[PlanetCCRMA] jackd xruns problem: solved? (FC2 2.6.8.1+P9 kernel)

Fernando Pablo Lopez-Lezcano nando@ccrma.Stanford.EDU
Wed Aug 25 19:16:01 2004


On Wed, 2004-08-25 at 03:11, Rui Nuno Capela wrote:
> Hi Fernando et al.
> >> In your opinion should I try this new kernel or wait?
> >> I have a smb pentium III, but If this is unstable..
> >
> > Well, I don't know (if it is stable). I have not had a chance to test on
> > a true multiprocessor machine yet (I'm not running fc2 on production
> > machines so I have to grab one, get another disk and install fc2 to be
> > able to test). Most probably tomorrow. Stay tuned :-)
>
> Just to add some of my experience here, although not precisely Planet+FC2
> related,.
> 
> I do have a HT enabled P4 currently test-running a SMP kernel patched with
> voluntary_preempt-2.6.8.1-P9. It's a SUSE 9.1 box, but that's not the
> point.
> 
> The thing to note is that I can't set voluntary_preemption>1 without
> getting a complete system lockup in no time. Almost everytime it just
> hangs on KDE 3.3 startup. Othertimes, is just suffices to start jackd -R
> and the complete box is frozen. No kyeboard magic SysRq can even work to
> the rescue :(

Yes, that is my experience so far with SMP + voluntary. Hard lock, no
sysrq. 

> OTHO on my laptop I have a Mdk 10.0 with same kernel setup but UP. With
> the default settings as kernel_preemption=1 and voluntary_preemption=3
> everything seems to work without a glitch.
> 
> So I think that SMP/HT kernel 2.6.8.1 and voluntary_preempt>1 is far from
> being stable. It's a showstopper.

I'd agree. 
This is what I currently see happening (all with 2.6.8.1 + P9):

1) UP kernel on UP machine (P4 Mobility / Athlon64 - Radeon, DRI):
  works fine, very good latency

2.1) UP kernel on SMP machine (dual Athlon MP, Matrox450, DRI):
  a) lockup problems with the dri driver

2.2) UP kernel on SMP machine (dual Athlon MP, Matrox450, no DRI):
  runs fine with DRI deactivated, very slow of course

3) SMP kernel on UP machine (P4 Mobility / Athlon64 - Radeon, DRI): 
  a) interrupt problems (acpi=on or acpi=off does not make a difference)
     P4 Mobility laptop = oops on mouse startup
     Athlon64 = IRQ#145 hangs, serial ata card (main disk, can't boot)

4.1) SMP kernel on SMP machine (dual Athlon MP, Matrox450, DRI):
  a) interrupt problems when acpi=on
  b) boots fine with acpi=off, hard lock when running jack -R

4.2) SMP kernel on SMP machine (dual Athlon MP, Matrox450, no DRI):
  a) boots fine with acpi=off, hard lock when running jack -R

-- Fernando

PS: Planet CCRMA users: sorry for the never ending kernel thread :-)