Re: [PATCH] exfat: optimize allocation bitmap loading time

[Namjae Jeon <linkinjeon@kernel.org>]

On Mon, Aug 4, 2025 at 5:57 PM Yuezhang.Mo@sony.com
<Yuezhang.Mo@sony.com> wrote:
>
> > On Fri, Aug 1, 2025 at 5:03 PM Yuezhang.Mo@sony.com
> > <Yuezhang.Mo@sony.com> wrote:
> > >
> > > > Loading the allocation bitmap is very slow if user set the small cluster
> > > > size on large partition.
> > > >
> > > > For optimizing it, This patch uses sb_breadahead() read the allocation
> > > > bitmap. It will improve the mount time.
> > > >
> > > > The following is the result of about 4TB partition(2KB cluster size)
> > > > on my target.
> > > >
> > > > without patch:
> > > > real 0m41.746s
> > > > user 0m0.011s
> > > > sys 0m0.000s
> > > >
> > > > with patch:
> > > > real 0m2.525s
> > > > user 0m0.008s
> > > > sys 0m0.008s
> > > >
> > > > Reviewed-by: Sungjong Seo <sj1557.seo@samsung.com>
> > > > Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
> > > > ---
> > > >  fs/exfat/balloc.c   | 12 +++++++++++-
> > > >  fs/exfat/dir.c      |  1 -
> > > >  fs/exfat/exfat_fs.h |  1 +
> > > >  3 files changed, 12 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/fs/exfat/balloc.c b/fs/exfat/balloc.c
> > > > index cc01556c9d9b..c40b73701941 100644
> > > > --- a/fs/exfat/balloc.c
> > > > +++ b/fs/exfat/balloc.c
> > > > @@ -30,9 +30,11 @@ static int exfat_allocate_bitmap(struct super_block *sb,
> > > >                 struct exfat_dentry *ep)
> > > >  {
> > > >         struct exfat_sb_info *sbi = EXFAT_SB(sb);
> > > > +       struct blk_plug plug;
> > > >         long long map_size;
> > > > -       unsigned int i, need_map_size;
> > > > +       unsigned int i, j, need_map_size;
> > > >         sector_t sector;
> > > > +       unsigned int max_ra_count = EXFAT_MAX_RA_SIZE >> sb->s_blocksize_bits;
> > > >
> > > >         sbi->map_clu = le32_to_cpu(ep->dentry.bitmap.start_clu);
> > > >         map_size = le64_to_cpu(ep->dentry.bitmap.size);
> > > > @@ -57,6 +59,14 @@ static int exfat_allocate_bitmap(struct super_block *sb,
> > > >
> > > >         sector = exfat_cluster_to_sector(sbi, sbi->map_clu);
> > > >         for (i = 0; i < sbi->map_sectors; i++) {
> > > > +               /* Trigger the next readahead in advance. */
> > > > +               if (0 == (i % max_ra_count)) {
> > > > +                       blk_start_plug(&plug);
> > > > +                       for (j = i; j < min(max_ra_count, sbi->map_sectors - i) + i; j++)
> > > > +                               sb_breadahead(sb, sector + j);
> > > > +                       blk_finish_plug(&plug);
> > > > +               }
> > > > +
> > > >                 sbi->vol_amap[i] = sb_bread(sb, sector + i);
> > > >                 if (!sbi->vol_amap[i]) {
> > > >                         /* release all buffers and free vol_amap */
> > > > diff --git a/fs/exfat/dir.c b/fs/exfat/dir.c
> > > > index ee060e26f51d..e7a8550c0346 100644
> > > > --- a/fs/exfat/dir.c
> > > > +++ b/fs/exfat/dir.c
> > > > @@ -616,7 +616,6 @@ static int exfat_find_location(struct super_block *sb, struct exfat_chain *p_dir
> > > >         return 0;
> > > >  }
> > > >
> > > > -#define EXFAT_MAX_RA_SIZE     (128*1024)
> > > >  static int exfat_dir_readahead(struct super_block *sb, sector_t sec)
> > > >  {
> > > >         struct exfat_sb_info *sbi = EXFAT_SB(sb);
> > > > diff --git a/fs/exfat/exfat_fs.h b/fs/exfat/exfat_fs.h
> > > > index f8ead4d47ef0..d1792d5c9eed 100644
> > > > --- a/fs/exfat/exfat_fs.h
> > > > +++ b/fs/exfat/exfat_fs.h
> > > > @@ -13,6 +13,7 @@
> > > >  #include <uapi/linux/exfat.h>
> > > >
> > > >  #define EXFAT_ROOT_INO         1
> > > > +#define EXFAT_MAX_RA_SIZE     (128*1024)
> > >
> > > Why is the max readahead size 128KiB?
> > > If the limit is changed to max_sectors_kb, so that a read request reads as much
> > > data as possible, will the performance be better?
> > This sets an appropriate readahead size for exfat. It's already used
> > elsewhere in exfat.
> > Getting ->max_sectors_kb from the block layer will result in a layer violation.
>
> I checked the code of read ahead, EXFAT_MAX_RA_SIZE is consistent with the
> default value(VM_READAHEAD_PAGES) of sb->s_bdi->ra_pages.
>
> Is it better to use sb->s_bdi->ra_pages instead?
> If so, users can set different values via 'blockdev --setra'.
I will change max_ra_count as follows:
max_ra_count = min(sb->s_bdi->ra_pages, sb->s_bdi->io_pages)
                << (PAGE_SHIFT - sb->s_blocksize_bits);
Thanks.