[PATCH RESEND 0/4] improve fadvise(POSIX_FADV_WILLNEED) with large folio

[PATCH RESEND 0/4] improve fadvise(POSIX_FADV_WILLNEED) with large folio

[Jaegeuk Kim]

This patch series aims to improve fadvise(POSIX_FADV_WILLNEED). The first patch
fixes the broken logic which was not reading the entire range ahead, and two
patches were converting the logic to adopt large folio supported by filesystem.
The last one is one example of how filesystem can set a desired folio order to
boost the readahead speed.

Jaegeuk Kim (4):
  mm/readahead: fix the broken readahead for POSIX_FADV_WILLNEED
  mm/readahead: use page_cache_sync_ra for FADVISE_FAV_WILLNEED
  mm/readahead: add a_ops->ra_folio_order to get a desired folio order
  f2fs: attach a_ops->ra_folio_order to allocate large folios for
    readahead

 fs/f2fs/data.c          |  9 +++++++++
 fs/f2fs/f2fs.h          |  3 +++
 fs/f2fs/super.c         |  1 +
 fs/f2fs/sysfs.c         |  9 +++++++++
 include/linux/fs.h      |  4 ++++
 include/linux/pagemap.h | 12 +++++++++++
 mm/readahead.c          | 44 ++++++++++++++++++++++++-----------------
 7 files changed, 64 insertions(+), 18 deletions(-)

-- 
2.52.0.107.ga0afd4fd5b-goog

[PATCH 1/4] mm/readahead: fix the broken readahead for POSIX_FADV_WILLNEED

[Jaegeuk Kim]

This patch fixes the broken readahead flow for POSIX_FADV_WILLNEED, where
the problem is, in force_page_cache_ra(nr_to_read), nr_to_read is cut by
the below code.

     max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
     nr_to_read = min_t(unsigned long, nr_to_read, max_pages);

IOWs, we are not able to read ahead larger than the above max_pages which
is most likely the range of 2MB and 16MB. Note, it doesn't make sense
to set ra->ra_pages to the entire file size. Instead, let's fix this logic.

Before:
f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
page_cache_ra_unbounded: dev=252:16 ino=e index=0 nr_to_read=512 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=512 nr_to_read=512 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=1024 nr_to_read=512 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=1536 nr_to_read=512 lookahead_size=0

After:
f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
page_cache_ra_unbounded: dev=252:16 ino=e index=0 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=2048 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=4096 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=6144 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=8192 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=10240 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=12288 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=14336 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=16384 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=18432 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=20480 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=22528 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=24576 nr_to_read=2048 lookahead_size=0
...
page_cache_ra_unbounded: dev=252:16 ino=e index=1042432 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=1044480 nr_to_read=2048 lookahead_size=0
page_cache_ra_unbounded: dev=252:16 ino=e index=1046528 nr_to_read=2048 lookahead_size=0

Cc: linux-mm@kvack.org
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 mm/readahead.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/mm/readahead.c b/mm/readahead.c
index 3a4b5d58eeb6..c0db049a5b7b 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -311,7 +311,7 @@ EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
  * behaviour which would occur if page allocations are causing VM writeback.
  * We really don't want to intermingle reads and writes like that.
  */
-static void do_page_cache_ra(struct readahead_control *ractl,
+static int do_page_cache_ra(struct readahead_control *ractl,
 		unsigned long nr_to_read, unsigned long lookahead_size)
 {
 	struct inode *inode = ractl->mapping->host;
@@ -320,45 +320,42 @@ static void do_page_cache_ra(struct readahead_control *ractl,
 	pgoff_t end_index;	/* The last page we want to read */
 
 	if (isize == 0)
-		return;
+		return -EINVAL;
 
 	end_index = (isize - 1) >> PAGE_SHIFT;
 	if (index > end_index)
-		return;
+		return -EINVAL;
 	/* Don't read past the page containing the last byte of the file */
 	if (nr_to_read > end_index - index)
 		nr_to_read = end_index - index + 1;
 
 	page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size);
+	return 0;
 }
 
 /*
- * Chunk the readahead into 2 megabyte units, so that we don't pin too much
- * memory at once.
+ * Chunk the readahead per the block device capacity, and read all nr_to_read.
  */
 void force_page_cache_ra(struct readahead_control *ractl,
 		unsigned long nr_to_read)
 {
 	struct address_space *mapping = ractl->mapping;
-	struct file_ra_state *ra = ractl->ra;
 	struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
-	unsigned long max_pages;
+	unsigned long this_chunk;
 
 	if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
 		return;
 
