Re: [PATCH] percpu_ref: call wake_up_all() after percpu_ref_put() completes

[Qi Zheng <zhengqi.arch@bytedance.com>]

On 2022/4/8 10:54 AM, Muchun Song wrote:
> On Thu, Apr 07, 2022 at 06:33:35PM +0800, Qi Zheng wrote:
>> In the percpu_ref_call_confirm_rcu(), we call the wake_up_all()
>> before calling percpu_ref_put(), which will cause the value of
>> percpu_ref to be unstable when percpu_ref_switch_to_atomic_sync()
>> returns.
>>
>> 	CPU0				CPU1
>>
>> percpu_ref_switch_to_atomic_sync(&ref)
>> --> percpu_ref_switch_to_atomic(&ref)
>>      --> percpu_ref_get(ref);	/* put after confirmation */
>> 	call_rcu(&ref->data->rcu, percpu_ref_switch_to_atomic_rcu);
>>
>> 					percpu_ref_switch_to_atomic_rcu
>> 					--> percpu_ref_call_confirm_rcu
>> 					    --> data->confirm_switch = NULL;
>> 						wake_up_all(&percpu_ref_switch_waitq);
>>
>>      /* here waiting to wake up */
>>      wait_event(percpu_ref_switch_waitq, !ref->data->confirm_switch);
>> 						(A)percpu_ref_put(ref);
>> /* The value of &ref is unstable! */
>> percpu_ref_is_zero(&ref)
>> 						(B)percpu_ref_put(ref);
>>
>> As shown above, assuming that the counts on each cpu add up to 0 before
>> calling percpu_ref_switch_to_atomic_sync(), we expect that after switching
>> to atomic mode, percpu_ref_is_zero() can return true. But actually it will
>> return different values in the two cases of A and B, which is not what
>> we expected.
>>
>> Maybe the original purpose of percpu_ref_switch_to_atomic_sync() is
>> just to ensure that the conversion to atomic mode is completed, but it
>> should not return with an extra reference count.
>>
>> Calling wake_up_all() after percpu_ref_put() ensures that the value of
>> percpu_ref is stable after percpu_ref_switch_to_atomic_sync() returns.
>> So just do it.
>>
>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> 
> Are any users affected by this?  If so, I think a Fixes tag
> is necessary.

Looks all current users(blk_pre_runtime_suspend() and set_in_sync()) are
affected by this.

I see that this patch has been merged into the mm tree, can Andrew help
me add the following Fixes tag?

Fixes: 490c79a65708 ("percpu_ref: decouple switching to atomic mode and 
killing")

Thanks,
Qi

> 
> The fix LGTM.
> 
> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
> 
> Thanks.
> 

-- 
Thanks,
Qi