Re: [v2 PATCH 7/10] rhashtable: Disable automatic shrinking

[Thomas Graf <tgraf@suug.ch>]

On 03/23/15 at 11:07am, Herbert Xu wrote:
> On Sun, Mar 22, 2015 at 01:06:30PM +0000, Thomas Graf wrote:
> > On 03/22/15 at 12:17pm, Thomas Graf wrote:
> > > On 03/22/15 at 07:04pm, Herbert Xu wrote:
> > > > +	struct bucket_table *new_tbl;
> > > > +	struct bucket_table *tbl;
> > > > +	int err;
> > > >  
> > > > -	ASSERT_RHT_MUTEX(ht);
> > > > +	if (size < ht->p.min_size)
> > > > +		size = ht->p.min_size;
> > > 
> > > We should only shrink if size < old_tbl->size
> > 
> > I found the check further down. Any particular reason why check
> > after allocation and then free again? Why do you want to avoid
> > the allocation inside the mutex?
> 
> It's just quality of code.  You should always try to minimise
> the locked sections.

So do you expect the user to replicate the new table size calculation
outside of rhashtable_shrink() to avoid the cost of a possible massive
memory allocation even if no shrinking will take place?

I think rhashtable_shrink() should fetch ht->tbl in an RCU section to
cheaply get the current table size and only do the allocation and take
the lock if the table size warrants for shrinking.