[1119] in linux-net channel archive
Interesting things with Linux AppleTalk
daemon@ATHENA.MIT.EDU (Baba Z Buehler)
Tue Sep 19 16:58:26 1995
Reply-to: Baba Z Buehler <baba@beckman.uiuc.edu>
From: Baba Z Buehler <baba@beckman.uiuc.edu>
To: linux-net@vger.rutgers.edu
Date: Tue, 19 Sep 1995 09:40:29 -0500
I have a problem with my SMC Ultra card on Linux in that sometimes, when there
is a high load on our local ethernet, the smc-ultra driver will get transmit
timeouts, generating messages like:
Sep 19 09:24:12 tigger kernel: eth0: transmit timed out, TX status 0x48, ISR 0x0.
Sep 19 09:24:12 tigger kernel: eth0: Possible network cable problem?
I've looked up the values of TX status, and generally they mean that the board
just couldn't get on the wire because of excess collisions. These messages
didn't seem to have any operational impact on my machine until I started
running atalkd and afpd from Netatalk-1.3.3b2.
I noticed that as soon as the SMC would generate these messages, atalkd would
think all of its routers went down, and I'd get something like:
Sep 19 09:24:30 tigger atalkd[85]: as_timer gateway 54272.1 down
Sep 19 09:24:30 tigger atalkd[85]: as_timer gateway 54272.152 down
Sep 19 09:25:00 tigger atalkd[85]: as_timer gateway 54272.1 down
Sep 19 09:25:00 tigger atalkd[85]: as_timer gateway 54272.152 down
Sep 19 09:25:00 tigger atalkd[85]: as_timer last gateway down
54272.1 is our Proteon CNX500, 54272.152 is a gatorbox on our net. After this,
AppleTalk becomes unusable until I reboot.
Now, the _really_ interesting thing happened the other day, when I noticed that
for some reason AppleTalk had started working again after this happened. I got
a series of messages like this:
Sep 19 09:26:55 tigger atalkd[85]: rtmp_packet router has become available
Sep 19 09:26:55 tigger atalkd[85]: zip_getnetinfo for eth0
Sep 19 09:26:55 tigger atalkd[85]: zip_getnetinfo for eth0
Sep 19 09:26:56 tigger atalkd[85]: zip_packet configured eth0 from 54272.1
Sep 19 09:27:05 tigger atalkd[85]: rtmp_packet gateway 54272.1 up
Sep 19 09:27:06 tigger atalkd[85]: rtmp_packet gateway 54272.152 up
I didn't realize what had caused it until today, when, just for the heck of it
I fired up sd to see if there was any interesting multicast channels up, and
just as soon as sd started up, I got the messages that the AppleTalk routers
were back up.
I tried this a couple of times. I can make the SMC generate the timeouts by
simply ftp'ing a big file between two other machines on our net. And I can
just as reliably get atalkd to come back up by starting sd.
This suggests that something is going on in the Linux net code that "breaks"
part of the AppleTalk code when the SMC times out, and that something happens
to "un-break" (I'm getting technical here , :-) ) that when an application
accesses multicast.
I haven't spent much time hacking at any of the net code other than the smc
driver, so I'm not really sure where to look for this. I'll, of course, be
happy to run any tests that anyone would like to suggest.
Thanks,
--
# Baba Z Buehler - 'Hackito Ergo Sum'
# Beckman Institute Systems Services, Urbana Illinois
#
# 355/113 -- Not the famous irrational number PI,
# but an incredible simulation!
#
# WWW: http://www.beckman.uiuc.edu/groups/biss/people/baba/
# PGP public key on WWW homepage and key servers (key id: C13D8EE1)
