Subject: Re: [linux-audio-dev] kernel preemption patch and 2.4.19
From: Michael Ost (most_AT_museresearch.com)
Date: Thu Jun 03 2004 - 08:00:25 EEST
On Wed, 2004-06-02 at 14:26, Roger Larsson wrote:
> > We found that a nasty system crash was fixed by turning off preemption.
> > The crash would happen fairly reliably by switching between virtual
> > terminals a number of times. It locked up the system hard. So hard that
> > we can't really find the problem in the kernel; we just found a
> > work-around basically by trial and error.
>
> Do you use a SMP computer? If not then you might have run into something
> that could hurt a SMP computer running without preemption...
>
> (Running SMP and preemption is kind of weird. And you may run into additional
> problems - per CPU data has also to be protected by spinlocks, etc... )
Nope. No SMP.
> >
> > We turned on kernel preemption basically because "everyone was doing
> > it". We haven't noticed any obvious, serious problems or differences
> > since turning preemption off.
> >
> > Does anyone have some suggestions about what differences to expect and
> > what we might have a closer look at?
>
> You won't notice much difference unless you run real time processes. But they
> will need to be suid root (or some wrapper/helper) to become RT processes,
> and many kernels do not ship with multimedia stuff suid...
We have a couple of SCHED_RR r/t threads running; and the app does run
as root. This is what I guess you would call an embedded Linux system: a
vst player on x86 hardware in a rack mount box (called Receptor; sketchy
info at http://www.museresearch.com). It's using Linux and Wine and a
bunch of our own code.
> >
> > Also, we are newbies about reporting this kind of crash. Any clues about
> > where to report it or ask about it?
>
> First - you mention virtual terminals
> What video driver are you using? Nvidia?
> If that is the case retest with xfree86 (nv)
Hmmm. It's onboard video. I'll have to look into that. I'll see if there
is any xfree86 more specific to our hardware.
> Second - try to what is happening (Magic SysRq)
Yes. We'll give that go.
> Third - 2.4.19 is kind of old now (not that many will be interested...),
> upgrade and try to reproduce.
We're stuck with it for the time being. It's too close to shipping to
switch over.
Thanks, Roger! ... mo
This archive was generated by hypermail 2b28 : Thu Jun 03 2004 - 07:56:30 EEST