Received-SPF: pass (google.com: domain of linux-kernel+bounces-28463-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Precedence: bulk
MIME-Version: 1.0
References: <20240116143422.1268-1-qwjhust@gmail.com> <CACOAw_zP_=w21p__X5hD8Br8tARLHozn5REA5maW8OU2M=AAhg@mail.gmail.com>
 <CACOAw_wzP2emXE_Os-eutBp9L3QRCwsDWg+qt-pQEHto_BvFUw@mail.gmail.com>
In-Reply-To: <CACOAw_wzP2emXE_Os-eutBp9L3QRCwsDWg+qt-pQEHto_BvFUw@mail.gmail.com>
From: Wenjie Qi <qwjhust@gmail.com>
Date: Wed, 17 Jan 2024 10:29:30 +0800
Message-ID: <CAGFpFsTJAJXjrkRT2n9oFHCMW-V3BgVmouR_8-GD4ZqxC2n_nw@mail.gmail.com>
Subject: Re: [f2fs-dev] [PATCH v1] f2fs: fix max open zone constraints
To: Daeho Jeong <daeho43@gmail.com>
Cc: jaegeuk@kernel.org, chao@kernel.org, 
	linux-f2fs-devel@lists.sourceforge.net, linux-kernel@vger.kernel.org, 
	hustqwj@hust.edu.cn
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Do you mean we down the semaphore when we write the first block of the zone
and up it when the last bio of the zone finishes? That seems a bit more con=
cise.

On Wed, Jan 17, 2024 at 7:12=E2=80=AFAM Daeho Jeong <daeho43@gmail.com> wro=
te:
>
> On Tue, Jan 16, 2024 at 2:58=E2=80=AFPM Daeho Jeong <daeho43@gmail.com> w=
rote:
> >
> > On Tue, Jan 16, 2024 at 6:36=E2=80=AFAM Wenjie Qi <qwjhust@gmail.com> w=
rote:
> > >
> > > We can get the number of remaining available zone
> > > resources by subtracting the number of active logs from
> > > the number of max active zones of zoned devices. We can
> > > use these available zone resources to reduce the number
> > > of pending bio when switching zones.
> > > If max active zones is 0, there is no limit.
> > >
> > > Signed-off-by: Wenjie Qi <qwjhust@gmail.com>
> > > ---
> > >  fs/f2fs/data.c  | 32 +++++++++++++++++++++++++-------
> > >  fs/f2fs/f2fs.h  |  2 ++
> > >  fs/f2fs/super.c | 11 +++++++++++
> > >  3 files changed, 38 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > > index dce8defdf4c7..0b62ca296074 100644
> > > --- a/fs/f2fs/data.c
> > > +++ b/fs/f2fs/data.c
> > > @@ -392,6 +392,19 @@ static void f2fs_zone_write_end_io(struct bio *b=
io)
> > >         complete(&io->zone_wait);
> > >         f2fs_write_end_io(bio);
> > >  }
> > > +
> > > +static void f2fs_zone_write_end_io_nowait(struct bio *bio)
> > > +{
> > > +#ifdef CONFIG_F2FS_IOSTAT
> > > +       struct bio_iostat_ctx *iostat_ctx =3D bio->bi_private;
> > > +       struct f2fs_sb_info *sbi =3D iostat_ctx->sbi;
> > > +#else
> > > +       struct f2fs_sb_info *sbi =3D (struct f2fs_sb_info *)bio->bi_p=
rivate;
> > > +#endif
> > > +
> > > +       atomic_inc(&sbi->available_active_zones);
> > > +       f2fs_write_end_io(bio);
> > > +}
> > >  #endif
> >
> > I don't think this works. Let's assume we start with 2
> > available_active_zones and 4 active logs.
> > How about 4 active logs reach at the end of the zones at the same time?
>
> Sorry, I was confused. It would work. However, it might be a little clums=
y.
> How about using a data structure like a counting semaphore?
> We can decrease it as we open a zone and increase it as we close a zone.
