On Wed, Apr 26, 2023 at 3:07 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > On Sat, 8 Apr 2023 at 18:43, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > For the common case of single lower layer, embed ovl_entry with a > > single lower path in ovl_inode, so no stack allocation is needed. > > This makes ovl_inode grow by 8 bytes, right? I can get rid of those extra bytes by stashing __numlower inside the union with the LSB set to indicate an external stack. > That's a win only in the numlower = 1 case, That's a *very* common case (i.e. non-dir without lowerdata inode) > in the other cases it's a net loss, so it might not > be worth it even without the added complexity. > I guess this is an optimization that should be justified with a benchmark. I'll drop it for now. Thanks, Amir.
