Re: [PATCH v2 1/2] nilfs2: fix 64-bit division operations in nilfs_bmap_find_target_in_group()

[Ryusuke Konishi]

On Sat, Mar 14, 2026 at 9:59 PM Jeff Layton wrote:
>
> On Sat, 2026-03-14 at 12:47 +0000, David Laight wrote:
> > On Fri, 13 Mar 2026 14:45:20 -0400
> > Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> >
> > > With the change to make inode->i_ino a u64, the build started failing on
> > > 32-bit ARM with:
> > >
> > >     ERROR: modpost: "__aeabi_uldivmod" [fs/nilfs2/nilfs2.ko] undefined!
> > >
> > > Fix this by using the 64-bit division interfaces in
> > > nilfs_bmap_find_target_in_group().
> > >
> > > Fixes: 998a59d371c2 ("treewide: fix missed i_ino format specifier conversions")
> > > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > Closes: https://lore.kernel.org/oe-kbuild-all/202603100602.KPxiClIO-lkp@xxxxxxxxx/
> > > Reviewed-by: Viacheslav Dubeyko <slava@xxxxxxxxxxx>
> > > Acked-by: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxx>
> > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > ---
> > >  fs/nilfs2/bmap.c | 9 ++++++---
> > >  1 file changed, 6 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
> > > index 824f2bd91c167965ec3a660202b6e6c5f1fe007e..abcf5252578ad24f694bfccf525893674bfcb4bc 100644
> > > --- a/fs/nilfs2/bmap.c
> > > +++ b/fs/nilfs2/bmap.c
> > > @@ -455,11 +455,14 @@ __u64 nilfs_bmap_find_target_in_group(const struct nilfs_bmap *bmap)
> > >  {
> > >     struct inode *dat = nilfs_bmap_get_dat(bmap);
> > >     unsigned long entries_per_group = nilfs_palloc_entries_per_group(dat);
> > > -   unsigned long group = bmap->b_inode->i_ino / entries_per_group;
> >
> > Are you sure entries_per_group can be more than 32 bits?
> > It looks like something that will be the same size on 32 and 64bit.
> >
>
> I'm not sure of anything here. I'm just want to get this to compile on
> all arches. FWIW, I'm not looking to optimize anything in this patch.
>
> > > +   unsigned long group;
> > > +   u32 index;
> > > +
> > > +   group = div_u64(bmap->b_inode->i_ino, entries_per_group);
> >
> > You don't need the full 64 by 64 divide.
> > IIRC there are both div_u64_u32() and div_u64_ulong().

Isn't the type of divisor in div_u64() u32?
Since entries_per_group cannot exceed 32 bits according to the current
specification, I think using div_u64() is fine.

> >
> > > +   div_u64_rem(bmap->b_inode->i_ino, NILFS_BMAP_GROUP_DIV, &index);
> >
> > NILFD_BMAP_GROUP_DIV is 8 (and probably has to be a power of 2).
> > So:
> >       index = bmap->b_inode->i_ino & (NILFS_BMAP_GROUP_DIV - 1);
> > is the same and likely much faster to calculate.
> > (The compiler will have done that optimisation before.)
> >
> >
>
> That all sounds reasonable to me. At this point though, it would be
> better if the NILFS2 folks stepped in with how they'd prefer this be
> done.

Yes, indeed.  It seems that the application of optimizations will
change, so this proposed correction is better.

Since NILFS_BMAP_GROUP_DIV is a fixed constant and cannot be anything
other than a power of 2, could you please adopt this proposed
correction with the following comment?

#define NILFS_BMAP_GROUP_DIV    8  /* must be a power of 2 */

Thanks,
Ryusuke Konishi

>
> >
> > >
> > >     return group * entries_per_group +
> > > -           (bmap->b_inode->i_ino % NILFS_BMAP_GROUP_DIV) *
> > > -           (entries_per_group / NILFS_BMAP_GROUP_DIV);
> > > +          index * (entries_per_group / NILFS_BMAP_GROUP_DIV);
> > >  }
> > >
> > >  static struct lock_class_key nilfs_bmap_dat_lock_key;
> > >
>
> --
> Jeff Layton <jlayton@xxxxxxxxxx>