Marc Singer wrote:

On Tue, Mar 23, 2004 at 01:42:43PM +0100, Wouter van Heyst wrote:

...

On Sat, Mar 20, 2004 at 08:00:18AM -0800, Marc Singer wrote:

<snip>

you are talking about smc91111 driver in rmk pxa tree (smc91x), right?

If so I have been trying to figure that one out for quite some time now - being a newbie at the whole thing doesn't make it easier either...

My pxa setup with smsc 911c11 runs fine at boot up and while my board is requesting info about nfs mounted root fs from the host. Every packet is successfuly handled until the big ones kick in. I'm talking about first 4k chunk of busybox. From here on there are several retries to resend the request packet (with the reply from host being recieved) with no luck and eventualy kernel oops. The thing is when the chip gets flooded with data weird things happen and retry packets don't even get out on the wire. On some other topic I saw a note about ARDY signal not being handled properly in CPLD and thats what I will investigate with our constructors further. I belive driver works, bacause enabling debug msgs from kernel renders connection usefull (for the price of annoying msgs on console) but there could be some more tuning done.

So could it be that kernel is running 'too fast' eg. is accessing smsc registers before it should (device not ready) ? Could this behaviour be caused by inproper ARDY handling ? Our ARDY from LAN chip drives pxa RDY signal, but after reading cpld vhdl description it seem that our RDY signal is set all the time - Could this be the answer for early register access and data corruption ?

rgrds, himba