Re: [PATCH v7 3/8] KVM: VMX: Detect Tertiary VM-Execution control when setup VMCS config

From: Zeng Guang
Date: Sat Apr 02 2022 - 08:58:58 EST



On 4/1/2022 6:41 AM, Sean Christopherson wrote:
On Fri, Mar 04, 2022, Zeng Guang wrote:
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index c569dc2b9192..8a5713d49635 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -2422,6 +2422,21 @@ static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
return 0;
}
+static __init int adjust_vmx_controls_64(u64 ctl_min, u64 ctl_opt,
I slightly prefer controls64 over controls_64. As usual, KVM is inconsistent as
a whole, but vmcs_read/write64 omit the underscore, so we can at least be somewhat
consistent within VMX.

+ u32 msr, u64 *result)
+{
+ u64 allowed1;
+
+ rdmsrl(msr, allowed1);
+
+ /* Ensure minimum (required) set of control bits are supported. */
+ if (ctl_min & ~allowed1)
Eh, just drop @ctl_min. Practically speaking, there is zero chance tertiary
controls or any other control of this nature will ever be mandatory. Secondary
controls would fall into the same boat, but specifying min=0 allows it to share
helpers, so it's the lesser of evils.

With the error return gone, this can be

static __init u64 adjust_vmx_controls64(u64 ctl_opt, u32 msr)
{
u64 allowed;

rdmsrl(msr, allowed);

return ctl_opt & allowed;
}

Make sense. I will change it.  Thanks.


Alternatively, we could take the control-to-modify directly and have no return,
but I like having the "u64 opt = ..." in the caller.