Re: unable to boot amd64-uefi-install from USB stick on a MacBook2,1

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Fri, Jan 17, 2020 at 08:53:39PM -0800, Greg A. Woods wrote:
> At Thu, 9 Jan 2020 01:19:39 +0100, Joerg Sonnenberger <joerg%bec.de@localhost> wrote:
> Subject: Re: unable to boot amd64-uefi-install from USB stick on a MacBook2,1
> >
> > On Wed, Jan 08, 2020 at 04:08:40PM -0800, Greg A. Woods wrote:
> > > However in searching related things I just came across a reminder about
> > > memory issues on this particular model of MacBook2,1.  It originally
> > > shipped with 2x1GB memory modules.  At some point I upgraded it to 2x2GB
> > > modules.  Under MacOS it only uses 3GB for some reason (though the
> > > hardware info clearly shows 4GB installed).
> >
> > Try building a kernel with PHYSMEM_MAX_ADDR=3096 in that case?
> 
> Thank you for the suggestion -- I was trying to remember what to twiddle
> to do this, though I had not yet looked at a config file for an example.
> 
> So, between bouts of shovelling snow off the driveway I tried the above,
> and with 2048, as well as (separately) PHYSMEM_MAX_SIZE=2048.
> 
> All with no difference from a plain kernel -- the machine still hangs
> hard immediately after clearing the screen.
> 
> If I were to post images or a transcription of the output of, for
> example, the EFI shell's "memmap" or other such commands, would that be
> of any help to anyone who might be able to diagnose this further?
> 
> For example it does seem to be a firmware and/or hardware limitation
> that the machine only makes available 3GB of the 4GB installed.
> "memmap" says (in part) "reserved: 4,384 Pages (17,956,864)" and
> "available:  770,146 Pages (3,154,518,016)" and "Total Memory: 3,054 MB
> (3,203,006,464) Bytes", for example.

On x86 memory mapped IO was put into a window somewhere between 3-3.5GB
and 4GB in terms of physical addresses, so that a 32bit operating system
can still access all devices without requiring 64bit DMA and using PAE.
That is still ignoring limitations of individual DMA engines of specific
PCI devices. I wouldn't be surprised if the hardware simply doesn't
properly deal with more than 3GB RAM and runs into various limits. The
PHYSMEM_MAX_SIZE would have helped if there are unknown limits beyond
that point, but it might also be a problem of MMIO still using the
fourth GB incorrectly. Difficult to say or analyze and generally, not a
very good use of time. Do you by chance still have the original 1GB
module? Might be easiest to just use the 3GB directly.

Joerg