Re: Query: PCI and Ethernet hardware/drivers

[Adrian Cox <apc@agelectronics.co.uk>]

Tom Roberts wrote:

> Nothing obscure, I hope. We need a local PCI bus to interface to
> two ethernet MACs, some HDLC interfaces, and the H.110 telephony bus
> (Lucent has a part which directly interfaces H.110 to PCI). While these
> other devices will be PCI masters (for DMA), they are all quite dumb
> and will be programmed from our 7410s (Linux).

Sounds totally uncontroversial.

> > > 4) We favor a 4-CPU SMP configuration. What not-so-obvious problems
> > >    are we likely to face?
> > Designing a good interrupt controller.
>
> We have done this for our current boards, but they have no PCI bus,
> only local peripherals directly connected to a 60x or MAX bus. "good"
> to us merely means it works reliably; we have only rather low data
> rates as this is primarily a compute engine -- 200 kilobytes/sec
> would be a high data rate to us (during boot it will be higher).

You may have a low enough and predictable enough interrupt rate that you
can take some short cuts instead of a complete SMP OpenPIC
implementation. The more general problem of delivering an interrupt to
the "best" available CPU is an interesting exercise.

- Adrian Cox

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/