On Mon, Dec 05, 2022, Isaku Yamahata wrote:

On Wed, Nov 30, 2022 at 11:09:15PM +0000,

quoted

index 66f16458aa97..3571bc968cf8 100644

--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c

@@ -9277,10 +9277,36 @@ static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
 	kvm_pmu_ops_update(ops->pmu_ops);
 }
 
+struct kvm_cpu_compat_check {
+	struct kvm_x86_init_ops *ops;
+	int *ret;

minor nitpick: just int ret. I don't see the necessity of the pointer.
Anyway overall it looks good to me.

...

quoted

@@ -9360,6 +9386,14 @@ static int __kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
 	if (r != 0)
 		goto out_mmu_exit;
 
+	c.ret = &r;
+	c.ops = ops;
+	for_each_online_cpu(cpu) {
+		smp_call_function_single(cpu, kvm_x86_check_cpu_compat, &c, 1);
+		if (r < 0)

Here it can be "c.ret < 0".

No, because the below goto leads to "return r", i.e. "c.ret" needs to be propagated
to "r".  That's why the code does the admittedly funky "int *ret" thing.

FWIW, this gets cleanup in the end.  "struct kvm_cpu_compat_check" goes away and
"&r" is passed directly to kvm_x86_check_cpu_compat.

quoted

+			goto out_hardware_unsetup;