ppc64 snapshot

[Nathan Whitehorn]

On 04/09/10 21:50, Justin Hibbits wrote:
> On Fri, Apr 9, 2010 at 9:20 PM, Nathan Whitehorn 
> <nwhitehorn at freebsd.org <mailto:nwhitehorn at freebsd.org>> wrote:
>
>     On 04/09/10 19:54, Justin Hibbits wrote:
>>     On Thu, Apr 8, 2010 at 9:57 PM, Justin Hibbits
>>     <jrh29 at alumni.cwru.edu <mailto:jrh29 at alumni.cwru.edu>> wrote:
>>
>>         On Tue, Apr 6, 2010 at 1:22 PM, Nathan Whitehorn
>>         <nwhitehorn at freebsd.org <mailto:nwhitehorn at freebsd.org>> wrote:
>>
>>             Justin Hibbits wrote:
>>
>>                 I just got my hands on a dual-core G5 (Late 2005),
>>                 and want to throw
>>                 -CURRENT on it.  Is there a snapshot available with
>>                 the recent ppc64 changes
>>                 that I could test out?
>>
>>                 - Justin
>>                 _______________________________________________
>>                 freebsd-ppc at freebsd.org
>>                 <mailto:freebsd-ppc at freebsd.org> mailing list
>>                 http://lists.freebsd.org/mailman/listinfo/freebsd-ppc
>>                 To unsubscribe, send any mail to
>>                 "freebsd-ppc-unsubscribe at freebsd.org
>>                 <mailto:freebsd-ppc-unsubscribe at freebsd.org>"
>>
>>             I just finished implementing the last missing feature in
>>             the 64-bit PowerPC port, and there are no more
>>             64-bit-specific bugs that I know about. Once M. Warner
>>             Losh's build system changes are in the tree, I will
>>             submit a final patch set for review, and merge it to
>>             head, but the port should be completely usable at this point.
>>
>>             System Compatibility:
>>             - Apple G5 machines
>>
>>             Caveats:
>>             - Do not run ofwdump on an SMP system, as it can cause
>>             hangs (also a 32-bit bug)
>>             - Many ports (e.g. X and GTK) need patches not currently
>>             in the ports tree to compile, since this is a new platform
>>
>>             Instructions:
>>             svn co http://svn.freebsd.org/base/projects/ppc64
>>             cd ppc64
>>             make buildworld buildkernel installkernel installworld
>>             distribution DESTDIR=/path/to/installation
>>             TARGET_ARCH=powerpc64
>>
>>             I would appreciate any feedback or tests, as well as
>>             testing on 32-bit Book-E systems to make sure I did not
>>             break anything. Many thanks to Andreas Tobler for his
>>             tireless testing efforts during development of this port.
>>             -Nathan
>>
>>
>>         I've finally had a chance to test it, but it hangs with the
>>         string
>>
>>         Kernel entry at 0x1034e0...
>>
>>         nothing more.  I tried booting verbose, but that gave
>>         nothing, it looks like it may not even be leaving the loader.
>>
>>         - Justin
>>
>>
>>     I just tried a fresh head boot, and I got the same thing loading
>>     a ppc32 kernel.  Trying with hw.physmem=512M (the machine has 4GB
>>     physical memory) failed as well, and loading a ppc32 kernel from
>>     loader.ppc64 same result.  Any ideas of how to continue debugging
>>     this?
>>
>>     - Justin
>     This sounds like an issue with syscons. Can you try setting
>     hw.syscons.disable=1 from the loader? That should make the kernel
>     fall back to the Open Firmware text console.
>     -Nathan
>
>
> Same result, with both ppc32 and ppc64 kernels.  Should I just start 
> riddling the kernel with printf()s to track this down?
>

That is really strange. One of the very first things the kernel does is 
to print out some lines from KDB.

You can try to add an OF_printf() to the line right after OF_bootstrap() 
in aim/machdep.c. That is the earliest you can use Open Firmware and get 
output from the kernel. But I suspect it's not even getting there.

The entry point looks a little wonky to me -- mine is 100160, and it 
should always be somewhere around there. Could you check if the printed 
entry point address corresponds to the first instructions in the text 
segment with objdump? You can use make buildenv TARGET_ARCH=powerpc64 to 
get a toolchain and objdump for PPC64 executables.
-Nathan