Re: [PATCH v6 44/44] KVM: VMX: Add mediated PMU support for CPUs without "save perf global ctrl"

["Mi, Dapeng" <dapeng1.mi@linux.intel.com>]

On 12/6/2025 8:17 AM, Sean Christopherson wrote:
> Extend mediated PMU support for Intel CPUs without support for saving
> PERF_GLOBAL_CONTROL into the guest VMCS field on VM-Exit, e.g. for Skylake
> and its derivatives, as well as Icelake.  While supporting CPUs without
> VM_EXIT_SAVE_IA32_PERF_GLOBAL_CTRL isn't completely trivial, it's not that
> complex either.  And not supporting such CPUs would mean not supporting 7+
> years of Intel CPUs released in the past 10 years.
>
> On VM-Exit, immediately propagate the saved PERF_GLOBAL_CTRL to the VMCS
> as well as KVM's software cache so that KVM doesn't need to add full EXREG
> tracking of PERF_GLOBAL_CTRL.  In practice, the vast majority of VM-Exits
> won't trigger software writes to guest PERF_GLOBAL_CTRL, so deferring the
> VMWRITE to the next VM-Enter would only delay the inevitable without
> batching/avoiding VMWRITEs.
>
> Note!  Take care to refresh VM_EXIT_MSR_STORE_COUNT on nested VM-Exit, as
> it's unfortunately possible that KVM could recalculate MSR intercepts
> while L2 is active, e.g. if userspace loads nested state and _then_ sets
> PERF_CAPABILITIES.  Eating the VMWRITE on every nested VM-Exit is
> unfortunate, but that's a pre-existing problem and can/should be solved
> separately, e.g. modifying the number of auto-load entries while L2 is
> active is also uncommon on modern CPUs.
>
> Signed-off-by: Sean Christopherson <seanjc@google.com>
> ---
>  arch/x86/kvm/vmx/nested.c    |  6 ++++-
>  arch/x86/kvm/vmx/pmu_intel.c |  7 -----
>  arch/x86/kvm/vmx/vmx.c       | 52 ++++++++++++++++++++++++++++++++----
>  3 files changed, 52 insertions(+), 13 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> index 614b789ecf16..1ee1edc8419d 100644
> --- a/arch/x86/kvm/vmx/nested.c
> +++ b/arch/x86/kvm/vmx/nested.c
> @@ -5142,7 +5142,11 @@ void __nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
>  
>  	kvm_nested_vmexit_handle_ibrs(vcpu);
>  
> -	/* Update any VMCS fields that might have changed while L2 ran */
> +	/*
> +	 * Update any VMCS fields that might have changed while vmcs02 was the
> +	 * active VMCS.  The tracking is per-vCPU, not per-VMCS.
> +	 */
> +	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.nr);
>  	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
>  	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
>  	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
> index 55249fa4db95..27eb76e6b6a0 100644
> --- a/arch/x86/kvm/vmx/pmu_intel.c
> +++ b/arch/x86/kvm/vmx/pmu_intel.c
> @@ -777,13 +777,6 @@ static bool intel_pmu_is_mediated_pmu_supported(struct x86_pmu_capability *host_
>  	if (WARN_ON_ONCE(!cpu_has_load_perf_global_ctrl()))
>  		return false;
>  
> -	/*
> -	 * KVM doesn't yet support mediated PMU on CPUs without support for
> -	 * saving PERF_GLOBAL_CTRL via a dedicated VMCS field.
> -	 */
> -	if (!cpu_has_save_perf_global_ctrl())
> -		return false;
> -
>  	return true;
>  }
>  
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 6a17cb90eaf4..ba1262c3e3ff 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -1204,6 +1204,17 @@ static bool update_transition_efer(struct vcpu_vmx *vmx)
>  	return true;
>  }
>  
