[PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[Like Xu]

In guest protected mode, if the current privilege level
is not 0 and the PCE flag in the CR4 register is cleared,
we will inject a #GP for RDPMC usage.

Signed-off-by: Like Xu <like.xu@linux.intel.com>
---
 arch/x86/kvm/pmu.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index b86346903f2e..d080d475c808 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -372,6 +372,11 @@ int kvm_pmu_rdpmc(struct kvm_vcpu *vcpu, unsigned idx, u64 *data)
 	if (!pmc)
 		return 1;
 
+	if ((kvm_x86_ops.get_cpl(vcpu) != 0) &&
+	    !(kvm_read_cr4(vcpu) & X86_CR4_PCE) &&
+	    (kvm_read_cr4(vcpu) & X86_CR0_PE))
+		return 1;
+
 	*data = pmc_read_counter(pmc) & mask;
 	return 0;
 }
-- 
2.21.3

Re: [PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[Sean Christopherson]

On Wed, Jul 08, 2020 at 03:44:09PM +0800, Like Xu wrote:
> in guest protected mode, if the current privilege level
> is not 0 and the pce flag in the cr4 register is cleared,
> we will inject a #gp for rdpmc usage.

Wrapping at ~58 characters is a bit aggressive.  checkpatch enforces 75
chars, something near that would be prefereable.

> Signed-off-by: Like Xu <like.xu@linux.intel.com>
> ---
>  arch/x86/kvm/pmu.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> index b86346903f2e..d080d475c808 100644
> --- a/arch/x86/kvm/pmu.c
> +++ b/arch/x86/kvm/pmu.c
> @@ -372,6 +372,11 @@ int kvm_pmu_rdpmc(struct kvm_vcpu *vcpu, unsigned idx, u64 *data)
>  	if (!pmc)
>  		return 1;
>  
> +	if ((kvm_x86_ops.get_cpl(vcpu) != 0) &&
> +	    !(kvm_read_cr4(vcpu) & X86_CR4_PCE) &&
> +	    (kvm_read_cr4(vcpu) & X86_CR0_PE))

This reads CR4 but checks CR0.PE.

And maybe put the X86_CR4_PCE check first so that it's the focus of the
statement?

> +		return 1;
> +
>  	*data = pmc_read_counter(pmc) & mask;
>  	return 0;
>  }
> -- 
> 2.21.3
> 

Re: [PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[Paolo Bonzini]

On 08/07/20 17:18, Sean Christopherson wrote:
> On Wed, Jul 08, 2020 at 03:44:09PM +0800, Like Xu wrote:
>> in guest protected mode, if the current privilege level
>> is not 0 and the pce flag in the cr4 register is cleared,
>> we will inject a #gp for rdpmc usage.
> 
> Wrapping at ~58 characters is a bit aggressive.  checkpatch enforces 75
> chars, something near that would be prefereable.
> 
>> Signed-off-by: Like Xu <like.xu@linux.intel.com>
>> ---
>>  arch/x86/kvm/pmu.c | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
>> index b86346903f2e..d080d475c808 100644
>> --- a/arch/x86/kvm/pmu.c
>> +++ b/arch/x86/kvm/pmu.c
>> @@ -372,6 +372,11 @@ int kvm_pmu_rdpmc(struct kvm_vcpu *vcpu, unsigned idx, u64 *data)
>>  	if (!pmc)
>>  		return 1;
>>  
>> +	if ((kvm_x86_ops.get_cpl(vcpu) != 0) &&
>> +	    !(kvm_read_cr4(vcpu) & X86_CR4_PCE) &&
>> +	    (kvm_read_cr4(vcpu) & X86_CR0_PE))
> 
> This reads CR4 but checks CR0.PE.
> 
> And maybe put the X86_CR4_PCE check first so that it's the focus of the
> statement?

I'll squash this to fix it (I'm OOO next week and would like to get kvm/queue
sorted out these few days that I've left).

diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index d080d475c808..67741d2a0308 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -372,9 +372,9 @@ int kvm_pmu_rdpmc(struct kvm_vcpu *vcpu, unsigned idx, u64 *data)
 	if (!pmc)
 		return 1;
 
-	if ((kvm_x86_ops.get_cpl(vcpu) != 0) &&
-	    !(kvm_read_cr4(vcpu) & X86_CR4_PCE) &&
-	    (kvm_read_cr4(vcpu) & X86_CR0_PE))
+	if (!(kvm_read_cr4(vcpu) & X86_CR4_PCE) &&
+	    (kvm_x86_ops.get_cpl(vcpu) != 0) &&
+	    (kvm_read_cr0(vcpu) & X86_CR0_PE))
 		return 1;
 
 	*data = pmc_read_counter(pmc) & mask;


The order follows the SDM.  I'm tempted to remove the CR0 check
altogether, since non-protected-mode always runs at CPL0 AFAIK, but let's
keep it close to what the manual says.

Paolo

Re: [PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[Sean Christopherson]

On Wed, Jul 08, 2020 at 05:31:14PM +0200, Paolo Bonzini wrote:
> The order follows the SDM.  I'm tempted to remove the CR0 check
> altogether, since non-protected-mode always runs at CPL0 AFAIK, but let's
> keep it close to what the manual says.

Heh, it wouldn't surprise me in the least if there's a way to get the SS
arbyte to hold a non-zero DPL in real mode :-).

Re: [PATCH] KVM/x86: pmu: Fix #GP condition check for RDPMC emulation

[Paolo Bonzini]

On 08/07/20 17:45, Sean Christopherson wrote:
> On Wed, Jul 08, 2020 at 05:31:14PM +0200, Paolo Bonzini wrote:
>> The order follows the SDM.  I'm tempted to remove the CR0 check
>> altogether, since non-protected-mode always runs at CPL0 AFAIK, but let's
>> keep it close to what the manual says.
> 
> Heh, it wouldn't surprise me in the least if there's a way to get the SS
> arbyte to hold a non-zero DPL in real mode :-).

I'm not sure if SMM lets you set non-zero SS.DPL in real mode.  It's one
of the few things that are checked with unrestricted guest mode so
there's hope; on the other hand I know for sure that in the past RSM
could get you to VM86 mode with CPL=0, while in VMX it causes vmentry to
fail.

It would be an interesting testcase to write for KVM, to see if you get
a vmentry failure after you set the hidden AR bytes that way and RSM...

Paolo