IntraServer Quad Ethernet

[Rick Kiessig]

Hi,

Thanks for responding.

tulip-diag reports that the driver is configured 100Mbit/full-duplex.

The indicator light on the switch (a Netgear FS105) also says full-duplex.

The collision counter in ifconfig only increments on the Tulip side of the
connection -- the other side doesn't seem to be seeing collisions (the other
machine is also connected full-duplex to the switch).

The collision lights on the switch never seem to come on.

I tried the following experiments:

-- plugged the card into a 10Mbit hub or a 100Mbit half-duplex switch,
forcing the connection to half-duplex.  This solved the performance problem,
but the collision counter still increments wildly.

-- Crossover cable directly between the two systems, avoiding the switch.
Full-duplex both directions.  This fails in the same way as the original.

-- Changed from compiled-in driver to a loadable module.  Still fails.

-- Switched to the old_tulip driver (module) in the 2.2.14 distribution.
Still fails.

-- Switched to the de4x5 driver (module).  Solved the performance problem in
full-duplex mode, but the collision counter still increments wildly as with
the half-duplex solution.  I also saw an odd SCSI reset while using this
driver, that I've never seen otherwise and that makes me very reluctant to
use this driver in a production system.

BTW, the IntraServer uses the 21143 chip, which seems to be properly
identified by the Tulip driver during initialization (as Rev 65).  I only
have one of the four ports on the board connected to anything.  This
particular board has a separate IRQ for each port.

In the failing full-duplex mode, tulip-diag reports (for port 2):

Digital DS21143 Tulip adapter at 0x2080
Port selection is 100mbps-SYM/PCS 100baseTx scrambler, full-duplex.
Transmit started, Receive started, full-duplex.
The Rx process state is 'Waiting for packets'.
The Tx process state is 'Idle'.
The transmit threshold is 128.
The NWay status register is 41e1d2c4.
Internal autonegotiation state is 'Negotiation complete'.

Are there compile-time or insmod-time flags/switches that might allow the
driver to work correctly in full-duplex mode?

Thanks,

Rick

-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-tulip-request@beowulf.org