Skip to contentSkip to conversation

Fix MSR_IA32_PERF_STATUS

PATCH In review Avi Kivity 2 participants 6 messages Sep 1, 2010

Subsystems (3)
lore
Share & tools
lore

Avi Kivity

author
#0 lore raw

cover Fix MSR_IA32_PERF_STATUS

MSR_IA32_PERF_STATUS is read-only, yet exposed for save/restore. This generates annoying printk()s when userspace attempts to initialize it.

The value returned is also bogus, the comments appear to refer to another msr.

Fix both issues. Alex, please test with a guest Mac.

Avi Kivity (2):
KVM: Don't save/restore MSR_IA32_PERF_STATUS
KVM: Fix bogus value returned for MSR_IA32_PERF_STATUS

 arch/x86/kvm/x86.c |    7 ++-----
 1 files changed, 2 insertions(+), 5 deletions(-)
author
#1 lore raw

PATCH 1/2 KVM: Don't save/restore MSR_IA32_PERF_STATUS

It is read/only; restoring it only results in annoying messages.

---
 arch/x86/kvm/x86.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kvm/x86.c +1 -1
index cc6d6cd..1cbf168 100644
@@ -739,7 +739,7 @@ static u32 msrs_to_save[] = {
 #ifdef CONFIG_X86_64
 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 #endif
-	MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
+	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 };
 
 static unsigned num_msrs_to_save;
Signature · 1 line
1.7.1
author
#2 lore raw

PATCH 2/2 KVM: Fix bogus value returned for MSR_IA32_PERF_STATUS

In reply to Avi Kivity

The returned value is completely bogus, and sets reserved bits. Return zero instead.

---
 arch/x86/kvm/x86.c |    5 +----
 1 files changed, 1 insertions(+), 4 deletions(-)
arch/x86/kvm/x86.c +1 -4
index 1cbf168..a2c03f1 100644
@@ -1641,10 +1641,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 		data = vcpu->arch.ia32_misc_enable_msr;
 		break;
 	case MSR_IA32_PERF_STATUS:
-		/* TSC increment by tick */
-		data = 1000ULL;
-		/* CPU multiplier */
-		data |= (((uint64_t)4ULL) << 40);
+		data = 0;
 		break;
 	case MSR_EFER:
 		data = vcpu->arch.efer;
Signature · 1 line
1.7.1
#3 lore raw

Re: [PATCH 2/2] KVM: Fix bogus value returned for MSR_IA32_PERF_STATUS

Quoted 14 lines · Avi Kivity

The returned value is completely bogus, and sets reserved bits. Return zero instead.

---
arch/x86/kvm/x86.c | 5 +----

1 files changed, 1 insertions(+), 4 deletions(-)
arch/x86/kvm/x86.c +0 -0
index 1cbf168..a2c03f1 100644
@@ -1641,10 +1641,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 		data = vcpu->arch.ia32_misc_enable_msr;

arch/x86/kvm/x86.c int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) · line 1647

Quoted 10 lines
arch/x86/kvm/x86.c +1 -4
 		break;
 	case MSR_IA32_PERF_STATUS:
-		/* TSC increment by tick */
-		data = 1000ULL;
-		/* CPU multiplier */
-		data |= (((uint64_t)4ULL) << 40);
+		data = 0;
 		break;
 	case MSR_EFER:
 		data = vcpu->arch.efer;

This is the respective code snippet from xnu:

        /*
         * Get the TSC increment.  The TSC is incremented by this
         * on every bus tick.  Calculate the TSC conversion factors
         * to and from nano-seconds.
         */
        if (cpuid_info()->cpuid_family == CPU_FAMILY_PENTIUM_M) {
                uint64_t        prfsts;

                prfsts = rdmsr64(IA32_PERF_STS);
                tscGranularity = (uint32_t)bitfield(prfsts, 44, 40);
                N_by_2_bus_ratio = prfsts & bit(46);

        } else {
                panic("rtclock_init: unknown CPU family: 0x%X\n",
                        cpuid_info()->cpuid_family);
        }

        if (N_by_2_bus_ratio)
                tscFCvtt2n = busFCvtt2n * 2 / (uint64_t)tscGranularity;
        else
                tscFCvtt2n = busFCvtt2n / (uint64_t)tscGranularity;

        tscFreq = ((1 * Giga)  << 32) / tscFCvtt2n;
        tscFCvtn2t = 0xFFFFFFFFFFFFFFFFULL / tscFCvtt2n;

So by passing in 0 here, you effectively make that code divide something by 0 which results in a panic.

Alex

author
#4 lore raw

Re: [PATCH 2/2] KVM: Fix bogus value returned for MSR_IA32_PERF_STATUS

Quoted 55 lines · Alexander Graf
Quoted 14 lines · Avi Kivity

The returned value is completely bogus, and sets reserved bits. Return zero instead.

