[PATCH v2] mm/THP: use hugepage_vma_check() in khugepaged_enter_vma_merge()

[PATCH v2] mm/THP: use hugepage_vma_check() in khugepaged_enter_vma_merge()

[Song Liu]

khugepaged_enter_vma_merge() is using a different approach to check
whether a vma is valid for khugepaged_enter():

    if (!vma->anon_vma)
            /*
             * Not yet faulted in so we will register later in the
             * page fault if needed.
             */
            return 0;
    if (vma->vm_ops || (vm_flags & VM_NO_KHUGEPAGED))
            /* khugepaged not yet working on file or special mappings */
            return 0;

This check has some problems. One of the obvious problems is that
it doesn't check shmem_file(), so that vma backed with shmem files
will not call khugepaged_enter(). Here is an example of failed madvise():

   /* mount /dev/shm with huge=advise:
    *     mount -o remount,huge=advise /dev/shm */
   /* create file /dev/shm/huge */
   #define HUGE_FILE "/dev/shm/huge"

   fd = open(HUGE_FILE, O_RDONLY);
   ptr = mmap(NULL, FILE_SIZE, PROT_READ, MAP_PRIVATE, fd, 0);
   ret = madvise(ptr, FILE_SIZE, MADV_HUGEPAGE);

madvise() will return 0, but this memory region is never put in huge
page (check from /proc/meminfo: ShmemHugePages).

This patch fixes these problems by reusing hugepage_vma_check() in
khugepaged_enter_vma_merge().

vma->vm_flags is not yet updated in khugepaged_enter_vma_merge(),
so we need to pass the new vm_flags to hugepage_vma_check() through
a separate argument.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 mm/khugepaged.c | 26 ++++++++++++--------------
 1 file changed, 12 insertions(+), 14 deletions(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index d7b2a4b..9f74e51 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -430,18 +430,15 @@ int __khugepaged_enter(struct mm_struct *mm)
 	return 0;
 }
 
+static bool hugepage_vma_check(struct vm_area_struct *vma,
+			       unsigned long vm_flags);
+
 int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
 			       unsigned long vm_flags)
 {
 	unsigned long hstart, hend;
-	if (!vma->anon_vma)
-		/*
-		 * Not yet faulted in so we will register later in the
-		 * page fault if needed.
-		 */
-		return 0;
-	if (vma->vm_ops || (vm_flags & VM_NO_KHUGEPAGED))
-		/* khugepaged not yet working on file or special mappings */
+
+	if (!hugepage_vma_check(vma, vm_flags))
 		return 0;
 	hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
 	hend = vma->vm_end & HPAGE_PMD_MASK;
@@ -819,10 +816,11 @@ khugepaged_alloc_page(struct page **hpage, gfp_t gfp, int node)
 }
 #endif
 
-static bool hugepage_vma_check(struct vm_area_struct *vma)
+static bool hugepage_vma_check(struct vm_area_struct *vma,
+			       unsigned long vm_flags)
 {
-	if ((!(vma->vm_flags & VM_HUGEPAGE) && !khugepaged_always()) ||
-	    (vma->vm_flags & VM_NOHUGEPAGE) ||
+	if ((!(vm_flags & VM_HUGEPAGE) && !khugepaged_always()) ||
+	    (vm_flags & VM_NOHUGEPAGE) ||
 	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
 		return false;
 	if (shmem_file(vma->vm_file)) {
@@ -835,7 +833,7 @@ static bool hugepage_vma_check(struct vm_area_struct *vma)
 		return false;
 	if (is_vma_temporary_stack(vma))
 		return false;
-	return !(vma->vm_flags & VM_NO_KHUGEPAGED);
+	return !(vm_flags & VM_NO_KHUGEPAGED);
 }
 
 /*
@@ -862,7 +860,7 @@ static int hugepage_vma_revalidate(struct mm_struct *mm, unsigned long address,
 	hend = vma->vm_end & HPAGE_PMD_MASK;
 	if (address < hstart || address + HPAGE_PMD_SIZE > hend)
 		return SCAN_ADDRESS_RANGE;
-	if (!hugepage_vma_check(vma))
+	if (!hugepage_vma_check(vma, vma->vm_flags))
 		return SCAN_VMA_CHECK;
 	return 0;
 }
@@ -1694,7 +1692,7 @@ static unsigned int khugepaged_scan_mm_slot(unsigned int pages,
 			progress++;
 			break;
 		}
-		if (!hugepage_vma_check(vma)) {
+		if (!hugepage_vma_check(vma, vma->vm_flags)) {
 skip:
 			progress++;
 			continue;
-- 
2.9.5

Re: [PATCH v2] mm/THP: use hugepage_vma_check() in khugepaged_enter_vma_merge()

[Kirill A. Shutemov]

