From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kiszka Subject: Re: [PATCH] KVM: nVMX: Fully support of nested VMX preemption timer Date: Sun, 25 Aug 2013 09:54:13 +0200 Message-ID: <5219B825.5000806@web.de> References: <1377369850-18583-1-git-send-email-root@Blade1-01.Blade1-01> <5219A7BA.8050602@web.de> <5219B590.7040304@web.de> Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="xHQPK62rwmrSmIqvcAVTDaafDQ7CFwpbB" Cc: gleb@redhat.com, kvm , pbonzini@redhat.com, =?ISO-2022-JP?B?IhskQk17PVU0cRsoQiA8QXJ0aHVyIENodW5xaSBMaT4i?= To: Abel Gordon Return-path: Received: from mout.web.de ([212.227.15.4]:49559 "EHLO mout.web.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756071Ab3HYHyR (ORCPT ); Sun, 25 Aug 2013 03:54:17 -0400 Received: from mchn199C.mchp.siemens.de ([95.157.58.223]) by smtp.web.de (mrweb003) with ESMTPSA (Nemesis) id 0LshXV-1WAVqh45zl-012Ko1 for ; Sun, 25 Aug 2013 09:54:15 +0200 In-Reply-To: Sender: kvm-owner@vger.kernel.org List-ID: This is an OpenPGP/MIME signed message (RFC 4880 and 3156) --xHQPK62rwmrSmIqvcAVTDaafDQ7CFwpbB Content-Type: text/plain; charset=ISO-2022-JP Content-Transfer-Encoding: quoted-printable On 2013-08-25 09:50, Abel Gordon wrote: >=20 >=20 > kvm-owner@vger.kernel.org wrote on 25/08/2013 10:43:12 AM: >=20 >> From: Jan Kiszka >> To: Abel Gordon/Haifa/IBM@IBMIL, >> Cc: gleb@redhat.com, kvm@vger.kernel.org, kvm-owner@vger.kernel.org, >> pbonzini@redhat.com, "=1B$BM{=3DU4q=1B(B " >> Date: 25/08/2013 10:43 AM >> Subject: Re: [PATCH] KVM: nVMX: Fully support of nested VMX preemption= > timer >> Sent by: kvm-owner@vger.kernel.org >> >> On 2013-08-25 09:37, Abel Gordon wrote: >>> >>> >>>> From: Jan Kiszka >>>> To: "=1B$BM{=3DU4q=1B(B " , >>>> Cc: kvm@vger.kernel.org, gleb@redhat.com, pbonzini@redhat.com >>>> Date: 25/08/2013 09:44 AM >>>> Subject: Re: [PATCH] KVM: nVMX: Fully support of nested VMX preempti= on >>> timer >>>> Sent by: kvm-owner@vger.kernel.org >>>> >>>> On 2013-08-24 20:44, root wrote: >>>>> This patch contains the following two changes: >>>>> 1. Fix the bug in nested preemption timer support. If vmexit L2->L0= >>>>> with some reasons not emulated by L1, preemption timer value should= >>>>> be save in such exits. >>>>> 2. Add support of "Save VMX-preemption timer value" VM-Exit control= s >>>>> to nVMX. >>>>> >>>>> With this patch, nested VMX preemption timer features are fully >>>>> supported. >>>>> >>>>> Signed-off-by: Arthur Chunqi Li >>>>> --- >>> >>>>> >>>>> @@ -7578,9 +7579,14 @@ static void prepare_vmcs02(struct kvm_vcpu >>>> *vcpu, struct vmcs12 *vmcs12) >>>>> (vmcs_config.pin_based_exec_ctrl | >>>>> vmcs12->pin_based_vm_exec_control)); >>>>> >>>>> - if (vmcs12->pin_based_vm_exec_control & >>> PIN_BASED_VMX_PREEMPTION_TIMER) >>>>> - vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, >>>>> - vmcs12->vmx_preemption_timer_value); >>>>> + if (vmcs12->pin_based_vm_exec_control & >>>> PIN_BASED_VMX_PREEMPTION_TIMER) { >>>>> + if (vmcs12->vm_exit_controls & >>> VM_EXIT_SAVE_VMX_PREEMPTION_TIMER) >>>>> + vmcs12->vmx_preemption_timer_value =3D >>>>> + vmcs_read32(VMX_PREEMPTION_TIMER_VALUE); >>>>> + else >>>>> + vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, >>>>> + vmcs12->vmx_preemption_timer_value); >>>>> + } >>>> >>>> This is not correct. We still need to set the vmcs to >>>> vmx_preemption_timer_value. The difference is that, on exit from L2,= >>>> vmx_preemption_timer_value has to be updated according to the saved >>>> hardware state. The corresponding code is missing in your patch so > far. >>> >>> I think something else maybe be missing here: assuming L0 handles exi= ts >>> for L2 without involving L1 (e.g. external interrupts or ept > violations), >>> then, we may spend some cycles in L0 handling these exits. Note L1 is= > not >>> aware of these exits and from L1 perspective L2 was running on the CP= U. >>> That means that we may need to reduce these cycles spent at >>> L0 from the preemtion timer or emulate a preemption timer exit to >>> force a transition to L1 instead of resuming L2. >> >> That's precisely what the logic I described should achieve: reload the= >> value we saved on L2 exit on reentry. >=20 > But don't you think we should also reduce the cycles spent at L0 from t= he > preemption timer ? I mean, if we spent X cycles at L0 handling a L2 exi= t > which was not forwarded to L1, then, before we resume L2, > the preemption timer should be: (previous_value_on_exit - X). > If (previous_value_on_exit - X) < 0, then we should force ("emulate") a= > preemption timer exit between L2 and L1. We ask the hardware to save the value of the preemption on L2 exit. This value will be exposed to L1 (if it asked for saving as well) and/or be written back to the hardware on L2 reenty (unless L1 had a chance to run and modified it). So the time spent in L0 is implicitly subtracted. Jan PS: You had kvm-owner in CC. --xHQPK62rwmrSmIqvcAVTDaafDQ7CFwpbB Content-Type: application/pgp-signature; name="signature.asc" Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.16 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/ iEYEARECAAYFAlIZuCUACgkQitSsb3rl5xQdSgCcDqpFpIRQwxM3n6E68qh6gFcR nQ4AoIvzsCILhzzAZOv9FBi1RLJaEkSj =BkfB -----END PGP SIGNATURE----- --xHQPK62rwmrSmIqvcAVTDaafDQ7CFwpbB--