Re: NetBSD 5.0.1 Networking Message

[David Young <dyoung%pobox.com@localhost>]

On Sun, Aug 16, 2009 at 01:33:05PM +0200, Manuel Bouyer wrote:
> On Sat, Aug 15, 2009 at 06:39:41PM -0400, Erik Winkler wrote:
> > I am running NetBSD 5.0.1 (Generic PPC Kernel) on PowerMac 7600 with a  
> > powerpc 7455 processor upgrade card running at 1 Ghz.  I have 2 of the  
> > following ethernet cards installed (per NetBSD dmesg output):
> > 
> > 3Com 3c905B-TX 10/100 Ethernet (rev. 0x30)
> > 
> > For some reason I keep getting the following messages and I don't see  
> > the cause:
> > 
> > ex1:  transmit underrun (9000) @20
> 
> This means the 3com chip couldn't get data fast enough from the PCI bus.
> THis is usually because another device (or possibly the CPU) holds the PCI
> bus for too long. It's probably happening only once in a while (no more than
> once every 256 packets) otherwise it would have bumped the TX threshold.

It's also possible that one or more PCI bridges are misconfigured.
For example, some PCI-PCI bridge between ex1 and the host may not do
multi-word memory read/write "bursts" because the latency timer is set
to a small value such as 0.

For example, I pcictl(8) shows that a PCI-PCI bridge on
my gateway is misconfigured,

*snip snip*
    Header Type: 0x01 (0x01)
    Latency Timer: 0x00
    Cache Line Size: 0x00

  Type 1 (PCI-PCI bridge) header:
    0x10: 0x00000000 0x00000000 0x20020100 0x2280d0c0
    0x20: 0xfea0fe60 0x0000fff0 0x00000000 0x00000000
    0x30: 0x00000000 0x00000000 0x00000000 0x00020000
*snip snip*

That may explain why NICs downstream of that bridge complain,

sf1: transmit FIFO underrun, new threshold: 400 bytes
sf1: transmit FIFO underrun, new threshold: 416 bytes

Dave

-- 
David Young             OJC Technologies
dyoung%ojctech.com@localhost      Urbana, IL * (217) 278-3933