[U-Boot-Users] PCI cards known to work on the LITE5200

[Jonas Mark-r50740]

Hello John and Olivier,

as I understand your post you've roughly done what I proposed and it should work according to my experience. 

What I do not understand is how you shifted the clock by 27ns. To my experience clock buffers only allow shifting a clock in the range of hundreds of Pico seconds in both directions. This is adjusted by the difference in capacitive load of the reference output and the load on the other outputs. Please refer to the datasheet of your clock buffer for more details. Please also make sure that the reference output is not connected to a PCI device but that you use one of the other outputs.

Are you sure that the resolution of your measurement system is high enough that you can detect timing differences less than Nano seconds?

Did you measure that you now have a Tval of 3ns or is that based on calculations? Did you consider the capacitive load of the PCI connector and card?

So far I measured Tval against the C/BE signals. Does that yield the same result? Can you send me a screenshot showing the PCI clock, C/BE and AD moving? Did you measure Tval at the other signals as well?

When adjusting Tval please try to reach 2ns as close as possible. If you adjust the clock to much MPC5200 will have problems when reading from a PCI target.

Regards
Mark

> Hi Mark,
> 
> > Hello,
> >
> >> I am currently using a LITE5200. There is a known (annoying) PCI
> >> issue with the MPC5200, and therdfore most of the PCI cards I have 
> >> tried on the LITE5200 were not recognized. For several of them, 
> >> u-boot even crashed (bus error). I was wondering if somedy 
> has tried
> >> with success common PCI graphics cards and gigabit
> ethernet cards...
> >
> > Please note that this problem needs two partners to be seen. On the
> > one hand, MPC5200 violates Tvalmin by having this about 1ns 
> instead of
> > at least 2 ns according to spec. On the other hand, a PCI
> card needs
> > to be able to handle a Tvalmin of down to 0 ns when somebody reads
> > from it. So if both sides behave badly the card is not recognized 
> > during Type-0 configuration.
> >
> >> I have tried a suggested workaround, using an ICS574 to adjust the
> >> MPC5200 PCI clock signal from the PCI bus, but still no 
> luck with the
> >> PCI cards I have tried (although it should work with the PCI SM712
> >> eval board).
> >
> > How did you adjust the clock using the clock buffer
> mentioned above?
> > Overdoing it will also cause harm especially during writes. My
> > suggestion is to measure the real Tvalmin and then adjust 
> it that it
> > just reaches 2 ns.
> 
> Our Hardware engineer, John, has the following response:
> 
> Mark - thanks for the reply.  I did try to do what you
> suggested without success but maybe I misunderstood 
> something.  If I look at the PCI clock and ADx signal, I see 
> a Tval of about 0ns (ADx changing right along with a rising 
> PCI clock).  So I try to delay the PCI clock by approximately 
> 27ns (33MHz PCI clock = 30ns clock period).  Since I think 
> that this is a free running clock, it should have the effect 
> of 'speeding up' the PCI  clock so that it gives a Tval of 
> about 3ns.  I've tried other delay values without success, 
> but what I have described above is what I read what I should 
> try to do.  Am I missing something?  Thanks.
> 
>