Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Sender: Paolo Bonzini <paolo.bonzini@gmail.com>
Message-ID: <2e60bd73-e4da-eb4e-4eae-e43be7fd5bcd@redhat.com>
Date:   Fri, 29 Jul 2022 00:25:16 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [PATCH v4 21/25] KVM: VMX: Move LOAD_IA32_PERF_GLOBAL_CTRL errata
 handling out of setup_vmcs_config()
Content-Language: en-US
To:     Sean Christopherson <seanjc@google.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>
Cc:     kvm@vger.kernel.org,
        Anirudh Rayabharam <anrayabh@linux.microsoft.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Maxim Levitsky <mlevitsk@redhat.com>,
        linux-hyperv@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20220714091327.1085353-1-vkuznets@redhat.com>
 <20220714091327.1085353-22-vkuznets@redhat.com> <YtnZmCutdd5tpUmz@google.com>
From:   Paolo Bonzini <pbonzini@redhat.com>
In-Reply-To: <YtnZmCutdd5tpUmz@google.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 7/22/22 00:56, Sean Christopherson wrote:
> Except the errata are based on FMS and the FMS exposed to the L1 hypervisor may
> not be the real FMS.
> 
> But that's moot, because they_should_  be fully emulated by KVM anyways; KVM
> runs L2 with a MSR value modified by perf, not the raw MSR value requested by L1.
> 
> Of course KVM screws things up and fails to clear the flag in entry controls...
> All exit controls are emulated so at least KVM gets those right.

Can you send this as a separate patch?

Paolo

> Untested, but I believe KVM the fix is:
> 
> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> index d0e781c7ac72..76926147b672 100644
> --- a/arch/x86/kvm/vmx/nested.c
> +++ b/arch/x86/kvm/vmx/nested.c
> @@ -2357,7 +2357,8 @@ static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs0
>           * we can avoid VMWrites during vmx_set_efer().
>           */
>          exec_control = __vm_entry_controls_get(vmcs01);
> -       exec_control |= vmcs12->vm_entry_controls;
> +       exec_control |= (vmcs12->vm_entry_controls &
> +                        ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
>          exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
>          if (cpu_has_load_ia32_efer()) {
>                  if (guest_efer & EFER_LMA)