svn commit: r292309 - in head/sys: modules modules/tcp modules/tcp/fastpath netinet netinet/tcp_stacks

[Navdeep Parhar]

On Wed, Dec 16, 2015 at 07:06:31AM -0800, John Baldwin wrote:
> On Wednesday, December 16, 2015 12:56:45 AM Randall Stewart wrote:
> > Author: rrs
> > Date: Wed Dec 16 00:56:45 2015
> > New Revision: 292309
> > URL: https://svnweb.freebsd.org/changeset/base/292309
> > 
> > Log:
> >   First cut of the modularization of our TCP stack. Still
> >   to do is to clean up the timer handling using the async-drain.
> >   Other optimizations may be coming to go with this. Whats here
> >   will allow differnet tcp implementations (one included).
> >   Reviewed by:	jtl, hiren, transports
> >   Sponsored by:	Netflix Inc.
> >   Differential Revision:	D4055
> 
> Have you considered treating TOE as a separate stack?  We already
> have a bit of a split to handle TOE specially.  It might be nice if
> it was able to plug in via this.  That might also help flesh out the
> abstraction a bit by giving another use case.

That's an interesting thought.  The TOE code should be a great test for
any KPI that aims to allow multiple TCP stacks to coexist (TOEs are
nothing but alternate TCP stacks running in the hardware).  The only
oddball thing about TOE is that the offload vs don't offload decision is
made on the fly (based on the route lookup) and not when the socket is
created.

Regards,
Navdeep

> 
> Also, note that the Differential Revision tag should be the full
> URL to the review (in which case it auto-closes it for you).
> 
> -- 
> John Baldwin
>