Re: (3c509) eth0: Missed interrupt

Paul Slootman (paul@waubel.ahwau.ahold.nl)
Tue, 12 Mar 1996 16:38:01 +0000 (WIN)


To: kaukasoi@elektroni.ee.tut.fi
Subject: Re: (3c509) eth0: Missed interrupt
Newsgroups: ahwau.linux-kernel
In-Reply-To: <Pine.LNX.3.91.960311100501.4880C-100000@elektroni.ee.tut.fi>
References: <199603102142.QAA01121@p90.pclark.com>
Organization: Albert Heijn Winkelautomatisering, Zaandam the Netherlands
Cc:
Bcc:

kaukasoi@elektroni.ee.tut.fi wrote:
>> > 1. The 3com driver is still giving error messages.
>> > Mar 8 11:40:11 borg kernel: eth0: Missed interrupt, status then 2011 now 2011
>> I'd *REALLY* like to see a real fix for this since with 15+
>> servers I'd hate to have to drop $1K+ in new ethernet cards :(

It most definitely is specific to 3c509, and started with 1.3.60 (I ran
1.3.59 without these messages, as soon as I had booted 1.3.60 it started).

>See the www-page of the author of the 3c509 driver (Donald Becker):
>http://cesdis.gsfc.nasa.gov/linux/drivers/3c509.html.
>The fix is to define final_version flag.

No, *that* mentions the situation with the message "status then 2011
now 2000"; in all the cases I've seen mentioned here status was in both
cases "2011". Donald Becker also states "Note that the "now" value has
the interrupt cleared, so it was handled after all." Again, that is not
the case here.

Please be more careful with these recommendations!

As far as I can see in the source defining the final_version flag
simply eliminates some error- checking code... As there appears to be
an error, and the code in question apparently handles the error,
removing this would seem to lead to trouble (an interrupt is
acknowledged here). Additionally this code was not changed when the
messages started to appear (again, after patching 1.3.59 to 1.3.60), so
something else in the kernel was changed that triggered this error
(interrupts being disabled too long?).

I would be extremely hesitant in simply eliminating this error-checking
code... However, someone here needed an ISA card, and was prepared to
swap a 3c59x PCI for it, so now I don't have the 509 anymore (or any
timeout messages, FWIW). When I did still have the 509, the missed
interrupts apparently had no other side effects than the message; the
error handling seems to work well. If the message is a problem, remove
*that*; again, I don't think simply eliminating the check is wise.

Paul Slootman

--
Murphy Software, Enschede, The Netherlands          There is no such thing
work: paul@ahwau.ahold.nl / paul@murphy.nl               as a coincidence.
home: paul@wurtel.hobby.nl