[PATCH v2] page cache: Store only head pages in i_pages

[PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

Transparent Huge Pages are currently stored in i_pages as pointers to
consecutive subpages.  This patch changes that to storing consecutive
pointers to the head page in preparation for storing huge pages more
efficiently in i_pages.

Large parts of this are "inspired" by Kirill's patch
https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/

Signed-off-by: Matthew Wilcox <willy@infradead.org>
---
v2: Rebase on top of linux-next 20190212
    Fixed a missing s/head/page/ in filemap_map_pages
    Include missing calls to xas_store() in __split_huge_page

 include/linux/pagemap.h |  9 ++++
 mm/filemap.c            | 99 +++++++++++++----------------------------
 mm/huge_memory.c        |  3 ++
 mm/khugepaged.c         |  4 +-
 mm/shmem.c              |  2 +-
 mm/swap_state.c         |  2 +-
 6 files changed, 46 insertions(+), 73 deletions(-)

diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index bcf909d0de5f..7d58e4e0b68e 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -333,6 +333,15 @@ static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
 			mapping_gfp_mask(mapping));
 }
 
+static inline struct page *find_subpage(struct page *page, pgoff_t offset)
+{
+	VM_BUG_ON_PAGE(PageTail(page), page);
+	VM_BUG_ON_PAGE(page->index > offset, page);
+	VM_BUG_ON_PAGE(page->index + (1 << compound_order(page)) <= offset,
+			page);
+	return page - page->index + offset;
+}
+
 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
diff --git a/mm/filemap.c b/mm/filemap.c
index 5673672fd444..ee28028c4323 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -1491,7 +1491,7 @@ EXPORT_SYMBOL(page_cache_prev_miss);
 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
 {
 	XA_STATE(xas, &mapping->i_pages, offset);
-	struct page *head, *page;
+	struct page *page;
 
 	rcu_read_lock();
 repeat:
@@ -1506,25 +1506,19 @@ struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
 	if (!page || xa_is_value(page))
 		goto out;
 
-	head = compound_head(page);
-	if (!page_cache_get_speculative(head))
-		goto repeat;
-
-	/* The page was split under us? */
-	if (compound_head(page) != head) {
-		put_page(head);
+	if (!page_cache_get_speculative(page))
 		goto repeat;
-	}
 
 	/*
-	 * Has the page moved?
+	 * Has the page moved or been split?
 	 * This is part of the lockless pagecache protocol. See
 	 * include/linux/pagemap.h for details.
 	 */
 	if (unlikely(page != xas_reload(&xas))) {
-		put_page(head);
+		put_page(page);
 		goto repeat;
 	}
+	page = find_subpage(page, offset);
 out:
 	rcu_read_unlock();
 
@@ -1706,7 +1700,6 @@ unsigned find_get_entries(struct address_space *mapping,
 
 	rcu_read_lock();
 	xas_for_each(&xas, page, ULONG_MAX) {
-		struct page *head;
 		if (xas_retry(&xas, page))
 			continue;
 		/*
@@ -1717,17 +1710,13 @@ unsigned find_get_entries(struct address_space *mapping,
 		if (xa_is_value(page))
 			goto export;
 
-		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto retry;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto put_page;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto put_page;
+		page = find_subpage(page, xas.xa_index);
 
 export:
 		indices[ret] = xas.xa_index;
@@ -1736,7 +1725,7 @@ unsigned find_get_entries(struct address_space *mapping,
 			break;
 		continue;
 put_page:
-		put_page(head);
+		put_page(page);
 retry:
 		xas_reset(&xas);
 	}
@@ -1778,33 +1767,27 @@ unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
 
 	rcu_read_lock();
 	xas_for_each(&xas, page, end) {
-		struct page *head;
 		if (xas_retry(&xas, page))
 			continue;
 		/* Skip over shadow, swap and DAX entries */
 		if (xa_is_value(page))
 			continue;
 
-		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto retry;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto put_page;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto put_page;
 
