On Tue, Nov 5, 2024 at 10:33 AM Xiaoyao Li <xiaoyao.li@xxxxxxxxx> wrote:
>
> On 11/5/2024 5:12 PM, Paolo Bonzini wrote:
> > On 11/5/24 07:24, Xiaoyao Li wrote:
> >> Otherwise, it gets warnings like below when number of vcpus > 1:
> >>
> >> warning: TDX enforces set the feature: CPUID.01H:EDX.ht [bit 28]
> >>
> >> This is because x86_confidential_guest_check_features() checks
> >> env->features[] instead of the cpuid date set up by cpu_x86_cpuid()
> >>
>
> It seems I mixed it up with no_autoenable_flags. /faceplam
>
> CPUID_HT doesn't get enabled by x86_cpu_expand_features() for "-cpu
> host/max". It won't be filtered by x86_cpu_filter_features() either
> because QEMU sets it in kvm_arch_get_supported_cpuid().
>
> yes, the comment is wrong and comment needs to be dropped. The code can
> be move up to just below x86_cpu_expand_features() or inside it?
Inside it seems okay, and you can then remove it from cpu_x86_cpuid().
However, let's also add qemu_early_init_vcpu() to the realize function
from all targets, and remove
MachineState *ms = MACHINE(qdev_get_machine());
cpu->nr_cores = machine_topo_get_cores_per_socket(ms);
cpu->nr_threads = ms->smp.threads;
from qemu_init_vcpu(). You can resend patches 58 and 59 separately
from the TDX series.
Paolo
