[PATCH linux-next] mm/huge_memory: mark racy access on huge_anon_orders_always

Posted by xu.xin16@zte.com.cn 1 year, 7 months ago

From: Ran Xiaokai <ran.xiaokai@zte.com.cn>

huge_anon_orders_always and huge_anon_orders_always are accessed
lockless, it is better to use the READ_ONCE() wrapper.
This is not fixing any visible bug, hopefully this can cease some
KCSAN complains in the future.
Also do that for huge_anon_orders_madvise.

Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
---
 include/linux/huge_mm.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
index de0c89105076..6573430ea600 100644
--- a/include/linux/huge_mm.h
+++ b/include/linux/huge_mm.h
@@ -122,8 +122,8 @@ static inline bool hugepage_flags_enabled(void)
 	 * So we don't need to look at huge_anon_orders_inherit.
 	 */
 	return hugepage_global_enabled() ||
-	       huge_anon_orders_always ||
-	       huge_anon_orders_madvise;
+			READ_ONCE(huge_anon_orders_always) ||
+			READ_ONCE(huge_anon_orders_madvise);
 }

 static inline int highest_order(unsigned long orders)
-- 
2.15.2

Re: [PATCH linux-next] mm/huge_memory: mark racy access on huge_anon_orders_always

Posted by Yang Shi 1 year, 7 months ago

On Sat, May 11, 2024 at 12:44 AM <xu.xin16@zte.com.cn> wrote:
>
> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>
> huge_anon_orders_always and huge_anon_orders_always are accessed
> lockless, it is better to use the READ_ONCE() wrapper.
> This is not fixing any visible bug, hopefully this can cease some
> KCSAN complains in the future.

A little bit confused here. Did you see complaints from KCSAN?

> Also do that for huge_anon_orders_madvise.
>
> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> ---
>  include/linux/huge_mm.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> index de0c89105076..6573430ea600 100644
> --- a/include/linux/huge_mm.h
> +++ b/include/linux/huge_mm.h
> @@ -122,8 +122,8 @@ static inline bool hugepage_flags_enabled(void)
>          * So we don't need to look at huge_anon_orders_inherit.
>          */
>         return hugepage_global_enabled() ||
> -              huge_anon_orders_always ||
> -              huge_anon_orders_madvise;
> +                       READ_ONCE(huge_anon_orders_always) ||
> +                       READ_ONCE(huge_anon_orders_madvise);
>  }
>
>  static inline int highest_order(unsigned long orders)
> --
> 2.15.2

Re: [PATCH linux-next] mm/huge_memory: mark racy access on huge_anon_orders_always

Posted by ran xiaokai 1 year, 7 months ago

>>
>> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>>
>> huge_anon_orders_always and huge_anon_orders_always are accessed
>> lockless, it is better to use the READ_ONCE() wrapper.
>> This is not fixing any visible bug, hopefully this can cease some
>> KCSAN complains in the future.
>
> A little bit confused here. Did you see complaints from KCSAN?

Not yet.
The only written access to huge_anon_orders_always is when
changing the mTHP sysfs "enabled" knob, I think peple do not
change the mTHP settings frequently.
But the possibility is not zero, please refer to the cgroup related knobs,
I think  wde'd better do the same and this annotation wont lead to any performance issue, right?

>> Also do that for huge_anon_orders_madvise.
>>
>> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>> ---
>>  include/linux/huge_mm.h | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index de0c89105076..6573430ea600 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -122,8 +122,8 @@ static inline bool hugepage_flags_enabled(void)
>>          * So we don't need to look at huge_anon_orders_inherit.
>>          */
>>         return hugepage_global_enabled() ||
>> -              huge_anon_orders_always ||
>> -              huge_anon_orders_madvise;
>> +                       READ_ONCE(huge_anon_orders_always) ||
>> +                       READ_ONCE(huge_anon_orders_madvise);
>>  }
>>
>>  static inline int highest_order(unsigned long orders)

Re: [PATCH linux-next] mm/huge_memory: mark racy access on huge_anon_orders_always

Posted by David Hildenbrand 1 year, 7 months ago

On 11.05.24 08:44, xu.xin16@zte.com.cn wrote:
> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> 
> huge_anon_orders_always and huge_anon_orders_always are accessed

"huge_anon_orders_always" mentioned twice.

> lockless, it is better to use the READ_ONCE() wrapper.
> This is not fixing any visible bug, hopefully this can cease some
> KCSAN complains in the future.
> Also do that for huge_anon_orders_madvise.
> 
> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> ---
>   include/linux/huge_mm.h | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> index de0c89105076..6573430ea600 100644
> --- a/include/linux/huge_mm.h
> +++ b/include/linux/huge_mm.h
> @@ -122,8 +122,8 @@ static inline bool hugepage_flags_enabled(void)
>   	 * So we don't need to look at huge_anon_orders_inherit.
>   	 */
>   	return hugepage_global_enabled() ||
> -	       huge_anon_orders_always ||
> -	       huge_anon_orders_madvise;
> +			READ_ONCE(huge_anon_orders_always) ||
> +			READ_ONCE(huge_anon_orders_madvise);

Don't mess up the alignment please.

>   }
> 
>   static inline int highest_order(unsigned long orders)

Acked-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb

Re: [PATCH linux-next] mm/huge_memory: mark racy access on huge_anon_orders_always

Posted by ran xiaokai 1 year, 7 months ago

Hi, David

Thanks for taking time to review!

>> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>> 
>> huge_anon_orders_always and huge_anon_orders_always are accessed
>
> "huge_anon_orders_always" mentioned twice.

yes, my mistake.

>> lockless, it is better to use the READ_ONCE() wrapper.
>> This is not fixing any visible bug, hopefully this can cease some
>> KCSAN complains in the future.
>> Also do that for huge_anon_orders_madvise.
>> 
>> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>> ---
>>   include/linux/huge_mm.h | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>> 
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index de0c89105076..6573430ea600 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -122,8 +122,8 @@ static inline bool hugepage_flags_enabled(void)
>>   	 * So we don't need to look at huge_anon_orders_inherit.
>>   	 */
>>   	return hugepage_global_enabled() ||
>> -	       huge_anon_orders_always ||
>> -	       huge_anon_orders_madvise;
>> +			READ_ONCE(huge_anon_orders_always) ||
>> +			READ_ONCE(huge_anon_orders_madvise);
>
> Don't mess up the alignment please.

yes.

I will wait for other feedbacks and then send out a v2 version.

>>   }
>> 
>>   static inline int highest_order(unsigned long orders)
>
> Acked-by: David Hildenbrand <david@redhat.com>