 	/*
-	 * If the request exceeds the readahead window, allow the read to
-	 * be up to the optimal hardware IO size
+	 * Consier the optimal hardware IO size for readahead chunk.
 	 */
-	max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
-	nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
+	this_chunk = max_t(unsigned long, bdi->io_pages, ractl->ra->ra_pages);
+
 	while (nr_to_read) {
-		unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
+		this_chunk = min_t(unsigned long, this_chunk, nr_to_read);
 
-		if (this_chunk > nr_to_read)
-			this_chunk = nr_to_read;
-		do_page_cache_ra(ractl, this_chunk, 0);
+		if (do_page_cache_ra(ractl, this_chunk, 0))
+			break;
 
 		nr_to_read -= this_chunk;
 	}
-- 
2.52.0.107.ga0afd4fd5b-goog

[PATCH 2/4] mm/readahead: use page_cache_sync_ra for FADVISE_FAV_WILLNEED

[Jaegeuk Kim]

This patch replaces page_cache_ra_unbounded() with page_cache_sync_ra() in
fadvise(FADVISE_FAV_WILLNEED) to support the large folio.

Before:
 f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
 page_cache_ra_unbounded: dev=252:16 ino=e index=0 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=2048 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=4096 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=6144 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=8192 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=10240 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=12288 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=14336 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=16384 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=18432 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=20480 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=22528 nr_to_read=2048 lookahead_size=0
 page_cache_ra_unbounded: dev=252:16 ino=e index=24576 nr_to_read=2048 lookahead_size=0
 ...
 page_cache_ra_unbounded: dev=252:16 ino=e index=1042432 nr_to_read=2048 lookahead_size=0

This is all zero-order page allocation.

After (order=0 by default):
 f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
 page_cache_sync_ra: dev=252:16 ino=e index=0 req_count=2048 order=0 size=0 async_size=0 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=0 order=0 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=2048 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=2048 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=4096 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=4096 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=6144 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=6144 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=8192 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=8192 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=10240 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=10240 nr_to_read=2048 lookahead_size=0
 ...
 page_cache_ra_unbounded: dev=252:16 ino=e index=1042432 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=1044480 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=1044480 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=1046528 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=1046528 nr_to_read=2048 lookahead_size=0

Cc: linux-mm@kvack.org
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 mm/readahead.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/mm/readahead.c b/mm/readahead.c
index c0db049a5b7b..5beaf7803554 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -340,6 +340,7 @@ void force_page_cache_ra(struct readahead_control *ractl,
 		unsigned long nr_to_read)
 {
 	struct address_space *mapping = ractl->mapping;
+	struct inode *inode = mapping->host;
 	struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
 	unsigned long this_chunk;
 
@@ -352,11 +353,19 @@ void force_page_cache_ra(struct readahead_control *ractl,
 	this_chunk = max_t(unsigned long, bdi->io_pages, ractl->ra->ra_pages);
 
 	while (nr_to_read) {
-		this_chunk = min_t(unsigned long, this_chunk, nr_to_read);
+		unsigned long index = readahead_index(ractl);
+		pgoff_t end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
 
-		if (do_page_cache_ra(ractl, this_chunk, 0))
+		if (index > end_index)
 			break;
 
+		if (nr_to_read > end_index - index)
+			nr_to_read = end_index - index + 1;
+
+		this_chunk = min_t(unsigned long, this_chunk, nr_to_read);
+
+		page_cache_sync_ra(ractl, this_chunk);
+
 		nr_to_read -= this_chunk;
 	}
 }
@@ -573,7 +582,7 @@ void page_cache_sync_ra(struct readahead_control *ractl,
 
 	/* be dumb */
 	if (do_forced_ra) {
-		force_page_cache_ra(ractl, req_count);
+		do_page_cache_ra(ractl, req_count, 0);
 		return;
 	}
 
-- 
2.52.0.107.ga0afd4fd5b-goog

[PATCH 3/4] mm/readahead: add a_ops->ra_folio_order to get a desired folio order

[Jaegeuk Kim]

This patch introduces a new address operation, a_ops->ra_folio_order(), which
proposes a new folio order based on the adjusted order for page_cache_sync_ra.

Hence, each filesystem can set the desired minimum order of folio allocation
when requesting fadvise(POSIX_FADV_WILLNEED).

