On Jun 26, 2005, at 12:55 PM, Manuel Bouyer wrote:
>
> It's possible that 2.0.2 is making an heavier usage of memory or PCI
> bandwith ? Maybe this SCB timeout is a problem caused by contention, 
> not
> handled very well by the fxp chip ?

Well, I think this is similar to what  Izumi Tsutsui has suggested as 
well:

"IMHO, the hardware design of fxp(4) by Intel guys doesn't suit
  most RISC CPUs (it can't prevent some race conditions
  because it has DMA descriptors in uint16_t)
  so I'd suggest to use other NIC (ex(4) etc) on such machines."

Since I use the fxp's built into the CS20, I've got two spare PCI slots
and I've got a number of ex boards I could pop in .. however, I was 
hoping
to use a CS20 as a new fileserver with two FC controllers.  
(Unfortunately
thats another story).

So if a CS20 can't run NetBSD 2.x and use its built in fxp, then I'll 
put in
some ex boards.

BTW, I did some double checking on my problem with port assertion 
issues under
1.6.  It seems I was using a 1.6MP kernel with lots of debugging.  I 
should just mention
now that the 2.0.2 kernels I'm using are built with -mbwx, no 
DIAGNOSTIC, no DEBUG=-g
and no MP, as suggested by other members off the list.