cvs commit: src/sys/net if.h

[Darren Reed]

Jack Vogel wrote:
> On 6/11/07, Andre Oppermann <andre at freebsd.org> wrote:
>> Jack Vogel wrote:
>> > On 6/11/07, Andre Oppermann <andre at freebsd.org> wrote:
>> >> Jack Vogel wrote:
>> >> > On 6/11/07, Jack Vogel <jfvogel at gmail.com> wrote:
>> >> >> On 6/11/07, Sam Leffler <sam at errno.com> wrote:
>> >> >> > Andre Oppermann wrote:
>> >> >> > > andre       2007-06-11 20:08:12 UTC
>> >> >> > >
>> >> >> > >   FreeBSD src repository
>> >> >> > >
>> >> >> > >   Modified files:
>> >> >> > >     sys/net              if.h
>> >> >> > >   Log:
>> >> >> > >   Add IFCAP_LRO flag for drivers to announce their TCP Large
>> >> >> Receive Offload
>> >> >> > >   capabilities.
>> >> >> > >
>> >> >> > >   Revision  Changes    Path
>> >> >> > >   1.108     +1 -0      src/sys/net/if.h
>> >> >> > >
>> http://cvsweb.FreeBSD.org/src/sys/net/if.h.diff?r1=1.107&r2=1.108
>> >> >> > >
>> >> >> > >
>> >> >> > There are many offload capabilities defined that are not well
>> >> thought
>> >> >> > out.  In particular we do not distinguish between ipv4 and
>> ipv6 for
>> >> >> > things like cksum and tso so there's no way to disable individual
>> >> >> > features.  The ability to tweak LRO is clearly needed and clearly
>> >> >> belong
>> >> >> > as an ifnet capability but unilateraly deciding this is the wrong
>> >> >> approach.
>> >> >> >
>> >> >> >         Sam
>> >> >>
>> >> >> We do distinguish between TSO4 and TSO6, its just that all the
>> pieces
>> >> >> for 6 arent actually there yet, you are right about the cksum
>> >> >> granularity,
>> >> >> but I'm not convinced its needed. You might be right on the need to
>> >> >> fine tune the functionality, but having a big ON/OFF doesnt seem a
>> >> >> bad thing to me.
>> >> >
>> >> > Oh, one other useful tidbit, for Oplin when RSS is enabled you
>> can't do
>> >> > CKSUM anyway, they reuse bits.
>> >>
>> >> What is Oplin?
>> >
>> > Our new 10G adapter, er, its code name, its really 82598.
>>
>> OK, then what is this 'RSS' feature?
> 
> I think its a windowism, stands for Receive Side Scaling, and what it means
> is multiple receive queues, each with an MSI/X vector so they can interrupt
> different CPUs, or they can be tied to virtual guests, or MACs, etc etc...

To the best of my knowledge, it is a Solarisism and the 10G cards
from Sun (now licensed to Marvel?) are already delivering this.

What would be interesting is if you could tie specific rx/tx rings,
interrupts, CPUs, etc, to jails...

Darren