Philippe Gerum wrote:
> On Thu, 2007-10-11 at 18:35 +0200, ROSSIER Daniel wrote:
> > Hi everyone,
> >
> > I'm taking over the original thread from Patrick concerning the
> > port of Xenomai on Xscale with Linux 2.6.20. Briefly summarized,
> > the boot process actually freezes after a while right after the
> > nucleus has been started.
> >
> > I've investigated the issue over the last hours, and I came up with
> > the following conclusion: it seems that the problem is due to a
> > endless loop in do_gettimeofday in arch/arm/kernel/time.c. Here is
> > the code:
> >
> > "... do { seq = read_seqbegin_irqsave(&xtime_lock, flags); usec =
> > system_timer->offset(); sec = xtime.tv_sec; usec += xtime.tv_nsec /
> > 1000; } while (read_seqretry_irqrestore(&xtime_lock, seq, flags));
> > ..."
> >
> > If I remove the do { } while loop with the call to
> > read_seqbegin_irqsave(), then the boot process is going ahead (I
> > got a suspicious error like "I-pipe: Detected illicit call from
> > domain 'Xenomai' " but it might well be normal with such a
> > modification.
> >
> > I checked with the code from a 2.6.15 kernel, and it is basically
> > the same structure.
> >
> > Any idea about the cause of this problem?
> >
>
> That's typical of someone from the real-time (Xenomai) domain
> spuriously calling into this routine, which is illegal.
>
> The usual scenario for the lockup to happen is: - Linux task A calls
> into some code grabbing the xtime lock for writing - Xenomai preempts
> task A before it releases the seqlock, e.g. to process an interrupt -
> some code on behalf of the Xenomai IRQ handler calls into
> do_gettimeofday, which is wrong - the culprit gets a reader lock
> sequence token, which will never match the quiescent state. This will
> cause an infinite loop, since task A can't get out the critical
> section to put the sequence in a quiescent state again, until the
> Xenomai code relinquishes the CPU. Catch #22.
>
> You likely need to chase the code who triggers the "Detected illicit
> domain" message, likely in some code also calling do_gettimeofday().
>
Ok I'll do it. What I find weird is that the the do_gettimeofday() is
called by xnpod_start_timer(kernel/xenomai/nucleus/pod.c). It is
therefore called from the RT domain
but at the initialization time. My interpretation is that the timer
interrupts arrived during the invokation
of xnpod_start_timer(), which are propagated to the Linux domain. There,
the timer handler
is probably accessing do_gettimeofday() - to be proved - before the
xnarch_start_timer() has finished, which may (but should not) lead to a
potential deadend...
> > I checked the timer IRQs and other pipeline activity, and
> > everything seems to be OK. By the way, if I disable the nucleus,
> > Linux is working properly (timer not being stolen by I-pipe).
> >
> > Thanks in advance for any help.
> >
> > Cheers Daniel
> >
> > _______________________________________________ Xenomai-core
> > mailing list [email protected]
> > https://mail.gna.org/listinfo/xenomai-core
Daniel
_______________________________________________
Xenomai-core mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-core