-		pages[ret] = page;
+		pages[ret] = find_subpage(page, xas.xa_index);
 		if (++ret == nr_pages) {
 			*start = page->index + 1;
 			goto out;
 		}
 		continue;
 put_page:
-		put_page(head);
+		put_page(page);
 retry:
 		xas_reset(&xas);
 	}
@@ -1849,7 +1832,6 @@ unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
 
 	rcu_read_lock();
 	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
-		struct page *head;
 		if (xas_retry(&xas, page))
 			continue;
 		/*
@@ -1859,24 +1841,19 @@ unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
 		if (xa_is_value(page))
 			break;
 
-		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto retry;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto put_page;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto put_page;
 
-		pages[ret] = page;
+		pages[ret] = find_subpage(page, xas.xa_index);
 		if (++ret == nr_pages)
 			break;
 		continue;
 put_page:
-		put_page(head);
+		put_page(page);
 retry:
 		xas_reset(&xas);
 	}
@@ -1912,7 +1889,6 @@ unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
 
 	rcu_read_lock();
 	xas_for_each_marked(&xas, page, end, tag) {
-		struct page *head;
 		if (xas_retry(&xas, page))
 			continue;
 		/*
@@ -1923,26 +1899,21 @@ unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
 		if (xa_is_value(page))
 			continue;
 
-		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto retry;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto put_page;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto put_page;
 
-		pages[ret] = page;
+		pages[ret] = find_subpage(page, xas.xa_index);
 		if (++ret == nr_pages) {
 			*index = page->index + 1;
 			goto out;
 		}
 		continue;
 put_page:
-		put_page(head);
+		put_page(page);
 retry:
 		xas_reset(&xas);
 	}
@@ -1991,7 +1962,6 @@ unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
 
 	rcu_read_lock();
 	xas_for_each_marked(&xas, page, ULONG_MAX, tag) {
-		struct page *head;
 		if (xas_retry(&xas, page))
 			continue;
 		/*
@@ -2002,17 +1972,13 @@ unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
 		if (xa_is_value(page))
 			goto export;
 
-		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto retry;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto put_page;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto put_page;
+		page = find_subpage(page, xas.xa_index);
 
 export:
 		indices[ret] = xas.xa_index;
@@ -2021,7 +1987,7 @@ unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
 			break;
 		continue;
 put_page:
-		put_page(head);
+		put_page(page);
 retry:
 		xas_reset(&xas);
 	}
@@ -2686,7 +2652,7 @@ void filemap_map_pages(struct vm_fault *vmf,
 	pgoff_t last_pgoff = start_pgoff;
 	unsigned long max_idx;
 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
-	struct page *head, *page;
+	struct page *page;
 
 	rcu_read_lock();
 	xas_for_each(&xas, page, end_pgoff) {
@@ -2695,24 +2661,19 @@ void filemap_map_pages(struct vm_fault *vmf,
 		if (xa_is_value(page))
 			goto next;
 
-		head = compound_head(page);
-
 		/*
 		 * Check for a locked page first, as a speculative
 		 * reference may adversely influence page migration.
 		 */
-		if (PageLocked(head))
+		if (PageLocked(page))
 			goto next;
-		if (!page_cache_get_speculative(head))
+		if (!page_cache_get_speculative(page))
 			goto next;
 
-		/* The page was split under us? */
-		if (compound_head(page) != head)
-			goto skip;
-
-		/* Has the page moved? */
+		/* Has the page moved or been split? */
 		if (unlikely(page != xas_reload(&xas)))
 			goto skip;
+		page = find_subpage(page, xas.xa_index);
 
 		if (!PageUptodate(page) ||
 				PageReadahead(page) ||
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index d4847026d4b1..7008174c033b 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2458,6 +2458,9 @@ static void __split_huge_page(struct page *page, struct list_head *list,
 			if (IS_ENABLED(CONFIG_SHMEM) && PageSwapBacked(head))
 				shmem_uncharge(head->mapping->host, 1);
 			put_page(head + i);
+		} else if (!PageAnon(page)) {
+			__xa_store(&head->mapping->i_pages, head[i].index,
+					head + i, 0);
 		}
 	}
 
