[SeaBIOS] [BUG?] Doesn't retrieve control after coreboot payload return

[Kevin O'Connor]

On Tue, Aug 02, 2016 at 04:53:16PM +0200, Antonello Dettori wrote:
> On 02/08/16 16:39, Kevin O'Connor wrote:
> > On Tue, Aug 02, 2016 at 03:37:44PM +0200, Antonello Dettori wrote:
> > > Hi everyone.
> > > 
> > > I'm currently working on coreboot but I stumbled on a strange SeaBIOS
> > > behaviour.
> > > After executing a payload and returning control to the caller SeaBIOS
> > > crashes.
> > > 
> > > The problem is currently solved by rebooting before the payload returns but
> > > doing so also makes chaining multiple payloads impossible so I'm trying to
> > > look into a solution.
> > The SeaBIOS payload support was not designed to robustly handle
> > returns from a payload.  It will crash if the payload alters the GDT,
> > overwrite the stack, or otherwise messes up the SeaBIOS state.  As far
> > as I know, payloads in general aren't designed to handle returns to
> > their caller.
> Some payloads can handle returns to the previous caller/payload.
> Would it be possible in theory to re-design the payload support so as to
> make it work (i.e. restoring the SeaBIOS state after returning)?
> Are there any reasons why it would be unfeasible?

Depends on what the payload is altering.  Which payload is it?

> > > I tried to increment the debug level to 8 to get more information and the
> > > attached log is what I got.
> > That log isn't from SeaBIOS.  It's unclear what hardware you have.  In
> > general, a serial port is the most robust way of obtaining a log.  See
> > also: http://www.seabios.org/Debugging
> > 
> > -Kevin
> I run the image on QEMU, didn't realise that SeaBIOS probably doesn't
> support QEMU debug port output.
> I'll try on a hardware target later.

SeaBIOS can support the qemu debug port, but it's probably easier to
compile seabios with serial support and instruct qemu to write the
serial log to a file: -serial file:foo.txt

-Kevin