PowerMac G5 Dual-core boot experiments report

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

PowerMac G5 Dual-core boot experiments report

Arnaud Bergeron
Disclaimer:  This machine is classified as unsupported and this is not
a complaint that things don't work.

I experimented a bit with my new G5 and found these interesting
things.  All the tests were run from the official 3.9 cd.

With a different screen(17in rather than 15in), screen-scrolling is
REALLY slow. Verbose boot takes more than 5 minutes because of all the
scrolling that takes place.  It seem like the kernel is repainting the
whole screen every time a line is added.

The panic described here:
http://marc.theaimsgroup.com/?l=openbsd-misc&m=115618623330732&w=2 has
disappeared with the new screen, possibly due to the lag the scrolling
adds.

Doing a verbose boot (with UKC) allows the machine to complete the
boot without crashing but the kernel thinks the root is wd0a and has
trouble reading from it.  (it does a read timout 2 times and hangs [no
panic]).

From the verbose boot I gathered some potentially useful bits:  The on
board nics (bge) are seen two times but the second time the attachment
fails with "can't find mem space".  I also noted that that the kernel
detects both PCI-X and PCI Express busses while the specs says that
there are only PCI Express busses on this machine.  The PCI-X busses
are detected first.  The failed attachment comes form the PCI Express
busses.

When smu is attached it shows this error "smu0 at mainbus0: too many
fans" and the fans turn off for a while.

The on-board SATA is recognized.

-current (22/08) hangs after printing "WARNING: unable to get
date/time -- CHECK AND RESET THE DATE!"

Also, if somebody wants to work toward support of this model, I might
be able to donate one next summer.  Please reply to me if it is the
case.

Arnaud

Reply | Threaded
Open this post in threaded view
|

Re: PowerMac G5 Dual-core boot experiments report

Mark Kettenis
> Disclaimer:  This machine is classified as unsupported and this is not
> a complaint that things don't work.

But we want to get it working ;-)

> I experimented a bit with my new G5 and found these interesting
> things.  All the tests were run from the official 3.9 cd.

Would be great if you could repeat this with a -current snapshot.  I
think I fixed a few things after 3.9 was released that might affect
your machine.

> With a different screen(17in rather than 15in), screen-scrolling is
> REALLY slow. Verbose boot takes more than 5 minutes because of all the
> scrolling that takes place.  It seem like the kernel is repainting the
> whole screen every time a line is added.

Unaccelerated framebuffers are sloooooow, and this gets worse with
higher resolutions of course.

> The panic described here:
> http://marc.theaimsgroup.com/?l=openbsd-misc&m=115618623330732&w=2 has
> disappeared with the new screen, possibly due to the lag the scrolling
> adds.

Yeah, there was a race in 3.9 that should be fixed now.

> Doing a verbose boot (with UKC) allows the machine to complete the
> boot without crashing but the kernel thinks the root is wd0a and has
> trouble reading from it.  (it does a read timout 2 times and hangs [no
> panic]).

This is booting from CD?  That should boot bsd.rd and use a ramdisk as
root device.

> From the verbose boot I gathered some potentially useful bits:  The on
> board nics (bge) are seen two times but the second time the attachment
> fails with "can't find mem space".  I also noted that that the kernel
> detects both PCI-X and PCI Express busses while the specs says that
> there are only PCI Express busses on this machine.  The PCI-X busses
> are detected first.  The failed attachment comes form the PCI Express
> busses.

Well, there probably is a PCI-X bridge in the machine for some of the
onboard devices.  But of the bge's should not show up twice.

> When smu is attached it shows this error "smu0 at mainbus0: too many
> fans" and the fans turn off for a while.

But they start again after that don't they?  The PowerMac probably has
a few more fans than the iMac G5 on which I wrote the smu(4) driver.

> -current (22/08) hangs after printing "WARNING: unable to get
> date/time -- CHECK AND RESET THE DATE!"

:(  Might be related to the bge's that show up twice.  You could try
disabling bge in UKC.

> Also, if somebody wants to work toward support of this model, I might
> be able to donate one next summer.  Please reply to me if it is the
> case.

Having the hardware would really help a lot.  But as a start it'd help
if you could provide me with a dump of the Open Firmware device tree
of the machine.  Below are the instructions to get the necessary info.

Cheers.

Mark


1. Boot into Open Firware

2. Type:

   " enet:telnet,192.168.0.1" io

   where 192.168.0.1 is the IP address for your Mac.

3. On another box, start script(8)

4. Telnet to the IP address specified above

5. At the Open Firmware prompt type

   dump-device-tree

6. Type

   reset-all

7. Disconnect

8. Send me the output

According to Dale, this doesn't work well on a busy net, so you might
want to make a direct connection.

If you forget 6, you get to power cycle your Mac ;-)