Hi, A blast from the past ! I'm following up on a very old problem that I had with the pcnet32 driver (1 year ago). See attached e-mails for the details of the discussion. Summary : I have a single board PC (Adastra, P233MMX, PCNET32). While upgrading from 2.2.13 to 2.2.14, the pcnet32 driver started to fail. Thomas identified that my hardware didn't has any SRAM, so when the driver was enabling it, it could not send any packet... Basically, what happened after the e-mail exchange is that my manager put the single board PC in the demo room (with 2.2.13) and told me "don't touch it, it works !". So, I was never able get the board back to test the fix of Thomas. This week, I was able to get the PC back ! Yes ! I'm supposed to upgrade it to 2.4.X to get the latest IrDA features. So, I did catch up on this problem. I guess that this hardware configuration is rather unusual, because it seem that nobody else did report/fix that bug. Conclusions : The fix of Thomas works ! Marvelous ! It survives the usual netperf test with 2.2.18. Of course, without Thomas's fix, 2.2.18 fails the same way. And I did a port of the fix for the version of the pcnet32.c driver in kernel 2.4.0, and this correct the problem as well (see attached). Unresolved Issues : My 2.4.0 patch could be cleaner. Currently, disabling NOUFLO is done only for the 79C971 (that's what the original fix from Thomas did). However, I believe that this fix should be generic (i.e. for all chip version). If the netperf is long, the driver fail and reset itself (see attached). After that, things are back to normal. I don't know if it's normal, hardware issue or whatever. But as the driver is usable for normal load, it's not critical. Sorry for the delay, thanks for the support, and I hope that it is still helpful... Jean