>
> >
> > >
> > >  struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
> > > @@ -1085,14 +1098,19 @@ void f2fs_submit_page_write(struct f2fs_io_in=
fo *fio)
> > >                 goto next;
> > >  out:
> > >  #ifdef CONFIG_BLK_DEV_ZONED
> > > -       if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
> > > +       if (f2fs_sb_has_blkzoned(sbi) && sbi->max_active_zones && bty=
pe < META &&
> > >                         is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
> > > -               bio_get(io->bio);
> > > -               reinit_completion(&io->zone_wait);
> > > -               io->bi_private =3D io->bio->bi_private;
> > > -               io->bio->bi_private =3D io;
> > > -               io->bio->bi_end_io =3D f2fs_zone_write_end_io;
> > > -               io->zone_pending_bio =3D io->bio;
> > > +               if (!atomic_add_negative(-1, &sbi->available_active_z=
ones)) {
> > > +                       io->bio->bi_end_io =3D f2fs_zone_write_end_io=
_nowait;
> > > +               } else {
> > > +                       atomic_inc(&sbi->available_active_zones);
> > > +                       bio_get(io->bio);
> > > +                       reinit_completion(&io->zone_wait);
> > > +                       io->bi_private =3D io->bio->bi_private;
> > > +                       io->bio->bi_private =3D io;
> > > +                       io->bio->bi_end_io =3D f2fs_zone_write_end_io=
;
> > > +                       io->zone_pending_bio =3D io->bio;
> > > +               }
> > >                 __submit_merged_bio(io);
> > >         }
> > >  #endif
> > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > index 65294e3b0bef..2aade367ac66 100644
> > > --- a/fs/f2fs/f2fs.h
> > > +++ b/fs/f2fs/f2fs.h
> > > @@ -1551,6 +1551,8 @@ struct f2fs_sb_info {
> > >
> > >  #ifdef CONFIG_BLK_DEV_ZONED
> > >         unsigned int blocks_per_blkz;           /* F2FS blocks per zo=
ne */
> > > +       unsigned int max_active_zones;          /* max zone resources=
 of the zoned device */
> > > +       atomic_t available_active_zones;        /* remaining zone res=
ources for zone switch */
> > >  #endif
> > >
> > >         /* for node-related operations */
> > > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > > index 206d03c82d96..6711283ff187 100644
> > > --- a/fs/f2fs/super.c
> > > +++ b/fs/f2fs/super.c
> > > @@ -2385,6 +2385,12 @@ static int f2fs_remount(struct super_block *sb=
, int *flags, char *data)
> > >         if (err)
> > >                 goto restore_opts;
> > >
> > > +#ifdef CONFIG_BLK_DEV_ZONED
> > > +       if (sbi->max_active_zones)
> > > +               atomic_set(&sbi->available_active_zones,
> > > +                               sbi->max_active_zones - F2FS_OPTION(s=
bi).active_logs);
> > > +#endif
> > > +
> > >         /* flush outstanding errors before changing fs state */
> > >         flush_work(&sbi->s_error_work);
> > >
> > > @@ -3932,6 +3938,11 @@ static int init_blkz_info(struct f2fs_sb_info =
*sbi, int devi)
> > >         if (!f2fs_sb_has_blkzoned(sbi))
> > >                 return 0;
> > >
> > > +       sbi->max_active_zones =3D bdev_max_active_zones(bdev);
> > > +       if (sbi->max_active_zones)
> > > +               atomic_set(&sbi->available_active_zones,
> > > +                               sbi->max_active_zones - F2FS_OPTION(s=
bi).active_logs);
> > > +
> >
> > Need to make available_active_zones not be negative, right?
> > Hmm, need to make sure active_logs should be equal or less than
> > max_active_zones.
> >
> > >         zone_sectors =3D bdev_zone_sectors(bdev);
> > >         if (!is_power_of_2(zone_sectors)) {
> > >                 f2fs_err(sbi, "F2FS does not support non power of 2 z=
one sizes\n");
> > > --
> > > 2.34.1
> > >
> > >
> > >
> > > _______________________________________________
> > > Linux-f2fs-devel mailing list
> > > Linux-f2fs-devel@lists.sourceforge.net
> > > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel