Subject: Re: Kern/25659 is not isolated to pcmcia attached drives.
To: Brian Buhrow <email@example.com>
From: Steven M. Bellovin <firstname.lastname@example.org>
Date: 04/22/2005 12:14:44
In message <200504212229.j3LMTIh9020321@lothlorien.nfbcal.org>, Brian Buhrow wr
> Hello. Has anyone else noticed that the wdc.c timeouts seem to be
>short enough that attaching drives to NetBSD-2.x systems can be somewhat
>challenging? Kern/25659 correctly identifies the problem as being in the
>sys/dev/ic/wdc.c file, but it doesn't look like any changes have been made
>to that file in the source trees which reflect the knowledge of this
>problem. If anything, changes over the past year or so have caused most
>timeouts in this file to shorten considerably. I've appended what I hope
>is a less hackish patch to this bug in the gnats database, and hope that
>someone with more knowledge than I could address the general problem. The
>patch I appended makes things work quite well, but I'd be happy to engage
>in a discussion on this topic if it might result in a permanent fix which
>will live on in the sources.
> Does anyone have any thoughts on this problem? Am I the only one
Per the original PR, I've only seen the problem with a PCMCIA drive. I
never committed the patch because there were objections -- it could
effectively stall the system for ~30 seconds under certain
My preference, though, is that my code should be incorporated, but with
the wdc_delay_cnt variable initialized by default to 0 -- that will
give the same timeout behavior as the current driver. The right way to
do it would involve timer interrupts to look for the results of various
activities; I suspect that that would actually require a new kernel
process to get it right, and I don't know that it's really worth it.
All that said, I no longer have access to the hardware that caused the
trouble for me, so I can't test any fixes.
--Prof. Steven M. Bellovin, http://www.cs.columbia.edu/~smb