David Brownlee wrote:

>         (setting reply-to to tech-embed)
>
> On Mon, 1 May 2000, John D. DeHart wrote:
>
> > We are currently trying to debug an embedded system of our own design
> > with a Pentium processor on which we are running a NetBSD kernel with
> > the file system in a memory disk.
> >
> > Under moderate ping traffic our system is crashing with kernel page
> > faults which I believe are caused by memory corruption. I think this,
> > because, if I turn on POOL_DIAGNOSTIC, that diagnostic code often
> > catches inconsistencies in the mbuf memory pools before it gets to the
> > point of a kernel page fault.
> >
> > At this point we cannot say whether the problem is hardware or
> > software. We are attacking it from many different angles.
> >
> > I'm wondering if there is anyone else who is working with an embedded
> > NetBSD system that might have some suggestions on anything special
> > that has to be configured for vm and paging to work right in such a
> > system.
>
>         I've not worked in an embedded environment, but on the offchance
>         I might be able to come up with a helpful suggestion...
>
>         Are you using an existing network driver, or have you written
>         your own? Is is possible you're missing an splnet() call?
>         You could start bracketing sections of code in splhigh() to ensure
>         you're not taking an interrupts at an unhelpful time.
>
>         What other hardware is hooked up to the machine at this point -
>         serial console?
>
>                 David/absolute
>                                    -- www.netbsd.org: Value design over hype --

David,

I'll also list the hardware that is on our system at the end here, but for anyone interested
in more details, here is a talk that I gave about the system at a recent workshop:
        http://www.arl.wustl.edu/arl/projects/ann/slides/spc.ppt
        http://www.arl.wustl.edu/arl/projects/ann/slides/spc.pdf

This system will plug into our prototype ATM switches which are already operational.
We also have an APIC Nic that plugs into a standard PCI bus that works with the
same driver in a normal PC. We have not had any similar problems in that environment.
I suspect we have splnet() in the right places, but I'll double check.

As we see it, the major differences in the two environments are:
    1. The hardware. Smart Port Card (SPC: our embedded system) in the switch vs.
            APIC Nic in a standard PC.

    2. The NetBSD kernel using a ramdisk filesystem

    3. Use of our own bootloader and BIOS


Thanks for any suggestions.
John

Here is the list of hardware components:

Network Device: APIC
        1 Gb/s ATM Nic Chip of our own design
        Driver we developed.

Pentium Embedded Module
        166 MHz MMX Pentium Processor
        L1 Cache: 16KB Data, 16KB Code
        L2 cache: 512 KB
        NorthBridge (430HX) - 33 MHz, 32 bit PCI Bus

System FPGA
        Xilinx XC4020XL-1 FPGA
        Replaces some of the PIIX3 (south bridge)
        Replaces some of the BIOS
        Replaces some of the Super IO Chip

Memory
        EDO DRAM
        64MB
        144 pin SODIMM

Switch Interface - 1 Gb Utopia

Link Interface - 1 Gb Utopia

UART
        Two Serial Ports
                NetBSD system console
                TTY port