> +static void vmx_add_autostore_msr(struct vcpu_vmx *vmx, u32 msr)
> +{
> +	vmx_add_auto_msr(&vmx->msr_autostore, msr, 0, VM_EXIT_MSR_STORE_COUNT,
> +			 vmx->vcpu.kvm);
> +}
> +
> +static void vmx_remove_autostore_msr(struct vcpu_vmx *vmx, u32 msr)
> +{
> +	vmx_remove_auto_msr(&vmx->msr_autostore, msr, VM_EXIT_MSR_STORE_COUNT);
> +}
> +
>  #ifdef CONFIG_X86_32
>  /*
>   * On 32-bit kernels, VM exits still load the FS and GS bases from the
> @@ -4225,6 +4236,8 @@ void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
>  
>  static void vmx_recalc_pmu_msr_intercepts(struct kvm_vcpu *vcpu)
>  {
> +	u64 vm_exit_controls_bits = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
> +				    VM_EXIT_SAVE_IA32_PERF_GLOBAL_CTRL;
>  	bool has_mediated_pmu = kvm_vcpu_has_mediated_pmu(vcpu);
>  	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
>  	struct vcpu_vmx *vmx = to_vmx(vcpu);
> @@ -4234,12 +4247,19 @@ static void vmx_recalc_pmu_msr_intercepts(struct kvm_vcpu *vcpu)
>  	if (!enable_mediated_pmu)
>  		return;
>  
> +	if (!cpu_has_save_perf_global_ctrl()) {
> +		vm_exit_controls_bits &= ~VM_EXIT_SAVE_IA32_PERF_GLOBAL_CTRL;
> +
> +		if (has_mediated_pmu)
> +			vmx_add_autostore_msr(vmx, MSR_CORE_PERF_GLOBAL_CTRL);
> +		else
> +			vmx_remove_autostore_msr(vmx, MSR_CORE_PERF_GLOBAL_CTRL);
> +	}
> +
>  	vm_entry_controls_changebit(vmx, VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
>  				    has_mediated_pmu);
>  
> -	vm_exit_controls_changebit(vmx, VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
> -					VM_EXIT_SAVE_IA32_PERF_GLOBAL_CTRL,
> -				   has_mediated_pmu);
> +	vm_exit_controls_changebit(vmx, vm_exit_controls_bits, has_mediated_pmu);
>  
>  	for (i = 0; i < pmu->nr_arch_gp_counters; i++) {
>  		vmx_set_intercept_for_msr(vcpu, MSR_IA32_PERFCTR0 + i,
> @@ -7346,6 +7366,29 @@ static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
>  					      msrs[i].host);
>  }
>  
> +static void vmx_refresh_guest_perf_global_control(struct kvm_vcpu *vcpu)
> +{
> +	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
> +	struct vcpu_vmx *vmx = to_vmx(vcpu);
> +
> +	if (msr_write_intercepted(vmx, MSR_CORE_PERF_GLOBAL_CTRL))
> +		return;
> +
> +	if (!cpu_has_save_perf_global_ctrl()) {
> +		int slot = vmx_find_loadstore_msr_slot(&vmx->msr_autostore,
> +						       MSR_CORE_PERF_GLOBAL_CTRL);
> +
> +		if (WARN_ON_ONCE(slot < 0))
> +			return;
> +
> +		pmu->global_ctrl = vmx->msr_autostore.val[slot].value;
> +		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL, pmu->global_ctrl);
> +		return;
> +	}
> +
> +	pmu->global_ctrl = vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL);
> +}
> +
>  static void vmx_update_hv_timer(struct kvm_vcpu *vcpu, bool force_immediate_exit)
>  {
>  	struct vcpu_vmx *vmx = to_vmx(vcpu);
> @@ -7631,8 +7674,7 @@ fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu, u64 run_flags)
>  
>  	vmx->loaded_vmcs->launched = 1;
>  
> -	if (!msr_write_intercepted(vmx, MSR_CORE_PERF_GLOBAL_CTRL))
> -		vcpu_to_pmu(vcpu)->global_ctrl = vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL);
> +	vmx_refresh_guest_perf_global_control(vcpu);
>  
>  	vmx_recover_nmi_blocking(vmx);
>  	vmx_complete_interrupts(vmx);

The change looks good to me, but I still have no bandwidth to test the code
on Ice lake, I would go back after I finish the tests. Thanks.