2021-12-14 18:24:47

by Jaegeuk Kim

Subject: [PATCH 1/2] f2fs: avoid down_write on nat_tree_lock during checkpoint

Let's cache nat entry if there's no lock contention only.

Signed-off-by: Jaegeuk Kim <[email protected]>
---
fs/f2fs/node.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index 556fcd8457f3..b1bc7d76da3b 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -430,6 +430,10 @@ static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
struct f2fs_nm_info *nm_i = NM_I(sbi);
struct nat_entry *new, *e;

+ /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
+ if (rwsem_is_locked(&sbi->cp_global_sem))
+ return;
+
new = __alloc_nat_entry(sbi, nid, false);
if (!new)
return;
--
2.34.1.173.g76aa8bc2d0-goog

2021-12-19 08:54:06

by Chao Yu

Subject: Re: [f2fs-dev] [PATCH 1/2] f2fs: avoid down_write on nat_tree_lock during checkpoint

On 2021/12/15 2:24, Jaegeuk Kim wrote:
> Let's cache nat entry if there's no lock contention only.
>
> Signed-off-by: Jaegeuk Kim <[email protected]>
> ---
> fs/f2fs/node.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index 556fcd8457f3..b1bc7d76da3b 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -430,6 +430,10 @@ static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
> struct f2fs_nm_info *nm_i = NM_I(sbi);
> struct nat_entry *new, *e;
>
> + /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
> + if (rwsem_is_locked(&sbi->cp_global_sem))

Why not down_write_trylock(nat_tree_lock)? cp_global_sem lock coverage is larger than
nat_tree_lock's in f2fs_write_checkpoint().

Thanks,

> + return;
> +
> new = __alloc_nat_entry(sbi, nid, false);
> if (!new)
> return;

2021-12-20 22:16:04

by Jaegeuk Kim

Subject: Re: [f2fs-dev] [PATCH 1/2] f2fs: avoid down_write on nat_tree_lock during checkpoint

On 12/19, Chao Yu wrote:
> On 2021/12/15 2:24, Jaegeuk Kim wrote:
> > Let's cache nat entry if there's no lock contention only.
> >
> > Signed-off-by: Jaegeuk Kim <[email protected]>
> > ---
> > fs/f2fs/node.c | 4 ++++
> > 1 file changed, 4 insertions(+)
> >
> > diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> > index 556fcd8457f3..b1bc7d76da3b 100644
> > --- a/fs/f2fs/node.c
> > +++ b/fs/f2fs/node.c
> > @@ -430,6 +430,10 @@ static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
> > struct f2fs_nm_info *nm_i = NM_I(sbi);
> > struct nat_entry *new, *e;
> > + /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
> > + if (rwsem_is_locked(&sbi->cp_global_sem))
>
> Why not down_write_trylock(nat_tree_lock)? cp_global_sem lock coverage is larger than
> nat_tree_lock's in f2fs_write_checkpoint().

I'm focusing on faster checkpoint by minimizing the nat_tree_lock holding time.

>
> Thanks,
>
> > + return;
> > +
> > new = __alloc_nat_entry(sbi, nid, false);
> > if (!new)
> > return;

2021-12-30 06:36:17

by Chao Yu

Subject: Re: [f2fs-dev] [PATCH 1/2] f2fs: avoid down_write on nat_tree_lock during checkpoint

On 2021/12/15 2:24, Jaegeuk Kim wrote:
> Let's cache nat entry if there's no lock contention only.
>
> Signed-off-by: Jaegeuk Kim <[email protected]>

Reviewed-by: Chao Yu <[email protected]>

Thanks,