Re: [PATCH v2 1/3] userfaultfd: use vma_pagesize for all huge page size calculation

From: Mike Kravetz <mike.kravetz@oracle.com>
To: Wei Yang <richardw.yang@linux.intel.com>,
	Andrew Morton <akpm@linux-foundation.org>
Cc: aarcange@redhat.com, hughd@google.com, linux-mm@kvack.org
Subject: Re: [PATCH v2 1/3] userfaultfd: use vma_pagesize for all huge page size calculation
Date: Mon, 7 Oct 2019 15:55:21 -0700	[thread overview]
Message-ID: <f405e537-1cc2-6d40-4d2b-f30e19d68ac9@oracle.com> (raw)
In-Reply-To: <20190929004507.GA11770@richard>

On 9/28/19 5:45 PM, Wei Yang wrote:
> On Fri, Sep 27, 2019 at 03:10:33PM -0700, Andrew Morton wrote:
>> On Fri, 27 Sep 2019 15:00:30 +0800 Wei Yang <richardw.yang@linux.intel.com> wrote:
>>
>>> In function __mcopy_atomic_hugetlb, we use two variables to deal with
>>> huge page size: vma_hpagesize and huge_page_size.
>>>
>>> Since they are the same, it is not necessary to use two different
>>> mechanism. This patch makes it consistent by all using vma_hpagesize.
>>>
>>> --- a/mm/userfaultfd.c
>>> +++ b/mm/userfaultfd.c
>>> @@ -262,7 +262,7 @@ static __always_inline ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm,
>>>  		pte_t dst_pteval;
>>>  
>>>  		BUG_ON(dst_addr >= dst_start + len);
>>> -		VM_BUG_ON(dst_addr & ~huge_page_mask(h));
>>> +		VM_BUG_ON(dst_addr & (vma_hpagesize - 1));
>>>  
>>>  		/*
>>>  		 * Serialize via hugetlb_fault_mutex
>>> @@ -273,7 +273,7 @@ static __always_inline ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm,
>>>  		mutex_lock(&hugetlb_fault_mutex_table[hash]);
>>>  
>>>  		err = -ENOMEM;
>>> -		dst_pte = huge_pte_alloc(dst_mm, dst_addr, huge_page_size(h));
>>> +		dst_pte = huge_pte_alloc(dst_mm, dst_addr, vma_hpagesize);
>>>  		if (!dst_pte) {
>>>  			mutex_unlock(&hugetlb_fault_mutex_table[hash]);
>>>  			goto out_unlock;
>>> @@ -300,7 +300,8 @@ static __always_inline ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm,
>>>  
>>>  			err = copy_huge_page_from_user(page,
>>>  						(const void __user *)src_addr,
>>> -						pages_per_huge_page(h), true);
>>> +						vma_hpagesize / PAGE_SIZE,
>>> +						true);
>>>  			if (unlikely(err)) {
>>>  				err = -EFAULT;
>>>  				goto out;
>>
>> Looks right.
>>
>> We could go ahead and remove local variable `h', given that
>> hugetlb_fault_mutex_hash() doesn't actually use its first arg..
> 
> Oops, haven't imagine h is not used in the function.
> 
> 
> Any historical reason to pass h in hugetlb_fault_mutex_hash()? Neither these
> two definition use it.

See 1b426bac66e6 ("hugetlb: use same fault hash key for shared and private
mappings").  Prior to that change, the hash key for private mappings was
created by:

	key[0] = (unsigned long) mm;
	key[1] = address >> huge_page_shift(h);

When removing that code, I should have removed 'h'.
-- 
Mike Kravetz