* [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
@ 2013-10-15 0:13 Ning Qu
2013-10-15 11:01 ` Kirill A. Shutemov
0 siblings, 1 reply; 6+ messages in thread
From: Ning Qu @ 2013-10-15 0:13 UTC (permalink / raw)
To: Andrea Arcangeli, Andrew Morton, Kirill A. Shutemov, Hugh Dickins
Cc: Al Viro, Hugh Dickins, Wu Fengguang, Jan Kara, Mel Gorman,
linux-mm, Andi Kleen, Matthew Wilcox, Hillf Danton, Dave Hansen,
Alexander Shishkin, linux-fsdevel, linux-kernel, Ning Qu
When comes to truncate file, add support to handle huge page in the
truncate range.
Signed-off-by: Ning Qu <quning@gmail.com>
---
mm/shmem.c | 97 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-------
1 file changed, 86 insertions(+), 11 deletions(-)
diff --git a/mm/shmem.c b/mm/shmem.c
index 0a423a9..90f2e0e 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -559,6 +559,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
struct shmem_inode_info *info = SHMEM_I(inode);
pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
+ /* Whether we have to do partial truncate */
unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
struct pagevec pvec;
@@ -570,12 +571,16 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
if (lend == -1)
end = -1; /* unsigned, so actually very big */
+ i_split_down_read(inode);
pagevec_init(&pvec, 0);
index = start;
while (index < end) {
+ bool thp = false;
+
pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
min(end - index, (pgoff_t)PAGEVEC_SIZE),
pvec.pages, indices);
+
if (!pvec.nr)
break;
mem_cgroup_uncharge_start();
@@ -586,6 +591,25 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
if (index >= end)
break;
+ thp = PageTransHugeCache(page);
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
+ if (thp) {
+ /* the range starts in middle of huge page */
+ if (index < start) {
+ partial_start = true;
+ start = index + HPAGE_CACHE_NR;
+ break;
+ }
+
+ /* the range ends on huge page */
+ if (index == (end & ~HPAGE_CACHE_INDEX_MASK)) {
+ partial_end = true;
+ end = index;
+ break;
+ }
+ }
+#endif
+
if (radix_tree_exceptional_entry(page)) {
if (unfalloc)
continue;
@@ -603,26 +627,52 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
}
}
unlock_page(page);
+ if (thp)
+ break;
}
shmem_deswap_pagevec(&pvec);
pagevec_release(&pvec);
mem_cgroup_uncharge_end();
cond_resched();
- index++;
+ if (thp)
+ index += HPAGE_CACHE_NR;
+ else
+ index++;
}
if (partial_start) {
struct page *page = NULL;
gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
+ int flags = 0;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
+ flags |= AOP_FLAG_TRANSHUGE;
+#endif
- shmem_getpage(inode, start - 1, &page, SGP_READ, gfp, 0, NULL);
+ shmem_getpage(inode, start - 1, &page, SGP_READ, gfp,
+ flags, NULL);
if (page) {
- unsigned int top = PAGE_CACHE_SIZE;
- if (start > end) {
- top = partial_end;
- partial_end = 0;
+ pgoff_t index_mask;
+ loff_t page_cache_mask;
+ unsigned pstart, pend;
+
+ index_mask = 0UL;
+ page_cache_mask = PAGE_CACHE_MASK;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
+ if (PageTransHugeCache(page)) {
+ index_mask = HPAGE_CACHE_INDEX_MASK;
+ page_cache_mask = HPAGE_PMD_MASK;
}
- zero_user_segment(page, partial_start, top);
+#endif
+
+ pstart = lstart & ~page_cache_mask;
+ if ((end & ~index_mask) == page->index) {
+ pend = (lend + 1) & ~page_cache_mask;
+ end = page->index;
+ partial_end = false; /* handled here */
+ } else
+ pend = PAGE_CACHE_SIZE << compound_order(page);
+
+ zero_pagecache_segment(page, pstart, pend);
set_page_dirty(page);
unlock_page(page);
page_cache_release(page);
@@ -631,20 +681,37 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
if (partial_end) {
struct page *page = NULL;
gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
+ int flags = 0;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
+ flags |= AOP_FLAG_TRANSHUGE;
+#endif
- shmem_getpage(inode, end, &page, SGP_READ, gfp, 0, NULL);
+ shmem_getpage(inode, end, &page, SGP_READ, gfp,
+ flags, NULL);
if (page) {
- zero_user_segment(page, 0, partial_end);
+ loff_t page_cache_mask;
+ unsigned pend;
+
+ page_cache_mask = PAGE_CACHE_MASK;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
+ if (PageTransHugeCache(page))
