On 1/10/18 4:47 AM, Ido Schimmel wrote:
> Hi David,
> 
> On Tue, Jan 09, 2018 at 08:48:37PM -0700, David Ahern wrote:
>> On 1/9/18 7:40 AM, Ido Schimmel wrote:
>>> The use of hash-threshold instead of modulo-N makes it trivial to add
>>> support for non-equal-cost multipath.
>>>
>>> Instead of dividing the multipath hash function's output space equally
>>> between the nexthops, each nexthop is assigned a region size which is
>>> proportional to its weight.
>>>
>>> Signed-off-by: Ido Schimmel <ido...@mellanox.com>
>>> ---
>>>  include/net/ip6_fib.h |  1 +
>>>  net/ipv6/route.c      | 11 +++++++----
>>>  2 files changed, 8 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/include/net/ip6_fib.h b/include/net/ip6_fib.h
>>> index 97cd05d87780..34ec321d6a03 100644
>>> --- a/include/net/ip6_fib.h
>>> +++ b/include/net/ip6_fib.h
>>> @@ -171,6 +171,7 @@ struct rt6_info {
>>>     u32                             rt6i_metric;
>>>     u32                             rt6i_pmtu;
>>>     /* more non-fragment space at head required */
>>> +   int                             rt6i_nh_weight;
>>>     unsigned short                  rt6i_nfheader_len;
>>>     u8                              rt6i_protocol;
>>>     u8                              exception_bucket_flushed:1,
>>
>> Since dst is cacheline aligned there is a hole after rt6i_nh_flags. In
>> patch 1 you put rt6i_nh_upper_bound in that hole. Putting the weight
>> there too keeps those variables together as well as using the open space.
> 
> Before patch 1 there's a hole of 4 bytes after rt6i_nh_flags which I use
> for rt6i_nh_upper_bound. If I put rt6i_nh_weight there as well, then I
> create a 60 bytes hole because the dst needs to be cached aligned.
> 
> Since rt6i_nh_weight isn't used in fast-path, I just put it at the end.
> 

Apparently, I was on a 4.14 branch when I ran pahole to dump the layout
of rt6_info. The patch looks good to me.

Acked-by: David Ahern <dsah...@gmail.com>

Reply via email to