[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210213202039.GC401513@shredder.lan>
Date: Sat, 13 Feb 2021 22:20:39 +0200
From: Ido Schimmel <idosch@...sch.org>
To: David Ahern <dsahern@...il.com>
Cc: Petr Machata <petrm@...dia.com>, netdev@...r.kernel.org,
David Ahern <dsahern@...nel.org>,
"David S. Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>,
Ido Schimmel <idosch@...dia.com>
Subject: Re: [RFC PATCH 04/13] nexthop: Add implementation of resilient
next-hop groups
On Sat, Feb 13, 2021 at 12:38:47PM -0700, David Ahern wrote:
> On 2/8/21 1:42 PM, Petr Machata wrote:
> > diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c
> > index 5d560d381070..4ce282b0a65f 100644
> > --- a/net/ipv4/nexthop.c
> > +++ b/net/ipv4/nexthop.c> @@ -734,6 +834,22 @@ static struct nexthop
> *nexthop_select_path_mp(struct nh_group *nhg, int hash)
> > return rc;
> > }
> >
> > +static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
> > +{
> > + struct nh_res_table *res_table = rcu_dereference(nhg->res_table);
> > + u32 bucket_index = hash % res_table->num_nh_buckets;
>
> Have you considered requiring the number of buckets to be a power of 2
> to avoid the modulo in the hot path? Seems like those are the more
> likely size options.
We thought about it, but I think it is overly limiting. Even in Spectrum
(for some sizes) it does not have to be a power of 2.
Powered by blists - more mailing lists