+ page_cache_mask = HPAGE_PMD_MASK;
+#endif
+ pend = (lend + 1) & ~page_cache_mask;
+ end = page->index;
+ zero_pagecache_segment(page, 0, pend);
set_page_dirty(page);
unlock_page(page);
page_cache_release(page);
}
}
if (start >= end)
- return;
+ goto out;
index = start;
for ( ; ; ) {
+ bool thp = false;
+
cond_resched();
pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
min(end - index, (pgoff_t)PAGEVEC_SIZE),
@@ -676,6 +743,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
continue;
}
+ thp = PageTransHugeCache(page);
lock_page(page);
if (!unfalloc || !PageUptodate(page)) {
if (page->mapping == mapping) {
@@ -684,17 +752,24 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
}
}
unlock_page(page);
+ if (thp)
+ break;
}
shmem_deswap_pagevec(&pvec);
pagevec_release(&pvec);
mem_cgroup_uncharge_end();
- index++;
+ if (thp)
+ index += HPAGE_CACHE_NR;
+ else
+ index++;
}
spin_lock(&info->lock);
info->swapped -= nr_swaps_freed;
shmem_recalc_inode(inode);
spin_unlock(&info->lock);
+out:
+ i_split_up_read(inode);
}
void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
--
1.8.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* RE: [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
2013-10-15 0:13 [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate Ning Qu
@ 2013-10-15 11:01 ` Kirill A. Shutemov
2013-10-15 18:48 ` Ning Qu
0 siblings, 1 reply; 6+ messages in thread
From: Kirill A. Shutemov @ 2013-10-15 11:01 UTC (permalink / raw)
To: Ning Qu
Cc: Andrea Arcangeli, Andrew Morton, Kirill A. Shutemov,
Hugh Dickins, Al Viro, Hugh Dickins, Wu Fengguang, Jan Kara,
Mel Gorman, linux-mm, Andi Kleen, Matthew Wilcox, Hillf Danton,
Dave Hansen, Alexander Shishkin, linux-fsdevel, linux-kernel,
Ning Qu
Ning Qu wrote:
> When comes to truncate file, add support to handle huge page in the
> truncate range.
>
> Signed-off-by: Ning Qu <quning@gmail.com>
> ---
> mm/shmem.c | 97 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-------
> 1 file changed, 86 insertions(+), 11 deletions(-)
>
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 0a423a9..90f2e0e 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -559,6 +559,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
> struct shmem_inode_info *info = SHMEM_I(inode);
> pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
> pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
> + /* Whether we have to do partial truncate */
> unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
> unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
> struct pagevec pvec;
> @@ -570,12 +571,16 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
> if (lend == -1)
> end = -1; /* unsigned, so actually very big */
>
> + i_split_down_read(inode);
> pagevec_init(&pvec, 0);
> index = start;
> while (index < end) {
> + bool thp = false;
> +
> pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
> min(end - index, (pgoff_t)PAGEVEC_SIZE),
> pvec.pages, indices);
> +
> if (!pvec.nr)
> break;
> mem_cgroup_uncharge_start();
> @@ -586,6 +591,25 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
> if (index >= end)
> break;
>
> + thp = PageTransHugeCache(page);
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
Again. Here and below ifdef is redundant: PageTransHugeCache() is zero
compile-time and thp case will be optimize out.
And do we really need a copy of truncate logic here? Is there a way to
share code?
--
Kirill A. Shutemov
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
2013-10-15 11:01 ` Kirill A. Shutemov
@ 2013-10-15 18:48 ` Ning Qu
2013-10-16 12:09 ` Kirill A. Shutemov
0 siblings, 1 reply; 6+ messages in thread
From: Ning Qu @ 2013-10-15 18:48 UTC (permalink / raw)
To: Kirill A. Shutemov
Cc: Andrea Arcangeli, Andrew Morton, Hugh Dickins, Al Viro,
Wu Fengguang, Jan Kara, Mel Gorman, linux-mm, Andi Kleen,
Matthew Wilcox, Hillf Danton, Dave Hansen, Alexander Shishkin,
linux-fsdevel, linux-kernel
Best wishes,
--
Ning Qu (曲宁) | Software Engineer | quning@google.com | +1-408-418-6066
On Tue, Oct 15, 2013 at 4:01 AM, Kirill A. Shutemov
<kirill.shutemov@linux.intel.com> wrote:
> Ning Qu wrote:
>> When comes to truncate file, add support to handle huge page in the
>> truncate range.
