Re: [pci] PCI target

[Duane Clark <dclark@akamail.com>]

Miha Dolenc wrote:

> Hi again,
> 
>     now, I would call that cheating ;-) . But there are still some issues
> that remain - this PCILOGIC block seems to control only clock enable signal
> depending on the state of IRDY and TRDY signals. But there are other signals
> that don't have any relationship to IRDY and TRDY handshaking - for example
> address phase ( no IRDY nor TRDY are asserted then ) and PAR and incoming
> byte enables for target reads and so on. I'm having problems here too,
> because let's say some external master is doing a read through a target
> interface. I have to calculate PAR from incoming CBE signals (7ns
> constraint) and outgoing data that target is driving. Here PCI logic block
> doesn't help if I'm not mistaking?
> 


Yep, you are absolutely right. I think the address timing is only a 
problem if you are trying to do "Fast" decode. I would go for the 
"Medium" decode in FPGAs. The PAR timing then appears to me to be the 
main signal where you really have to push things.

I was partly through a PCI design 7 or 8 months ago (I've been buried in 
work since then). But the way I figured the PAR timing could be made was 
to not use the output FF in the IOB for the PAR signal, or the input FFs 
for the CBE signals.

Instead, I had the CBE signals come in directly to a LUT, along with a 
single bit for the rest of the signals on which parity had already been 
calculated. Fortunately, there is an extra XOR gate in the Virtex slice, 
so a 5 input XOR is easy. I latched that with the FF in the same slice, 
and then output that to the PAR pin. I think I even made a small test 
circuit and ran place and route and timing, to verify that this would 
meet the timing specs.

Mike wrote:

> You are absolutely right, the Xilinx PCI core cheats using exactly those
> 'hidden' resources.
> Not sure how we can compete against that.
> 


There is nothing wrong with using them yourself. All the Xilinx tools 
understand the PCILOGIC primitive just fine, so go ahead and use it.

Duane



--
To unsubscribe from pci mailing list please visit http://www.opencores.org/mailinglists.shtml