Subject: Re: le0: overflow
To: None <email@example.com>
From: Robert Elz <kre@munnari.OZ.AU>
Date: 12/19/1998 04:14:31
Date: Fri, 18 Dec 1998 08:31:07 +0200
From: Jukka Marin <firstname.lastname@example.org>
| Uh, Ethernet DMA has a lower access priority to RAM than the CPU? What
| good does DMA do then in the first place? :) How about the SCSI controller
| in SS1/SS1+ - does it use DMA?
Sorry, I can't comment on Sparc hardware design, I'm no expert at that (though
I think that scsi does do DMA). I do know a fair bit about the lance ethernet
chip however, and what its error status bits mean. Which has priority isn't
always the issue - its the latency that matters (exactly how long some previous
cycle takes before the lance's access is permitted)
I also suspect that there was a reason why Sun created the "buffered ethernet"
Sbus card - there may be something about all of this in old sun related
discussions somewhere. Or, if you can find someone who is a sparcstation
expert, they may be able to comment.
| I tried the "comment-out-this-annoying-kernel-msg"-method which worked
| quite well but it sucks.
Yes, however I suspect it is likely to be about the best that you will
achieve with that hardware. It isn't impossible that there's something
broken, or marginal, in your hardware, though you'd think that anything
marginal in this particular area would manifest itself with far more serious
and annoying effects that lance oferflow errors.