Good news I now have the network working. I'm seeing a number of transmit timeouts (see below) but don't seem to be losing packets. I found that as well as clearing bit 21 in csr0 that I also had to clear bit 24. That makes the csr0 assignment
static int csr0 = 0x00800000 | 0x9000 I suspect that the timeouts that I'm seeing probably have something to do with other dm9102A/pci effects. I will keep trying differing settings to see what effects they have. As for the timeouts I'm seeing NETDEV WATCHDOG: eth1 transmit timed out Whilst doing pings I get this message around 8 times a minute. Sometimes I see a number of pings sitting waiting to go and then suddenly I get a flurry of them (maybe up to 15 seconds late). I think that this is consistent with the transmit timeouts. Richard P.S. This is running 2.4.6-pre7 or thereabouts. I didn't need to go back to earlier versions to get the network errors removed. I guess that I ought to update to current cvs now that I seem to be making definite progress. Richard Mortimer writes: > Ben Collins writes: > > On Sat, Jul 21, 2001 at 12:14:49AM +0100, Richard Mortimer wrote: > > > We may be able to scrap the dmfe.c idea and keep with tulip. I did a > > > bit of looking into why I was seeing pci errors with the tulip > > > driver. After a bit of scratching around and a bit of a helpful > > > suggestion from a colleague it seems that the dm9102a chip has a bit > > > of a problem with Memory Read Multiple pci transactions (That is CSR0 > > > bit 21). If the MRM bit is cleared the pci errors go away and the chip > > > starts to act a bit more reasonably. > > > > This sounds an awful lot like the problems I am having with tulip on my > > g4, so it may not be an X1 issue. I had to back down to a version from > > around 2.4.[34]. Maybe you could try copying the tulip directory from > > that kernel version to 2.4.7, apply your patch and see how that works? > > I tried that and got the same result. Do you think it might be worth > trying 2.4.[34] on the X1 (with hand crafted patches to get the > machine to run at all?). > > Richard > > -- > Richard Mortimer - [EMAIL PROTECTED] dot netscapeonline dot co dot uk >