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 449044378782..7ba7a1e4fa79 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1374,7 +1374,7 @@ static void collapse_shmem(struct mm_struct *mm,
 				result = SCAN_FAIL;
 				goto xa_locked;
 			}
-			xas_store(&xas, new_page + (index % HPAGE_PMD_NR));
+			xas_store(&xas, new_page);
 			nr_none++;
 			continue;
 		}
@@ -1450,7 +1450,7 @@ static void collapse_shmem(struct mm_struct *mm,
 		list_add_tail(&page->lru, &pagelist);
 
 		/* Finally, replace with the new page. */
-		xas_store(&xas, new_page + (index % HPAGE_PMD_NR));
+		xas_store(&xas, new_page);
 		continue;
 out_unlock:
 		unlock_page(page);
diff --git a/mm/shmem.c b/mm/shmem.c
index c8cdaa012f18..a78d4f05a51f 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -614,7 +614,7 @@ static int shmem_add_to_page_cache(struct page *page,
 		if (xas_error(&xas))
 			goto unlock;
 next:
-		xas_store(&xas, page + i);
+		xas_store(&xas, page);
 		if (++i < nr) {
 			xas_next(&xas);
 			goto next;
diff --git a/mm/swap_state.c b/mm/swap_state.c
index 85245fdec8d9..c5da342b5dba 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -132,7 +132,7 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp)
 		for (i = 0; i < nr; i++) {
 			VM_BUG_ON_PAGE(xas.xa_index != idx + i, page);
 			set_page_private(page + i, entry.val + i);
-			xas_store(&xas, page + i);
+			xas_store(&xas, page);
 			xas_next(&xas);
 		}
 		address_space->nrpages += nr;
-- 
2.20.1

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Jan Kara]

On Tue 12-02-19 10:34:54, Matthew Wilcox wrote:
> Transparent Huge Pages are currently stored in i_pages as pointers to
> consecutive subpages.  This patch changes that to storing consecutive
> pointers to the head page in preparation for storing huge pages more
> efficiently in i_pages.
> 
> Large parts of this are "inspired" by Kirill's patch
> https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> 
> Signed-off-by: Matthew Wilcox <willy@infradead.org>

I like the idea!

> @@ -1778,33 +1767,27 @@ unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
>  
>  	rcu_read_lock();
>  	xas_for_each(&xas, page, end) {
> -		struct page *head;
>  		if (xas_retry(&xas, page))
>  			continue;
>  		/* Skip over shadow, swap and DAX entries */
>  		if (xa_is_value(page))
>  			continue;
>  
> -		head = compound_head(page);
> -		if (!page_cache_get_speculative(head))
> +		if (!page_cache_get_speculative(page))
>  			goto retry;
>  
> -		/* The page was split under us? */
> -		if (compound_head(page) != head)
> -			goto put_page;
> -
> -		/* Has the page moved? */
> +		/* Has the page moved or been split? */
>  		if (unlikely(page != xas_reload(&xas)))
>  			goto put_page;
>  
> -		pages[ret] = page;
> +		pages[ret] = find_subpage(page, xas.xa_index);
>  		if (++ret == nr_pages) {
>  			*start = page->index + 1;
>  			goto out;
>  		}

So this subtly changes the behavior because now we will be returning in
'*start' a different index. So you should rather use 'pages[ret]->index'
instead.

> @@ -1923,26 +1899,21 @@ unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
>  		if (xa_is_value(page))
>  			continue;
>  
> -		head = compound_head(page);
> -		if (!page_cache_get_speculative(head))
> +		if (!page_cache_get_speculative(page))
>  			goto retry;
>  
> -		/* The page was split under us? */
> -		if (compound_head(page) != head)
> -			goto put_page;
> -
> -		/* Has the page moved? */
> +		/* Has the page moved or been split? */
>  		if (unlikely(page != xas_reload(&xas)))
>  			goto put_page;
>  
> -		pages[ret] = page;
> +		pages[ret] = find_subpage(page, xas.xa_index);
>  		if (++ret == nr_pages) {
>  			*index = page->index + 1;
>  			goto out;
>  		}

Ditto here.

