[tulip] stall on tulip card

[Donald Becker]

On Mon, 8 Jan 2001, Dan Hollis wrote:
> On Mon, 8 Jan 2001, Donald Becker wrote:
> > The only difference is with the transmit threshold, which was increased in
> > the error case.  There was a report that manipulating Tx threshold was
> > causing a problem with the Centaur, but I was unable to reproduce the
> > problem.  If this really is the problem, we will have to figure out if the
> > hang is caused by the initial underrun, or the attempt to increase the Tx
> > threshold value.
> 
> Donald, I have about 10 data points now from people who have used the
> CSR18 bit 1 patch. For most of them (9 out of 10) it fixed the problem
> and for the last it didnt change anything, eg it didnt make things worse.

Does this mask the TxUnderrun event?
If so, the Tx threshold will never be increased, and the Tx underruns will
continue.

> I know you cant reproduce the problem but lots of us have the problem and
> CSR18 bit 1 really does fix it. Please consider adding it to the driver.

It's in, but I'm fully comfortable with it.
Please send reports...
I added this code in the TxFIFOUnderrun handler:
	if (tulip_debug > 1)
		printk(KERN_WARNING "%s: Tx threshold increased, "
			   "new CSR6 %x.\n", dev->name, tp->csr6);

Look for 0.92q in the test directory today.  If it passes the tests it will
become either 0.93 or 1.00 at the end of the week.

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993