mm/khugepaged: use round_{up,down}() for PMD alignment in madvise_collapse()

[PATCH] mm/khugepaged: use round_{up,down}() for PMD alignment in madvise_collapse()

Posted by Ye Liu 2 months, 1 week ago

From: Ye Liu <liuye@kylinos.cn>

The PMD alignment in madvise_collapse() is currently implemented
using open-coded bitmask operations:

    (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK
    end & HPAGE_PMD_MASK

Replace them with round_up() and round_down(), which express the
same alignment semantics more directly and match other PMD-alignment
code paths in khugepaged.

No functional change intended.

Signed-off-by: Ye Liu <liuye@kylinos.cn>
---
 mm/khugepaged.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 1e2bff40d014..80d0b7136b58 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -2835,8 +2835,8 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
 	mmgrab(mm);
 	lru_add_drain_all();
 
-	hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
-	hend = end & HPAGE_PMD_MASK;
+	hstart = round_up(start, HPAGE_PMD_SIZE);
+	hend = round_down(end, HPAGE_PMD_SIZE);
 
 	for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
 		enum scan_result result = SCAN_FAIL;
-- 
2.43.0

Re: [PATCH] mm/khugepaged: use round_{up,down}() for PMD alignment in madvise_collapse()

Posted by David Hildenbrand (Arm) 2 months, 1 week ago

On 4/8/26 11:35, Ye Liu wrote:
> From: Ye Liu <liuye@kylinos.cn>
> 
> The PMD alignment in madvise_collapse() is currently implemented
> using open-coded bitmask operations:
> 
>     (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK
>     end & HPAGE_PMD_MASK
> 
> Replace them with round_up() and round_down(), which express the
> same alignment semantics more directly and match other PMD-alignment
> code paths in khugepaged.
> 
> No functional change intended.
> 
> Signed-off-by: Ye Liu <liuye@kylinos.cn>
> ---
>  mm/khugepaged.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 1e2bff40d014..80d0b7136b58 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -2835,8 +2835,8 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
>  	mmgrab(mm);
>  	lru_add_drain_all();
>  
> -	hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
> -	hend = end & HPAGE_PMD_MASK;
> +	hstart = round_up(start, HPAGE_PMD_SIZE);
> +	hend = round_down(end, HPAGE_PMD_SIZE);
>  
>  	for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
>  		enum scan_result result = SCAN_FAIL;

I tend to prefer ALIGN / ALIGN_DOWN when dealing with sizes + addresses.

-- 
Cheers,

David

Re: [PATCH] mm/khugepaged: use round_{up,down}() for PMD alignment in madvise_collapse()

Posted by liuye 2 months, 1 week ago

在 2026/4/8 18:34, David Hildenbrand (Arm) 写道:
> On 4/8/26 11:35, Ye Liu wrote:
>> From: Ye Liu <liuye@kylinos.cn>
>>
>> The PMD alignment in madvise_collapse() is currently implemented
>> using open-coded bitmask operations:
>>
>>     (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK
>>     end & HPAGE_PMD_MASK
>>
>> Replace them with round_up() and round_down(), which express the
>> same alignment semantics more directly and match other PMD-alignment
>> code paths in khugepaged.
>>
>> No functional change intended.
>>
>> Signed-off-by: Ye Liu <liuye@kylinos.cn>
>> ---
>>  mm/khugepaged.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>> index 1e2bff40d014..80d0b7136b58 100644
>> --- a/mm/khugepaged.c
>> +++ b/mm/khugepaged.c
>> @@ -2835,8 +2835,8 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
>>  	mmgrab(mm);
>>  	lru_add_drain_all();
>>  
>> -	hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
>> -	hend = end & HPAGE_PMD_MASK;
>> +	hstart = round_up(start, HPAGE_PMD_SIZE);
>> +	hend = round_down(end, HPAGE_PMD_SIZE);
>>  
>>  	for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
>>  		enum scan_result result = SCAN_FAIL;
> I tend to prefer ALIGN / ALIGN_DOWN when dealing with sizes + addresses.
>
Thanks, that makes sense.                                          
                                                                   
I'll switch this to ALIGN()/ALIGN_DOWN() in v2 for the address/size
alignment style preference.                                        

-- 
Thanks,
Ye Liu