On 25 October 2016 at 08:33, Andrew Jones <drjo...@redhat.com> wrote: > On Mon, Oct 24, 2016 at 11:39:59PM -0500, Wei Huang wrote: >> >> V6->V7: >> >> * change has_pmu variable type from OnOffAuto to Boolean >> >> * only add "pmu" property to CPU under kvm mode, default ON >> > >> > Hmm, if we don't allow the property with TCG then switching a guest from >> > KVM to TCG will require more than just an accelerator switch. That's a >> > bit annoying and I think we'd have to teach it to libvirt too. I'd prefer >> > >> > -M virt-2.8,accel=tcg -cpu cortex-a57 NO NO >> > -M virt-2.8,accel=tcg -cpu cortex-a57,pmu=off NO NO >> > -M virt-2.8,accel=tcg -cpu cortex-a57,pmu=on NO "Warning: PMU not >> > yet supported with TCG" (or something) >> >> I am fine with this request. But note that, if we enforce >> pmu-default=ON, we can't tell "-cpu cortex-a57" apart from >> "cortex-a57,pmu=on", implying that we have to print the warning msg for >> the case of "cortex-a57" as well (this was why we switch to tri-state >> before). To solve this problem, we have to switch from pmu-default=ON to >> pmu-default=OFF under TCG mode, something like: >> >> if (arm_feature(&cpu->env, ARM_FEATURE_PMU)) { >> qdev_property_add_static(DEVICE(obj), >> &arm_cpu_has_pmu_property, &error_abort); >> if (!kvm_enabled()) >> object_property_set_bool(obj, false, "pmu", NULL); >> } >> >> Then we do: >> if (cpu->has_pmu && !kvm_enabled()) { >> cpu->has_pmu = false; >> if (!pmu_warned && !qtest_enabled()) { >> error_report("warning: pmu not supported under TCG"); >> pmu_warned = true; >> } >> } >> >> This will work. Are you and Peter OK with this solution? >> > > Yes, I prefer it to not being able to easily switch a command line from > KVM to TCG. Granted, right now we mostly need to run with -cpu host for > KVM, which also needs to be switched for TCG. However that will hopefully > change with Shannon's work.
If the TCG default is different from the KVM default then you will have problems with switching it anyway, because it will silently change behaviour underfoot. I think we should avoid that... thanks -- PMM