Re: [PATCH 2/3] cgroup: add lockless fast-path checks to cgroup_file_notify()

[Chen Ridong <chenridong@huaweicloud.com>]

On 2026/3/3 12:01, Shakeel Butt wrote:
> On Tue, Mar 03, 2026 at 11:18:17AM +0800, Chen Ridong wrote:
>>
>>
> [...]
>>> @@ -4689,6 +4689,12 @@ void cgroup_file_notify(struct cgroup_file *cfile)
>>>  	unsigned long flags;
>>>  	struct kernfs_node *kn = NULL;
>>>  
>>> +	if (!READ_ONCE(cfile->kn))
>>> +		return;
>>> +
>>> +	if (timer_pending(&cfile->notify_timer))
>>> +		return;
>>> +
>>
>> The added timer_pending() check seems problematic. According to the function's
>> comment, callers must ensure serialization with other timer operations. Here
>> we're checking timer_pending() locklessly, which means we might:
>>
> 
> That comment seems outdated. Check the commit 90c018942c2ba ("timer: Use
> hlist_unhashed_lockless() in timer_pending()").
> 

Maybe we should update the comment.

>> 1. See an inconsistent state if another CPU is concurrently modifying the timer
>>
> 
> It will not see inconsistent state but it can see stale state which is totally
> fine. At worst we will take the lock and recheck but we will never miss the
> notifications.
> 

You are right. Thank you for explanation.


>> 2. Race with del_timer() or mod_timer() from other contexts
>>
>>>  	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
>>>  	if (cfile->kn) {
>>>  		unsigned long last = cfile->notified_at;
>>
>> -- 
>> Best regards,
>> Ridong
>>

-- 
Best regards,
Ridong