[RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[Yang Shi]

Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
filesystem with huge page support anymore. tmpfs can use huge page via
THP when mounting by "huge=" mount option.

When applications use huge page on hugetlbfs, it just need check the
filesystem magic number, but it is not enough for tmpfs. Make
stat.st_blksize return huge page size if it is mounted by appropriate
"huge=" option to give applications a hint to optimize the behavior with
THP.

Some applications may not do wisely with THP. For example, QEMU may mmap
file on non huge page aligned hint address with MAP_FIXED, which results
in no pages are PMD mapped even though THP is used. Some applications
may mmap file with non huge page aligned offset. Both behaviors make THP
pointless.

statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and it
also may fallback to 4KB page silently if there is not enough huge page.
Furthermore, different f_bsize makes max_blocks and free_blocks
calculation harder but without too much benefit. Returning huge page
size via stat.st_blksize sounds good enough.

Since PUD size huge page for THP has not been supported, now it just
returns HPAGE_PMD_SIZE.

Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Suggested-by: Christoph Hellwig <hch@infradead.org>
---
v3 --> v4:
* Rework the commit log per the education from Michal and Kirill
* Fix build error if CONFIG_TRANSPARENT_HUGEPAGE is disabled
v2 --> v3:
* Use shmem_sb_info.huge instead of global variable per Michal's comment
v2 --> v1:
* Adopted the suggestion from hch to return huge page size via st_blksize
  instead of creating a new flag.

 mm/shmem.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/mm/shmem.c b/mm/shmem.c
index b859192..19b8055 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -988,6 +988,7 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
 {
 	struct inode *inode = path->dentry->d_inode;
 	struct shmem_inode_info *info = SHMEM_I(inode);
+	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 
 	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
 		spin_lock_irq(&info->lock);
@@ -995,6 +996,11 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
 		spin_unlock_irq(&info->lock);
 	}
 	generic_fillattr(inode, stat);
+#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
+	if (sbinfo->huge > 0)
+		stat->blksize = HPAGE_PMD_SIZE;
+#endif
+	
 	return 0;
 }
 
-- 
1.8.3.1

Re: [RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[Kirill A. Shutemov]

On Tue, Apr 24, 2018 at 12:00:50PM +0800, Yang Shi wrote:
> Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
> filesystem with huge page support anymore. tmpfs can use huge page via
> THP when mounting by "huge=" mount option.
> 
> When applications use huge page on hugetlbfs, it just need check the
> filesystem magic number, but it is not enough for tmpfs. Make
> stat.st_blksize return huge page size if it is mounted by appropriate
> "huge=" option to give applications a hint to optimize the behavior with
> THP.
> 
> Some applications may not do wisely with THP. For example, QEMU may mmap
> file on non huge page aligned hint address with MAP_FIXED, which results
> in no pages are PMD mapped even though THP is used. Some applications
> may mmap file with non huge page aligned offset. Both behaviors make THP
> pointless.
> 
> statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and it
> also may fallback to 4KB page silently if there is not enough huge page.
> Furthermore, different f_bsize makes max_blocks and free_blocks
> calculation harder but without too much benefit. Returning huge page
> size via stat.st_blksize sounds good enough.
> 
> Since PUD size huge page for THP has not been supported, now it just
> returns HPAGE_PMD_SIZE.
> 
> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
> Cc: Hugh Dickins <hughd@google.com>
> Cc: Michal Hocko <mhocko@kernel.org>
> Cc: Alexander Viro <viro@zeniv.linux.org.uk>
> Suggested-by: Christoph Hellwig <hch@infradead.org>
> ---
> v3 --> v4:
> * Rework the commit log per the education from Michal and Kirill
> * Fix build error if CONFIG_TRANSPARENT_HUGEPAGE is disabled
> v2 --> v3:
> * Use shmem_sb_info.huge instead of global variable per Michal's comment
> v2 --> v1:
> * Adopted the suggestion from hch to return huge page size via st_blksize
>   instead of creating a new flag.
> 
>  mm/shmem.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index b859192..19b8055 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -988,6 +988,7 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>  {
>  	struct inode *inode = path->dentry->d_inode;
>  	struct shmem_inode_info *info = SHMEM_I(inode);
> +	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
>  
>  	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
>  		spin_lock_irq(&info->lock);
> @@ -995,6 +996,11 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>  		spin_unlock_irq(&info->lock);
>  	}
>  	generic_fillattr(inode, stat);
> +#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
> +	if (sbinfo->huge > 0)

No ifdeffery, please.

And we probably want to check if shmem_huge is 'force'.

Something like this?

