[PlanetCCRMA] kernel-rt in new core components

[Fernando Lopez-Lezcano]

On Tue, 2007-03-06 at 15:50 -0500, Craig Bourne wrote:
> One problem I have experienced with the most recent release of the 
> 2.6.19 rt kernel, including the one in the core package that you just 
> released, is what looks to be thrashing during shutdown. Based around 
> one out of memory  condition,  this resource may be compromised in some 
> way that was not evident as recently as the version just prior in Ingo 
> Molnar's release sequence.

Trashing in the traditional sense of hitting swap constantly?

> The whole shutdown sequence is dragged out with the shutdown of many 
> resources now taking more than 1 minute where before this took one 
> second or less. The first shutdown that takes more than one minute on my 
> system is mpd (the music player daemon). 

Have you tried to shut down that while you are logged in? (ie: before
rebooting?). Maybe that will trigger error messages in dmesg
or /var/log/messages that will give us a clue on what's happening...

> Though all the preceding 
> shutdowns are abnormally slow as well.
> 
> I have several times seen that shutdown is normal if done immediately 
> after boot. So I get the impression that uptime is a factor.
> 
> There has been no crash even with the (transient) out of memory 
> condition. If I am patient enough it seems that the system will, after 
> some very long period, shut down.
> 
> If there is some diagnostic that you ( or I.Molnar) would like me to run 
> to help isolate the cause of this instability I will try to do that for you.

It would be great. I have not experienced those stops during shutdown. 

When you reboot after the long shutdown sequence, do you see anything in
the contents of /var/log/messages that might indicate trouble? (a kernel
Oops, a BUG notice, etc, etc... I don't know how familiar you are with
the contents of that log file). 

I'm copying Ingo to see if he has any hints on what info you could
send...

I'd start with the output of dmesg after a normal boot sequence. Maybe
Ingo can spot something right away in the startup sequence. 

-- Fernando