Re: [PATCH] mm: remove a redundant check in do_munmap()

[Wei Yang <richard.weiyang@gmail.com>]

Well, this change is not correct.

On Wed, Oct 10, 2018 at 07:13:55AM -0700, Matthew Wilcox wrote:
>On Wed, Oct 10, 2018 at 08:53:27PM +0800, Wei Yang wrote:
>> A non-NULL vma returned from find_vma() implies:
>> 
>>    vma->vm_start <= start
>> 

My misunderstanding of find_vma(), the non-NULL return value from
find_vma() doesn't impley vma->vm_start <= start. Instead it just
implies addr < vma->vm_end.

This means the original check between vm_start and end is necessary.

Thanks for testing from Rong.

>> Since len != 0, the following condition always hods:
>> 
>>    vma->vm_start < start + len = end
>> 
>> This means the if check would never be true.
>
>This is true because earlier in the function, start + len is checked to
>be sure that it does not wrap.
>
>> This patch removes this redundant check and fix two typo in comment.
>
>> @@ -2705,12 +2705,8 @@ int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
>> -	/* we have  start < vma->vm_end  */
>> -
>> -	/* if it doesn't overlap, we have nothing.. */
>> +	/* we have vma->vm_start <= start < vma->vm_end */
>>  	end = start + len;
>> -	if (vma->vm_start >= end)
>> -		return 0;
>
>I agree that it's not currently a useful check, but it's also not going
>to have much effect on anything to delete it.  I think there are probably
>more worthwhile places to look for inefficiencies.

-- 
Wei Yang
Help you, Help me