Re: [PATCH] simplify b_inode usage

[Andrew Morton]

Christoph Hellwig wrote:
> 
> ...
> > Also, Joe Thornber needs to add another pointer to struct buffer_head
> > for LVM2 reasons.  If we collapse b_inode into a b_flags bit then
> > Joe gets his pointer for free (bh stays at 48 bytes on ia32).
> 
> We also need to make b_size an 32 bit quantity, otherwise 64k buffers
> on architectures like ia64 will get us horrible overflows.  And yes,
> people use that big pages - Nathan just added an ugly workaround to XFS,
> for splitting 64k pages into multiple bh, because of that exactly that
> limitation.

OK.  We'd need to prevent that from taking bh beyond the current 96 bytes.
The hashed wakeups, perhaps.

> >
> > So I'd suggest you just go ahead and do it that way.  (I had a patch
> > for that but seem to have misplaced it).
> 
> As the patch is already large enough I'd be happy if Marcelo applies the
> current patch, once it's in I'll move the indicator to b_flags, okay?

Yup.