[PATCH] KVM: X86: #GP when guest attempts to write MCi banks w/o 0

[PATCH] KVM: X86: #GP when guest attempts to write MCi banks w/o 0

[Wanpeng Li]

From: Wanpeng Li <wanpeng.li@hotmail.com>

SDM section 15.3.2.2~15.3.2.4 mentioned that MCi_STATUS/ADDR/MISC, when the 
registers are implemented, these registers can be cleared by explicitly writing 
0s to these registers. Writing 1s to these registers will cause a 
general-protection exception.

The mce is emulated in qemu, so just the guest attempts to write 1 to these 
registers should cause a #GP, this patch does it.

Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Jim Mattson <jmattson@google.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
---
 arch/x86/kvm/x86.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 5669af0..a8680ea 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2006,10 +2006,12 @@ static void kvmclock_sync_fn(struct work_struct *work)
 					KVMCLOCK_SYNC_PERIOD);
 }
 
-static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
+static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 {
 	u64 mcg_cap = vcpu->arch.mcg_cap;
 	unsigned bank_num = mcg_cap & 0xff;
+	u32 msr = msr_info->index;
+	u64 data = msr_info->data;
 
 	switch (msr) {
 	case MSR_IA32_MCG_STATUS:
@@ -2034,6 +2036,9 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 			if ((offset & 0x3) == 0 &&
 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
 				return -1;
+			if (!msr_info->host_initiated &&
+				(offset & 0x3) != 0 && data != 0)
+				return -1;
 			vcpu->arch.mce_banks[offset] = data;
 			break;
 		}
@@ -2283,7 +2288,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	case MSR_IA32_MCG_CTL:
 	case MSR_IA32_MCG_STATUS:
 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
-		return set_msr_mce(vcpu, msr, data);
+		return set_msr_mce(vcpu, msr_info);
 
 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
-- 
2.7.4

Re: [PATCH] KVM: X86: #GP when guest attempts to write MCi banks w/o 0

[Jim Mattson]

The AMD APM says, "For each error-reporting register bank, software
should set the enable bits to 1 in the MCi_CTL register for the error
types it wants the processor to report. Software can write each
MCi_CTL with all 1s to enable all error-reporting mechanisms.' It does
not say that only all 1's or all 0's are allowed, and it implies that
any value is allowed.

Since this is a vendor-agnostic function, the Intel-specific
constraints should only be applied when virtualizing Intel CPUs (in
particular, Intel P6 family CPUs). The same comment applies to the
existing constraints from commit 890ca9aefa78 ("KVM: Add MCE
support"), which were only partially relaxed by commit 114be429c8cd4
("KVM: allow bit 10 to be cleared in MSR_IA32_MC4_CTL").

On Wed, Oct 18, 2017 at 2:49 AM, Wanpeng Li <kernellwp@gmail.com> wrote:
> From: Wanpeng Li <wanpeng.li@hotmail.com>
>
> SDM section 15.3.2.2~15.3.2.4 mentioned that MCi_STATUS/ADDR/MISC, when the
> registers are implemented, these registers can be cleared by explicitly writing
> 0s to these registers. Writing 1s to these registers will cause a
> general-protection exception.
>
> The mce is emulated in qemu, so just the guest attempts to write 1 to these
> registers should cause a #GP, this patch does it.
>
> Cc: Radim Krčmář <rkrcmar@redhat.com>
> Cc: Jim Mattson <jmattson@google.com>
> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
> ---
>  arch/x86/kvm/x86.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 5669af0..a8680ea 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2006,10 +2006,12 @@ static void kvmclock_sync_fn(struct work_struct *work)
>                                         KVMCLOCK_SYNC_PERIOD);
>  }
>
> -static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
> +static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>  {
>         u64 mcg_cap = vcpu->arch.mcg_cap;
>         unsigned bank_num = mcg_cap & 0xff;
> +       u32 msr = msr_info->index;
> +       u64 data = msr_info->data;
>
>         switch (msr) {
>         case MSR_IA32_MCG_STATUS:
> @@ -2034,6 +2036,9 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
>                         if ((offset & 0x3) == 0 &&
>                             data != 0 && (data | (1 << 10)) != ~(u64)0)
>                                 return -1;
> +                       if (!msr_info->host_initiated &&
> +                               (offset & 0x3) != 0 && data != 0)
> +                               return -1;
>                         vcpu->arch.mce_banks[offset] = data;
>                         break;
>                 }
> @@ -2283,7 +2288,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>         case MSR_IA32_MCG_CTL:
>         case MSR_IA32_MCG_STATUS:
>         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
> -               return set_msr_mce(vcpu, msr, data);
> +               return set_msr_mce(vcpu, msr_info);
>
>         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
>         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
> --
> 2.7.4
>

Re: [PATCH] KVM: X86: #GP when guest attempts to write MCi banks w/o 0

[Jim Mattson]

Right. I was side-tracked by the code above yours for MCi_CTL.
However, does writing a non-zero value to MCi_STATUS/ADDR/MISC raise
#GP on AMD hardware? It is not clear from the APM. For MCi_MISC0, the
APM says:
"In some implementations, the MCi_MISC0 register is used for error
thresholding." Figure 9-8: Miscellaneous Information Register
(Thresholding Register Format) suggests that non-zero value can, in
fact, be written to MCi_MISC0 in such implementations.

