On Wed, 2015-09-23 at 10:08 -0600, David Ahern wrote:
> On 9/23/15 9:30 AM, Eric Dumazet wrote:
> >
> > You also remove a lot of goto, but do not place likely() or unlikely()
> > clauses that would help compiler to emit the same optimal code.
>
> I do not see where I have changed the likely/unlikely aspect of the
> logic. Can you point to an example in the patches?
>
Sure, I did for PATCH net-next 7/9
Copied here :
> diff --git a/net/ipv4/route.c b/net/ipv4/route.c
> index 0fdcb0539795..c23bb0965966 100644
> --- a/net/ipv4/route.c
> +++ b/net/ipv4/route.c
> @@ -1790,15 +1790,19 @@ static int ip_route_input_slow(struct sk_buff
*skb, __be32 daddr, __be32 saddr,
> if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
> goto brd_input;
>
> - if (ipv4_is_zeronet(daddr))
> - goto martian_destination;
> + if (ipv4_is_zeronet(daddr)) {
> + ip_handle_martian_dest(dev, in_dev, daddr, saddr);
> + goto out;
> + }
>
As requested, here is an example of code for which compiler might
compile in a non efficient way, ie inserting code at the wrong place,
increasing icache pressure.
Removing a goto is removing an implicit unlikely.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