Otherwise the patch looks good to me so feel free to add:

Acked-by: Jan Kara <jack@suse.cz>

after fixing these two.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Wed, Feb 13, 2019 at 03:41:02PM +0100, Jan Kara wrote:
> On Tue 12-02-19 10:34:54, Matthew Wilcox wrote:
> > Transparent Huge Pages are currently stored in i_pages as pointers to
> > consecutive subpages.  This patch changes that to storing consecutive
> > pointers to the head page in preparation for storing huge pages more
> > efficiently in i_pages.
> > 
> > Large parts of this are "inspired" by Kirill's patch
> > https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> > 
> > Signed-off-by: Matthew Wilcox <willy@infradead.org>
> 
> I like the idea!

Thanks!  It's a step towards reducing the overhead of huge pages in the
page cache from (on x86) 520 pointers to a mere 8.  Still not as good as
hugetlbfs, but I'm working on that too.

> > -		pages[ret] = page;
> > +		pages[ret] = find_subpage(page, xas.xa_index);
> >  		if (++ret == nr_pages) {
> >  			*start = page->index + 1;
> >  			goto out;
> >  		}
> 
> So this subtly changes the behavior because now we will be returning in
> '*start' a different index. So you should rather use 'pages[ret]->index'
> instead.

You're right, I made a mistake there.  However, seeing this:
https://lore.kernel.org/lkml/20190110030838.84446-1-yuzhao@google.com/

makes me think that I should be using xa_index + 1 there.

> Otherwise the patch looks good to me so feel free to add:
> 
> Acked-by: Jan Kara <jack@suse.cz>
> 
> after fixing these two.

Thanks!

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Jan Kara]

On Wed 13-02-19 12:17:15, Matthew Wilcox wrote:
> > > -		pages[ret] = page;
> > > +		pages[ret] = find_subpage(page, xas.xa_index);
> > >  		if (++ret == nr_pages) {
> > >  			*start = page->index + 1;
> > >  			goto out;
> > >  		}
> > 
> > So this subtly changes the behavior because now we will be returning in
> > '*start' a different index. So you should rather use 'pages[ret]->index'
> > instead.
> 
> You're right, I made a mistake there.  However, seeing this:
> https://lore.kernel.org/lkml/20190110030838.84446-1-yuzhao@google.com/
> 
> makes me think that I should be using xa_index + 1 there.

Yeah, you're right. Thanks!

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Tue, Feb 12, 2019 at 10:34:54AM -0800, Matthew Wilcox wrote:
> Transparent Huge Pages are currently stored in i_pages as pointers to
> consecutive subpages.  This patch changes that to storing consecutive
> pointers to the head page in preparation for storing huge pages more
> efficiently in i_pages.
> 
> Large parts of this are "inspired" by Kirill's patch
> https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> 
> Signed-off-by: Matthew Wilcox <willy@infradead.org>

I believe I found few missing pieces:

 - page_cache_delete_batch() will blow up on

			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
					!= pvec->pages[i]->index, page);

 - migrate_page_move_mapping() has to be converted too.

The rest *looks* fine to me. But it needs a lot of testing.

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
>  - page_cache_delete_batch() will blow up on
> 
> 			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> 					!= pvec->pages[i]->index, page);

Quite right.  I decided to rewrite page_cache_delete_batch.  What do you
(and Jan!) think to this?  Compile-tested only.

