Re: [RFC v11 08/14] mm: page_frag: some minor refactoring before adding new API
From: Yunsheng Lin
Date: Thu Aug 01 2024 - 08:54:23 EST
On 2024/8/1 1:02, Alexander H Duyck wrote:
> On Wed, 2024-07-31 at 20:35 +0800, Yunsheng Lin wrote:
>> On 2024/7/30 23:12, Alexander H Duyck wrote:
>>
>> ...
>>
>>>> }
>>>>
>>>> nc->pagecnt_bias--;
>>>> nc->remaining = remaining - fragsz;
>>>>
>>>> return encoded_page_address(encoded_va) +
>>>> (page_frag_cache_page_size(encoded_va) - remaining);
>>>
>>> Parenthesis here shouldn't be needed, addition and subtractions
>>> operations can happen in any order with the result coming out the same.
>>
>> I am playing safe to avoid overflow here, as I am not sure if the allocator
>> will give us the last page. For example, '0xfffffffffffff000 + 0x1000' will
>> have a overflow.
>
> So what if it does though? When you subtract remaining it will
> underflow and go back to the correct value shouldn't it?
I guess that it is true that underflow will bring back the correct value.
But I am not sure what does it hurt to have a parenthesis here, doesn't having
a parenthesis make it more obvious that 'size - remaining' indicate the offset
of allocated fragment and not having to scratch my head and wondering if there
is overflow/underflow problem? Or is there any performance trick behind the above
comment?