[eepro100] eth0: card reports no resources

[Donald Becker]

On Thu, 17 Oct 2002, Ben Greear wrote:
> Irene Koo wrote:
> > recently several of our linux machines are experiencing
> > the eth0:card reports no resources problem.
...
> > i have read through the previous emails in the archive
> > and updated the bdflush with
> >    echo 100 >> /proc/sys/vm/bdflush
> > (i read from the documentations that the 2nd, 3rd
> >    parameters are dummy)

What documentation.  (That's only half a joke.)
It depends on the kernel verison, and I don't know of any documentation
that lists the behavior changes between kernels.

> I have run sustained 96Mbps send + receive on two ports for 4+ days.
> 
> However, this was using the tulip driver, and it was using a low level
> packet generator (pktgen).  This does lead me to be a little skeptical
> about the excuse that it is the operating system that is screwing up, and
> not the hardware/driver...

The Tulip chip handles PCI bus bandwidth shortages and OS running short of
receive buffers better than the eepro100 chip.  In some cases, with some
chip versions, the eepro100 appears to get confused by running out of
either.  And all errors are reported as out-of-resources.

> I have also run a sustained 85Mbps stream send + receive over two ports,
> from user space, on tulip and have seen no problems (12 hour run, so far)
>
> Mr Becker, I would be glad to hear of a reason why the e100 seems to have
> these problems but the tulip does not.  Is the tulip driver/chipset that
> much better?  (I had always assumed that the e100 hardware was supperior).

The bugs and changes in the real Tulip chips are better documented (see
my postings about "that's the only bug").  But there are some versions
of the eepro100 chip with versions of my driver that are similarly robust.

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