Subject: Re: kern/21831: bridge(4) should take care of checksum offloading
To: None <tech-net@netbsd.org>
From: Matthias Scheler <tron@zhadum.de>
List: tech-net
Date: 06/09/2003 18:06:22
In article <200306091358.h59Dwv8M001020@lyssa.zhadum.de>,
	tron@zhadum.de (Matthias Scheler) writes:
> In article <1055162698.134409.936.nullmailer@yamt.dyndns.org>,
> 	yamt@mwd.biglobe.ne.jp writes:
>> 	- bridge_output() and bridge_broadcast() should do checksum
>> 	  if the interface has no h/w checksum support.
>> 	  (or don't delay checksumming for bridge member.)
> 
> The last suggestion sounds like a bad idea from a performance point of view.

To clarify this: the bridge driver should only do software checksumming
if at least one of the the physical interfaces doesn't support hardware
checksumming.

An alternative would be to ...
a) ... automatically turn off all checksumming features which are not
   supported (and enabled?) for all bridge members once a bridge
   is configured with "ifconfig".
b) ... only allow checksumming features which are supported by all
   interface to be turned on for an existing bridge interface.

	Kind regards

-- 
Matthias Scheler                                  http://scheler.de/~matthias/