On Tue, May 22, 2018 at 12:44:30PM -0700, Song Liu wrote:
> khugepaged_enter_vma_merge() is using a different approach to check
> whether a vma is valid for khugepaged_enter():
> 
>     if (!vma->anon_vma)
>             /*
>              * Not yet faulted in so we will register later in the
>              * page fault if needed.
>              */
>             return 0;
>     if (vma->vm_ops || (vm_flags & VM_NO_KHUGEPAGED))
>             /* khugepaged not yet working on file or special mappings */
>             return 0;
> 
> This check has some problems. One of the obvious problems is that
> it doesn't check shmem_file(), so that vma backed with shmem files
> will not call khugepaged_enter(). Here is an example of failed madvise():
> 
>    /* mount /dev/shm with huge=advise:
>     *     mount -o remount,huge=advise /dev/shm */
>    /* create file /dev/shm/huge */
>    #define HUGE_FILE "/dev/shm/huge"
> 
>    fd = open(HUGE_FILE, O_RDONLY);
>    ptr = mmap(NULL, FILE_SIZE, PROT_READ, MAP_PRIVATE, fd, 0);
>    ret = madvise(ptr, FILE_SIZE, MADV_HUGEPAGE);
> 
> madvise() will return 0, but this memory region is never put in huge
> page (check from /proc/meminfo: ShmemHugePages).
> 
> This patch fixes these problems by reusing hugepage_vma_check() in
> khugepaged_enter_vma_merge().
> 
> vma->vm_flags is not yet updated in khugepaged_enter_vma_merge(),
> so we need to pass the new vm_flags to hugepage_vma_check() through
> a separate argument.
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>  mm/khugepaged.c | 26 ++++++++++++--------------
>  1 file changed, 12 insertions(+), 14 deletions(-)
> 
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index d7b2a4b..9f74e51 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -430,18 +430,15 @@ int __khugepaged_enter(struct mm_struct *mm)
>  	return 0;
>  }
>  
> +static bool hugepage_vma_check(struct vm_area_struct *vma,
> +			       unsigned long vm_flags);
> +

The patch looks good to me.

But can we move hugepage_vma_check() here to avoid forward declaration of
the function?

-- 
 Kirill A. Shutemov

Re: [PATCH v2] mm/THP: use hugepage_vma_check() in khugepaged_enter_vma_merge()

[Song Liu]

> On May 28, 2018, at 3:57 AM, Kirill A. Shutemov <kirill@shutemov.name> wrote:
> 
> On Tue, May 22, 2018 at 12:44:30PM -0700, Song Liu wrote:
>> khugepaged_enter_vma_merge() is using a different approach to check
>> whether a vma is valid for khugepaged_enter():
>> 
>>    if (!vma->anon_vma)
>>            /*
>>             * Not yet faulted in so we will register later in the
>>             * page fault if needed.
>>             */
>>            return 0;
>>    if (vma->vm_ops || (vm_flags & VM_NO_KHUGEPAGED))
>>            /* khugepaged not yet working on file or special mappings */
>>            return 0;
>> 
>> This check has some problems. One of the obvious problems is that
>> it doesn't check shmem_file(), so that vma backed with shmem files
>> will not call khugepaged_enter(). Here is an example of failed madvise():
>> 
>>   /* mount /dev/shm with huge=advise:
>>    *     mount -o remount,huge=advise /dev/shm */
>>   /* create file /dev/shm/huge */
>>   #define HUGE_FILE "/dev/shm/huge"
>> 
>>   fd = open(HUGE_FILE, O_RDONLY);
>>   ptr = mmap(NULL, FILE_SIZE, PROT_READ, MAP_PRIVATE, fd, 0);
>>   ret = madvise(ptr, FILE_SIZE, MADV_HUGEPAGE);
>> 
>> madvise() will return 0, but this memory region is never put in huge
>> page (check from /proc/meminfo: ShmemHugePages).
>> 
>> This patch fixes these problems by reusing hugepage_vma_check() in
>> khugepaged_enter_vma_merge().
>> 
>> vma->vm_flags is not yet updated in khugepaged_enter_vma_merge(),
>> so we need to pass the new vm_flags to hugepage_vma_check() through
>> a separate argument.
>> 
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> mm/khugepaged.c | 26 ++++++++++++--------------
>> 1 file changed, 12 insertions(+), 14 deletions(-)
>> 
>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>> index d7b2a4b..9f74e51 100644
>> --- a/mm/khugepaged.c
>> +++ b/mm/khugepaged.c
>> @@ -430,18 +430,15 @@ int __khugepaged_enter(struct mm_struct *mm)
>> 	return 0;
>> }
>> 
>> +static bool hugepage_vma_check(struct vm_area_struct *vma,
>> +			       unsigned long vm_flags);
>> +
> 
> The patch looks good to me.
> 
> But can we move hugepage_vma_check() here to avoid forward declaration of
> the function?

Thanks for the feedback! I will send v3 with this change. 

Song