On Wed, 4 Jul 2018, Sebastian Bauer wrote:
Am 2018-07-04 06:50, schrieb Mark Cave-Ayland:
Either to give up the vga cirrus preference or to apply the same as was
done with the spapr machine. I would prefer the first variant but it would
also cause some differences on other machines.
I understand that the decision was taken to add VGA cirrus to the PPC
build and therefore accept the default VGA device change, however your
the commit message only mentions spapr and not all the other PPC
machines whose defaults are also potentially affected:
$ ./qemu-system-ppc -M ?
Supported machines are:
40p IBM RS/6000 7020 (40p)
bamboo bamboo
g3beige Heathrow based PowerMAC (default)
mac99 Mac99 based PowerMAC
mpc8544ds mpc8544ds
none empty machine
ppce500 generic paravirt e500 platform
prep PowerPC PREP platform
ref405ep ref405ep
sam460ex aCube Sam460ex
taihu taihu
virtex-ml507 Xilinx Virtex ML507 reference design
I'm not able to personally validate all of these, but a quick check
shows that at least both PReP machines (-M prep and -M 40p) are also
broken by this patch which tells me it hasn't had much testing.
David, do you know what the defaults are for the other PPC machines in
order for Sebastian to test and send a follow-up patch?
I don't know much how to validate these machines (I did a 'make check' but it
does not show up any problems) and what impact these change has. But I
suggested in the note section (which is not the commit message) that some of
the machines properly behave different now (OTOH CirrusGD is a vga card
nonetheless, I would have expected that at least the Mac can deal with that
card). This can either be fixed by following the 'spapr' approach (as it
explicitly forbids this card as the primary display and was therefore visible
immediately) or by generally rethinking whether CirrusGD really should be the
preferred card over std vga (after all it is considered as obsolete so why
should it be preferred?). I think it would be nice if a consensus can be
found about the "correct approach" way to fix it. I personally think that
adding more exceptions is not the right way :)
Maybe including Gerd Hoffmann in this discussion would help finding a
preferred approach.
Last but not least, all of the targets should still work as before if you use
-vga std option.
That's not an acceptable solution because it wasn't needed before and now
people who depended on that would need to change their ways. I think
changing the default to std at least for ppc machines if that's possible
before adding cirrus would be a good solution but I don't know the
details. Although I think this patch is not critical so if it's not easy
to fix now it could also be reverted and then we can address it in next
devel cycle in a better way.
Regards,
BALATON Zoltan
