From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752384AbdCPOGt (ORCPT ); Thu, 16 Mar 2017 10:06:49 -0400 Received: from mx1.redhat.com ([209.132.183.28]:34176 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751842AbdCPOGr (ORCPT ); Thu, 16 Mar 2017 10:06:47 -0400 DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com D06A181F03 Authentication-Results: ext-mx03.extmail.prod.ext.phx2.redhat.com; dmarc=none (p=none dis=none) header.from=redhat.com Authentication-Results: ext-mx03.extmail.prod.ext.phx2.redhat.com; spf=pass smtp.mailfrom=mst@redhat.com DKIM-Filter: OpenDKIM Filter v2.11.0 mx1.redhat.com D06A181F03 Date: Thu, 16 Mar 2017 16:04:12 +0200 From: "Michael S. Tsirkin" To: "Gabriel L. Somlo" Cc: linux-kernel@vger.kernel.org, Paolo Bonzini , Radim =?utf-8?B?S3LEjW3DocWZ?= , Jonathan Corbet , Thomas Gleixner , Ingo Molnar , "H. Peter Anvin" , x86@kernel.org, Joerg Roedel , kvm@vger.kernel.org, linux-doc@vger.kernel.org Subject: Re: [PATCH v5 untested] kvm: better MWAIT emulation for guests Message-ID: <20170316155550-mutt-send-email-mst@kernel.org> References: <1489612895-12799-1-git-send-email-mst@redhat.com> <20170315233534.GG2239@HEDWIG.INI.CMU.EDU> <20170316013903-mutt-send-email-mst@kernel.org> <20170316132426.GB4085@HEDWIG.INI.CMU.EDU> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: <20170316132426.GB4085@HEDWIG.INI.CMU.EDU> X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.27]); Thu, 16 Mar 2017 14:04:17 +0000 (UTC) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Mar 16, 2017 at 09:24:27AM -0400, Gabriel L. Somlo wrote: > After studying your patch a bit more carefully (sorry, it's crazy > around here right now :) ) I realized you're simply trying to > (selectively) decide when to exit L1 and emulate as NOP vs. when to > just allow L1 to execute MONITOR & MWAIT natively. > > Is that right ? Because if so, the issues I saw on my MacPro1,1 are > weird and inexplicable, given that allowing L>=1 to run MONITOR/MWAIT > natively was one of the options Alex Graf and Rene Rebe used back in > the very early days of OS X on QEMU, at the time I got involved with > that project. Here's part of an out of tree patch against 3.4 which did > just that, and worked as far as I remember on *any* MWAIT capable > intel chip I had access to back in 2010: > > ############################################################################## > # 99-mwait.patch.kvm-kmod (Rene Rebe ) 2010-04-27 > ############################################################################## > diff -pNarU5 linux-3.4/arch/x86/kvm/cpuid.c linux-3.4-mac/arch/x86/kvm/cpuid.c > --- linux-3.4/arch/x86/kvm/cpuid.c 2012-05-20 18:29:13.000000000 -0400 > +++ linux-3.4-mac/arch/x86/kvm/cpuid.c 2012-10-09 11:42:59.921215750 -0400 > @@ -222,11 +222,11 @@ static int do_cpuid_ent(struct kvm_cpuid > f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | > F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | > 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); > /* cpuid 1.ecx */ > const u32 kvm_supported_word4_x86_features = > - F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | > + F(XMM3) | F(PCLMULQDQ) | F(MWAIT) /* DTES64, MONITOR */ | > 0 /* DS-CPL, VMX, SMX, EST */ | > 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | > F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | > 0 /* Reserved, DCA */ | F(XMM4_1) | > F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | > diff -pNarU5 linux-3.4/arch/x86/kvm/svm.c linux-3.4-mac/arch/x86/kvm/svm.c > --- linux-3.4/arch/x86/kvm/svm.c 2012-05-20 18:29:13.000000000 -0400 > +++ linux-3.4-mac/arch/x86/kvm/svm.c 2012-10-09 11:44:41.598997481 -0400 > @@ -1102,12 +1102,10 @@ static void init_vmcb(struct vcpu_svm *s > set_intercept(svm, INTERCEPT_VMSAVE); > set_intercept(svm, INTERCEPT_STGI); > set_intercept(svm, INTERCEPT_CLGI); > set_intercept(svm, INTERCEPT_SKINIT); > set_intercept(svm, INTERCEPT_WBINVD); > - set_intercept(svm, INTERCEPT_MONITOR); > - set_intercept(svm, INTERCEPT_MWAIT); > set_intercept(svm, INTERCEPT_XSETBV); > > control->iopm_base_pa = iopm_base; > control->msrpm_base_pa = __pa(svm->msrpm); > control->int_ctl = V_INTR_MASKING_MASK; > diff -pNarU5 linux-3.4/arch/x86/kvm/vmx.c linux-3.4-mac/arch/x86/kvm/vmx.c > --- linux-3.4/arch/x86/kvm/vmx.c 2012-05-20 18:29:13.000000000 -0400 > +++ linux-3.4-mac/arch/x86/kvm/vmx.c 2012-10-09 11:42:59.925215977 -0400 > @@ -1938,11 +1938,11 @@ static __init void nested_vmx_setup_ctls > nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high); > nested_vmx_procbased_ctls_low = 0; > nested_vmx_procbased_ctls_high &= > CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING | > CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING | > - CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING | > + CPU_BASED_CR3_LOAD_EXITING | > CPU_BASED_CR3_STORE_EXITING | > #ifdef CONFIG_X86_64 > CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING | > #endif > CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING | > @@ -2404,12 +2404,10 @@ static __init int setup_vmcs_config(stru > CPU_BASED_CR3_LOAD_EXITING | > CPU_BASED_CR3_STORE_EXITING | > CPU_BASED_USE_IO_BITMAPS | > CPU_BASED_MOV_DR_EXITING | > CPU_BASED_USE_TSC_OFFSETING | > - CPU_BASED_MWAIT_EXITING | > - CPU_BASED_MONITOR_EXITING | > CPU_BASED_INVLPG_EXITING | > CPU_BASED_RDPMC_EXITING; > > opt = CPU_BASED_TPR_SHADOW | > CPU_BASED_USE_MSR_BITMAPS | > > If all you're trying to do is (selectively) revert to this behavior, > that "shouldn't" mess it up for the MacPro either, so I'm thoroughly > confused at this point :) Yes. Me too. Want to try that other patch and see what happens? > Back in 2010, running MWAIT in L>=1 behaved 100% exactly like a NOP, > didn't power down the physical CPU, just immediately moved on to the > next instruction. As such, there was no power saving and no > opportunity to yield to another L0 thread either, unlike with NOP > emulation at L0. > > Did that change on newer Intel chips (i.e., is guest-mode MWAIT now > doing something smarter than just acting as a guest-mode NOP) ? > > Thanks, > --Gabriel Interesting. What it seems to say is this: MWAIT. Behavior of the MWAIT instruction (which always causes an invalid- opcode exception—#UD—if CPL > 0) is determined by the setting of the “MWAIT exiting” VM-execution control: — If the “MWAIT exiting” VM-execution control is 1, MWAIT causes a VM exit (see Section 22.1.3). — If the “MWAIT exiting” VM-execution control is 0, MWAIT operates normally if any of the following is true: (1) the “interrupt-window exiting” VM-execution control is 0; (2) ECX[0] is 0; or (3) RFLAGS.IF = 1. — If the “MWAIT exiting” VM-execution control is 0, the “interrupt-window exiting” VM-execution control is 1, ECX[0] = 1, and RFLAGS.IF = 0, MWAIT does not cause the processor to enter an implementation-dependent optimized state; instead, control passes to the instruction following the MWAIT instruction. And since interrupt-window exiting is 0 most of the time for KVM, I would expect MWAIT to behave normally. -- MST