Re: [Qemu-devel] Missing interrupts on g3beige and mac99 machines

[Mark Cave-Ayland]

On 16/04/15 19:03, Cormac O'Brien wrote:

> While trying to implement OS 9 support on QEMU, I've run into the following
> errors from the operating system:
> 
> g3beige:
>     ******************* MacOS: Fatal Error!  (0xF3B37FDB) *******************
>     Official Apple copyright message missing.
>     ******************* MacOS: Fatal Error!  (0xF3B37FDB) *******************
>     MacOS: missing cpu "reservation-granule-size" property.
>     MacOS: unable to find a usable NVRAM partition - using offset 0x1400.
>     ******************* MacOS: Fatal Error!  (0xF3B37FDB) *******************
>     MacOS: Neither RTAS nor plug-ins found and installed!

Heh that's interesting - so do this mean that g3beige needs a /rtas node?

>     MacOS: Boot Failure!  (0xF3C481F6)
>     MacOS: unable to find an interrupt controller node.
> 
> mac99:
>     ******************* MacOS: Fatal Error!  (0xF3B37FDB) *******************
>     Official Apple copyright message missing.
>     ******************* MacOS: Fatal Error!  (0xF3B37FDB) *******************
>     MacOS: missing cpu "reservation-granule-size" property.
>     MacOS: unable to find a usable NVRAM partition - using offset 0x1400.
>     MacOS: Boot Failure!  (0xF3C481F6)
>     MacOS: device has < 1 interrupts!
> 
> The g3beige machine doesn't list an interrupt controller in either '/aliases'
> or '/pci/mac-io', which might explain why Mac OS can't find it.
> 
> The mac99 machine does list a PIC:
> 
>     0 > dev /pci/mac-io/interrupt-controller  ok
>     0 > .properties
>     name                      "interrupt-controller"
>     device_type               "open-pic"
>     compatible                "chrp,open-pic"
>     built-in                  <empty>
>     reg                       00040000   00040000
>     #interrupt-cells          2
>     #address-cells            0
>     interrupt-controller      <empty>
>     clock-frequency           3f940a
> 
> This location and property list is identical to the Mac-on-Linux device tree
> and the device tree of a UniNorth machine that Alex sent me (the tree, not the
> machine). So it seems that the mac99 PIC is implemented correctly but has no
> interrupts registered.

Depending upon the OS, devices are normally detected via the
"device_type" and "compatible" properties. Some bad OSs will attempt to
locate a device using a fixed path, or the "name" property or other
methods that may not match the device tree generated by OpenBIOS.

If MOL is working then it shouldn't be too difficult to figure out the
differences between these two sets of device nodes for the interrupt
controllers.


ATB,

Mark.