Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 11 Mar 2005 13:57:56 +0000
From:      Jason Thomson <jason.thomson@mintel.com>
To:        Emmanuel OTTON <otton@enstimac.fr>
Cc:        freebsd-stable@freebsd.org
Subject:   Re: PAE+SMP panic with "page fault while in kernel mode" in many files	opened situations
Message-ID:  <4231A3E4.4070605@mintel.com>
In-Reply-To: <942faaf51f9210be76c25158da4d1b86@enstimac.fr>
References:  <6290d171fce4e3b4ee88f403a4aac711@enstimac.fr> <42317CF3.20903@mintel.com> <942faaf51f9210be76c25158da4d1b86@enstimac.fr>

next in thread | previous in thread | raw e-mail | index | archive | help
Emmanuel OTTON wrote:

> 
> Le 11 mars 05, à 12:11, Jason Thomson a écrit :
> 
>> If you disable the logical processors in the BIOS,  this problem 
>> should go away.
>>
>> We ended up reverting to 4-STABLE for other reasons.
> 
> 
> Thanks for your answer.
> 
> The problem is still there, unfortunately.
> 
> Crashes no more on pagedaemon though, for a change, but on irq48:em0...
> 
> Would you have KVA_PAGES / maxvnodes recommendation for this case ?
> 
> Was your answer based on painful trial-and-error, or is there a logical 
> reason (known bad HTT or PAE or Xeon support) ?
> 
> Should I revert to 4-STABLE to get solid PAE+SMP support ?
> 
> -- 
> Emmanuel OTTON - Ecole des Mines d'Albi-Carmaux - Responsable informatique
> Tél: 05 63 49 30 86
> 


We had KVA_PAGES=768.  It's now at KVA_PAGES=672 to give us the right
balance between KVM & usable process size.

We spent a few (4 hours) trying to get 5-STABLE to work properly.

It would reliably crash when running make -j 4 buildworld.


When we disabled the logical processors in the BIOS,  make -j 4
buildworld would succeed.

Also,  we disabled the USB subsystem in the BIOS.  I don't know whether
you have already.  It looks like you haven't from your first post.  This
is advisable.


In any case,  we reverted to 4-STABLE.  We needed the patches posted to
this list to enable us to use PAE + amr driver.

Since reverting it has been completely stable.  (And very very fast:-)

I didn't post anything to the list regarding our problems,  because we
weren't able to get a crashdump.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4231A3E4.4070605>