Re: Successful install of a current NetBSD/bebox system

On 9 June 2014 13:33, Frank Wille <frank%phoenix.owl.de@localhost> wrote:

On Sun, 1 Jun 2014 23:46:38 +0100
David Brownlee <abs%absd.org@localhost> wrote:

> > 1. Reboot. All blinken lights (left and right) are activated. Then the
> > system hangs, until I manually press the reset buttons.
>
> Linux ran on the BeBox at one point - I wonder if it would be easy to
> check what it did to reboot?

I was unable to find the source for Linux on BeBox. And in recent source
archives it is no longer present.

It does appear to have completely vanished. Not an issue since things started being kept in a revision control system, but Linux/BeBox pre-dated git... :/

But I guess there is no way to reset the system in software. There is a
board register to assert /HRESET and /SRESET to CPU 1, but not to CPU 0.

Maybe I can reset as many hardware registers as possible and then jump
into the boot ROM's reset vector?

Sounds like its worth a try...

[...]

> > 6. mcclock(4) doesn't work correctly. The time after a cold-start is
> > wrong (e.g. "WARNING: preposterous TOD clock time"). Maybe I need a
> > new battery, but IIRC it worked under BeOS.
>
> It could be oddly parsed. If NetBSD writes something back to the TOD
> does BeOS see it as gibberish?

I have to replace the disk to start BeOS. Didn't have the time to do that.

Can you have two disks hooked up in it?

I inserted some debugging output into sys/dev/ic/mc146818.c and saw
the driver was reading a time like 20:17:87. Maybe there is a problem
with a mis-detected BCD mode (87 would be 0x57)?

But before investigating any further I should check the RTC under BeOS
again and replace the battery.

> > 7. X11/Xorg doesn't work yet. Cannot map I/O space. But the correct
> > driver (S3) is recognized:
> > [ 888.123] (WW) xf86EnableIO -1
> > [ 888.126] (II) xf86EnableIO: ffffffff
> > [ 888.128] (WW) Can't map IO space!

> > [...]

>
> Wouldn't it be mem space rather than io space on a powerpc?

Sounds sensible. Probaby macallan@ would bring up X11 in no time. ;)

He would have accelerated anti-aliased fonts and alpha blended blitting in no time :)

> > 9. Multiprocessor support. I compiled the GENERIC.MP kernel, but the
> > boot loader doesn't want to load it. A renamed GENERIC kernel works,
> > which is strange.
> > Boot: /dev/disk/scsi/000/0_0:netbsdmp
> > Loading /dev/disk/scsi/000/0_0:netbsdmp
> > illegal request, data = 0 0 0 0 24 0 0 cb 0 1 0 0 0 0 0 0 ff ff ff
> > ff 0 0 0 0
> > PHASE MSGIN: siop_clearfifo...
> > invalid status code 128
> > generic HBA error
> > siop_intr: I shouldn't be there !
>
> Could this be related to kernel size? Could you strip stuff out of
> GENERIC.MP until its smaller than GENERIC as a quick test?

Tried that, but didn't help. Now I even got this error when loading
a modified single-CPU kernel. :|

I noticed in siop.c that the siop_clearfifo() function was not yet
implemented. Probably this is the reason for all the following errors.

After implementing it, it looks like this:

illegal request, data = 0 0 0 0 24 0 0 cb 0 1 0 0 0 0 0 0 ff ff ff ff 0 0 0 0

DMA FIFO not empty!
illegal request, data = 0 0 0 0 24 0 0 ce 0 3 0 0 0 0 0 0 ff ff ff ff 0 0 0 0
open /dev/disk/scsi/000/0_0:netbsdtst: Input/ouput error

That's better (I'm printing "DMA FIFO not empty" inside siop_clearfifo)
although there is still a problem.

Maybe something need time to settle? (waving of hands at this point)

I have no idea about the other points yet. So no comment from me at this
time. I will continue my investigations whenever I find some time for it.

Good hunting! :)