Re: [PATCH] KVM: VMX: Do not trap VMFUNC instructions for L1 guests.

From: Yu Zhang
Date: Tue Nov 08 2022 - 05:42:01 EST


On Mon, Nov 07, 2022 at 06:20:23PM +0100, Paolo Bonzini wrote:
> On 11/7/22 09:27, Yu Zhang wrote:
> > VMFUNC is not supported for L1 guests, and executing VMFUNC in
> > L1 shall generate a #UD directly. Just disable it in secondary
> > proc-based execution control for L1, instead of intercepting it
> > and inject the #UD again.
> >
> > Signed-off-by: Yu Zhang<yu.c.zhang@xxxxxxxxxxxxxxx>
>
> Is this for TDX or similar? The reason for a patch should be mentioned in
> the commit message.

Thanks for your quick reply, Paolo.

It is not a new feature. Just a clean up for VMFUNC, which is not
supported by KVM for L1 guest.

According to Intel SDM 25.5.6.2 - "General Operation of the VMFUNC
Instruction", The VMFUNC instruction causes an invalid-opcode exception
(#UD) if the “enable VM functions” VM-execution controls is 0 or the
value of EAX is greater than 63 (only VM functions 0–63 can be enable).
Otherwise, the instruction causes a VM exit if the bit at position
EAX is 0 in the VM-function controls (the selected VM function is not
enabled)

And since KVM only provides emulation of VMFUNC for nested guests,
it is uncessary for KVM to intercept it and reinject a #UD. So just
disable VMFUNC in VM-execution control for L1 guests.

But please feel free to educate me if I missed some backgrounds about
why this is enabled in the first place. Thanks!

B.R.
Yu