From: Vitaly Kuznetsov <vkuznets@redhat.com> To: Sean Christopherson <sean.j.christopherson@intel.com> Cc: Paolo Bonzini <pbonzini@redhat.com>, kvm@vger.kernel.org, Jim Mattson <jmattson@google.com>, linux-kernel@vger.kernel.org, Liran Alon <liran.alon@oracle.com>, Roman Kagan <rkagan@virtuozzo.com> Subject: Re: [PATCH RFC 2/3] x86/kvm/hyper-v: move VMX controls sanitization out of nested_enable_evmcs() Date: Wed, 22 Jan 2020 17:29:49 +0100 Message-ID: <87blqvsbcy.fsf@vitty.brq.redhat.com> (raw) In-Reply-To: <20200122155108.GA7201@linux.intel.com> Sean Christopherson <sean.j.christopherson@intel.com> writes: > On Wed, Jan 22, 2020 at 04:08:55PM +0100, Vitaly Kuznetsov wrote: >> Paolo Bonzini <pbonzini@redhat.com> writes: >> >> > On 22/01/20 06:47, Sean Christopherson wrote: >> >>> Yes, it most likely is and it would be nice if Microsoft fixed it, but I >> >>> guess we're stuck with it for existing Windows versions. Well, for one >> >>> we found a bug in Hyper-V and not the converse. :) >> >>> >> >>> There is a problem with this approach, in that we're stuck with it >> >>> forever due to live migration. But I guess if in the future eVMCS v2 >> >>> adds an apic_address field we can limit the hack to eVMCS v1. Another >> >>> possibility is to use the quirks mechanism but it's overkill for now. >> >>> >> >>> Unless there are objections, I plan to apply these patches. >> >> Doesn't applying this patch contradict your earlier opinion? This patch >> >> would still hide the affected controls from the guest because the host >> >> controls enlightened_vmcs_enabled. >> > >> > It does. Unfortunately the key sentence is "we're stuck with it for >> > existing Windows versions". :( > > Ah, I didn't understand what "it" referred to :-) > >> >> Rather than update vmx->nested.msrs or filter vmx_get_msr(), what about >> >> manually adding eVMCS consistency checks on the disallowed bits and handle >> >> SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES as a one-off case by simply >> >> clearing it from the eVMCS? Or alternatively, squashing all the disallowed >> >> bits. >> > >> > Hmm, that is also a possibility. It's a very hacky one, but I guess >> > adding APIC virtualization to eVMCS would require bumping the version to >> > 2. Vitaly, what do you think? >> >> As I already replied to Sean I like the idea to filter out unsupported >> controls from eVMCS but unfortunately it doesn't work: Hyper-V actually >> expects APIC virtualization to work when it enables >> SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES (I have no idea how without >> apic_access_addr field but). I checked and at least Hyper-V 2016 doesn't >> boot (when >1 vCPU). > > Nice. > > I still don't see what we gain from applying this patch. Once eVMCS is > enabled by userspace, which presumably happens before the guest is launched, > the guest will see the eVMCS-unfriendly controls as being unsupported, both > for eVMCS and regular VMCS. AFAICT, we're adding a fairly ugly hack to KVM > just so that KVM can lie to userspace about what controls will be exposed to > the guest. > > Can we extend the API to use cap->args[1] to control whether or not the > unsupported controls are removed from vmx->nested.msrs? Userspace could > pass '1' to leave the controls untouched and then surgically hide the > controls that the guest is too dumb to know it shouldn't use by writing the > appropriate MSRs. Assuming existing userspace is expected/required to zero > out args[1..3], this would be fully backwards compatible. Yes, in case we're back to the idea to filter things out in QEMU we can do this. What I don't like is that every other userspace which decides to enable eVMCS will have to perform the exact same surgery as in case it sets allow_unsupported_controls=0 it'll have to know (hardcode) the filtering (or KVM_SET_MSRS will fail) and in case it opts for allow_unsupported_controls=1 Windows guests just won't boot without the filtering. It seems to be 1:1, eVMCSv1 requires the filter. > > > diff --git a/arch/x86/kvm/vmx/evmcs.c