[PATCH 17/23] KVM: x86/pmu: Handle GLOBAL_CTRL under PerfMon masking

[Zide Chen]

Guest writes must be gated, and guest reads masked, with
pmu->global_ctrl_rsvd rather than pmu->perfmon_mask, because bits
63:52 in global_ctrl_rsvd are guaranteed to be cleared, and bits 51:0
are the complement between the two masks.

Since the host could schedule !exclude_guest events on host-owned
resources in non-root mode, OR the host-owned bits into
GUEST_IA32_PERF_GLOBAL_CTRL when loading guest PMU state. The host-
owned bits are supposed to be set during perf_load_guest_context().

Add a host_global_ctrl parameter to the mediated_load() callback so
that the host value can be available before the register is cleared.
This allows intel_mediated_pmu_load() to preserve host-owned bits.

When caching pmu->global_ctrl at VM exit, mask the guest value with
pmu->global_ctrl_rsvd to retain guest-owned bits only.

Signed-off-by: Zide Chen <zide.chen@xxxxxxxxx>
---
 arch/x86/kvm/pmu.c           |  7 ++++++-
 arch/x86/kvm/pmu.h           |  2 +-
 arch/x86/kvm/svm/pmu.c       |  2 +-
 arch/x86/kvm/vmx/pmu_intel.c | 13 ++++++++++++-
 arch/x86/kvm/vmx/vmx.c       |  4 ++++
 5 files changed, 24 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index 337d2f55a216..b051ba66edce 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -1406,6 +1406,8 @@ static void kvm_pmu_load_guest_pmcs(struct kvm_vcpu *vcpu)
 
 void kvm_mediated_pmu_load(struct kvm_vcpu *vcpu)
 {
+	u64 host_global_ctrl = 0;
+
 	if (!kvm_vcpu_has_mediated_pmu(vcpu) ||
 	    KVM_BUG_ON(!lapic_in_kernel(vcpu), vcpu->kvm))
 		return;
@@ -1429,13 +1431,16 @@ void kvm_mediated_pmu_load(struct kvm_vcpu *vcpu)
 	 * even for SVM to minimize the damage if a perf event is left enabled,
 	 * and to ensure a consistent starting state.
 	 */
+	if (kvm_vcpu_has_perfmon_mask(vcpu))
+		rdmsrq(kvm_pmu_ops.PERF_GLOBAL_CTRL, host_global_ctrl);
+
 	wrmsrq(kvm_pmu_ops.PERF_GLOBAL_CTRL, 0);
 
 	perf_load_guest_lvtpc(kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVTPC));
 
 	kvm_pmu_load_guest_pmcs(vcpu);
 
-	kvm_pmu_call(mediated_load)(vcpu);
+	kvm_pmu_call(mediated_load)(vcpu, host_global_ctrl);
 }
 
 static void kvm_pmu_put_guest_pmcs(struct kvm_vcpu *vcpu)
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index c5feeb60bcf6..30e456302c65 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -39,7 +39,7 @@ struct kvm_pmu_ops {
 	bool (*pmc_is_disabled_in_current_mode)(struct kvm_pmc *pmc);
 
 	bool (*is_mediated_pmu_supported)(struct x86_pmu_capability *host_pmu);
-	void (*mediated_load)(struct kvm_vcpu *vcpu);
+	void (*mediated_load)(struct kvm_vcpu *vcpu, u64 host_global_ctrl);
 	void (*mediated_put)(struct kvm_vcpu *vcpu);
 	void (*write_global_ctrl)(u64 global_ctrl);
 
diff --git a/arch/x86/kvm/svm/pmu.c b/arch/x86/kvm/svm/pmu.c
index f81817606baa..d9cd2a5ab411 100644
--- a/arch/x86/kvm/svm/pmu.c
+++ b/arch/x86/kvm/svm/pmu.c
@@ -246,7 +246,7 @@ static bool amd_pmu_is_mediated_pmu_supported(struct x86_pmu_capability *host_pm
 	return host_pmu->version >= 2;
 }
 
-static void amd_mediated_pmu_load(struct kvm_vcpu *vcpu)
+static void amd_mediated_pmu_load(struct kvm_vcpu *vcpu, u64 host_global_ctrl)
 {
 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
 	u64 global_status;
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index 9236bfa15c41..d0373de951d5 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -947,11 +947,22 @@ static u64 intel_fixed_ctrl_host_bits(struct kvm_pmu *pmu)
 	return fixed_ctl;
 }
 
-static void intel_mediated_pmu_load(struct kvm_vcpu *vcpu)
+static void intel_mediated_pmu_load(struct kvm_vcpu *vcpu, u64 host_global_ctrl)
 {
 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
 	u64 global_status, toggle;
 
+	/*
+	 * Preserve host-owned bits: perf may schedule !exclude_guest events on
+	 * host-owned counters in non-root mode.
+	 * PerfMon masking requires VM_EXIT_SAVE_IA32_PERF_GLOBAL_CTRL, so the
+	 * MSR-store/load path does not apply here.
+	 */
+	if (kvm_vcpu_has_perfmon_mask(vcpu)) {
+		host_global_ctrl &= pmu->global_ctrl_rsvd;
+		intel_pmu_write_global_ctrl(pmu->global_ctrl | host_global_ctrl);
+	}
+
 	if (kvm_vcpu_has_perf_metrics(vcpu))
 		wrmsrq(MSR_PERF_METRICS, pmu->perf_metrics);
 
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 1b42c9d6f168..a241efb99b3d 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -7467,6 +7467,10 @@ static void vmx_refresh_guest_perf_global_control(struct kvm_vcpu *vcpu)
 	}
 
 	pmu->global_ctrl = vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL);
+
+	/* Strip host-owned bits that were ORed into the VMCS on VMX entry. */
+	if (kvm_vcpu_has_perfmon_mask(vcpu))
+		pmu->global_ctrl &= ~pmu->global_ctrl_rsvd;
 }
 
 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
-- 
2.55.0