On Fri, 24 Aug 2001, Noah Meyerhans wrote: > On Fri, Aug 24, 2001 at 10:54:18PM -0400, Donald Becker wrote: > > > However, it is > > > > behaving strangely. I repeatedly get the error message: > > > > > > > > eth1: Something Wicked Happened! 18000 > > > > Hmmm, the Rx status steck overflowed. > > I don't see this with my driver release. > > This is very unusual -- either the driver misconfigured the chip, or the > > chip couldn't get any PCI bus bandwidth to transfer Rx descriptors. > > Could that be the fault of a buggy BIOS or chipset? For what it's > worth, I saw this problem on both a FIC PA2013 motherboard (I forget the > pci chipset used there) and an ABIT KT7A with a VIA Apollo KT133 > chipset.
It could be the fault of a BIOS that configures the card with a too-low PCI latency timer. "Too low" is referenced to other devices on the PCI bus, not an absolute value. > > That's pretty unlikely, unless the interrupt is blocked for a really > > long time. That's a problem with another driver on the system, not with > > sharing IRQs. > > Interesting. I should swap the card back in and see which other devices > the card is sharing IRQs with. I did move it to a different slot at one > pointin an attempt to make sure it was sharing interrupts with different > devices, but maybe it still got stuck sharing with a buggy driver. It doesn't matter (much) which other devices the natsemi chip is sharing IRQs with. The problem is any other driver which blocks interrupt service. > > As usual, my primary reply is that I don't support modified drivers. If > > you touch it, you should be willing to test and maintain the whole > > driver. > > I will debug problems that you can reproduce on an original driver. > > I had no trouble duplicating the problem in the original driver > (assuming, of course, that the version of natsemi.c included in Linux > 2.4 counts as "original"). No, it doesn't. Read the source code and you'll see it was modified. > ... There was a SCSI card in the configuration (using the > sym53c8xx driver), so that may have been involved. But as I said, I > don't know right now which cards were sharing IRQs. I suspect the SCSI driver played a role in the problem. Donald Becker [EMAIL PROTECTED] Scyld Computing Corporation http://www.scyld.com 410 Severn Ave. Suite 210 Second Generation Beowulf Clusters Annapolis MD 21403 410-990-9993
