Re: Failing back to INSANE timesource :) Time stopped today.

[john stultz <johnstul@us.ibm.com>]

On Tue, 2004-04-20 at 05:40, Maciej W. Rozycki wrote:
> On Tue, 20 Apr 2004, Niclas Gustafsson wrote:
> 
> > I've now been running the system since last week, about 6 days now with
> > sometimes quite high load, both in regard to CPU usage and network
> > traffic.
> > And it seems to be running just fine with the patch from Maciej.
> 
>  I'm glad to read this.

It appears to be working here in our labs as well.


> > I've got a couple of questions, 
> > 
> > When was this bug introduced? Was it 2.6.1 ( or rather somewhere in
> > 2.5)? Or was it already present in 2.4?
> 
>  Well, the bug has been introduced by IBM in their firmware (SMM code).  
> ;-)  The patch only works it around.  Functionally the changed code is the
> same for your configuration.
> 
>  If you are asking about the problematic code, then it's there since
> 2.3.x, so it's in 2.4, too.  It's a part of the NMI watchdog support,
> though it's used for ordinary timer interrupts for certain systems as
> well.

Are you saying that 2.4 will exhibit this problem as well, or that 2.4
already has an equivalent workaround?


> > When will this patch be merged into the 2.6-tree?  I don't have to
> > stress the impact of this problem on IBM servers as they are rendered
> > quite useless.
> 
>  Apparently there are problems with the workaround on certain AMD
> Athlon-based systems.  I suppose they need to be resolved somehow first.

Can you point me to any threads on this issue. I'd like to do what I can
to help get this workaround in.

> > Which other IBM models are affected? Can I run 2.6.5 on my 345:s or
> > 335:s?  Do they use the same buggy SMM firmware?
> 
>  Ask IBM.  The reason is an incorrect handling of PIC (8259A) state
> saving/restoration.

I'm following up w/ our hardware group about this issue.

Thanks so much for the help!
-john