[eepro100] "Transmit timed out:" when activating the interface

[Donald Becker]

On Wed, 27 Dec 2000, erik gostischa franta wrote:

> I just just compiled a 2.2.18 kernel from the vanilla source .
> on a dual pentium box with two Intel EEPRO interfaces.
> rev 5 & rev 9 I think
> and I am using the newest 1.12 eepro driver from donald becker.
> 
> well, everything seems to work fine..except:
> at boot up (or everytime I restart the network of course)
> I get folowing error message:
> 
> eth0: Transmit timed out: status 0000  0090 at 0/7 commands 0001a000 
> 00020000 00030000
..
> a dmesg dump gives me:
> 
> eth0: Tx ring dump,  Tx queue 7 / 0:
> eth0: * 0 0001a000.
> eth0:   1 00020000.
> eth0:   2 00030000.
> eth0:   3 00030000.
> eth0:   4 00030000.
> eth0:   5 00030000.
> eth0:   6 40030000.
> eth0:  =7 00000000.

There are three commands types on the command list.
0001  Station Address setup
0002  Configure command
0003  Multicast setup (several)

Only the first command has completed, and not the subsequent commands.
The multicast setup commands are queued every two seconds whenever the no
packets have been received to work around an obscure Rx bug.

The status is
  Tx command unit active
  Rx unit ready.

Usually I would suggest that the card is having bus-master problems, but the
first command completed and wrote back an 'OK' status.  The chip is hanging
on the configure command.  This command is complex, but it shouldn't cause
any bus traffic.

> eth0:Printing Rx ring (next to receive into 0).
> Rx ring entry 0 00000001.

Sure 'nuf, no packets have been received.

> eth0: Tx ring dump,  Tx queue 7 / 0:
> eth0: * 0 0001a000.
> eth0:   1 00020000.

Some story after the reset.  The configure command is taking forever to
complete.

Does the interface eventually start after this problem?  Or is it
permanently hosed?

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