Do any of the AMD contributors want to weigh in?

On Wed, Oct 18, 2017 at 6:39 PM, Wanpeng Li <wanpeng.li@hotmail.com> wrote:
> Hi Jim，
> On 10/19/17 12:28 AM, Jim Mattson wrote:
>> The AMD APM says, "For each error-reporting register bank, software
>> should set the enable bits to 1 in the MCi_CTL register for the error
>> types it wants the processor to report. Software can write each
>> MCi_CTL with all 1s to enable all error-reporting mechanisms.' It does
>> not say that only all 1's or all 0's are allowed, and it implies that
>> any value is allowed.
>>
>> Since this is a vendor-agnostic function, the Intel-specific
>> constraints should only be applied when virtualizing Intel CPUs (in
>> particular, Intel P6 family CPUs). The same comment applies to the
>> existing constraints from commit 890ca9aefa78 ("KVM: Add MCE
>
> I have a discuss with the author of ("KVM: Add MCE support") face to
> face today in a kernel meeting held in our country. He told me his patch
> is against Intel architecture and not consider AMD when he introduced
> the patch. In addition, the difference which you mentioned is about
> MCi_CTL, however, my patches just focus on MCi_STATUS/ADDR/MISC.
>
> Regards,
> Wanpeng Li
>
>> support"), which were only partially relaxed by commit 114be429c8cd4
>> ("KVM: allow bit 10 to be cleared in MSR_IA32_MC4_CTL").
>>
>> On Wed, Oct 18, 2017 at 2:49 AM, Wanpeng Li <kernellwp@gmail.com> wrote:
>>> From: Wanpeng Li <wanpeng.li@hotmail.com>
>>>
>>> SDM section 15.3.2.2~15.3.2.4 mentioned that MCi_STATUS/ADDR/MISC, when the
>>> registers are implemented, these registers can be cleared by explicitly writing
>>> 0s to these registers. Writing 1s to these registers will cause a
>>> general-protection exception.
>>>
>>> The mce is emulated in qemu, so just the guest attempts to write 1 to these
>>> registers should cause a #GP, this patch does it.
>>>
>>> Cc: Radim Krčmář <rkrcmar@redhat.com>
>>> Cc: Jim Mattson <jmattson@google.com>
>>> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
>>> ---
>>>   arch/x86/kvm/x86.c | 9 +++++++--
>>>   1 file changed, 7 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>>> index 5669af0..a8680ea 100644
>>> --- a/arch/x86/kvm/x86.c
>>> +++ b/arch/x86/kvm/x86.c
>>> @@ -2006,10 +2006,12 @@ static void kvmclock_sync_fn(struct work_struct *work)
>>>                                          KVMCLOCK_SYNC_PERIOD);
>>>   }
>>>
>>> -static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
>>> +static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>   {
>>>          u64 mcg_cap = vcpu->arch.mcg_cap;
>>>          unsigned bank_num = mcg_cap & 0xff;
>>> +       u32 msr = msr_info->index;
>>> +       u64 data = msr_info->data;
>>>
>>>          switch (msr) {
>>>          case MSR_IA32_MCG_STATUS:
>>> @@ -2034,6 +2036,9 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
>>>                          if ((offset & 0x3) == 0 &&
>>>                              data != 0 && (data | (1 << 10)) != ~(u64)0)
>>>                                  return -1;
>>> +                       if (!msr_info->host_initiated &&
>>> +                               (offset & 0x3) != 0 && data != 0)
>>> +                               return -1;
>>>                          vcpu->arch.mce_banks[offset] = data;
>>>                          break;
>>>                  }
>>> @@ -2283,7 +2288,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>          case MSR_IA32_MCG_CTL:
>>>          case MSR_IA32_MCG_STATUS:
>>>          case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
>>> -               return set_msr_mce(vcpu, msr, data);
>>> +               return set_msr_mce(vcpu, msr_info);
>>>
>>>          case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
>>>          case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
>>> --
>>> 2.7.4
>>>
>

Re: [PATCH] KVM: X86: #GP when guest attempts to write MCi banks w/o 0

[Jim Mattson]

MCi_STATUS sounds fine, or any other Intel constraints that are
guarded by checks for virtualizing an Intel CPU.