diff --git a/mm/filemap.c b/mm/filemap.c
index 0d71b1acf811..facaa6913ffa 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -279,11 +279,11 @@ EXPORT_SYMBOL(delete_from_page_cache);
  * @pvec: pagevec with pages to delete
  *
  * The function walks over mapping->i_pages and removes pages passed in @pvec
- * from the mapping. The function expects @pvec to be sorted by page index.
+ * from the mapping. The function expects @pvec to be sorted by page index
+ * and is optimised for it to be dense.
  * It tolerates holes in @pvec (mapping entries at those indices are not
  * modified). The function expects only THP head pages to be present in the
- * @pvec and takes care to delete all corresponding tail pages from the
- * mapping as well.
+ * @pvec.
  *
  * The function expects the i_pages lock to be held.
  */
@@ -292,40 +292,36 @@ static void page_cache_delete_batch(struct address_space *mapping,
 {
 	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
 	int total_pages = 0;
-	int i = 0, tail_pages = 0;
+	int i = 0;
 	struct page *page;
 
 	mapping_set_update(&xas, mapping);
 	xas_for_each(&xas, page, ULONG_MAX) {
-		if (i >= pagevec_count(pvec) && !tail_pages)
+		if (i >= pagevec_count(pvec))
 			break;
+
+		/* A swap/dax/shadow entry got inserted? Skip it. */
 		if (xa_is_value(page))
 			continue;
-		if (!tail_pages) {
-			/*
-			 * Some page got inserted in our range? Skip it. We
-			 * have our pages locked so they are protected from
-			 * being removed.
-			 */
-			if (page != pvec->pages[i]) {
-				VM_BUG_ON_PAGE(page->index >
-						pvec->pages[i]->index, page);
-				continue;
-			}
-			WARN_ON_ONCE(!PageLocked(page));
-			if (PageTransHuge(page) && !PageHuge(page))
-				tail_pages = HPAGE_PMD_NR - 1;
+		/*
+		 * A page got inserted in our range? Skip it. We have our
+		 * pages locked so they are protected from being removed.
+		 */
+		if (page != pvec->pages[i]) {
+			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
+					page);
+			continue;
+		}
+
+		WARN_ON_ONCE(!PageLocked(page));
+
+		if (page->index == xas.xa_index)
 			page->mapping = NULL;
-			/*
-			 * Leave page->index set: truncation lookup relies
-			 * upon it
-			 */
+		/* Leave page->index set: truncation lookup relies on it */
+
+		if (page->index + (1UL << compound_order(page)) - 1 ==
+				xas.xa_index)
 			i++;
-		} else {
-			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
-					!= pvec->pages[i]->index, page);
-			tail_pages--;
-		}
 		xas_store(&xas, NULL);
 		total_pages++;
 	}

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Thu, Feb 14, 2019 at 12:53:31PM -0800, Matthew Wilcox wrote:
> On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> >  - page_cache_delete_batch() will blow up on
> > 
> > 			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> > 					!= pvec->pages[i]->index, page);
> 
> Quite right.  I decided to rewrite page_cache_delete_batch.  What do you
> (and Jan!) think to this?  Compile-tested only.
> 
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 0d71b1acf811..facaa6913ffa 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -279,11 +279,11 @@ EXPORT_SYMBOL(delete_from_page_cache);
>   * @pvec: pagevec with pages to delete
>   *
>   * The function walks over mapping->i_pages and removes pages passed in @pvec
> - * from the mapping. The function expects @pvec to be sorted by page index.
> + * from the mapping. The function expects @pvec to be sorted by page index
> + * and is optimised for it to be dense.
>   * It tolerates holes in @pvec (mapping entries at those indices are not
>   * modified). The function expects only THP head pages to be present in the
> - * @pvec and takes care to delete all corresponding tail pages from the
> - * mapping as well.
> + * @pvec.
>   *
>   * The function expects the i_pages lock to be held.
>   */
> @@ -292,40 +292,36 @@ static void page_cache_delete_batch(struct address_space *mapping,
>  {
>  	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
>  	int total_pages = 0;
> -	int i = 0, tail_pages = 0;
> +	int i = 0;
>  	struct page *page;
>  
>  	mapping_set_update(&xas, mapping);
>  	xas_for_each(&xas, page, ULONG_MAX) {
> -		if (i >= pagevec_count(pvec) && !tail_pages)
> +		if (i >= pagevec_count(pvec))
>  			break;
> +
> +		/* A swap/dax/shadow entry got inserted? Skip it. */
>  		if (xa_is_value(page))
>  			continue;
> -		if (!tail_pages) {
> -			/*
> -			 * Some page got inserted in our range? Skip it. We
> -			 * have our pages locked so they are protected from
> -			 * being removed.
> -			 */
> -			if (page != pvec->pages[i]) {
> -				VM_BUG_ON_PAGE(page->index >
> -						pvec->pages[i]->index, page);
> -				continue;
> -			}
> -			WARN_ON_ONCE(!PageLocked(page));
> -			if (PageTransHuge(page) && !PageHuge(page))
> -				tail_pages = HPAGE_PMD_NR - 1;
> +		/*
> +		 * A page got inserted in our range? Skip it. We have our
> +		 * pages locked so they are protected from being removed.
> +		 */
> +		if (page != pvec->pages[i]) {