Cc: linux-mm@kvack.org
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 include/linux/fs.h      |  4 ++++
 include/linux/pagemap.h | 12 ++++++++++++
 mm/readahead.c          |  6 ++++--
 3 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index c895146c1444..ddab68b7e03b 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -472,6 +472,10 @@ struct address_space_operations {
 	void (*is_dirty_writeback) (struct folio *, bool *dirty, bool *wb);
 	int (*error_remove_folio)(struct address_space *, struct folio *);
 
+	/* Min folio order to allocate pages. */
+	unsigned int (*ra_folio_order)(struct address_space *mapping,
+			unsigned int order);
+
 	/* swapfile support */
 	int (*swap_activate)(struct swap_info_struct *sis, struct file *file,
 				sector_t *span);
diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 09b581c1d878..e1fe07477220 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -476,6 +476,18 @@ mapping_min_folio_order(const struct address_space *mapping)
 	return (mapping->flags & AS_FOLIO_ORDER_MIN_MASK) >> AS_FOLIO_ORDER_MIN;
 }
 
+static inline unsigned int
+mapping_ra_folio_order(struct address_space *mapping, unsigned int order)
+{
+	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
+		return 0;
+
+	if (!mapping->a_ops->ra_folio_order)
+		return order;
+
+	return mapping->a_ops->ra_folio_order(mapping, order);
+}
+
 static inline unsigned long
 mapping_min_folio_nrpages(const struct address_space *mapping)
 {
diff --git a/mm/readahead.c b/mm/readahead.c
index 5beaf7803554..8c7d08af6e00 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -592,8 +592,10 @@ void page_cache_sync_ra(struct readahead_control *ractl,
 	 * A start of file, oversized read, or sequential cache miss:
 	 * trivial case: (index - prev_index) == 1
 	 * unaligned reads: (index - prev_index) == 0
+	 * if filesystem sets high-order allocation
 	 */
-	if (!index || req_count > max_pages || index - prev_index <= 1UL) {
+	if (!index || req_count > max_pages || index - prev_index <= 1UL ||
+	    mapping_ra_folio_order(ractl->mapping, 0)) {
 		ra->start = index;
 		ra->size = get_init_ra_size(req_count, max_pages);
 		ra->async_size = ra->size > req_count ? ra->size - req_count :
@@ -627,7 +629,7 @@ void page_cache_sync_ra(struct readahead_control *ractl,
 	ra->size = min(contig_count + req_count, max_pages);
 	ra->async_size = 1;
 readit:
-	ra->order = 0;
+	ra->order = mapping_ra_folio_order(ractl->mapping, 0);
 	ractl->_index = ra->start;
 	page_cache_ra_order(ractl, ra);
 }
-- 
2.52.0.107.ga0afd4fd5b-goog

[PATCH 4/4] f2fs: attach a_ops->ra_folio_order to allocate large folios for readahead

[Jaegeuk Kim]

This patch adds a sysfs entry to change the folio order. Given ra_folio_order=9,
we can see page_cache_ra_order getting order=9 when we submit readahead() as
below.

==== folio_order=0 ====
 f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
 page_cache_sync_ra: dev=252:16 ino=e index=0 req_count=2048 order=0 size=0 async_size=0 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=0 order=0 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=2048 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=2048 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=4096 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=4096 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=6144 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=6144 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=8192 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=8192 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=10240 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=10240 nr_to_read=2048 lookahead_size=0
 ...
 page_cache_ra_unbounded: dev=252:16 ino=e index=1042432 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=1044480 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=1044480 nr_to_read=2048 lookahead_size=0
 page_cache_sync_ra: dev=252:16 ino=e index=1046528 req_count=2048 order=0 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_unbounded: dev=252:16 ino=e index=1046528 nr_to_read=2048 lookahead_size=0

==== folio_order=9 ====
 f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
 page_cache_sync_ra: dev=252:16 ino=e index=0 req_count=2048 order=0 size=0 async_size=0 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=0 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=2048 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=2048 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=4096 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=4096 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=6144 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=6144 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=8192 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=8192 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=10240 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=10240 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=12288 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 ...
 page_cache_sync_ra: dev=252:16 ino=e index=1040384 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=1040384 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=1042432 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=1042432 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=1044480 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=1044480 order=9 size=2048 async_size=1024 ra_pages=2048
 page_cache_sync_ra: dev=252:16 ino=e index=1046528 req_count=2048 order=9 size=2048 async_size=1024 ra_pages=2048 mmap_miss=0 prev_pos=-1
 page_cache_ra_order: dev=252:16 ino=e index=1046528 order=9 size=2048 async_size=1024 ra_pages=2048

