[net.periphs] Clock loses time on 11-70 with UNET/Ethernet.

jerry@oliveb.UUCP (Jerry Aguirre) (09/16/83)

We have installed 3Com's UNET software on two of our PDP 11-70s.
The physical link is 3Com's Ethernet controller.  Ever since we
installed the Ethernet driver on those two systems their clocks
have been losing about 15 minutes a day.  We are running Berkeley
2.8 Unix and 3Com UNET 2.0.

3Com's installation instructions say that for Berkeley 2.8 you
should fix a "bug" in the mapin/mapout routines.

    The "spl5() call in mapin must be changed to "mapspl = spl6()",
    and the "spl0()" call in mapout must be changed to splx(mapspl)".

Where "mapspl" is a global int variable.  The save/restore seems
reasonable but the change from 5 to 6 makes me suspicious since the
original code contains the comment:

	spl5();	/* NOTE: ONLY CLOCK AND DU ARE CERTIFIED SAFE.  M000 */

Setting the priority to 6 would seem to block the clock interrupts.
I am not sure about changing it back to a 5 since they call the
mapin/mapout from interrupt level and the documentation gives the
priority levels for the Ethernet controller as:

	Receive completed	6
	Transmit jam/collision	5
	Transmit completed	4

I know of one other site that has the same problem on their 11-70
(but not on their VAX).  Does anyone have any ideas?  Is the
"spl6()" change the problem?  Is there a fix?

			    Jerry Aguirre
	    fortune | ios | hplabs !oliveb!jerry