Re: [PATCH 01/10] btrfs: remove a bunch of pointles stripe_len arguments

[David Sterba <dsterba@suse.cz>]

On Wed, Jun 22, 2022 at 06:19:15AM +0200, Christoph Hellwig wrote:
> On Mon, Jun 20, 2022 at 07:38:34PM +0200, Christoph Hellwig wrote:
> > > I'd rather keep it there so it gets used eventually, we have ongoing
> > > work to fix the corner cases of raid56 so removing and adding it back
> > > causes churn, but I'll give it another thought.
> > 
> > Well, right now it is very much dead code and complicates a lot
> > of the argument passing as well as bloating the code size.
> > 
> > IFF the superblock member were to be actually used in the future,
> > it would make sense to expose it in the btrfs_fs_info and use that
> > instead of the constant, but still skip all the argument passing.
> > 
> > hch@brick:~/work/linux$ size btrfs.o.*
> >    text	   data	    bss	    dec	    hex	filename
> > 1502453	 125590	  28776	1656819	 1947f3	btrfs.o.new
> > 1502599	 125590	  28776	1656965	 194885	btrfs.o.old
> 
> So I guess this wasn't convincing enough. My plan B would be the fs_info
> member.  The related member seems to be the sectorsize field in the super
> block, which is checked if it is a power of two but otherwise completely
> ignored.  The fs_info has a stripe_size member that is initialized to the
> sectorsize after reading the super_block but then also mostly ignored.
> 
> So this is a ll a bit of a mess unfortunately :(

Hold on, it would be better to remove the code, as Qu also noted it's
not used consistently everywhere. My initial objection is more of a
general direction of such changes, but cleaning up the raid56 code to
the level of current functionality should work.