Subject: de looping in interrupt on heavy load
To: None <current-users@NetBSD.ORG>
From: William Coldwell <billc@warped.com>
List: current-users
Date: 03/03/1998 19:27:00
Problem: Under heavy ftp/scp transfers, the machine will stop responding on
the network... and become pretty much a doorstop. Sometimes I would find
the keyboard LEDs stuck off and non-responsive.. other times, just the getty
was there, and wouldn't accept anything, just echoing ^C ^Z, etc. I started
ripping things out of the conf file.. to no avail.
After fighting with this problem for about 2 weeks, I finally was able to
catch it in DDB and do a trace.
What we're seeing is it's spinning inside of the tulip_rx_intr code jumping
into malloc, eintrnames, m_free, and some other vm and kernel calls before
coming right back to the interrupt. Now, I'm not a DDB jock, and I don't
play one on TV, so I can't really be of much more help :/
I _believe_ this started about the time that Matt Thomas moved the de driver
to the bus_dma architecture, and I suspect something got biffed. Anyway,
this is really a pisser, because I can't even FTP the 1.3 archive to bring
the system back up since it tickles it. It only happens under heavy network
load, and there doesn't seem to be a specific amount of time (the worse kinds
to track down).
Frustrated as hell. Is anyone else seeing this with the de 21040?
--billc
--
William J. Coldwell MediaPlex Communications
Gizmo Attitude Adjuster Internet Extra Corporation
http://www.warped.com/~billc WebMaster, Inc.
v:408.346.0106 p:415.423.8147 Warped Communications