25 11:52, Andrea Righi wrote:
> On Sat, Aug 09, 2025 at 09:03:46AM -1000, Tejun Heo wrote:
>> Hello,
>>
>> On Tue, Aug 05, 2025 at 12:10:36PM +0100, Christian Loehle wrote:
>>> @@ -7420,10 +7420,20 @@ __bpf_kfunc s32 scx_bpf_task_cpu(const struct task_struct *p)
>>>   */
>>>  __bpf_kfunc struct rq *scx_bpf_cpu_rq(s32 cpu)
>>>  {
>>> +	struct rq *rq;
>>> +
>>>  	if (!kf_cpu_valid(cpu, NULL))
>>>  		return NULL;
>>>  
>>> -	return cpu_rq(cpu);
>>> +	preempt_disable();
>>> +	rq = cpu_rq(cpu);
>>> +	if (rq != scx_locked_rq()) {
>>> +		scx_kf_error("Accessing not locked rq %d", cpu);
>>> +		rq = NULL;
>>> +	}
>>> +	preempt_enable();
>>
>> So, this will break the existing scheduler binaries immediately, which I
>> don't think is a good way to go about it. Can you add a pr_warn_once() to
>> print deprecation warning and add e.g. scx_bpf_locked_cpu_rq() instead?
> 
> Yeah, this is not nice, but they would be still broken though, in PATCH 1/3
> we force schedulers to check for NULL and, if they don't, the verifier
> won't be happy, so this already breaks existing binaries.
> 
> Even if a scheduler performs the NULL check, this change might still cause
> incorrect behaviors, which can be worse than triggering an error.
> 
> How about we introduce scx_bpf_locked_cpu_rq() and we still trigger an
> error in scx_bpf_cpu_rq(), mentioning about the new locked kfunc and
> scx_bpf_task_acquire_remote_curr()?

If we still trigger an error in scx_bpf_cpu_rq() what's the difference
between scx_bpf_cpu_rq() and scx_bpf_cpu_rq_locked() then?
Just the error message?