Re: Current ofwboot failure

[Michael <macallan1888%gmail.com@localhost>]

Hello,

On Sun, 4 Dec 2022 11:21:32 +0100
Martin Husemann <martin%duskware.de@localhost> wrote:

> current ofwboot fails for me on a G5 powermac. It loads the kernel
> (skipping several memory regions beyound 4G physical address) and
> then lists two available regions, jumps to the kernel and crashes
> (manual transcript, please excuse the left out parts):
> 
> avail region 2 start ...
> Invalid memory access at %SRR0: ...00134b3c  %SRR1: 100000000000.0000003030

This looks like fallout from the locking-around-console-output dance,
which now needs things like curcpu() and the like to work. Problem with
that - macppc sets up its console *very* early compared to other ports.

> That would be:
> 
> (gdb) list *(0x00134b3c)
> 0x134b3c is in splraise (../../../../arch/powerpc/pic/intr.c:622).
> 617     splraise(int ncpl)
> 618     {
> 619             struct cpu_info *ci = curcpu();
> 620             int ocpl;
> 621     
> 622             if (ncpl == ci->ci_cpl)
> 623                     return ncpl;
> 624             REORDER_PROTECT();
> 625             ocpl = ci->ci_cpl;
> 626             KASSERT(ncpl < NIPL);
> 
> 
> Using an older ofwboot the same kernel just boots fine.
> Only relevant recent changes were in libsa it seems (but none looks
> related to this). Maybe something in the kernel changed and older ofwboot
> just is lucky?

No idea how ofwboot plays into this, my guess is it got lucky.
I started converting early debug output to ofprint() in order to avoid this.

have fun
Michael