[PATCH] readahead: Add trace points

[PATCH] readahead: Add trace points

[Jan Kara]

Add a couple of trace points to make debugging readahead logic easier.

Signed-off-by: Jan Kara <jack@suse.cz>
---
 include/trace/events/readahead.h | 132 +++++++++++++++++++++++++++++++
 mm/readahead.c                   |   7 ++
 2 files changed, 139 insertions(+)
 create mode 100644 include/trace/events/readahead.h

I've added these when looking into how bs > page size interacts with readahead
code (and got bored with placing kprobes to dump info). I think they are useful
enough to warrant staying in the code...

diff --git a/include/trace/events/readahead.h b/include/trace/events/readahead.h
new file mode 100644
index 000000000000..992a6ce5c154
--- /dev/null
+++ b/include/trace/events/readahead.h
@@ -0,0 +1,132 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM readahead
+
+#if !defined(_TRACE_FILEMAP_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_READAHEAD_H
+
+#include <linux/types.h>
+#include <linux/tracepoint.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+
+TRACE_EVENT(do_page_cache_ra,
+	TP_PROTO(struct inode *inode, pgoff_t index, unsigned long nr_to_read,
+		 unsigned long lookahead_size),
+
+	TP_ARGS(inode, index, nr_to_read, lookahead_size),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, i_ino)
+		__field(dev_t, s_dev)
+		__field(pgoff_t, index)
+		__field(unsigned long, nr_to_read)
+		__field(unsigned long, lookahead_size)
+	),
+
+	TP_fast_assign(
+		__entry->i_ino = inode->i_ino;
+		__entry->s_dev = inode->i_sb->s_dev;
+		__entry->index = index;
+		__entry->nr_to_read = nr_to_read;
+		__entry->lookahead_size = lookahead_size;
+	),
+
+	TP_printk(
+		"dev=%d:%d ino=%lx index=%lu nr_to_read=%lu lookahead_size=%lu",
+		MAJOR(__entry->s_dev), MINOR(__entry->s_dev), __entry->i_ino,
+		__entry->index, __entry->nr_to_read, __entry->lookahead_size
+	)
+);
+
+TRACE_EVENT(page_cache_ra_order,
+	TP_PROTO(struct inode *inode, pgoff_t index, struct file_ra_state *ra),
+
+	TP_ARGS(inode, index, ra),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, i_ino)
+		__field(dev_t, s_dev)
+		__field(pgoff_t, index)
+		__field(unsigned int, order)
+		__field(unsigned int, size)
+		__field(unsigned int, async_size)
+		__field(unsigned int, ra_pages)
+	),
+
+	TP_fast_assign(
+		__entry->i_ino = inode->i_ino;
+		__entry->s_dev = inode->i_sb->s_dev;
+		__entry->index = index;
+		__entry->order = ra->order;
+		__entry->size = ra->size;
+		__entry->async_size = ra->async_size;
+		__entry->ra_pages = ra->ra_pages;
+	),
+
+	TP_printk(
+		"dev=%d:%d ino=%lx index=%lu order=%u size=%u async_size=%u ra_pages=%u",
+		MAJOR(__entry->s_dev), MINOR(__entry->s_dev), __entry->i_ino,
+		__entry->index, __entry->order, __entry->size,
+		__entry->async_size, __entry->ra_pages
+	)
+);
+
+DECLARE_EVENT_CLASS(page_cache_ra_op,
+	TP_PROTO(struct inode *inode, pgoff_t index, struct file_ra_state *ra,
+		 unsigned long req_count),
+
+	TP_ARGS(inode, index, ra, req_count),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, i_ino)
+		__field(dev_t, s_dev)
+		__field(pgoff_t, index)
+		__field(unsigned int, order)
+		__field(unsigned int, size)
+		__field(unsigned int, async_size)
+		__field(unsigned int, ra_pages)
+		__field(unsigned int, mmap_miss)
+		__field(loff_t, prev_pos)
+		__field(unsigned long, req_count)
+	),
+
+	TP_fast_assign(
+		__entry->i_ino = inode->i_ino;
+		__entry->s_dev = inode->i_sb->s_dev;
+		__entry->index = index;
+		__entry->order = ra->order;
+		__entry->size = ra->size;
+		__entry->async_size = ra->async_size;
+		__entry->ra_pages = ra->ra_pages;
+		__entry->mmap_miss = ra->mmap_miss;
+		__entry->prev_pos = ra->prev_pos;
+		__entry->req_count = req_count;
+	),
+
+	TP_printk(
+		"dev=%d:%d ino=%lx index=%lu req_count=%lu order=%u size=%u async_size=%u ra_pages=%u mmap_miss=%u prev_pos=%lld",
+		MAJOR(__entry->s_dev), MINOR(__entry->s_dev), __entry->i_ino,
+		__entry->index, __entry->req_count, __entry->order,
+		__entry->size, __entry->async_size, __entry->ra_pages,
+		__entry->mmap_miss, __entry->prev_pos
+	)
+);
+
+DEFINE_EVENT(page_cache_ra_op, page_cache_sync_ra,
+	TP_PROTO(struct inode *inode, pgoff_t index, struct file_ra_state *ra,
+		 unsigned long req_count),
+	TP_ARGS(inode, index, ra, req_count)
+);
+
+DEFINE_EVENT(page_cache_ra_op, page_cache_async_ra,
+	TP_PROTO(struct inode *inode, pgoff_t index, struct file_ra_state *ra,
+		 unsigned long req_count),
+	TP_ARGS(inode, index, ra, req_count)
+);
+
+#endif /* _TRACE_FILEMAP_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/mm/readahead.c b/mm/readahead.c
index 406756d34309..210395fe1044 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -129,6 +129,9 @@
 #include <linux/fadvise.h>
 #include <linux/sched/mm.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/readahead.h>
+
 #include "internal.h"
 