Cc: linux-mm@kvack.org
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 fs/f2fs/data.c  | 9 +++++++++
 fs/f2fs/f2fs.h  | 3 +++
 fs/f2fs/super.c | 1 +
 fs/f2fs/sysfs.c | 9 +++++++++
 4 files changed, 22 insertions(+)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 7a4f0f2d60cf..addef5a1fdb1 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -3995,6 +3995,14 @@ static bool f2fs_dirty_data_folio(struct address_space *mapping,
 	return false;
 }
 
+static unsigned int f2fs_ra_folio_order(struct address_space *mapping,
+				     unsigned int order)
+{
+	if (!mapping_large_folio_support(mapping))
+		return order;
+
+	return max(order, F2FS_M_SB(mapping)->ra_folio_order);
+}
 
 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
 {
@@ -4313,6 +4321,7 @@ const struct address_space_operations f2fs_dblock_aops = {
 	.dirty_folio	= f2fs_dirty_data_folio,
 	.migrate_folio	= filemap_migrate_folio,
 	.invalidate_folio = f2fs_invalidate_folio,
+	.ra_folio_order	= f2fs_ra_folio_order,
 	.release_folio	= f2fs_release_folio,
 	.bmap		= f2fs_bmap,
 	.swap_activate  = f2fs_swap_activate,
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index d7600979218e..06f90d510a01 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -1932,6 +1932,9 @@ struct f2fs_sb_info {
 	/* carve out reserved_blocks from total blocks */
 	bool carve_out;
 
+	/* enable large folio for readahead. */
+	unsigned int ra_folio_order;
+
 #ifdef CONFIG_F2FS_FS_COMPRESSION
 	struct kmem_cache *page_array_slab;	/* page array entry */
 	unsigned int page_array_slab_size;	/* default page array slab size */
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index ccb477086444..bae02ca96c1f 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -4287,6 +4287,7 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
 			NAT_ENTRY_PER_BLOCK));
 	sbi->allocate_section_hint = le32_to_cpu(raw_super->section_count);
 	sbi->allocate_section_policy = ALLOCATE_FORWARD_NOHINT;
+	sbi->ra_folio_order = 0;
 	F2FS_ROOT_INO(sbi) = le32_to_cpu(raw_super->root_ino);
 	F2FS_NODE_INO(sbi) = le32_to_cpu(raw_super->node_ino);
 	F2FS_META_INO(sbi) = le32_to_cpu(raw_super->meta_ino);
diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
index c42f4f979d13..2537a25986a6 100644
--- a/fs/f2fs/sysfs.c
+++ b/fs/f2fs/sysfs.c
@@ -906,6 +906,13 @@ static ssize_t __sbi_store(struct f2fs_attr *a,
 		return count;
 	}
 
+	if (!strcmp(a->attr.name, "ra_folio_order")) {
+		if (t < 0 || t > MAX_PAGECACHE_ORDER)
+			return -EINVAL;
+		sbi->ra_folio_order = t;
+		return count;
+	}
+
 	*ui = (unsigned int)t;
 
 	return count;
@@ -1180,6 +1187,7 @@ F2FS_SBI_GENERAL_RW_ATTR(migration_window_granularity);
 F2FS_SBI_GENERAL_RW_ATTR(dir_level);
 F2FS_SBI_GENERAL_RW_ATTR(allocate_section_hint);
 F2FS_SBI_GENERAL_RW_ATTR(allocate_section_policy);
+F2FS_SBI_GENERAL_RW_ATTR(ra_folio_order);
 #ifdef CONFIG_F2FS_IOSTAT
 F2FS_SBI_GENERAL_RW_ATTR(iostat_enable);
 F2FS_SBI_GENERAL_RW_ATTR(iostat_period_ms);
@@ -1422,6 +1430,7 @@ static struct attribute *f2fs_attrs[] = {
 	ATTR_LIST(reserved_pin_section),
 	ATTR_LIST(allocate_section_hint),
 	ATTR_LIST(allocate_section_policy),
+	ATTR_LIST(ra_folio_order),
 	NULL,
 };
 ATTRIBUTE_GROUPS(f2fs);
-- 
2.52.0.107.ga0afd4fd5b-goog

Re: [PATCH 3/4] mm/readahead: add a_ops->ra_folio_order to get a desired folio order

[Matthew Wilcox]

On Mon, Dec 01, 2025 at 09:01:26PM +0000, Jaegeuk Kim wrote:
> This patch introduces a new address operation, a_ops->ra_folio_order(), which
> proposes a new folio order based on the adjusted order for page_cache_sync_ra.
> 
> Hence, each filesystem can set the desired minimum order of folio allocation
> when requesting fadvise(POSIX_FADV_WILLNEED).

