Subject: Re: DECchip 21143 and the tlp driver.
To: None <>
From: Brad Spencer <>
List: current-users
Date: 10/29/1999 17:11:42 (Brad Spencer) writes:
   >    > 2) Falling back on the de driver...  I have a 10/100 ethernet switch that
   >    > can do full duplex and the auto probe finds that it is connected to a
   >    > 100Mbits full duplex port, as expected.  However, it appears that an
   >    > output error is generated [or at least recorded] for every every packet
   >    > sent down the wire.  Slightly annoying, but I don't know if this is
   >    > significant.
   >    What swicth do you have ?
   >    I have a cisco 10/100 switch I don't have this behavior under 1.4.x.
   >    Maybe your board and switch dissagree on the half/full duplex settings ?
   > The switch is a Linksys 5 port.  I also have an 8 port down the hall in
   > the closet.

   When I had an early SMC etherpower card that didn't autonegotiate I
   found that I could never get it to switch media correctly without the
   interface being down for a while before and after the media change.

	   # de1 comes up 10BaseT
	   ifconfig de1 down
	   ifconfig de1 media 100BaseTX
	   sleep 1				# this is crucial
	   ifconfig de1 up

I tried the above, but it didn't seem to matter.  No amount of ifconfiging
can cause the media type to change.

   BTW my D-link 4-port DFE-570TX comes up ok so I never had to force the
   media to upshift.  I do see the fifo heuristics being changed if I
   really beat on the PCI bus.  This will only happen to me if I tar(1)
   on an NFS-mounted system.  The combination of hot-n-heavy disk I/O and
   network I/O appears to starve the PCI bus.  I do make sure I put my
   ethernet card in the slot closest to the CPU, which I am told on my
   intel al440lx board gives it a higher priority.  That might be why I
   don't see it more often.

The fifo changes start to occure around the time that the quotas are being
checked.  The machine in question has 3 scsi busses on it, and a couple of
them would be active checking the quotas.  It appears that any amount of
traffic at this time causes the fifos to adjust.

The set up seems to work fine.  Although the amount of Oerrs from 'netstat
-in' are quite high.


	  Wolfgang Rupprecht <>
   DGPS signals via the Internet

Brad Spencer -
[finger for PGP public key]