 /*
@@ -314,6 +317,7 @@ static void do_page_cache_ra(struct readahead_control *ractl,
 	loff_t isize = i_size_read(inode);
 	pgoff_t end_index;	/* The last page we want to read */
 
+	trace_do_page_cache_ra(inode, index, nr_to_read, lookahead_size);
 	if (isize == 0)
 		return;
 
@@ -470,6 +474,7 @@ void page_cache_ra_order(struct readahead_control *ractl,
 	gfp_t gfp = readahead_gfp_mask(mapping);
 	unsigned int new_order = ra->order;
 
+	trace_page_cache_ra_order(mapping->host, start, ra);
 	if (!mapping_large_folio_support(mapping)) {
 		ra->order = 0;
 		goto fallback;
@@ -554,6 +559,7 @@ void page_cache_sync_ra(struct readahead_control *ractl,
 	unsigned long max_pages, contig_count;
 	pgoff_t prev_index, miss;
 
+	trace_page_cache_sync_ra(ractl->mapping->host, index, ra, req_count);
 	/*
 	 * Even if readahead is disabled, issue this request as readahead
 	 * as we'll need it to satisfy the requested range. The forced
@@ -638,6 +644,7 @@ void page_cache_async_ra(struct readahead_control *ractl,
 	if (folio_test_writeback(folio))
 		return;
 
+	trace_page_cache_async_ra(ractl->mapping->host, index, ra, req_count);
 	folio_clear_readahead(folio);
 
 	if (blk_cgroup_congested())
-- 
2.51.0

Re: [PATCH] readahead: Add trace points

[Pankaj Raghav (Samsung)]

On Mon, Sep 08, 2025 at 04:55:34PM +0200, Jan Kara wrote:
>  
>  /*
> @@ -314,6 +317,7 @@ static void do_page_cache_ra(struct readahead_control *ractl,
>  	loff_t isize = i_size_read(inode);
>  	pgoff_t end_index;	/* The last page we want to read */
>  
> +	trace_do_page_cache_ra(inode, index, nr_to_read, lookahead_size);

Any reason why put a probe here instead of page_cache_ra_unbounded as
that is where the actual readahead happens?

-- 
Pankaj Raghav

Re: [PATCH] readahead: Add trace points

[Jan Kara]

On Tue 09-09-25 14:41:33, Pankaj Raghav (Samsung) wrote:
> On Mon, Sep 08, 2025 at 04:55:34PM +0200, Jan Kara wrote:
> >  
> >  /*
> > @@ -314,6 +317,7 @@ static void do_page_cache_ra(struct readahead_control *ractl,
> >  	loff_t isize = i_size_read(inode);
> >  	pgoff_t end_index;	/* The last page we want to read */
> >  
> > +	trace_do_page_cache_ra(inode, index, nr_to_read, lookahead_size);
> 
> Any reason why put a probe here instead of page_cache_ra_unbounded as
> that is where the actual readahead happens?

Hum, no. Originally I had it in force_page_cache_ra() but then I've decided
do_page_cache_ra() is better because it captures also other places issuing
non-standard readahead. But you're right that placing the tracepoint in
page_cache_ra_unbounded() will achieve that as well and will be a more
standard place. I'll respin the patch. Thanks for suggestion.

								Honza

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