[ixp1200] excessive collisions on 10 Mbit/s ethernet with BridalVeil Card

[Colin Flanagan]

Hello,

I am working with the IXP1200 book by Johnson & Kunze, and am trying to 
get the "single thread receive" code from chapter 5 to work successfully 
on the BridalVeil board.  To this end I have written my own drivers for 
the octal MAC and have modified ue.o to reflect the 64-bit 
bi-directional IX bus.  

I can get everything to work and have the IXP1200 receiving packets 
successfully from a Smartbits over a 10Mbit/s simplex Ethernet link. 
However, I get excessive collisions on the link, these can range from 
0.14 collisions per packet sent to 1.2 collisions per packet sent.  This 
seems like very unusual behaviour. I wonder if anyone else has seen 
anything like this?

Interestingly, the rate of collisions on the link seems to be related 
quite sensitively to the program running on the ready bus sequencer. 
 With 12 RxRdy MAC0 instructions running on the RBS, there are no 
collisions at all. With a single RxRdy MAC0 followed by 11 NOP1 
instructions, the rate of collisions per packet sent is 0.14. With 
alternating RxRdy MAC0 / TxRdy MAC0 pairs (repeated 6 times), I get 1.2 
collisions per packet transmitted from the Smartbits.  With 11 RxRdy 
MAC0 instructions and a single TxRdy MAC0 instruction, the rate is 0.7 
collisions per packet transmitted from the Smartbits. In all cases the 
RDYBUS_SYNCH_COUNT_DEFAULT register is set to 0. The IX bus is running 
at 66MHz.

I find this quite confusing, surely there should be no need for any 
collisions on a link with only one source (the Smartbits)?  One thought 
that has occurred to me, is the IXF440 applying backpressure to the 
link?  I can't see how this would happen though, in my understanding 
only a Flwctl instruction can cause backpressure to be applied. 
 Alternatively, is the repetition rate of the ready bus program 
significant?  All the RBS programs I have tried take 36 cycles, at 66MHz 
this should allow an adequate rate of sampling of the status bits.

If anyone else has encountered problems like this, or can give me 
pointers to where I might be going wrong, I'd be very grateful to 
receive them.

Thanks,

Colin Flanagan
Dept of Electronic & Computer Engineering
University of Limerick
Ireland

PS

FBI setup I am using:

RCV_RDY_CTL = 0
XMIT_RDY_CTL = 0
FLOWCTL_MASK = 0
RDYBUS_SYNCH_COUNT_DEFAULT = 0
RDYBUS_TEMPLATE_CTL = 0xf00