kern/60526: Post-PAE stable SMP machine freezes

[Søren Schmidt]

Doug White wrote:
  >>According to
>>http://freebsd.rambler.ru/bsdmail/freebsd-hackers_2003/msg03936.html
>>the problem may be related to the <ServerWorks ROSB4 ATA33 controller>
>>and ATA write cache (hw.ata.wc=1).
> The problem is with DMA mode on these controllers, not write caching.
> Write caching may help to get the data rate up fast enough to trigger the
> bug, and turning it off slows things down.
> 
> I HIGHLY recommend using a DIFFERENT controller for disks. The ROSB4 is OK
> for CDROMs and the like, but use a different controller for the system
> drive.  Promise controllers work great. :)

Cant disagree with the last sentence, however my only i386 SMP machine 
has exactly the ROSB4 chip in it, and I use that for disks etc, newer 
had a problem with it, but again this is an ASUS board and they 
supposedly fixed the problem in HW (the *only* way to fix it btw).

> ATA tagging is known to work with only a few disk models.  You should
> enable ATA tags ONLY if you KNOW your drive supports it, and test
> extensively to ensure stability.

To be more exact, only a very few disks supports tags, in fact I only 
know of IBM and a select few WDC disks that actually anounces support 
for tags. The ATA driver (in -stable) only tries to enable tags if the 
HW says it supports it and its a known to work drive. However the PAE 
import has broken tags support so it doesn't work on any disks :(


-- 
-Søren