Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1407423 > unrolled thread

[PATCH] KVM: Handle MSR_IA32_PERF_CTL

Back to article view | Back to linux.kernel

Contents

  [PATCH] KVM: Handle MSR_IA32_PERF_CTL kmeaw@yandex-team.ru - 2016-05-26 09:50 +0200
    Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL Radim Krčmář <rkrcmar@redhat.com> - 2016-05-26 22:40 +0200
      Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL "Gabriel L. Somlo" <gsomlo@gmail.com> - 2016-05-26 22:50 +0200
        Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL Radim Krčmář <rkrcmar@redhat.com> - 2016-05-27 17:30 +0200
          Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL Radim Krčmář <rkrcmar@redhat.com> - 2016-05-27 17:40 +0200
            Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL Paolo Bonzini <pbonzini@redhat.com> - 2016-05-31 10:00 +0200
    Re: [PATCH] KVM: Handle MSR_IA32_PERF_CTL Radim Krčmář <rkrcmar@redhat.com> - 2016-05-27 17:30 +0200

#1407423 — [PATCH] KVM: Handle MSR_IA32_PERF_CTL

From: Dmitry Bilunov <kmeaw@yandex-team.ru>

Intel CPUs having Turbo Boost feature implement an MSR to provide a
control interface via rdmsr/wrmsr instructions. One could detect the
presence of this feature by issuing one of these instructions and
handling the #GP exception which is generated in case the referenced MSR
is not implemented by the CPU.

KVM's vCPU model behaves exactly as a real CPU in this case by injecting
a fault when MSR_IA32_PERF_CTL is called (which KVM does not support).
However, some operating systems use this register during an early boot
stage in which their kernel is not capable of handling #GP correctly,
causing #DP and finally a triple fault effectively resetting the vCPU.

This patch implements a dummy handler for MSR_IA32_PERF_CTL to avoid the
crashes. Most notably it fixes an issue with MacOS X 10.10 kernel.

Signed-off-by: Dmitry Bilunov <kmeaw@yandex-team.ru>
---
 arch/x86/kvm/x86.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index c805cf4..9f38c7f 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -983,6 +983,7 @@ static u32 emulated_msrs[] = {
        MSR_IA32_MCG_STATUS,
        MSR_IA32_MCG_CTL,
        MSR_IA32_SMBASE,
+       MSR_IA32_PERF_CTL,
 };
 
 static unsigned num_emulated_msrs;
@@ -2050,6 +2051,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
        case MSR_VM_HSAVE_PA:
        case MSR_AMD64_PATCH_LOADER:
        case MSR_AMD64_BU_CFG2:
+       case MSR_IA32_PERF_CTL:
                break;
 
        case MSR_EFER:
@@ -2314,6 +2316,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
        case MSR_AMD64_NB_CFG:
        case MSR_FAM10H_MMIO_CONF_BASE:
        case MSR_AMD64_BU_CFG2:
+       case MSR_IA32_PERF_CTL:
                msr_info->data = 0;
                break;
        case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
-- 
2.8.2

[toc] | [next] | [standalone]

#1407704

2016-05-26 10:32+0300, kmeaw@yandex-team.ru:
> From: Dmitry Bilunov <kmeaw@yandex-team.ru>
> 
> Intel CPUs having Turbo Boost feature implement an MSR to provide a
> control interface via rdmsr/wrmsr instructions. One could detect the
> presence of this feature by issuing one of these instructions and
> handling the #GP exception which is generated in case the referenced MSR
> is not implemented by the CPU.
> 
> KVM's vCPU model behaves exactly as a real CPU in this case by injecting
> a fault when MSR_IA32_PERF_CTL is called (which KVM does not support).
> However, some operating systems use this register during an early boot
> stage in which their kernel is not capable of handling #GP correctly,
> causing #DP and finally a triple fault effectively resetting the vCPU.
> 
> This patch implements a dummy handler for MSR_IA32_PERF_CTL to avoid the
> crashes. Most notably it fixes an issue with MacOS X 10.10 kernel.
> 
> Signed-off-by: Dmitry Bilunov <kmeaw@yandex-team.ru>
> ---
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> @@ -2050,6 +2051,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>         case MSR_VM_HSAVE_PA:
>         case MSR_AMD64_PATCH_LOADER:
>         case MSR_AMD64_BU_CFG2:
> +       case MSR_IA32_PERF_CTL:
>                 break;

Does MacOS X write it too?

Thanks.

[toc] | [prev] | [next] | [standalone]

#1407709

On Thu, May 26, 2016 at 10:39:31PM +0200, Radim Krčmář wrote:
> 2016-05-26 10:32+0300, kmeaw@yandex-team.ru:
> > From: Dmitry Bilunov <kmeaw@yandex-team.ru>
> > 
> > Intel CPUs having Turbo Boost feature implement an MSR to provide a
> > control interface via rdmsr/wrmsr instructions. One could detect the
> > presence of this feature by issuing one of these instructions and
> > handling the #GP exception which is generated in case the referenced MSR
> > is not implemented by the CPU.
> > 
> > KVM's vCPU model behaves exactly as a real CPU in this case by injecting
> > a fault when MSR_IA32_PERF_CTL is called (which KVM does not support).
> > However, some operating systems use this register during an early boot
> > stage in which their kernel is not capable of handling #GP correctly,
> > causing #DP and finally a triple fault effectively resetting the vCPU.
> > 
> > This patch implements a dummy handler for MSR_IA32_PERF_CTL to avoid the
> > crashes. Most notably it fixes an issue with MacOS X 10.10 kernel.
> > 
> > Signed-off-by: Dmitry Bilunov <kmeaw@yandex-team.ru>
> > ---
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > @@ -2050,6 +2051,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> >         case MSR_VM_HSAVE_PA:
> >         case MSR_AMD64_PATCH_LOADER:
> >         case MSR_AMD64_BU_CFG2:
> > +       case MSR_IA32_PERF_CTL:
> >                 break;
> 
> Does MacOS X write it too?

