Patchwork KVM: x86: Work around buggy MPX platform

login
register
mail settings
Submitter Liu, Jinsong
Date March 10, 2014, 8:48 a.m.
Message ID <DE8DF0795D48FD4CA783C40EC82923350155757D@SHSMSX152.ccr.corp.intel.com>
Download mbox | patch
Permalink /patch/328511/
State New
Headers show

Comments

Liu, Jinsong - March 10, 2014, 8:48 a.m.
From 5854070994c5002b3a37577165ed3e82f36f712d Mon Sep 17 00:00:00 2001
From: Liu Jinsong <jinsong.liu@intel.com>
Date: Sat, 8 Mar 2014 04:40:02 +0800
Subject: [PATCH] KVM: x86: Work around buggy MPX platform

Work around buggy MPX platform which support MSR_IA32_BNDCFGS
but has issue at, say, VMX ucode.

Signed-off-by: Liu Jinsong <jinsong.liu@intel.com>
---
 arch/x86/kvm/x86.c |   14 +++++++++++++-
 1 files changed, 13 insertions(+), 1 deletions(-)
Paolo Bonzini - March 10, 2014, 8:53 a.m.
Il 10/03/2014 09:48, Liu, Jinsong ha scritto:
> From 5854070994c5002b3a37577165ed3e82f36f712d Mon Sep 17 00:00:00 2001
> From: Liu Jinsong <jinsong.liu@intel.com>
> Date: Sat, 8 Mar 2014 04:40:02 +0800
> Subject: [PATCH] KVM: x86: Work around buggy MPX platform
>
> Work around buggy MPX platform which support MSR_IA32_BNDCFGS
> but has issue at, say, VMX ucode.
>
> Signed-off-by: Liu Jinsong <jinsong.liu@intel.com>
> ---
>  arch/x86/kvm/x86.c |   14 +++++++++++++-
>  1 files changed, 13 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 1e91a24..1fc184d 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -3936,6 +3936,16 @@ static void kvm_init_msr_list(void)
>  	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
>  		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
>  			continue;
> +
> +		/*
> +		 * Work around some buggy MPX platform which support
> +		 * MSR_IA32_BNDCFGS but has issue at, say, VMX ucode.
> +		 */
> +		if ((msrs_to_save[i] == MSR_IA32_BNDCFGS) &&

Too many parentheses in this "if".


> +			(kvm_x86_ops->mpx_supported   ?
> +			!kvm_x86_ops->mpx_supported() : 1))

Better: "!kvm_x86_ops->mpx_supported || !kvm_x86_ops->mpx_supported()" 
but I'm leaning towards adding an implementation of mpx_supported to SVM 
too (it will always return false).

> +				continue;
> +
>  		if (j < i)
>  			msrs_to_save[j] = msrs_to_save[i];
>  		j++;
> @@ -5576,9 +5586,11 @@ int kvm_arch_init(void *opaque)
>  		goto out_free_percpu;
>
>  	kvm_set_mmio_spte_mask();
> -	kvm_init_msr_list();
>
>  	kvm_x86_ops = ops;
> +
> +	kvm_init_msr_list();
> +
>  	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
>  			PT_DIRTY_MASK, PT64_NX_MASK, 0);
>
>

Thanks Liu.  Indeed I could not reproduce Robert's issue, though it 
might be possible to get it with nested virtualization.  I'll send a 
similar patch that I was testing.

Paolo

Patch

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 1e91a24..1fc184d 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3936,6 +3936,16 @@  static void kvm_init_msr_list(void)
 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
 			continue;
+
+		/*
+		 * Work around some buggy MPX platform which support
+		 * MSR_IA32_BNDCFGS but has issue at, say, VMX ucode.
+		 */
+		if ((msrs_to_save[i] == MSR_IA32_BNDCFGS) &&
+			(kvm_x86_ops->mpx_supported   ?
+			!kvm_x86_ops->mpx_supported() : 1))
+				continue;
+
 		if (j < i)
 			msrs_to_save[j] = msrs_to_save[i];
 		j++;
@@ -5576,9 +5586,11 @@  int kvm_arch_init(void *opaque)
 		goto out_free_percpu;
 
 	kvm_set_mmio_spte_mask();
-	kvm_init_msr_list();
 
 	kvm_x86_ops = ops;
+
+	kvm_init_msr_list();
+
 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
 			PT_DIRTY_MASK, PT64_NX_MASK, 0);