[PATCH 18/23] KVM: x86/pmu: Handle GLOBAL_STATUS MSRs under PerfMon masking

[Zide Chen]

For IA32_PERF_GLOBAL_STATUS{,_SET,_RESET}, the effective mask under
PerfMon masking is perfmon_mask. Writes that set bits outside the
effective mask must #GP, so use ~perfmon_mask as the effective
reserved-bit mask instead of global_status_rsvd.

At guest context load, keep host-owned GLOBAL_STATUS bits unchanged
because system-wide events may be scheduled on host-owned resources
while running in VMX non-root mode.

Likewise, at guest context put, preserve host-owned bits and clear
only the guest-owned subset from hardware GLOBAL_STATUS, leaving
pmu->global_status containing only guest-owned bits.

Signed-off-by: Zide Chen <zide.chen@xxxxxxxxx>
---
 arch/x86/kvm/pmu.c           | 13 ++++++++-----
 arch/x86/kvm/vmx/pmu_intel.c | 17 ++++++++++++++++-
 2 files changed, 24 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index b051ba66edce..23177c38f286 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -922,12 +922,15 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
 	u32 msr = msr_info->index;
 	u64 data = msr_info->data;
-	u64 diff;
+	u64 global_status_rsvd, diff;
 
 	/*
 	 * Note, AMD ignores writes to reserved bits and read-only PMU MSRs,
 	 * whereas Intel generates #GP on attempts to write reserved/RO MSRs.
 	 */
+	global_status_rsvd = kvm_vcpu_has_perfmon_mask(vcpu) ?
+			     ~pmu->perfmon_mask : pmu->global_status_rsvd;
+
 	switch (msr) {
 	case MSR_CORE_PERF_GLOBAL_STATUS:
 		if (!msr_info->host_initiated)
@@ -938,7 +941,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		if (!msr_info->host_initiated)
 			break;
 
-		if (data & pmu->global_status_rsvd)
+		if (data & global_status_rsvd)
 			return 1;
 
 		pmu->global_status = data;
@@ -967,7 +970,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		 * GLOBAL_OVF_CTRL, a.k.a. GLOBAL STATUS_RESET, clears bits in
 		 * GLOBAL_STATUS, and so the set of reserved bits is the same.
 		 */
-		if (data & pmu->global_status_rsvd)
+		if (data & global_status_rsvd)
 			return 1;
 		fallthrough;
 	case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR:
@@ -975,14 +978,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 			pmu->global_status &= ~data;
 		break;
 	case MSR_CORE_PERF_GLOBAL_STATUS_SET:
-		if (data & pmu->global_status_rsvd)
+		if (data & global_status_rsvd)
 			return 1;
 		if (!msr_info->host_initiated)
 			pmu->global_status |= data;
 		break;
 	case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_SET:
 		if (!msr_info->host_initiated)
-			pmu->global_status |= data & ~pmu->global_status_rsvd;
+			pmu->global_status |= data & ~global_status_rsvd;
 		break;
 	default:
 		kvm_pmu_mark_pmc_in_use(vcpu, msr_info->index);
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index d0373de951d5..e43380f79c72 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -968,6 +968,14 @@ static void intel_mediated_pmu_load(struct kvm_vcpu *vcpu, u64 host_global_ctrl)
 
 	rdmsrq(MSR_CORE_PERF_GLOBAL_STATUS, global_status);
 	toggle = pmu->global_status ^ global_status;
+
+	/*
+	 * Restrict OVF_CTRL/STATUS_SET writes to guest-owned bits under
+	 * PerfMon masking.
+	 */
+	if (kvm_vcpu_has_perfmon_mask(vcpu))
+		toggle &= pmu->perfmon_mask;
+
 	if (global_status & toggle)
 		wrmsrq(MSR_CORE_PERF_GLOBAL_OVF_CTRL, global_status & toggle);
 	if (pmu->global_status & toggle)
@@ -984,7 +992,14 @@ static void intel_mediated_pmu_put(struct kvm_vcpu *vcpu)
 	/* MSR_CORE_PERF_GLOBAL_CTRL is already saved at VM-exit. */
 	rdmsrq(MSR_CORE_PERF_GLOBAL_STATUS, pmu->global_status);
 
-	/* Clear hardware MSR_CORE_PERF_GLOBAL_STATUS MSR, if non-zero. */
+	/*
+	 * Clear only the guest-owned bits from the hardware GLOBAL_STATUS
+	 * if any are set. pmu->global_status is then left holding just the
+	 * guest-owned subset.
+	 */
+	if (kvm_vcpu_has_perfmon_mask(vcpu))
+		pmu->global_status &= pmu->perfmon_mask;
+
 	if (pmu->global_status)
 		wrmsrq(MSR_CORE_PERF_GLOBAL_OVF_CTRL, pmu->global_status);
 
-- 
2.55.0