[linux-audio-dev] debugging realtime issues with instrumented kernel

Robert Jonsson rj at spamatica.se
Tue Apr 19 09:03:28 UTC 2005


Great,

thanks for all the info everybody!

/Robert

On Tuesday 12 Apr 2005 13:01, Florian Schmidt wrote:
> On Tue, 12 Apr 2005 12:28:21 +0200
>
> Robert Jonsson <rj at spamatica.se> wrote:
> > Hi,
> >
> > I recall someone (Lee?) mentioning something along the lines of using a
> > recent kernel with Ingos patches to get feedback on realtime problems of
> > your own code. It sounded very interesting!
> >
> > Now, I can't for the life of me find the message, if someone knows
> > anything about this I would very much appreciate some more information,
> > are there any links?
> >
> > Thanks in advance,
> > Robert
>
> Hi,
>
> you need a realtime preemptive kernel. Read about them here (i just
> updated the RP-patch-script to create a current kernel (the realtime lsm
> version i use is a tad bit older. it should still work though, will
> update when i come home from university)):
>
> http://www.affenbande.org/~tapas/wiki/index.php?Realtime%20Preemption
>
> You will need to enable user triggered tracing in the kernel hacking
> section of the kernel config (i just patched up a recent RP kernel and i
> cannot find the option anymore - anyone please fill me in.. maybe you
> just need to enable the other tracing mechanisms there and then enable
> the user triggered tracing via one of the control in /proc/sys/kernel/.
> Ah damn, that's what you get for not following kernel development
> closely)..
>
> You also need to compile jackd (i think you need the CVS version for
> this) with the
>
> --enable-preemption-check
>
> flag during ./configure
>
> If an app is preempted now during its process() callback it will receive
> a signal SIGUSR2 (iirc), and since most apps don't handle this signal,
> they will exit.. The syslog will contain a report about what exactly
> triggered the preemption..
>
> Note though that this check is not garanteed to trigger for possibly
> unsafe code.. It will trigger only if the process really is preempted.
>
> Flo
>
> P.S.: Please correct me and fill in missing details. I will then create
> a page on my site about this..

-- 
http://spamatica.se/musicsite/



More information about the Linux-audio-dev mailing list