Re: Crashes with qemu-system-ppc64

[Peter Maydell]

On Thu, 25 Mar 2021 at 12:57, Mark Cave-Ayland
<mark.cave-ayland@ilande.co.uk> wrote:
> Thanks for the analysis: I can certainly see how the above commit would have 
> changed
> the behaviour. Looking at hw/ppc/e590plat.c in e500plat_machine_class_init() 
> I see
> that line 101 reads "machine_class_allow_dynamic_sysbus_dev(mc, 
> TYPE_ETSEC_COMMON);"
> which looks like it is intended to add a class restriction to this 
> functionality.

Aha. I thought there was a restriction somewhere but hadn't found that
function. I think the device plug callback functions should check that
list rather than saying "any TYPE_SYS_BUS_DEVICE is fine".

> In machine_initfn() a callback for machine_init_notify() is added to perform 
> the
> check but the macio-oldworld device is realized first

Also, this check is only for "did the user dynamically create a sysbus
device that isn't an allowed one", which is a necessary check, but
not quite the same thing as "is this device we're looking at one we want to
assume belongs to the platform bus".

I'll put together a patch...

thanks
-- PMM