	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
		 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge))

> +		stat->blksize = HPAGE_PMD_SIZE;
> +#endif
> +	
>  	return 0;
>  }
>  
> -- 
> 1.8.3.1
> 

-- 
 Kirill A. Shutemov

Re: [RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[Yang Shi]

On 4/24/18 5:23 AM, Kirill A. Shutemov wrote:
> On Tue, Apr 24, 2018 at 12:00:50PM +0800, Yang Shi wrote:
>> Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
>> filesystem with huge page support anymore. tmpfs can use huge page via
>> THP when mounting by "huge=" mount option.
>>
>> When applications use huge page on hugetlbfs, it just need check the
>> filesystem magic number, but it is not enough for tmpfs. Make
>> stat.st_blksize return huge page size if it is mounted by appropriate
>> "huge=" option to give applications a hint to optimize the behavior with
>> THP.
>>
>> Some applications may not do wisely with THP. For example, QEMU may mmap
>> file on non huge page aligned hint address with MAP_FIXED, which results
>> in no pages are PMD mapped even though THP is used. Some applications
>> may mmap file with non huge page aligned offset. Both behaviors make THP
>> pointless.
>>
>> statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and it
>> also may fallback to 4KB page silently if there is not enough huge page.
>> Furthermore, different f_bsize makes max_blocks and free_blocks
>> calculation harder but without too much benefit. Returning huge page
>> size via stat.st_blksize sounds good enough.
>>
>> Since PUD size huge page for THP has not been supported, now it just
>> returns HPAGE_PMD_SIZE.
>>
>> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
>> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
>> Cc: Hugh Dickins <hughd@google.com>
>> Cc: Michal Hocko <mhocko@kernel.org>
>> Cc: Alexander Viro <viro@zeniv.linux.org.uk>
>> Suggested-by: Christoph Hellwig <hch@infradead.org>
>> ---
>> v3 --> v4:
>> * Rework the commit log per the education from Michal and Kirill
>> * Fix build error if CONFIG_TRANSPARENT_HUGEPAGE is disabled
>> v2 --> v3:
>> * Use shmem_sb_info.huge instead of global variable per Michal's comment
>> v2 --> v1:
>> * Adopted the suggestion from hch to return huge page size via st_blksize
>>    instead of creating a new flag.
>>
>>   mm/shmem.c | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/mm/shmem.c b/mm/shmem.c
>> index b859192..19b8055 100644
>> --- a/mm/shmem.c
>> +++ b/mm/shmem.c
>> @@ -988,6 +988,7 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>>   {
>>   	struct inode *inode = path->dentry->d_inode;
>>   	struct shmem_inode_info *info = SHMEM_I(inode);
>> +	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
>>   
>>   	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
>>   		spin_lock_irq(&info->lock);
>> @@ -995,6 +996,11 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>>   		spin_unlock_irq(&info->lock);
>>   	}
>>   	generic_fillattr(inode, stat);
>> +#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
>> +	if (sbinfo->huge > 0)
> No ifdeffery, please.
>
> And we probably want to check if shmem_huge is 'force'.
>
> Something like this?
>
> 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
> 		 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge))

Yes, looks good, will do that. I missed "force" part, just realized it 
is applicable to all mounts.

Thanks,
Yang

>
>> +		stat->blksize = HPAGE_PMD_SIZE;
>> +#endif
>> +	
>>   	return 0;
>>   }
>>   
>> -- 
>> 1.8.3.1
>>

Re: [RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[Yang Shi]

On 4/24/18 5:23 AM, Kirill A. Shutemov wrote:
> On Tue, Apr 24, 2018 at 12:00:50PM +0800, Yang Shi wrote:
>> Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
>> filesystem with huge page support anymore. tmpfs can use huge page via
>> THP when mounting by "huge=" mount option.
>>
>> When applications use huge page on hugetlbfs, it just need check the
>> filesystem magic number, but it is not enough for tmpfs. Make
>> stat.st_blksize return huge page size if it is mounted by appropriate
>> "huge=" option to give applications a hint to optimize the behavior with
>> THP.
>>
>> Some applications may not do wisely with THP. For example, QEMU may mmap
>> file on non huge page aligned hint address with MAP_FIXED, which results
>> in no pages are PMD mapped even though THP is used. Some applications
>> may mmap file with non huge page aligned offset. Both behaviors make THP
>> pointless.
>>
>> statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and it
>> also may fallback to 4KB page silently if there is not enough huge page.
>> Furthermore, different f_bsize makes max_blocks and free_blocks
>> calculation harder but without too much benefit. Returning huge page
>> size via stat.st_blksize sounds good enough.
>>
>> Since PUD size huge page for THP has not been supported, now it just
>> returns HPAGE_PMD_SIZE.
>>
>> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
>> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
>> Cc: Hugh Dickins <hughd@google.com>
>> Cc: Michal Hocko <mhocko@kernel.org>
>> Cc: Alexander Viro <viro@zeniv.linux.org.uk>
>> Suggested-by: Christoph Hellwig <hch@infradead.org>
>> ---
>> v3 --> v4:
>> * Rework the commit log per the education from Michal and Kirill
>> * Fix build error if CONFIG_TRANSPARENT_HUGEPAGE is disabled
>> v2 --> v3:
>> * Use shmem_sb_info.huge instead of global variable per Michal's comment
>> v2 --> v1:
>> * Adopted the suggestion from hch to return huge page size via st_blksize
>>    instead of creating a new flag.
>>
>>   mm/shmem.c | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/mm/shmem.c b/mm/shmem.c
>> index b859192..19b8055 100644
>> --- a/mm/shmem.c
>> +++ b/mm/shmem.c
>> @@ -988,6 +988,7 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>>   {
>>   	struct inode *inode = path->dentry->d_inode;
>>   	struct shmem_inode_info *info = SHMEM_I(inode);
>> +	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
>>   
>>   	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
>>   		spin_lock_irq(&info->lock);
>> @@ -995,6 +996,11 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
>>   		spin_unlock_irq(&info->lock);
>>   	}
>>   	generic_fillattr(inode, stat);
>> +#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
>> +	if (sbinfo->huge > 0)
> No ifdeffery, please.
>
> And we probably want to check if shmem_huge is 'force'.
>
> Something like this?
>
> 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
> 		 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge))