On Wed, Oct 18, 2017 at 8:04 PM, Wanpeng Li <wanpeng.li@hotmail.com> wrote:
> On 10/19/17 10:49 AM, Jim Mattson wrote:
>
> Right. I was side-tracked by the code above yours for MCi_CTL.
> However, does writing a non-zero value to MCi_STATUS/ADDR/MISC raise
> #GP on AMD hardware? It is not clear from the APM. For MCi_MISC0, the
>
>
> AMD Volume 2:
>
> 9.3.2 Error-Reporting Register Bank:
>
> Attempting to write a value other than 0 to an MCi_STATUS register will
> raise a general protection (#GP) exception.
>
> So maybe my patch can just handle MCi_STATUS or current patch is ok, what's
> your opinion?
>
>
> Regards,
> Wanpeng Li
>
>
> APM says:
> "In some implementations, the MCi_MISC0 register is used for error
> thresholding." Figure 9-8: Miscellaneous Information Register
> (Thresholding Register Format) suggests that non-zero value can, in
> fact, be written to MCi_MISC0 in such implementations.
>
> Do any of the AMD contributors want to weigh in?
>
> On Wed, Oct 18, 2017 at 6:39 PM, Wanpeng Li <wanpeng.li@hotmail.com> wrote:
>
> Hi Jim，
> On 10/19/17 12:28 AM, Jim Mattson wrote:
>
> The AMD APM says, "For each error-reporting register bank, software
> should set the enable bits to 1 in the MCi_CTL register for the error
> types it wants the processor to report. Software can write each
> MCi_CTL with all 1s to enable all error-reporting mechanisms.' It does
> not say that only all 1's or all 0's are allowed, and it implies that
> any value is allowed.
>
> Since this is a vendor-agnostic function, the Intel-specific
> constraints should only be applied when virtualizing Intel CPUs (in
> particular, Intel P6 family CPUs). The same comment applies to the
> existing constraints from commit 890ca9aefa78 ("KVM: Add MCE
>
> I have a discuss with the author of ("KVM: Add MCE support") face to
> face today in a kernel meeting held in our country. He told me his patch
> is against Intel architecture and not consider AMD when he introduced
> the patch. In addition, the difference which you mentioned is about
> MCi_CTL, however, my patches just focus on MCi_STATUS/ADDR/MISC.
>
> Regards,
> Wanpeng Li
>
> support"), which were only partially relaxed by commit 114be429c8cd4
> ("KVM: allow bit 10 to be cleared in MSR_IA32_MC4_CTL").
>
> On Wed, Oct 18, 2017 at 2:49 AM, Wanpeng Li <kernellwp@gmail.com> wrote:
>
> From: Wanpeng Li <wanpeng.li@hotmail.com>
>
> SDM section 15.3.2.2~15.3.2.4 mentioned that MCi_STATUS/ADDR/MISC, when the
> registers are implemented, these registers can be cleared by explicitly
> writing
> 0s to these registers. Writing 1s to these registers will cause a
> general-protection exception.
>
> The mce is emulated in qemu, so just the guest attempts to write 1 to these
> registers should cause a #GP, this patch does it.
>
> Cc: Radim Krčmář <rkrcmar@redhat.com>
> Cc: Jim Mattson <jmattson@google.com>
> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
> ---
>   arch/x86/kvm/x86.c | 9 +++++++--
>   1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 5669af0..a8680ea 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2006,10 +2006,12 @@ static void kvmclock_sync_fn(struct work_struct
> *work)
>                                          KVMCLOCK_SYNC_PERIOD);
>   }
>
> -static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
> +static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>   {
>          u64 mcg_cap = vcpu->arch.mcg_cap;
>          unsigned bank_num = mcg_cap & 0xff;
> +       u32 msr = msr_info->index;
> +       u64 data = msr_info->data;
>
>          switch (msr) {
>          case MSR_IA32_MCG_STATUS:
> @@ -2034,6 +2036,9 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr,
> u64 data)
>                          if ((offset & 0x3) == 0 &&
>                              data != 0 && (data | (1 << 10)) != ~(u64)0)
>                                  return -1;
> +                       if (!msr_info->host_initiated &&
> +                               (offset & 0x3) != 0 && data != 0)
> +                               return -1;
>                          vcpu->arch.mce_banks[offset] = data;
>                          break;
>                  }
> @@ -2283,7 +2288,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct
> msr_data *msr_info)
>          case MSR_IA32_MCG_CTL:
>          case MSR_IA32_MCG_STATUS:
>          case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
> -               return set_msr_mce(vcpu, msr, data);
> +               return set_msr_mce(vcpu, msr_info);
>
>          case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
>          case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
> --
> 2.7.4
>
>