On 22.08.2017 03:09, David Gibson wrote: > On Wed, Aug 16, 2017 at 12:18:07PM +0100, Peter Maydell wrote: >> On 16 August 2017 at 11:51, Paolo Bonzini wrote: >>> On 16/08/2017 10:26, Cornelia Huck wrote: >>>> Prefer to use the tcg accelarator if it is available: This is our only >>>> real smoke test for tcg, and fast enough to use it for that. >>> >>> I'm not sure this is required for 2.10. Yes, it means the coverage from >>> "make check" is worse, but that's it. >> >> Yes, I'd put it under "if we need to roll an rc4 anyway for >> some more significant bug we might as well put this in too, >> but it doesn't merit cutting rc4 by itself." > > It does entirely break "make check" on a ppc host. And that in turn > has held up my testing cycle for a couple of ppc regressions from 2.9 > that I was hoping to squeeze in. Does that change your calculations? Uuh, right. The problem is that we can only run the "pseries" machine with kvm-hv, but not the other ppc machines (and the boot-serial test tries to start a couple of these). So the "accel=tcg:kvm" fix should currently be the best way to ease the situation - since ppc does not support "--disable-tcg" yet, it will always be tested with tcg. But when we support "--disable-tcg" one day on ppc, too, we might have to come up with a more clever solution here instead... Thomas