Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 31 Jul 1997 23:50:43 -0400 (EDT)
From:      StevenR362@aol.com
To:        bde@zeta.org.au, cgull@smoke.marlboro.vt.us, freebsd-current@freebsd.org
Subject:   Re: problems with IDE DMA
Message-ID:  <970731235042_1593313789@emout10.mail.aol.com>

next in thread | raw e-mail | index | archive | help
In a message dated 97-07-31 06:02:42 EDT, bde@zeta.org.au (Bruce Evans)
writes:

> After successfully testing IDE DMA for a little while (reading rwd0,
>  fsck -n and bonnie -s 64 worked fine) I got the following messages
>  after the write stages for bonnie -s 64:
>  
>      wd0: interrupt timeout:
>      wd0: status 50<rdy,seekdone> error 0
>      wd0: interrupt timeout:
>      wd0: status 58<rdy,seekdone,drq> error 1<no_dam>
>      wd0: interrupt timeout:
>      wd0: status 58<rdy,seekdone,drq> error 1<no_dam>
>      wd0: interrupt timeout:
>      wd0: status 58<rdy,seekdone,drq> error 1<no_dam>
>      wd0: Last time I say: interrupt timeout.  Probably a portable PC.:
>      wd0: status 58<rdy,seekdone,drq> error 1<no_dam>
>  
>  The system then drifted South:
>  - no more interrupt timeouts, but wd seemed to get stuck in single
>    sector mode.  ls on the bonnie output file was very slow, and
>    systat reported about 20 interrupts/sec, 10 "blks", and often
>    "***** msps" (division by 0?).
>  - long before the 13107 seconds required to read 128MB at 10K/sec,
>    the system hung (the keyboard LEDs didn't work).
>  - the test drive was dead after hard reset.
>  - the test drive was OK after cycling the power.
>  
I very much doubt that this is a problem with the new DMA code.  As I 
have been seeing this happen off and on since 386BSD with patchkit 2.4
days.  It seems to happen when you have multiple processes hitting the
drive hard and perhaps an interrupt gets lost.  I sometimes see it during
make worlds or when doing a cvs update in /usr/src and /usr/ports 
simultaneously.  These errors appear on the console and the hard drive
LED goes solid on with the whole system locked up inside of 30 seconds.

I believe it is a timing related race condition in the wd.c driver.   As
various
FreeBSD editions exhibit it where others don't.  Current, checked out a
couple
of hours before the Lite2 merge at the beginning of the year was very bad
in this regard.  Whereas current from the past few months has been very
solid.

Steve



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?970731235042_1593313789>