Re: [RFC PATCH v3] mm: use READ_ONCE in page_cpupid_xchg_last()

[Christian Borntraeger]

On 12/07/2016 10:29 AM, Vlastimil Babka wrote:
> On 12/07/2016 09:58 AM, Michal Hocko wrote:
>> On Wed 07-12-16 09:48:52, Vlastimil Babka wrote:
>>> On 12/07/2016 09:43 AM, Michal Hocko wrote:
>>>> On Tue 06-12-16 09:53:14, Xishi Qiu wrote:
>>>>> A compiler could re-read "old_flags" from the memory location after reading
>>>>> and calculation "flags" and passes a newer value into the cmpxchg making 
>>>>> the comparison succeed while it should actually fail.
>>>>>
>>>>> Signed-off-by: Xishi Qiu <qiuxishi@xxxxxxxxxx>
>>>>> Suggested-by: Christian Borntraeger <borntraeger@xxxxxxxxxx>
>>>>> ---
>>>>>  mm/mmzone.c | 2 +-
>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/mm/mmzone.c b/mm/mmzone.c
>>>>> index 5652be8..e0b698e 100644
>>>>> --- a/mm/mmzone.c
>>>>> +++ b/mm/mmzone.c
>>>>> @@ -102,7 +102,7 @@ int page_cpupid_xchg_last(struct page *page, int cpupid)
>>>>>  	int last_cpupid;
>>>>>  
>>>>>  	do {
>>>>> -		old_flags = flags = page->flags;
>>>>> +		old_flags = flags = READ_ONCE(page->flags);
>>>>>  		last_cpupid = page_cpupid_last(page);
>>>>
>>>> what prevents compiler from doing?
>>>> 		old_flags = READ_ONCE(page->flags);
>>>> 		flags = READ_ONCE(page->flags);
>>>
>>> AFAIK, READ_ONCE tells the compiler that page->flags is volatile. It
>>> can't read from volatile location more times than being told?
>>
>> But those are two different variables which we assign to so what
>> prevents the compiler from applying READ_ONCE on each of them
>> separately?
> 
> I would naively expect that it's assigned to flags first, and then from
> flags to old_flags. But I don't know exactly the C standard evaluation
> rules that apply here.
> 
>> Anyway, this could be addressed easily by
> 
> Yes, that way there should be no doubt.

That change would make it clearer, but the code is correct anyway,
as assignments in C are done from right to left, so 
old_flags = flags = READ_ONCE(page->flags);

is equivalent to 

flags = READ_ONCE(page->flags);
old_flags = flags;


> 
>> diff --git a/mm/mmzone.c b/mm/mmzone.c
>> index 5652be858e5e..b4e093dd24c1 100644
>> --- a/mm/mmzone.c
>> +++ b/mm/mmzone.c
>> @@ -102,10 +102,10 @@ int page_cpupid_xchg_last(struct page *page, int cpupid)
>>  	int last_cpupid;
>>  
>>  	do {
>> -		old_flags = flags = page->flags;
>> +		old_flags = READ_ONCE(page->flags);
>>  		last_cpupid = page_cpupid_last(page);
>>  
>> -		flags &= ~(LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT);
>> +		flags = old_flags & ~(LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT);
>>  		flags |= (cpupid & LAST_CPUPID_MASK) << LAST_CPUPID_PGSHIFT;
>>  	} while (unlikely(cmpxchg(&page->flags, old_flags, flags) != old_flags));
>>  
>>
>>>> Or this doesn't matter?
>>>
>>> I think it would matter.
>>>
>>>>>  
>>>>>  		flags &= ~(LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT);
>>>>> -- 
>>>>> 1.8.3.1
>>>>>
>>>>
>>
>