2023-06-30 09:29:21

by Andrew Yang

Subject: [PATCH] fs: drop_caches: draining pages before dropping caches

We expect a file page access after dropping caches should be a major
fault, but sometimes it's still a minor fault. That's because a file
page can't be dropped if it's in a per-cpu pagevec. Draining all pages
from per-cpu pagevec to lru list before trying to drop caches.

Signed-off-by: Andrew Yang <[email protected]>
---
fs/drop_caches.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/fs/drop_caches.c b/fs/drop_caches.c
index e619c31b6bd9..b9575957a7c2 100644
--- a/fs/drop_caches.c
+++ b/fs/drop_caches.c
@@ -10,6 +10,7 @@
#include <linux/writeback.h>
#include <linux/sysctl.h>
#include <linux/gfp.h>
+#include <linux/swap.h>
#include "internal.h"

/* A global variable is a bit ugly, but it keeps the code simple */
@@ -59,6 +60,7 @@ int drop_caches_sysctl_handler(struct ctl_table *table, int write,
static int stfu;

if (sysctl_drop_caches & 1) {
+ lru_add_drain_all();
iterate_supers(drop_pagecache_sb, NULL);
count_vm_event(DROP_PAGECACHE);
}
--
2.18.0

2023-07-03 10:31:17

by Christian Brauner

Subject: Re: [PATCH] fs: drop_caches: draining pages before dropping caches

Really more suitable for -mm to review.

On Fri, Jun 30, 2023 at 05:22:02PM +0800, Andrew Yang wrote:
> We expect a file page access after dropping caches should be a major
> fault, but sometimes it's still a minor fault. That's because a file
> page can't be dropped if it's in a per-cpu pagevec. Draining all pages
> from per-cpu pagevec to lru list before trying to drop caches.
>
> Signed-off-by: Andrew Yang <[email protected]>
> ---
> fs/drop_caches.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/fs/drop_caches.c b/fs/drop_caches.c
> index e619c31b6bd9..b9575957a7c2 100644
> --- a/fs/drop_caches.c
> +++ b/fs/drop_caches.c
> @@ -10,6 +10,7 @@
> #include <linux/writeback.h>
> #include <linux/sysctl.h>
> #include <linux/gfp.h>
> +#include <linux/swap.h>
> #include "internal.h"
>
> /* A global variable is a bit ugly, but it keeps the code simple */
> @@ -59,6 +60,7 @@ int drop_caches_sysctl_handler(struct ctl_table *table, int write,
> static int stfu;
>
> if (sysctl_drop_caches & 1) {
> + lru_add_drain_all();
> iterate_supers(drop_pagecache_sb, NULL);
> count_vm_event(DROP_PAGECACHE);
> }
> --
> 2.18.0
>

2023-07-04 18:28:37

by Andrew Morton

Subject: Re: [PATCH] fs: drop_caches: draining pages before dropping caches

On Fri, 30 Jun 2023 17:22:02 +0800 Andrew Yang <[email protected]> wrote:

> We expect a file page access after dropping caches should be a major
> fault, but sometimes it's still a minor fault. That's because a file
> page can't be dropped if it's in a per-cpu pagevec. Draining all pages
> from per-cpu pagevec to lru list before trying to drop caches.
>
> --- a/fs/drop_caches.c
> +++ b/fs/drop_caches.c
> @@ -10,6 +10,7 @@
> #include <linux/writeback.h>
> #include <linux/sysctl.h>
> #include <linux/gfp.h>
> +#include <linux/swap.h>
> #include "internal.h"
>
> /* A global variable is a bit ugly, but it keeps the code simple */
> @@ -59,6 +60,7 @@ int drop_caches_sysctl_handler(struct ctl_table *table, int write,
> static int stfu;
>
> if (sysctl_drop_caches & 1) {
> + lru_add_drain_all();
> iterate_supers(drop_pagecache_sb, NULL);
> count_vm_event(DROP_PAGECACHE);
> }

um, yes. I'm suprised that this oversight has survived 20+ years.