Re: ofwboot.xcf problem (port-macppc/44895)

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

On Sun, Jun 04, 2017 at 19:22:56 +0300, Valery Ushakov wrote:

> On Sun, Jun 04, 2017 at 11:06:41 -0400, Michael wrote:
> 
> > On Fri, 2 Jun 2017 23:33:20 +0300
> > Valery Ushakov <uwe%stderr.spb.ru@localhost> wrote:
> > 
> > > On Fri, Jun 02, 2017 at 21:08:24 +0200, Havard Eidnes wrote:
> > > 
> > > > >> The workaround is already in -6 and -7 and it's called ofwboot.elf :)  
> > > > 
> > > > Well, the installation notes say
> > > > 
> > > >   Note:
> > > >   ofwboot.elf is obsoleted.  All users should be using ofwboot.xcf
> > > >   instead of ofwboot.elf now.
> > > > 
> > > > and there's no other mention of ofwboot.elf.  If that note is no
> > > > longer true, the documentation should be updated.  
> > > 
> > > Do we have a portmaster that is not AWOL? :)
> > 
> > Awful, but not AWOL ;)
> > I'll update the docs as soon as
> > - there is clear consent what works where - to me it looks like
> >   ofwboot.xcf works on old world machines, some new world machines need
> >   ofwboot.elf. Does that match everyone's experiences? Do we need to
> >   put ofwboot.elf on the install ISOs?
> 
> Sevan says that his mini is happy with the xcoff, which makes the
> puzzle even more interesting.
> 
> Debugging this without a serial console is a bit of a pain.  E.g. I
> have some free time now, but I'm physically away from the machine, so
> I can't do anyhting.  IIRC there's an option to replace the modem card
> with normal serial port.  Is it standard or does that require a custom
> board?
> 
> Compilng ofwboot with libsa's alloc seems to work on the mini.  We
> might want to commit this and pull up to -6 and -7.  Portmaster's
> call, I'd say.
> 
> Installation cd has ofwboot and ofwboot.xcf in the root, but
> ofwboot.xcf and ofwboot.elf in /macppc/installation - is that
> intended?  We should probably put all three in both places. 

Quoting my update to http://gnats.netbsd.org/44895

I've just committed a bit of trivial code that I had sitting in my
tree to use libsa alloc.c instead of ofwboot's own ./alloc.c.  It's
not yet enabled and the same object code is generated for now.  You
can edit the makefile to change to the libsa allocator.

The new (disabled) code makes ofwboot.xcf happy, apparently.  Since
the root cause of the bug was never discovered it may be just masking
the bug, but whatever :)


./alloc.c is very naive as it rounds every allocation up to the page
size that it gets with OF_claim.

The new code OF_claim's HEAP_SIZE heap (default 0x20000) and lets
libsa manage it.

Since libsa allocator cannot request additional memory anyway we may
as well just use an array in .bss for heap instead.

But I wanted to commit what I had and what's known to fix or at least
mask the issue and then someone who's interested can experiment and
test on other machines too.


To enabled it and pull it up, it would be nice if people could test it
on various machines first.

It would also be nice if someone can find the root cause of the
problem.  Part of the problem might be that our OF_chain doesn't
really call "chain" method (since 1.1), so it's a bit of mess overall.

-uwe