Re: [PATCH] netdevsim: fib: Make use of rhashtable_iter

[Ido Schimmel]

On Fri, Apr 28, 2023 at 11:11:55PM +0800, Cai Huoqing wrote:
> On 26 4月 23 17:37:49, Herbert Xu wrote:
> > Cai Huoqing <cai.huoqing@xxxxxxxxx> wrote:
> > > Iterating 'fib_rt_ht' by rhashtable_walk_next and rhashtable_iter directly
> > > instead of using list_for_each, because each entry of fib_rt_ht can be
> > > found by rhashtable API. And remove fib_rt_list.
> > > 
> > > Signed-off-by: Cai Huoqing <cai.huoqing@xxxxxxxxx>
> > > ---
> > > drivers/net/netdevsim/fib.c | 37 ++++++++++++++++++-------------------
> > > 1 file changed, 18 insertions(+), 19 deletions(-)
> > 
> > What is the rationale for this patch? Are you trying to save
> > memory?
> Hi 
> Thanks for your reply,
> 
> I think not need to use two structs to link fib_rt node, 
> fib_rt_list is redundant.

There are cases where we want to iterate over the objects without
destroying the hashtable itself.

> 
> Thanks,
> Cai-
> 
> > 
> > > @@ -1099,9 +1090,12 @@ static void nsim_fib_dump_inconsistent(struct notifier_block *nb)
> > >        /* The notifier block is still not registered, so we do not need to
> > >         * take any locks here.
> > >         */
> > > -       list_for_each_entry_safe(fib_rt, fib_rt_tmp, &data->fib_rt_list, list) {
> > > -               rhashtable_remove_fast(&data->fib_rt_ht, &fib_rt->ht_node,
> > > +       rhashtable_walk_enter(&data->fib_rt_ht, &hti);
> > > +       rhashtable_walk_start(&hti);
> > > +       while ((pos = rhashtable_walk_next(&hti))) {
> > > +               rhashtable_remove_fast(&data->fib_rt_ht, hti.p,
> > >                                       nsim_fib_rt_ht_params);
> > > +               fib_rt = rhashtable_walk_peek(&hti);
> > >                nsim_fib_rt_free(fib_rt, data);
> > >        }
> > 
> > In general rhashtable walks are not stable.  You may miss entries
> > or see entries twice.  They should be avoided unless absolutely
> > necessary.
> Agree, but how about using rhashtable_free_and_destroy here
> instead of rhashtable_walk_next in this patch.

We don't want to destroy the hashtable in this case, only free the
objects.