Anthony Liguori <anth...@codemonkey.ws> writes:
> Markus Armbruster <arm...@redhat.com> writes:
>
>> Very basic smoke test: start QEMU with -monitor stdio, quit immediately.
>> Wouldn't it be nice if that worked for all targets and machine types?
>>
>> Many targets have mandatory options (fun oxymoron), such as -kernel or
>> -pflash. Can't stop me, I just try a bunch until something works.
>>
>> Many targets expect various files to be present, and some of them need
>> to have the right size. Can't stop me, I hack up the file loaders until
>> it works (silly patch appended). To do this right, we'd need the
>> required files or suitable mock-ups in-tree.
>
> I attempted something similar in the past and ran into similar results.
>
>>
>> Test script:
>>
>> #!/bin/sh
>> for i in ../qemu/bld/*-softmmu/qemu-system-*
>> do
>> echo "= $i ="
>> for m in `$i -M help | sed -n '2,$s/ .*//gp'`
>> do
>> echo "== $m =="
>> for k in "" "-kernel /dev/null" "-pflash /dev/null" "-pflash /dev/null
>> -pflash /dev/null -kernel /dev/null"
>> do
>> echo "=== ${k:-(default)} ==="
>> if echo q | QEMU_AUDIO_DRV=none $i -S -vnc :0 -M $m $k -monitor
>> stdio | fgrep -q '(qemu)'
>> then break
>> else false
>> fi
>> done
>> if [ $? -eq 0 ]
>> then echo "*** Success $k ***"
>> else echo '*** Fail'
>> fi
>> done
>> done
>>
>> Summary of results:
>>
>> * Bad unexplained
>>
>> qemu-system-arm lm3s811evb
>> qemu-system-arm lm3s6965evb
>> qemu-system-arm: /work/armbru/qemu/hw/qdev.c:310: qdev_get_gpio_in:
>> Assertion `n >= 0 && n < dev->num_gpio_in' failed.
>>
>> qemu-system-ppc64 prep
>> qemu: hardware error: Unknown device 'i82378' for bus 'PCI'
>>
>> qemu-system-ppcemb ref405ep
>> qemu-system-ppcemb taihu
>> Unable to find PowerPC 405ep CPU definition
>>
>> qemu-system-ppcemb mac99
>> qemu-system-ppcemb g3beige
>> qemu-system-ppcemb prep
>> Unable to find PowerPC CPU definition
>>
>> qemu-system-xtensaeb lx60
>> qemu-system-xtensaeb lx200
>> qemu-system-xtensaeb sim
>> Unable to find CPU definition
>>
>> I'm not saying these are all busted. If you know how to "start to
>> monitor" one of these, let us know.
>
> Perhaps we could add a QEMUMachine parameter that indicates that the
> machine doesn't start without special options.
Recommend to make it a string that lists the mandatory options.
> At least a handful of these machines cannot be run without the use of
> non-free binaries firmware :-(
Yes, and that's bad.
However, my test isn't trying to execute any guest code. It *should* be
possible to pull off with mocked-up firmware.
The mock-ups could then serve as documentation on what firmware binaries
are required.
