Problems booting 9.1-STABLE on Netra X1

[Marius Strobl]

On Mon, Feb 25, 2013 at 04:05:42PM -0500, Chris Ross wrote:
> 
> On Feb 25, 2013, at 5:13 AM, Marius Strobl wrote:
> > This means that the machine is generating a power failure interrupt,
> > which causes FreeBSD to initiate a graceful shutdown rather than
> > waiting for the power to supply to suddenly die, which could cause
> > data loss.
> > In general, it's very model specific whether Sun hooked up that
> > interrupt to anything and to what. F.e., in U5/U10 it's just
> > connected to the power button rather than a circuit that monitors
> > the power supply. I don't know for certain what it is connected to
> > in X1. However, given that these latter use an ACPI-style power
> > button and were intended as servers, it's quite likely that their
> > power failure interrupt actually is connected to a power supply
> > monitoring circuit.
> > You could hack psycho(4) to just not register the power failure
> > interrupt handler. Whether you really want to use that machine in
> > this configuration (it could be either the power supply actually
> > starting to fail or also just the monitoring circuit being broken)
> > is something you have to decide on your own.
> 
>   Okay.  My memory is sketchy at best, but email is forever.  The following
> thread, and the second URL to a notable piece there-in where I found a
> pointer to the core problem, documents my efforts to get NetBSD running
> on this same system 3 years ago.
> 
> http://mail-index.netbsd.org/port-sparc64/2010/05/07/msg001259.html
> http://mail-index.netbsd.org/port-sparc64/2010/05/08/msg001267.html
> 
>   The solution that I think was to ignore the power fail on this hardware.
> A reference in that thread mentions a OpenBSD commit:
> 
> | revision 1.11
> | date: 2002/01/29 20:33:19;  author: jason;  state: Exp;  lines: +3 -1
> | Don't install a handler for powerfail... this causes weird problems with
> | the Netra X1 (interactions with lom)
> 
>   Numerous things were attempted during the course of that port-sparc64
> netbsd thread, some of which were tied to not installing the handler
> in the case of a machine identified as a Netra X1.
> 
>   It appears the code in place now in psycho.c is:
> 
> 		/*
> 		 * Netra X1 may hang when the powerfail interrupt is enabled.
> 		 */
> 		if (strcmp(machine_model, "SUNW,UltraAX-i2") != 0) {
> 			psycho_set_intr(sc, 15, psycho_powerfail,
> 				&sc->sc_regs->power_int_map,
> 				&sc->sc_regs->power_clr_int);
> 			psycho_register_power_button(sc);
> 		}
>   Would that be unacceptable for placing into FreeBSD?  Clearly, I can
> put it in my own tree for testing and use, but would prefer it live in the
> main sources long-term.  :-)
> 

Yes, IMO this approach is inappropriate as there really seems to be
something broken with this particular machine rather than this issue
being a general problem with that model, i.e. nobody else running
FreeBSD on an X1 reported this so far. I don't own an X1 but I do
have a v100, which employ a mainboard pretty much identical to that
of X1 and even also identify themselves as SUNW,UltraAX-i2, but don't
see such a problem, i.e. neither the power fail interrupt firing nor
strange hangs, either. Once I have a replacement for the broken
hardware of my committatron I can make registering the power fail
interrupt a tunable defaulting to on though.

Marius