Maybe a comment for the VM_BUG while you're there?

> +			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
> +					page);
> +			continue;
> +		}
> +
> +		WARN_ON_ONCE(!PageLocked(page));
> +
> +		if (page->index == xas.xa_index)
>  			page->mapping = NULL;
> -			/*
> -			 * Leave page->index set: truncation lookup relies
> -			 * upon it
> -			 */
> +		/* Leave page->index set: truncation lookup relies on it */
> +
> +		if (page->index + (1UL << compound_order(page)) - 1 ==
> +				xas.xa_index)

It's 1am here and I'm slow, but it took me few minutes to understand how
it works. Please add a comment.

>  			i++;
> -		} else {
> -			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> -					!= pvec->pages[i]->index, page);
> -			tail_pages--;
> -		}
>  		xas_store(&xas, NULL);
>  		total_pages++;
>  	}

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Fri, Feb 15, 2019 at 01:03:44AM +0300, Kirill A. Shutemov wrote:
> > +		/*
> > +		 * A page got inserted in our range? Skip it. We have our
> > +		 * pages locked so they are protected from being removed.
> > +		 */
> > +		if (page != pvec->pages[i]) {
> 
> Maybe a comment for the VM_BUG while you're there?

Great idea.  I didn't understand it the first time I looked at it either,
but I forgot to write a comment when I figured it out.

                /*
                 * A page got inserted in our range? Skip it. We have our
                 * pages locked so they are protected from being removed.
+                * If we see a page whose index is higher than ours, it
+                * means our page has been removed, which shouldn't be
+                * possible because we're holding the PageLock.
                 */
                if (page != pvec->pages[i]) {

> > +		/* Leave page->index set: truncation lookup relies on it */
> > +
> > +		if (page->index + (1UL << compound_order(page)) - 1 ==
> > +				xas.xa_index)
> 
> It's 1am here and I'm slow, but it took me few minutes to understand how
> it works. Please add a comment.

I should get you to review at 1am more often!  You're quite right.
Sleep-deprived Kirill spots problems that normal people would encounter.


                /* Leave page->index set: truncation lookup relies on it */
 
+               /*
+                * Move to the next page in the vector if this is a small page
+                * or the index is of the last page in this compound page).
+                */
                if (page->index + (1UL << compound_order(page)) - 1 ==
                                xas.xa_index)
                        i++;


Thanks for the review.

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
>  - migrate_page_move_mapping() has to be converted too.

I think that's as simple as:

+++ b/mm/migrate.c
@@ -465,7 +465,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
 
                for (i = 1; i < HPAGE_PMD_NR; i++) {
                        xas_next(&xas);
-                       xas_store(&xas, newpage + i);
+                       xas_store(&xas, newpage);
                }
        }
 

or do you see something else I missed?

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Thu, Feb 14, 2019 at 01:17:57PM -0800, Matthew Wilcox wrote:
> On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> >  - migrate_page_move_mapping() has to be converted too.
> 
> I think that's as simple as:
> 
> +++ b/mm/migrate.c
> @@ -465,7 +465,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
>  
>                 for (i = 1; i < HPAGE_PMD_NR; i++) {
>                         xas_next(&xas);
> -                       xas_store(&xas, newpage + i);
> +                       xas_store(&xas, newpage);
>                 }
>         }
>  
> 
> or do you see something else I missed?

