Re: [PATCH 2/3 net-next] rhashtable: Use spin_lock_bh_nested() consistently

[Thomas Graf <tgraf@suug.ch>]

On 03/14/15 at 01:21pm, Herbert Xu wrote:
> On Fri, Mar 13, 2015 at 12:54:11PM -0400, David Miller wrote:
> > From: Thomas Graf <tgraf@suug.ch>
> > Date: Fri, 13 Mar 2015 15:45:20 +0100
> > 
> > > No change in behaviour as the outer lock already disables softirq
> > > but it prevents bugs down the line as this lock logically requires
> > > the BH variant.
> > > 
> > > Signed-off-by: Thomas Graf <tgraf@suug.ch>
> > 
> > I would prefer you don't do this.

OK, I'm dropping this patch.

> > x_bh() may be relatively cheap, but it is not zero cost.
> > 
> > If there is an invariant that when we are called here BH
> > is disabled, make it explicit.

I assume you are referring to the preempt disabled case. Fair enough.

> Agreed.  I dropped the _bh precisely for this reason when I did
> the arbitrary rehash.  Please don't add it back because it serves
> zero purpose.  Only the outside lock should do _bh while the
> nested one should not.

A note in the commit would have helped, it looked like an
accidental change.