Re: mad mad interrupts(bridging)

[Artur Grabowski <art@stacken.kth.se>]

Jake Luck <openbsd@10k.org> writes:

> trying to bridge...
> 
> on sparc4 running Dec1's -current. with a sbus quad-fast-ethernet board.
> here is the layout
> 
>                                                -+---to internet--
>                                                 |
>           +--------------------------+    +----++---+
>           |   openbsd-current sparc  |    |router   |
>           |     +------------------+ |    +--+------+
>           +--+--+-  quadfast ether +-+       |
>           le0|  +------------+-----+
>              |          hme2 |   |hme3       |
>              |               |   |           |
>              |               |   +----+------+
>              |               |
>              |               |
>              |               |         LAN
>   +----------+---------------+------------------------------------+
> 
> 
> the bridge is composed of hme2 and hme3. le0 has a real IP(so I can reach
> the box over the network from the internal LAN). the openbsd box has a
> default route to the router(i also tried without it). however as soon as I
> bind the 2 hme interface to the bridge, the box slows to a grinding halt.
> the first 10 seconds it seems to pass packets(was able to ping the router
> from another box on the LAN. but after that, network traffic goes
> berserk(i can see it on the hubs). check the openbsd box: it has a load of
> 1.4 and the cpu is loaded 99% with system interrupts(from top). I check
> the wire via arp, and all 4 interfaces from the quad card seems to get
> unique mac addresses.
> 
> would the wise here please shed some light on this? is my topology asking
> for trouble? what would be a better way? 

This is asking for trouble.
You shouldn't have one machine connected to the same ethernet segment.
Especially with sparcs this is begging for trouble, since all ethernet
interfaces on a sparc share the same mac-address.

//art