Again, you've said what but not why.  Does the mm code not ramp up the
folio order sufficiently quickly?  What are you trying to accomplish?

Re: [PATCH 3/4] mm/readahead: add a_ops->ra_folio_order to get a desired folio order

[Jaegeuk Kim]

On 12/01, Matthew Wilcox wrote:
> On Mon, Dec 01, 2025 at 09:01:26PM +0000, Jaegeuk Kim wrote:
> > This patch introduces a new address operation, a_ops->ra_folio_order(), which
> > proposes a new folio order based on the adjusted order for page_cache_sync_ra.
> > 
> > Hence, each filesystem can set the desired minimum order of folio allocation
> > when requesting fadvise(POSIX_FADV_WILLNEED).
> 
> Again, you've said what but not why.  Does the mm code not ramp up the
> folio order sufficiently quickly?  What are you trying to accomplish?

That's why I posted a series of the patches to provide more details. Could you
please check the last patch in the series to show fadvise() does not increase
the folio order?

https://lore.kernel.org/linux-f2fs-devel/20251201210152.909339-5-jaegeuk@kernel.org/T/#u

Re: [f2fs-dev] [PATCH 1/4] mm/readahead: fix the broken readahead for POSIX_FADV_WILLNEED

[Chao Yu]

On 2025/12/2 05:01, Jaegeuk Kim via Linux-f2fs-devel wrote:
> This patch fixes the broken readahead flow for POSIX_FADV_WILLNEED, where
> the problem is, in force_page_cache_ra(nr_to_read), nr_to_read is cut by
> the below code.
> 
>       max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
>       nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
> 
> IOWs, we are not able to read ahead larger than the above max_pages which
> is most likely the range of 2MB and 16MB. Note, it doesn't make sense
> to set ra->ra_pages to the entire file size. Instead, let's fix this logic.
> 
> Before:
> f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
> page_cache_ra_unbounded: dev=252:16 ino=e index=0 nr_to_read=512 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=512 nr_to_read=512 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=1024 nr_to_read=512 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=1536 nr_to_read=512 lookahead_size=0
> 
> After:
> f2fs_fadvise: dev = (252,16), ino = 14, i_size = 4294967296 offset:0, len:4294967296, advise:3
> page_cache_ra_unbounded: dev=252:16 ino=e index=0 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=2048 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=4096 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=6144 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=8192 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=10240 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=12288 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=14336 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=16384 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=18432 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=20480 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=22528 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=24576 nr_to_read=2048 lookahead_size=0
> ...
> page_cache_ra_unbounded: dev=252:16 ino=e index=1042432 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=1044480 nr_to_read=2048 lookahead_size=0
> page_cache_ra_unbounded: dev=252:16 ino=e index=1046528 nr_to_read=2048 lookahead_size=0
> 
> Cc: linux-mm@kvack.org
> Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>   mm/readahead.c | 27 ++++++++++++---------------
>   1 file changed, 12 insertions(+), 15 deletions(-)
> 
> diff --git a/mm/readahead.c b/mm/readahead.c
> index 3a4b5d58eeb6..c0db049a5b7b 100644
> --- a/mm/readahead.c
> +++ b/mm/readahead.c
> @@ -311,7 +311,7 @@ EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
>    * behaviour which would occur if page allocations are causing VM writeback.
>    * We really don't want to intermingle reads and writes like that.
>    */
> -static void do_page_cache_ra(struct readahead_control *ractl,
> +static int do_page_cache_ra(struct readahead_control *ractl,
>   		unsigned long nr_to_read, unsigned long lookahead_size)
>   {
>   	struct inode *inode = ractl->mapping->host;
> @@ -320,45 +320,42 @@ static void do_page_cache_ra(struct readahead_control *ractl,
>   	pgoff_t end_index;	/* The last page we want to read */
>   
>   	if (isize == 0)
> -		return;
> +		return -EINVAL;
>   
>   	end_index = (isize - 1) >> PAGE_SHIFT;
>   	if (index > end_index)
> -		return;
> +		return -EINVAL;
>   	/* Don't read past the page containing the last byte of the file */
>   	if (nr_to_read > end_index - index)
>   		nr_to_read = end_index - index + 1;
>   
>   	page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size);
> +	return 0;
>   }
>   
>   /*
> - * Chunk the readahead into 2 megabyte units, so that we don't pin too much
> - * memory at once.
> + * Chunk the readahead per the block device capacity, and read all nr_to_read.
>    */
>   void force_page_cache_ra(struct readahead_control *ractl,
>   		unsigned long nr_to_read)
>   {
>   	struct address_space *mapping = ractl->mapping;
> -	struct file_ra_state *ra = ractl->ra;
>   	struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
> -	unsigned long max_pages;
> +	unsigned long this_chunk;
>   
>   	if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
>   		return;
>   
>   	/*
> -	 * If the request exceeds the readahead window, allow the read to
> -	 * be up to the optimal hardware IO size
> +	 * Consier the optimal hardware IO size for readahead chunk.

