On 13.06.2018 15:54, Paolo Bonzini wrote: > On 13/06/2018 14:53, Thomas Huth wrote: >> On 13.06.2018 14:48, Paolo Bonzini wrote: >>> On 13/06/2018 07:05, Thomas Huth wrote: >>>> diff --git a/tests/vmgenid-test.c b/tests/vmgenid-test.c >>>> index 8d915c6..4324034 100644 >>>> --- a/tests/vmgenid-test.c >>>> +++ b/tests/vmgenid-test.c >>>> @@ -131,7 +131,7 @@ static void read_guid_from_monitor(QemuUUID *guid) >>>> static char disk[] = "tests/vmgenid-test-disk-XXXXXX"; >>>> >>>> #define GUID_CMD(guid) \ >>>> - "-machine accel=kvm:tcg " \ >>>> + "-accel kvm:tcg " \ >>>> "-device vmgenid,id=testvgid,guid=%s " \ >>> >>> "-accel kvm:tcg" works, but it really shouldn't (and I think we can >>> change it without a deprecation period). The right syntax would be >>> "-accel kvm -accel tcg", so that you can specify options that are valid >>> only for KVM, or onlty for TCG. >> >> I see your point, but this would break these qtests that are trying to >> override the "-machine accel=qtest" from libqtest.c this way... > > The solution would be to put "-accel kvm -accel tcg" before "-accel qtest". > >> and if >> any other tool out there in the wild is already depending on this >> behavior, too, we can not change it so easily anymore. > > I think -accel is new and obscure enough that it's unlikely to be used > with a colon. In my opinion, the risk is worth the benefit of finally > getting a proper interface for accelerators.
I agree with you. Let's try to get this right before it's too late! Thomas