Looks right to me.

BTW, maybe some add syntax sugar from XArray side?

Replace the loop and xas_store() before it with:

		xas_fill(&xas, newpage, 1UL << compound_order(newpage));

or something similar?

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Fri, Feb 15, 2019 at 01:08:10AM +0300, Kirill A. Shutemov wrote:
> On Thu, Feb 14, 2019 at 01:17:57PM -0800, Matthew Wilcox wrote:
> > On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> > >  - migrate_page_move_mapping() has to be converted too.
> > 
> > I think that's as simple as:
> > 
> > +++ b/mm/migrate.c
> > @@ -465,7 +465,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
> >  
> >                 for (i = 1; i < HPAGE_PMD_NR; i++) {
> >                         xas_next(&xas);
> > -                       xas_store(&xas, newpage + i);
> > +                       xas_store(&xas, newpage);
> >                 }
> >         }
> >  
> > 
> > or do you see something else I missed?
> 
> Looks right to me.
> 
> BTW, maybe some add syntax sugar from XArray side?
> 
> Replace the loop and xas_store() before it with:
> 
> 		xas_fill(&xas, newpage, 1UL << compound_order(newpage));
> 
> or something similar?

If we were keeping this code longterm, then yes, something like that
would be great.  I'm hoping this code is a mere stepping stone towards
using multi-slot entries for the page cache.

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Thu, Feb 14, 2019 at 02:11:58PM -0800, Matthew Wilcox wrote:
> On Fri, Feb 15, 2019 at 01:08:10AM +0300, Kirill A. Shutemov wrote:
> > On Thu, Feb 14, 2019 at 01:17:57PM -0800, Matthew Wilcox wrote:
> > > On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> > > >  - migrate_page_move_mapping() has to be converted too.
> > > 
> > > I think that's as simple as:
> > > 
> > > +++ b/mm/migrate.c
> > > @@ -465,7 +465,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
> > >  
> > >                 for (i = 1; i < HPAGE_PMD_NR; i++) {
> > >                         xas_next(&xas);
> > > -                       xas_store(&xas, newpage + i);
> > > +                       xas_store(&xas, newpage);
> > >                 }
> > >         }
> > >  
> > > 
> > > or do you see something else I missed?
> > 
> > Looks right to me.
> > 
> > BTW, maybe some add syntax sugar from XArray side?
> > 
> > Replace the loop and xas_store() before it with:
> > 
> > 		xas_fill(&xas, newpage, 1UL << compound_order(newpage));
> > 
> > or something similar?
> 
> If we were keeping this code longterm, then yes, something like that
> would be great.  I'm hoping this code is a mere stepping stone towards
> using multi-slot entries for the page cache.

Fair enough.

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> On Tue, Feb 12, 2019 at 10:34:54AM -0800, Matthew Wilcox wrote:
> > Transparent Huge Pages are currently stored in i_pages as pointers to
> > consecutive subpages.  This patch changes that to storing consecutive
> > pointers to the head page in preparation for storing huge pages more
> > efficiently in i_pages.
> > 
> > Large parts of this are "inspired" by Kirill's patch
> > https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> > 
> > Signed-off-by: Matthew Wilcox <willy@infradead.org>
> 
> I believe I found few missing pieces:
> 
>  - page_cache_delete_batch() will blow up on
> 
> 			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> 					!= pvec->pages[i]->index, page);
> 
>  - migrate_page_move_mapping() has to be converted too.

Other missing pieces are memfd_wait_for_pins() and memfd_tag_pins()
We need to call page_mapcount() for tail pages there.