s/Consier/Consider

Thanks,

>   	 */
> -	max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages);
> -	nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
> +	this_chunk = max_t(unsigned long, bdi->io_pages, ractl->ra->ra_pages);
> +
>   	while (nr_to_read) {
> -		unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE;
> +		this_chunk = min_t(unsigned long, this_chunk, nr_to_read);
>   
> -		if (this_chunk > nr_to_read)
> -			this_chunk = nr_to_read;
> -		do_page_cache_ra(ractl, this_chunk, 0);
> +		if (do_page_cache_ra(ractl, this_chunk, 0))
> +			break;
>   
>   		nr_to_read -= this_chunk;
>   	}

Re: [PATCH 3/4] mm/readahead: add a_ops->ra_folio_order to get a desired folio order

[Matthew Wilcox]

On Mon, Dec 01, 2025 at 09:24:41PM +0000, Jaegeuk Kim wrote:
> On 12/01, Matthew Wilcox wrote:
> > On Mon, Dec 01, 2025 at 09:01:26PM +0000, Jaegeuk Kim wrote:
> > > This patch introduces a new address operation, a_ops->ra_folio_order(), which
> > > proposes a new folio order based on the adjusted order for page_cache_sync_ra.
> > > 
> > > Hence, each filesystem can set the desired minimum order of folio allocation
> > > when requesting fadvise(POSIX_FADV_WILLNEED).
> > 
> > Again, you've said what but not why.  Does the mm code not ramp up the
> > folio order sufficiently quickly?  What are you trying to accomplish?
> 
> That's why I posted a series of the patches to provide more details. Could you
> please check the last patch in the series to show fadvise() does not increase
> the folio order?
> 
> https://lore.kernel.org/linux-f2fs-devel/20251201210152.909339-5-jaegeuk@kernel.org/T/#u

So what you're trying to say is that readahead() currently only uses
order-0 pages and you want it to use larger order pages?  I agree with
that!  But I don't think this is the way to do it.  We should just use
larger order allocations, always.  None of this "call the filesystem,
check a sysfs parameter".  Just use the largest order page that fits.

Re: [PATCH 3/4] mm/readahead: add a_ops->ra_folio_order to get a desired folio order

[Jaegeuk Kim]

On 12/01, Matthew Wilcox wrote:
> On Mon, Dec 01, 2025 at 09:24:41PM +0000, Jaegeuk Kim wrote:
> > On 12/01, Matthew Wilcox wrote:
> > > On Mon, Dec 01, 2025 at 09:01:26PM +0000, Jaegeuk Kim wrote:
> > > > This patch introduces a new address operation, a_ops->ra_folio_order(), which
> > > > proposes a new folio order based on the adjusted order for page_cache_sync_ra.
> > > > 
> > > > Hence, each filesystem can set the desired minimum order of folio allocation
> > > > when requesting fadvise(POSIX_FADV_WILLNEED).
> > > 
> > > Again, you've said what but not why.  Does the mm code not ramp up the
> > > folio order sufficiently quickly?  What are you trying to accomplish?
> > 
> > That's why I posted a series of the patches to provide more details. Could you
> > please check the last patch in the series to show fadvise() does not increase
> > the folio order?
> > 
> > https://lore.kernel.org/linux-f2fs-devel/20251201210152.909339-5-jaegeuk@kernel.org/T/#u
> 
> So what you're trying to say is that readahead() currently only uses
> order-0 pages and you want it to use larger order pages?  I agree with
> that!  But I don't think this is the way to do it.  We should just use
> larger order allocations, always.  None of this "call the filesystem,
> check a sysfs parameter".  Just use the largest order page that fits.

I got that, and posted v2.

Thanks,