Re: [PATCH v2 1/3] KVM: nVMX: Fix nested VPID vmx exec control
From: Paolo Bonzini
Date: Tue Mar 21 2017 - 12:22:49 EST
On 21/03/2017 05:18, Wanpeng Li wrote:
> From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
>
> This can be reproduced by running kvm-unit-tests/vmx.flat on L0 w/ vpid disabled.
>
> Test suite: VPID
> Unhandled exception 6 #UD at ip 00000000004051a6
> error_code=0000 rflags=00010047 cs=00000008
> rax=0000000000000000 rcx=0000000000000001 rdx=0000000000000047 rbx=0000000000402f79
> rbp=0000000000456240 rsi=0000000000000001 rdi=0000000000000000
> r8=000000000000000a r9=00000000000003f8 r10=0000000080010011 r11=0000000000000000
> r12=0000000000000003 r13=0000000000000708 r14=0000000000000000 r15=0000000000000000
> cr0=0000000080010031 cr2=0000000000000000 cr3=0000000007fff000 cr4=0000000000002020
> cr8=0000000000000000
> STACK: @4051a6 40523e 400f7f 402059 40028f
>
> We should hide and forbid VPID in L1 if it is disabled on L0. However, nested VPID
> enable bit is set unconditionally during setup nested vmx exec controls though VPID
> is not exposed through nested VMX capablity. This patch fixes it by don't set nested
> VPID enable bit if it is disabled on L0.
>
> Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> Cc: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
> Fixes: 5c614b3583e (KVM: nVMX: nested VPID emulation)
> Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> ---
> arch/x86/kvm/vmx.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 98e82ee..8795a70 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2753,7 +2753,6 @@ static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
> SECONDARY_EXEC_RDTSCP |
> SECONDARY_EXEC_DESC |
> SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
> - SECONDARY_EXEC_ENABLE_VPID |
> SECONDARY_EXEC_APIC_REGISTER_VIRT |
> SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
> SECONDARY_EXEC_WBINVD_EXITING |
> @@ -2781,10 +2780,12 @@ static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
> * though it is treated as global context. The alternative is
> * not failing the single-context invvpid, and it is worse.
> */
> - if (enable_vpid)
> + if (enable_vpid) {
> + vmx->nested.nested_vmx_secondary_ctls_high |=
> + SECONDARY_EXEC_ENABLE_VPID;
> vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
> VMX_VPID_EXTENT_SUPPORTED_MASK;
> - else
> + } else
> vmx->nested.nested_vmx_vpid_caps = 0;
>
> if (enable_unrestricted_guest)
>
Applied patch 1 and 3 to kvm/master (well, locally for now).
Paolo