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Fri, Feb 15, 2019 at 01:29:44AM +0300, Kirill A. Shutemov wrote:
> On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> > On Tue, Feb 12, 2019 at 10:34:54AM -0800, Matthew Wilcox wrote:
> > > Transparent Huge Pages are currently stored in i_pages as pointers to
> > > consecutive subpages.  This patch changes that to storing consecutive
> > > pointers to the head page in preparation for storing huge pages more
> > > efficiently in i_pages.
> > > 
> > > Large parts of this are "inspired" by Kirill's patch
> > > https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> > > 
> > > Signed-off-by: Matthew Wilcox <willy@infradead.org>
> > 
> > I believe I found few missing pieces:
> > 
> >  - page_cache_delete_batch() will blow up on
> > 
> > 			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> > 					!= pvec->pages[i]->index, page);
> > 
> >  - migrate_page_move_mapping() has to be converted too.
> 
> Other missing pieces are memfd_wait_for_pins() and memfd_tag_pins()
> We need to call page_mapcount() for tail pages there.

@@ -39,6 +39,7 @@ static void memfd_tag_pins(struct xa_state *xas)
        xas_for_each(xas, page, ULONG_MAX) {
                if (xa_is_value(page))
                        continue;
+               page = find_subpage(page, xas.xa_index);
                if (page_count(page) - page_mapcount(page) > 1)
                        xas_set_mark(xas, MEMFD_TAG_PINNED);
 
@@ -88,6 +89,7 @@ static int memfd_wait_for_pins(struct address_space *mapping)
                        bool clear = true;
                        if (xa_is_value(page))
                                continue;
+                       page = find_subpage(page, xas.xa_index);
                        if (page_count(page) - page_mapcount(page) != 1) {
                                /*
                                 * On the last scan, we clean up all those tags

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Thu, Feb 14, 2019 at 04:30:04PM +0300, Kirill A. Shutemov wrote:
> On Tue, Feb 12, 2019 at 10:34:54AM -0800, Matthew Wilcox wrote:
> > Transparent Huge Pages are currently stored in i_pages as pointers to
> > consecutive subpages.  This patch changes that to storing consecutive
> > pointers to the head page in preparation for storing huge pages more
> > efficiently in i_pages.
> > 
> > Large parts of this are "inspired" by Kirill's patch
> > https://lore.kernel.org/lkml/20170126115819.58875-2-kirill.shutemov@linux.intel.com/
> > 
> > Signed-off-by: Matthew Wilcox <willy@infradead.org>
> 
> I believe I found few missing pieces:
> 
>  - page_cache_delete_batch() will blow up on
> 
> 			VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
> 					!= pvec->pages[i]->index, page);
> 
>  - migrate_page_move_mapping() has to be converted too.

  - __delete_from_swap_cache() will blow up on

	VM_BUG_ON_PAGE(entry != page + i, entry);

-- 
 Kirill A. Shutemov

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Matthew Wilcox]

On Fri, Feb 15, 2019 at 01:41:15AM +0300, Kirill A. Shutemov wrote:
>   - __delete_from_swap_cache() will blow up on
> 
> 	VM_BUG_ON_PAGE(entry != page + i, entry);

Right.

@@ -167,7 +167,7 @@ void __delete_from_swap_cache(struct page *page, swp_entry_t entry)
 
        for (i = 0; i < nr; i++) {
                void *entry = xas_store(&xas, NULL);
-               VM_BUG_ON_PAGE(entry != page + i, entry);
+               VM_BUG_ON_PAGE(entry != page, entry);
                set_page_private(page + i, 0);
                xas_next(&xas);
        }

Re: [PATCH v2] page cache: Store only head pages in i_pages

[Kirill A. Shutemov]

On Fri, Feb 15, 2019 at 12:20:57PM -0800, Matthew Wilcox wrote:
> On Fri, Feb 15, 2019 at 01:41:15AM +0300, Kirill A. Shutemov wrote:
> >   - __delete_from_swap_cache() will blow up on
> > 
> > 	VM_BUG_ON_PAGE(entry != page + i, entry);
> 
> Right.

Thanks. I think this is the last one I found.

Could you send v3 with all fixups applied?

-- 
 Kirill A. Shutemov