>>
>> Signed-off-by: Ning Qu <quning@gmail.com>
>> ---
>> mm/shmem.c | 97 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-------
>> 1 file changed, 86 insertions(+), 11 deletions(-)
>>
>> diff --git a/mm/shmem.c b/mm/shmem.c
>> index 0a423a9..90f2e0e 100644
>> --- a/mm/shmem.c
>> +++ b/mm/shmem.c
>> @@ -559,6 +559,7 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
>> struct shmem_inode_info *info = SHMEM_I(inode);
>> pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
>> pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
>> + /* Whether we have to do partial truncate */
>> unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
>> unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
>> struct pagevec pvec;
>> @@ -570,12 +571,16 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
>> if (lend == -1)
>> end = -1; /* unsigned, so actually very big */
>>
>> + i_split_down_read(inode);
>> pagevec_init(&pvec, 0);
>> index = start;
>> while (index < end) {
>> + bool thp = false;
>> +
>> pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
>> min(end - index, (pgoff_t)PAGEVEC_SIZE),
>> pvec.pages, indices);
>> +
>> if (!pvec.nr)
>> break;
>> mem_cgroup_uncharge_start();
>> @@ -586,6 +591,25 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
>> if (index >= end)
>> break;
>>
>> + thp = PageTransHugeCache(page);
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE
>
> Again. Here and below ifdef is redundant: PageTransHugeCache() is zero
> compile-time and thp case will be optimize out.
The problem is actually from HPAGE_CACHE_INDEX_MASK, it is marked as
build bug when CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE is false. So we
either wrap some logic inside a inline function, or we have to be like
this .. Or we don't treat the HPAGE_CACHE_INDEX_MASK as a build bug?
>
> And do we really need a copy of truncate logic here? Is there a way to
> share code?
>
The truncate between tmpfs and general one is similar but not exactly
the same (no readahead), so share the whole function might not be a
good choice from the perspective of tmpfs? Anyway, there are other
similar functions in tmpfs, e.g. the one you mentioned for
shmem_add_to_page_cache. It is possible to share the code, I am just
worried it will make the logic more complicated?
Maybe Hugh is in better position to judge on this? Thanks!
> --
> Kirill A. Shutemov
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
2013-10-15 18:48 ` Ning Qu
@ 2013-10-16 12:09 ` Kirill A. Shutemov
2013-10-16 18:48 ` Ning Qu
2013-10-17 20:58 ` Ning Qu
0 siblings, 2 replies; 6+ messages in thread
From: Kirill A. Shutemov @ 2013-10-16 12:09 UTC (permalink / raw)
To: Ning Qu
Cc: Kirill A. Shutemov, Andrea Arcangeli, Andrew Morton,
Hugh Dickins, Al Viro, Wu Fengguang, Jan Kara, Mel Gorman,
linux-mm, Andi Kleen, Matthew Wilcox, Hillf Danton, Dave Hansen,
Alexander Shishkin, linux-fsdevel, linux-kernel
Ning Qu wrote:
> > Again. Here and below ifdef is redundant: PageTransHugeCache() is zero
> > compile-time and thp case will be optimize out.
>
> The problem is actually from HPAGE_CACHE_INDEX_MASK, it is marked as
> build bug when CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE is false. So we
> either wrap some logic inside a inline function, or we have to be like
> this .. Or we don't treat the HPAGE_CACHE_INDEX_MASK as a build bug?
HPAGE_CACHE_INDEX_MASK shouldn't be a problem.
If it's wrapped into 'if PageTransHugeCache(page)' or similar it will be
eliminated by compiler if thp-pc disabled and build bug will not be
triggered.
>
> >
> > And do we really need a copy of truncate logic here? Is there a way to
> > share code?
> >
> The truncate between tmpfs and general one is similar but not exactly
> the same (no readahead), so share the whole function might not be a
> good choice from the perspective of tmpfs? Anyway, there are other
> similar functions in tmpfs, e.g. the one you mentioned for
> shmem_add_to_page_cache. It is possible to share the code, I am just
> worried it will make the logic more complicated?
I think introducing thp-pc is good opportunity to refactor all these code.
--
Kirill A. Shutemov
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
2013-10-16 12:09 ` Kirill A. Shutemov
@ 2013-10-16 18:48 ` Ning Qu
2013-10-17 20:58 ` Ning Qu
1 sibling, 0 replies; 6+ messages in thread
From: Ning Qu @ 2013-10-16 18:48 UTC (permalink / raw)
To: Kirill A. Shutemov
Cc: Andrea Arcangeli, Andrew Morton, Hugh Dickins, Al Viro,
Wu Fengguang, Jan Kara, Mel Gorman, linux-mm, Andi Kleen,
Matthew Wilcox, Hillf Danton, Dave Hansen, Alexander Shishkin,
linux-fsdevel, linux-kernel
Best wishes,
--
Ning Qu (曲宁) | Software Engineer | quning@google.com | +1-408-418-6066
On Wed, Oct 16, 2013 at 5:09 AM, Kirill A. Shutemov
<kirill.shutemov@linux.intel.com> wrote:
> Ning Qu wrote:
>> > Again. Here and below ifdef is redundant: PageTransHugeCache() is zero
>> > compile-time and thp case will be optimize out.