BTW, it sounds we also need check shmem_huge != SHMEM_HUGE_DENY, right? 
Otherwise it still return huge page size, but in fact no huge page can 
be allocated with 'deny'.

So, how about:

if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
          (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
          shmem_huge != SHMEM_HUGE_DENY)

Thanks,
Yang

>
>> +		stat->blksize = HPAGE_PMD_SIZE;
>> +#endif
>> +	
>>   	return 0;
>>   }
>>   
>> -- 
>> 1.8.3.1
>>

Re: [RFC v4 PATCH] mm: shmem: make stat.st_blksize return huge page size if THP is on

[Kirill A. Shutemov]

On Tue, Apr 24, 2018 at 08:14:37AM -0600, Yang Shi wrote:
> 
> 
> On 4/24/18 5:23 AM, Kirill A. Shutemov wrote:
> > On Tue, Apr 24, 2018 at 12:00:50PM +0800, Yang Shi wrote:
> > > Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
> > > filesystem with huge page support anymore. tmpfs can use huge page via
> > > THP when mounting by "huge=" mount option.
> > > 
> > > When applications use huge page on hugetlbfs, it just need check the
> > > filesystem magic number, but it is not enough for tmpfs. Make
> > > stat.st_blksize return huge page size if it is mounted by appropriate
> > > "huge=" option to give applications a hint to optimize the behavior with
> > > THP.
> > > 
> > > Some applications may not do wisely with THP. For example, QEMU may mmap
> > > file on non huge page aligned hint address with MAP_FIXED, which results
> > > in no pages are PMD mapped even though THP is used. Some applications
> > > may mmap file with non huge page aligned offset. Both behaviors make THP
> > > pointless.
> > > 
> > > statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and it
> > > also may fallback to 4KB page silently if there is not enough huge page.
> > > Furthermore, different f_bsize makes max_blocks and free_blocks
> > > calculation harder but without too much benefit. Returning huge page
> > > size via stat.st_blksize sounds good enough.
> > > 
> > > Since PUD size huge page for THP has not been supported, now it just
> > > returns HPAGE_PMD_SIZE.
> > > 
> > > Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
> > > Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
> > > Cc: Hugh Dickins <hughd@google.com>
> > > Cc: Michal Hocko <mhocko@kernel.org>
> > > Cc: Alexander Viro <viro@zeniv.linux.org.uk>
> > > Suggested-by: Christoph Hellwig <hch@infradead.org>
> > > ---
> > > v3 --> v4:
> > > * Rework the commit log per the education from Michal and Kirill
> > > * Fix build error if CONFIG_TRANSPARENT_HUGEPAGE is disabled
> > > v2 --> v3:
> > > * Use shmem_sb_info.huge instead of global variable per Michal's comment
> > > v2 --> v1:
> > > * Adopted the suggestion from hch to return huge page size via st_blksize
> > >    instead of creating a new flag.
> > > 
> > >   mm/shmem.c | 6 ++++++
> > >   1 file changed, 6 insertions(+)
> > > 
> > > diff --git a/mm/shmem.c b/mm/shmem.c
> > > index b859192..19b8055 100644
> > > --- a/mm/shmem.c
> > > +++ b/mm/shmem.c
> > > @@ -988,6 +988,7 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
> > >   {
> > >   	struct inode *inode = path->dentry->d_inode;
> > >   	struct shmem_inode_info *info = SHMEM_I(inode);
> > > +	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
> > >   	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
> > >   		spin_lock_irq(&info->lock);
> > > @@ -995,6 +996,11 @@ static int shmem_getattr(const struct path *path, struct kstat *stat,
> > >   		spin_unlock_irq(&info->lock);
> > >   	}
> > >   	generic_fillattr(inode, stat);
> > > +#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
> > > +	if (sbinfo->huge > 0)
> > No ifdeffery, please.
> > 
> > And we probably want to check if shmem_huge is 'force'.
> > 
> > Something like this?
> > 
> > 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
> > 		 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge))
> 
> BTW, it sounds we also need check shmem_huge != SHMEM_HUGE_DENY, right?
> Otherwise it still return huge page size, but in fact no huge page can be
> allocated with 'deny'.
> 
> So, how about:
> 
> if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
> ��� ��� �(shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
> �������� shmem_huge != SHMEM_HUGE_DENY)

I think at this point it worth moving it to a helper function.

-- 
 Kirill A. Shutemov