---
arch/x86/kvm/x86.c | 5 +----

1 files changed, 1 insertions(+), 4 deletions(-)
arch/x86/kvm/x86.c +0 -0
index 1cbf168..a2c03f1 100644
@@ -1641,10 +1641,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 		data = vcpu->arch.ia32_misc_enable_msr;

arch/x86/kvm/x86.c int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) · line 1647

Quoted 10 lines
arch/x86/kvm/x86.c +1 -4
 		break;
 	case MSR_IA32_PERF_STATUS:
-		/* TSC increment by tick */
-		data = 1000ULL;
-		/* CPU multiplier */
-		data |= (((uint64_t)4ULL)<<  40);
+		data = 0;
 		break;
 	case MSR_EFER:
 		data = vcpu->arch.efer;

This is the respective code snippet from xnu:

         /*
          * Get the TSC increment.  The TSC is incremented by this
          * on every bus tick.  Calculate the TSC conversion factors
          * to and from nano-seconds.
          */
         if (cpuid_info()->cpuid_family == CPU_FAMILY_PENTIUM_M) {
                 uint64_t        prfsts;

                 prfsts = rdmsr64(IA32_PERF_STS);
                 tscGranularity = (uint32_t)bitfield(prfsts, 44, 40);
                 N_by_2_bus_ratio = prfsts&  bit(46);

         } else {
                 panic("rtclock_init: unknown CPU family: 0x%X\n",
                         cpuid_info()->cpuid_family);
         }

         if (N_by_2_bus_ratio)
                 tscFCvtt2n = busFCvtt2n * 2 / (uint64_t)tscGranularity;
         else
                 tscFCvtt2n = busFCvtt2n / (uint64_t)tscGranularity;

         tscFreq = ((1 * Giga)<<  32) / tscFCvtt2n;
         tscFCvtn2t = 0xFFFFFFFFFFFFFFFFULL / tscFCvtt2n;
Quoted 3 lines

So by passing in 0 here, you effectively make that code divide something by 0 which results in a panic.

Right. Searching again I find that this is actually documented, (under MSR_PERF_STATU s instead of IA32_PERF_STATUS). I'll just drop this patch.

Signature · 2 lines
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.
#5 lore raw

Re: [PATCH 2/2] KVM: Fix bogus value returned for MSR_IA32_PERF_STATUS

Quoted 56 lines · Avi Kivity
Quoted 55 lines · Alexander Graf
Quoted 14 lines · Avi Kivity

The returned value is completely bogus, and sets reserved bits. Return zero instead.

---
arch/x86/kvm/x86.c | 5 +----

1 files changed, 1 insertions(+), 4 deletions(-)
arch/x86/kvm/x86.c +0 -0
index 1cbf168..a2c03f1 100644
@@ -1641,10 +1641,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 		data = vcpu->arch.ia32_misc_enable_msr;

arch/x86/kvm/x86.c int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) · line 1647

Quoted 10 lines
arch/x86/kvm/x86.c +1 -4
 		break;
 	case MSR_IA32_PERF_STATUS:
-		/* TSC increment by tick */
-		data = 1000ULL;
-		/* CPU multiplier */
-		data |= (((uint64_t)4ULL)<<  40);
+		data = 0;
 		break;
 	case MSR_EFER:
 		data = vcpu->arch.efer;

This is the respective code snippet from xnu:

        /*
         * Get the TSC increment.  The TSC is incremented by this
         * on every bus tick.  Calculate the TSC conversion factors
         * to and from nano-seconds.
         */
        if (cpuid_info()->cpuid_family == CPU_FAMILY_PENTIUM_M) {
                uint64_t        prfsts;

                prfsts = rdmsr64(IA32_PERF_STS);
                tscGranularity = (uint32_t)bitfield(prfsts, 44, 40);
                N_by_2_bus_ratio = prfsts&  bit(46);

        } else {
                panic("rtclock_init: unknown CPU family: 0x%X\n",
                        cpuid_info()->cpuid_family);
        }

        if (N_by_2_bus_ratio)
                tscFCvtt2n = busFCvtt2n * 2 / (uint64_t)tscGranularity;
        else
                tscFCvtt2n = busFCvtt2n / (uint64_t)tscGranularity;

        tscFreq = ((1 * Giga)<<  32) / tscFCvtt2n;
        tscFCvtn2t = 0xFFFFFFFFFFFFFFFFULL / tscFCvtt2n;
Quoted 6 lines
Quoted 3 lines

So by passing in 0 here, you effectively make that code divide something by 0 which results in a panic.

Right. Searching again I find that this is actually documented, (under MSR_PERF_STATU s instead of IA32_PERF_STATUS). I'll just drop this patch.

Sounds good. The other one seems sane to me. I don't have a working setup for Mac guests atm. That was on the TODO list once the AHCI stuff gets upstream :).

Alex