After setting /sys/module/kvm/parameters/ignore_msrs, all I get in
dmesg after firing up OS X is:

	vcpu0 ignored rdmsr: 0x199

So no, I don't think it would try to write it.

HTH,
--Gabriel

[toc] | [prev] | [next] | [standalone]

#1408140

2016-05-27 09:49+0300, kmeaw@yandex-team.ru:
> 26.05.2016, 23:44, "Gabriel L. Somlo" <gsomlo@gmail.com>:
>> On Thu, May 26, 2016 at 10:39:31PM +0200, Radim Krčmář wrote:
>>>  2016-05-26 10:32+0300, kmeaw@yandex-team.ru:
>>>  > This patch implements a dummy handler for MSR_IA32_PERF_CTL to avoid the
>>>  > crashes. Most notably it fixes an issue with MacOS X 10.10 kernel.
>>>  Does MacOS X write it too?
>>
>> After setting /sys/module/kvm/parameters/ignore_msrs, all I get in
>> dmesg after firing up OS X is:
>>
>>         vcpu0 ignored rdmsr: 0x199
>>
>> So no, I don't think it would try to write it.
> 
> That's right, OS X does not issue an wrmsr to 0x199. More specifically, I have not
> observed that on my KVM instances. Should I remove the "wrmsr" portion from the
> patch?

Yes, please.  Silently ignoring the write is worse than #GP and #GP is
not a problem, so I wouldn't bother with a phony implementation.
Returning 0 on read is ok as seems to mean P-state=0, which is within
expectations.

(I wonder why MacOS X doesn't read IA32_PERF_STATUS, though.)

[toc] | [prev] | [next] | [standalone]

#1408144

2016-05-27 17:22+0200, Radim Krčmář:
> (I wonder why MacOS X doesn't read IA32_PERF_STATUS, though.)

Oh, it maybe does ... we already emulate status and return 0x1000 in its
bottom 16 bits.  I have no idea what is that supposed to mean, but I
think we should return 0x1000 in IA32_PERF_CTL then.

(Would be nice to understand how that 0x1000 happened ... we might want
 0 in both.)

[toc] | [prev] | [next] | [standalone]

#1409946

> 2016-05-27 17:22+0200, Radim Krčmář:
> > (I wonder why MacOS X doesn't read IA32_PERF_STATUS, though.)
> 
> Oh, it maybe does ... we already emulate status and return 0x1000 in its
> bottom 16 bits.  I have no idea what is that supposed to mean, but I
> think we should return 0x1000 in IA32_PERF_CTL then.

It's 1000, not 0x1000 (instead, on real hardware the value is typically a
multiple of 256).  It was added for Darwin too.

Returning different values is okay, because they are different on real
hardware too:

(sudo dd if=/dev/cpu/0/msr skip=$((0x198)) iflag=skip_bytes bs=8 count=1;
 sudo dd if=/dev/cpu/0/msr skip=$((0x199)) iflag=skip_bytes bs=8 count=1) | od -tx8
0000000 00001f3900001100 0000000000001300
        ^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^
        PERF_STATUS      PERF_CTL

And perhaps if we returned non-zero values for PERF_CTL Darwin would try to
write to it.  So returning zero is fine, I think.  There is no correct answer...

Paolo

> (Would be nice to understand how that 0x1000 happened ... we might want
>  0 in both.)
> 

[toc] | [prev] | [next] | [standalone]

#1408139

2016-05-26 10:32+0300, kmeaw@yandex-team.ru:
> From: Dmitry Bilunov <kmeaw@yandex-team.ru>
> 
> Intel CPUs having Turbo Boost feature implement an MSR to provide a
> control interface via rdmsr/wrmsr instructions. One could detect the
> presence of this feature by issuing one of these instructions and
> handling the #GP exception which is generated in case the referenced MSR
> is not implemented by the CPU.
> 
> KVM's vCPU model behaves exactly as a real CPU in this case by injecting
> a fault when MSR_IA32_PERF_CTL is called (which KVM does not support).
> However, some operating systems use this register during an early boot
> stage in which their kernel is not capable of handling #GP correctly,
> causing #DP and finally a triple fault effectively resetting the vCPU.
> 
> This patch implements a dummy handler for MSR_IA32_PERF_CTL to avoid the
> crashes. Most notably it fixes an issue with MacOS X 10.10 kernel.
> 
> Signed-off-by: Dmitry Bilunov <kmeaw@yandex-team.ru>
> ---
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> @@ -983,6 +983,7 @@ static u32 emulated_msrs[] = {
>         MSR_IA32_MCG_STATUS,
>         MSR_IA32_MCG_CTL,
>         MSR_IA32_SMBASE,
> +       MSR_IA32_PERF_CTL,

The MSR value is always 0, so there is no point in putting it here.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web