From: Marcus Glocker <marcus@nazgul.ch>
Subject: Re: Prevent crash on MacPro6,1 during power-down
To: Mark Kettenis <mark.kettenis@xs4all.nl>
Cc: tech@openbsd.org
Date: Wed, 30 Apr 2025 23:29:46 +0200

On Wed, Apr 30, 2025 at 11:13:46PM GMT, Mark Kettenis wrote:

> > Date: Wed, 30 Apr 2025 17:05:21 +0200
> > From: Marcus Glocker <marcus@nazgul.ch>
> > 
> > I've installed an amd64 snap on an MacPro6,1 (trash can), and noticed 
> > that during power-down the system is crashing at the very last steps:
> > 
> > 	syncing disks... done
> > 	Stopped at	x86_bus_space_io_write_4+0x1d:  leave
> 
> So why did it stop here?  Can you get a backtrace?

The keyboard doesn't work in ddb at this point.
 
> > After some investigations I found that this happens when we try to
> > write to the I/O port 0x80.  This seems to be the POST debug port used
> > by the BIOS and firmware to output POST status codes, and is most
> > likely not implement on systems like the MacPro6,1.
> 
> Writing to an address that doesn't "exist" typically doesn't do
> anything in x86.
> 
> > I couldn't come up with a more elegant solution, but this diff lets me
> > nicely power down this machine.
> 
> So this write is initiated from ACPI code.  That suggests that the
> machine explicitly directs us to make that write.

It happens when we call the "_PTS":

        /* 1st powerdown AML step: _PTS(tostate) */
        aml_node_setval(sc, sc->sc_pts, state);

> > Better ideas or OKs?
> 
> Maybe start with a proper bug report?

I'll send something to bugs@, and put the DSDT.dsl in my home.
 
> > Index: sys/dev/acpi/acpi.c
> > ===================================================================
> > RCS file: /cvs/src/sys/dev/acpi/acpi.c,v
> > diff -u -p -u -p -r1.444 acpi.c
> > --- sys/dev/acpi/acpi.c	24 Mar 2025 09:53:20 -0000	1.444
> > +++ sys/dev/acpi/acpi.c	30 Apr 2025 14:47:29 -0000
> > @@ -253,6 +253,14 @@ acpi_gasio(struct acpi_softc *sc, int io
> >  		else
> >  			iot = sc->sc_iot;
> >  
> > +		/*
> > +		 * Accessing the I/O port 0x80 (POST debug port) on a
> > +		 * MacPro6,1 will crash the system since this port is likely
> > +		 * not implemented.
> > +		 */
> > +		if (address == 0x80 && strcmp(hw_prod, "MacPro6,1") == 0)
> > +			return (-1);
> > +
> >  		if (acpi_bus_space_map(iot, address, len, 0, &ioh) != 0) {
> >  			printf("%s: unable to map iospace\n", DEVNAME(sc));
> >  			return (-1);
> > 
> > 
>