b/arch/x86/kvm/vmx/evmcs.c > index 72359709cdc1..241a769be738 100644 > --- a/arch/x86/kvm/vmx/evmcs.c > +++ b/arch/x86/kvm/vmx/evmcs.c > @@ -346,8 +346,8 @@ uint16_t nested_get_evmcs_version(struct kvm_vcpu *vcpu) > return 0; > } > > -int nested_enable_evmcs(struct kvm_vcpu *vcpu, > - uint16_t *vmcs_version) > +int nested_enable_evmcs(struct kvm_vcpu *vcpu, uint16_t *vmcs_version, > + bool allow_unsupported_controls) Personally, I'd call it 'keep_unsupported_controls'. > { > struct vcpu_vmx *vmx = to_vmx(vcpu); > bool evmcs_already_enabled = vmx->nested.enlightened_vmcs_enabled; > @@ -358,7 +358,7 @@ int nested_enable_evmcs(struct kvm_vcpu *vcpu, > *vmcs_version = nested_get_evmcs_version(vcpu); > > /* We don't support disabling the feature for simplicity. */ > - if (evmcs_already_enabled) > + if (evmcs_already_enabled || allow_unsupported_controls) > return 0; > > vmx->nested.msrs.pinbased_ctls_high &= ~EVMCS1_UNSUPPORTED_PINCTRL; > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c > index 0cccc52e2d0a..5e1b8d51277b 100644 > --- a/arch/x86/kvm/x86.c > +++ b/arch/x86/kvm/x86.c > @@ -4005,7 +4005,8 @@ static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, > case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: > if (!kvm_x86_ops->nested_enable_evmcs) > return -ENOTTY; > - r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version); > + r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version, > + cap->args[1]); > if (!r) { > user_ptr = (void __user *)(uintptr_t)cap->args[0]; > if (copy_to_user(user_ptr, &vmcs_version, > -- Vitaly
next prev parent reply index Thread overview: 34+ messages / expand[flat|nested] mbox.gz Atom feed top 2020-01-15 17:10 [PATCH RFC 0/3] x86/kvm/hyper-v: fix enlightened VMCS & QEMU4.2 Vitaly Kuznetsov 2020-01-15 17:10 ` [PATCH RFC 1/3] x86/kvm/hyper-v: remove stale evmcs_already_enabled check from nested_enable_evmcs() Vitaly Kuznetsov 2020-01-15 22:50 ` Liran Alon 2020-01-15 17:10 ` [PATCH RFC 2/3] x86/kvm/hyper-v: move VMX controls sanitization out of nested_enable_evmcs() Vitaly Kuznetsov 2020-01-15 22:49 ` Liran Alon 2020-01-16 8:37 ` Vitaly Kuznetsov 2020-02-03 15:11 ` Vitaly Kuznetsov 2020-01-15 23:27 ` Sean Christopherson 2020-01-15 23:30 ` Liran Alon 2020-01-16 8:51 ` Vitaly Kuznetsov 2020-01-16 16:19 ` Sean Christopherson 2020-01-16 16:57 ` Vitaly Kuznetsov 2020-01-17 6:31 ` Sean Christopherson 2020-01-18 21:42 ` Paolo Bonzini 2020-01-19 8:54 ` Paolo Bonzini 2020-01-22 5:47 ` Sean Christopherson 2020-01-22 9:37 ` Vitaly Kuznetsov 2020-01-22 14:33 ` Paolo Bonzini 2020-01-22 15:08 ` Vitaly Kuznetsov 2020-01-22 15:51 ` Sean Christopherson 2020-01-22 16:29 ` Vitaly Kuznetsov [this message] 2020-01-22 16:40 ` Paolo Bonzini 2020-01-23 9:15 ` Vitaly Kuznetsov 2020-01-23 19:09 ` Vitaly Kuznetsov 2020-01-24 17:25 ` Sean Christopherson 2020-01-27 15:38 ` Vitaly Kuznetsov 2020-01-27 17:53 ` Paolo Bonzini 2020-01-27 21:52 ` Vitaly Kuznetsov 2020-01-27 18:17 ` Sean Christopherson 2020-01-15 17:10 ` [PATCH RFC 3/3] x86/kvm/hyper-v: don't allow to turn on unsupported VMX controls for nested guests Vitaly Kuznetsov 2020-01-15 22:59 ` Liran Alon 2020-01-16 8:55 ` Vitaly Kuznetsov 2020-01-16 16:21 ` Sean Christopherson 2020-01-19 8:57 ` Paolo Bonzini
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=87blqvsbcy.fsf@vitty.brq.redhat.com \ --to=vkuznets@redhat.com \ --cc=jmattson@google.com \ --cc=kvm@vger.kernel.org \ --cc=linux-kernel@vger.kernel.org \ --cc=liran.alon@oracle.com \ --cc=pbonzini@redhat.com \ --cc=rkagan@virtuozzo.com \ --cc=sean.j.christopherson@intel.com \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
KVM Archive on lore.kernel.org Archives are clonable: git clone --mirror https://lore.kernel.org/kvm/0 kvm/git/0.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 kvm kvm/ https://lore.kernel.org/kvm \ kvm@vger.kernel.org public-inbox-index kvm Example config snippet for mirrors Newsgroup available over NNTP: nntp://nntp.lore.kernel.org/org.kernel.vger.kvm AGPL code for this site: git clone https://public-inbox.org/public-inbox.git