@@ -473,10 +473,17 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp)
trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu));
+ /*
+ * tdx_pre_create_vcpu() may call cpu_x86_cpuid(). It in turn may call
+ * kvm_vm_ioctl(). Set cpu->kvm_state in advance to avoid NULL pointer
+ * dereference.
+ */
+ cpu->kvm_state = s;
ret = kvm_arch_pre_create_vcpu(cpu);
if (ret < 0) {
error_setg_errno(errp, -ret,
"kvm_init_vcpu: kvm_arch_pre_create_vcpu() failed");
+ cpu->kvm_state = NULL;
goto err;
}
@@ -484,11 +491,11 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp)
if (ret < 0) {
error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)",
kvm_arch_vcpu_id(cpu));
+ cpu->kvm_state = NULL;
goto err;
}
cpu->kvm_fd = ret;
- cpu->kvm_state = s;
cpu->vcpu_dirty = true;
cpu->dirty_pages = 0;
cpu->throttle_us_per_full = 0;
@@ -2197,6 +2197,14 @@ int kvm_arch_init_vcpu(CPUState *cs)
return r;
}
+int kvm_arch_pre_create_vcpu(CPUState *cpu)
+{
+ if (is_tdx_vm())
+ return tdx_pre_create_vcpu(cpu);
+
+ return 0;
+}
+
int kvm_arch_destroy_vcpu(CPUState *cs)
{
X86CPU *cpu = X86_CPU(cs);
@@ -7,3 +7,8 @@ int tdx_kvm_init(MachineState *ms, Error **errp)
{
return -EINVAL;
}
+
+int tdx_pre_create_vcpu(CPUState *cpu)
+{
+ return -EINVAL;
+}
@@ -458,6 +458,38 @@ int tdx_kvm_init(MachineState *ms, Error **errp)
return 0;
}
+int tdx_pre_create_vcpu(CPUState *cpu)
+{
+ MachineState *ms = MACHINE(qdev_get_machine());
+ X86CPU *x86cpu = X86_CPU(cpu);
+ CPUX86State *env = &x86cpu->env;
+ struct kvm_tdx_init_vm init_vm;
+ int r = 0;
+
+ qemu_mutex_lock(&tdx_guest->lock);
+ if (tdx_guest->initialized) {
+ goto out;
+ }
+
+ memset(&init_vm, 0, sizeof(init_vm));
+ init_vm.cpuid.nent = kvm_x86_arch_cpuid(env, init_vm.entries, 0);
+
+ init_vm.attributes = tdx_guest->attributes;
+ init_vm.max_vcpus = ms->smp.cpus;
+
+ r = tdx_vm_ioctl(KVM_TDX_INIT_VM, 0, &init_vm);
+ if (r < 0) {
+ error_report("KVM_TDX_INIT_VM failed %s", strerror(-r));
+ goto out;
+ }
+
+ tdx_guest->initialized = true;
+
+out:
+ qemu_mutex_unlock(&tdx_guest->lock);
+ return r;
+}
+
/* tdx guest */
OBJECT_DEFINE_TYPE_WITH_INTERFACES(TdxGuest,
tdx_guest,
@@ -470,6 +502,8 @@ static void tdx_guest_init(Object *obj)
{
TdxGuest *tdx = TDX_GUEST(obj);
+ qemu_mutex_init(&tdx->lock);
+
tdx->attributes = 0;
}
@@ -17,6 +17,9 @@ typedef struct TdxGuestClass {
typedef struct TdxGuest {
ConfidentialGuestSupport parent_obj;
+ QemuMutex lock;
+
+ bool initialized;
uint64_t attributes; /* TD attributes */
} TdxGuest;
@@ -29,5 +32,6 @@ bool is_tdx_vm(void);
int tdx_kvm_init(MachineState *ms, Error **errp);
void tdx_get_supported_cpuid(uint32_t function, uint32_t index, int reg,
uint32_t *ret);
+int tdx_pre_create_vcpu(CPUState *cpu);
#endif /* QEMU_I386_TDX_H */
Invoke KVM_TDX_INIT in kvm_arch_pre_create_vcpu() that KVM_TDX_INIT configures global TD state, e.g. the canonical CPUID config, and must be executed prior to creating vCPUs. Use kvm_x86_arch_cpuid() to setup the CPUID settings for TDX VM and tie x86cpu->enable_pmu with TD's attributes. Note, this doesn't address the fact that QEMU may change the CPUID configuration when creating vCPUs, i.e. punts on refactoring QEMU to provide a stable CPUID config prior to kvm_arch_init(). Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> --- accel/kvm/kvm-all.c | 9 ++++++++- target/i386/kvm/kvm.c | 8 ++++++++ target/i386/kvm/tdx-stub.c | 5 +++++ target/i386/kvm/tdx.c | 34 ++++++++++++++++++++++++++++++++++ target/i386/kvm/tdx.h | 4 ++++ 5 files changed, 59 insertions(+), 1 deletion(-)