Subject: Re: HW-assisted VLAN handling
To: David Laight <david@l8s.co.uk>
From: Jonathan Stone <jonathan@dsg.stanford.edu>
List: tech-kern
Date: 01/30/2005 14:39:09
In message <20050130220424.GF1271@snowdrop.l8s.co.uk>,
David Laight writes:

>Most chipsets don't have exact multicast filters.  They tend to work
>by using a hash to one of 64 buckets (using the crc).

Many old chips did this. Some (most?) new chips still have this as an
intermediate point between ``accept all multicast'' and exact-match
MAC filters.

>Some modern chips may support multiple local addresses - which could
>be set to multicast addresses....

I dunno about you, but Most datasheets I've seen recently have between
4 and 16 exact-match MAC filters.  and your assertion about common
multicast use may reflect sites or subnets that only use local-subnet;
but it conflicts *wildly* with, say, use of mbone multicast tools to
watch a conference.

But let's try to keep focused on the point under discussion.  What
should our API to hardware VLAN tagging look like?  What should the
kernel do, on NIC hardware that can filter a small number of VLAN tags
in hardware ---say 16 tags, as per another message--- when the
sysadmin has enabled 64 distinct VLANs?  Or 1024, or 4096?