>>
>> The problem is actually from HPAGE_CACHE_INDEX_MASK, it is marked as
>> build bug when CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE is false. So we
>> either wrap some logic inside a inline function, or we have to be like
>> this .. Or we don't treat the HPAGE_CACHE_INDEX_MASK as a build bug?
>
> HPAGE_CACHE_INDEX_MASK shouldn't be a problem.
> If it's wrapped into 'if PageTransHugeCache(page)' or similar it will be
> eliminated by compiler if thp-pc disabled and build bug will not be
> triggered.
>
Yes, you are totally right about this. I have remove all the ifdef for
CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE now. Thanks!
>>
>> >
>> > And do we really need a copy of truncate logic here? Is there a way to
>> > share code?
>> >
>> The truncate between tmpfs and general one is similar but not exactly
>> the same (no readahead), so share the whole function might not be a
>> good choice from the perspective of tmpfs? Anyway, there are other
>> similar functions in tmpfs, e.g. the one you mentioned for
>> shmem_add_to_page_cache. It is possible to share the code, I am just
>> worried it will make the logic more complicated?
>
> I think introducing thp-pc is good opportunity to refactor all these code.
>
> --
> Kirill A. Shutemov
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate
2013-10-16 12:09 ` Kirill A. Shutemov
2013-10-16 18:48 ` Ning Qu
@ 2013-10-17 20:58 ` Ning Qu
1 sibling, 0 replies; 6+ messages in thread
From: Ning Qu @ 2013-10-17 20:58 UTC (permalink / raw)
To: Kirill A. Shutemov
Cc: Andrea Arcangeli, Andrew Morton, Hugh Dickins, Al Viro,
Wu Fengguang, Jan Kara, Mel Gorman, linux-mm, Andi Kleen,
Matthew Wilcox, Hillf Danton, Dave Hansen, Alexander Shishkin,
linux-fsdevel, linux-kernel
Best wishes,
--
Ning Qu (曲宁) | Software Engineer | quning@google.com | +1-408-418-6066
On Wed, Oct 16, 2013 at 5:09 AM, Kirill A. Shutemov
<kirill.shutemov@linux.intel.com> wrote:
> Ning Qu wrote:
>> > Again. Here and below ifdef is redundant: PageTransHugeCache() is zero
>> > compile-time and thp case will be optimize out.
>>
>> The problem is actually from HPAGE_CACHE_INDEX_MASK, it is marked as
>> build bug when CONFIG_TRANSPARENT_HUGEPAGE_PAGECACHE is false. So we
>> either wrap some logic inside a inline function, or we have to be like
>> this .. Or we don't treat the HPAGE_CACHE_INDEX_MASK as a build bug?
>
> HPAGE_CACHE_INDEX_MASK shouldn't be a problem.
> If it's wrapped into 'if PageTransHugeCache(page)' or similar it will be
> eliminated by compiler if thp-pc disabled and build bug will not be
> triggered.
>
>>
>> >
>> > And do we really need a copy of truncate logic here? Is there a way to
>> > share code?
>> >
>> The truncate between tmpfs and general one is similar but not exactly
>> the same (no readahead), so share the whole function might not be a
>> good choice from the perspective of tmpfs? Anyway, there are other
>> similar functions in tmpfs, e.g. the one you mentioned for
>> shmem_add_to_page_cache. It is possible to share the code, I am just
>> worried it will make the logic more complicated?
>
> I think introducing thp-pc is good opportunity to refactor all these code.
I agree, I review the code of generate truncate and shmem_undo_range again.
There are just too many differences in almost every major piece of
logic. It's really
not possible to extract any meaningful common function to share between them.
And I agree, we will try to refactor any other functions which are
possible. Thanks!
>
> --
> Kirill A. Shutemov
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2013-10-17 20:58 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-10-15 0:13 [PATCH 07/12] mm, thp, tmpfs: handle huge page in shmem_undo_range for truncate Ning Qu
2013-10-15 11:01 ` Kirill A. Shutemov
2013-10-15 18:48 ` Ning Qu
2013-10-16 12:09 ` Kirill A. Shutemov
2013-10-16 18:48 ` Ning Qu
2013-10-17 20:58 ` Ning Qu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).