[PlanetCCRMA] fc6/7: new rt kernel; fc6/7: updated libgig

Fernando Lopez-Lezcano nando@ccrma.Stanford.EDU
Sun Jul 8 22:18:01 2007

On Sun, 2007-07-08 at 21:49 -0700, Tracey Hytry wrote:
> Fernando said:
> > On Sun, 2007-07-08 at 15:26 -0400, Hector Centeno wrote:
> > > Ok, so just to confirm, that 2.6.21-0182.rt19.1.fc7.ccrmart works fine
> > > on my desktop but on my laptop it makes Firefox and Tomboy to crash.
> > > On the same laptop using 2.6.21-0182.rt17.1.fc7.ccrmart there is no
> > > problem.
> > Turns out that since rt11 the rt patch includes a new scheduler (the
> > CFS, Completely Fair Scheduler, also by Ingo Molnar) and may be causing
> > this problem. I posted to lkml and asked Ingo if it is possible to split
> > out this added functionality... sigh...
> I can get firefox to crash on just about any kernel(FC6).  It's quite 
> easy actually;  I just open a few tabs and especially while a closed 
> one is loading(or sometimes the current one refreshing) I just scroll 
> down the page real fast with the mouse wheel.  It took me a while to 
> see some kind of pattern on it, but it seems that if it crashes, it 
> crashes there.  It's also taken me a bit of time to remember to be careful, 
> which is real silly to be modifying my behavior for a computer program(oh well).

I never got that one. The one I'm getting happens with the rt20 kernel
and does not happen with the f7 kernel...

> Also, please don't remove the CFS scheduler, it's great!  

Well, it depends on whether it is the culprit or not for this problem. I
managed to hang both Firefox with Flash 9 and Tomboy (as reported by
Hector), in both cases they wait forever - or at least I ran out of
patience before the system call returned - in sched_getaffinity, that
would suggest something in the scheduler, but what do I know?

I also got a similar hang with OpenOffice but in a read system call and
could not repeat it afterwards. Weird. In all cases the processes are
unkillable and a reboot is needed. 

My immediate goal is to provide good low latency kernels, CFS would be
icing on the cake if it is good (I appreciate your feedback!). But I
would like to decide whether to include it or not and not have it
"bundled" with the rt patch. But then again maybe at this point both are
commited to be together forever :-)

> I don't have any load balancing running with the AMD dual core and 
> from watching the cpu loads in gkrellm, CFS totally outshines the 
> the stock fedora's kernel.  Fernando, if you get a chance, please 
> tell Ingo that I can't wait until CFS gets into the mainline kernel.  
> (I've been following the kernel scheduling stuff for some time now).

I will... he was in fact asking for this type of feedback in the
relevant lkml thread where I'm reporting this. He'll be happy...

-- Fernando