Subject: Re: how to crash a diskless client (i386)
To: Manuel Bouyer <>
From: None <>
List: current-users
Date: 05/11/1999 09:58:33

> This is in bus_space_read_multi_4() I guess (it's inline, so doesn't appear
> in the stack trace).
ok, I see.

> Do you get this immediatly after starting the command, or at random?
no, immediately after starting the command...

> Do you always get this same panic, at the same place 
> (_wdc_ata_bio_intr+0x372) ?
...and allways in the same place!!

> I suspect the 32bit I/O here, could you try to recompile a kernel with the
> following patch: 
> [...patch snipped...]

Hey! That's it! I applied your patch and the beast is singing and
dancing! I get data rates of about 3.40MB/sec to /dev/null (bs=64k).
piping dd throu gzip on the slow dx4/100 to a file in the nfs-mounted
root directory gives data rates of about 0.12MB/sec...

hm... now there should be a way the set WDC_CAPABILITY_DATA32
conditionally for the hardware supporting it... or some compiletime
option? (flags perhaps?)

thanks a lot for helping with this! now I can clone my prototype
client easily to the other 20 or so boxes w/o using any pc-crap! (we
used ghost the clone pc's...) ha! ;-)