From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754143AbbCBTQl (ORCPT ); Mon, 2 Mar 2015 14:16:41 -0500 Received: from mx1.redhat.com ([209.132.183.28]:45179 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752602AbbCBTQj (ORCPT ); Mon, 2 Mar 2015 14:16:39 -0500 Date: Mon, 2 Mar 2015 19:49:54 +0100 From: Radim =?utf-8?B?S3LEjW3DocWZ?= To: Joel Schopp Cc: Gleb Natapov , Paolo Bonzini , kvm@vger.kernel.org, David Kaplan , Joerg Roedel , linux-kernel@vger.kernel.org, Borislav Petkov Subject: Re: [PATCH v2] x86: svm: use kvm_fast_pio_in() Message-ID: <20150302184953.GB25123@potion.brq.redhat.com> References: <20150302164004.17276.77376.stgit@joelvmguard2.amd.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20150302164004.17276.77376.stgit@joelvmguard2.amd.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 2015-03-02 10:40-0600, Joel Schopp: > From: David Kaplan > > We can make the in instruction go faster the same way the out instruction is > already. > > Changes from v1 > * Added kvm_fast_pio_in() implementation that was left out of v1 > > Signed-off-by: David Kaplan > [extracted from larger unlrelated patch, forward ported, tested] > Signed-off-by: Joel Schopp > --- > diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c > index d319e0c..f8c906b 100644 > --- a/arch/x86/kvm/svm.c > +++ b/arch/x86/kvm/svm.c > @@ -1899,7 +1899,7 @@ static int io_interception(struct vcpu_svm *svm) > ++svm->vcpu.stat.io_exits; > string = (io_info & SVM_IOIO_STR_MASK) != 0; > in = (io_info & SVM_IOIO_TYPE_MASK) != 0; > - if (string || in) > + if (string) (I guess that most accesses are covered now, so we don't need to make REP case a bit faster ...) > return emulate_instruction(vcpu, 0) == EMULATE_DONE; > > port = io_info >> 16; > @@ -1907,6 +1907,8 @@ static int io_interception(struct vcpu_svm *svm) > svm->next_rip = svm->vmcb->control.exit_info_2; > skip_emulated_instruction(&svm->vcpu); > > + if (in) > + return kvm_fast_pio_in(vcpu, size, port); > return kvm_fast_pio_out(vcpu, size, port); (kvm_fast_pio() comes to mind.) > } > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c > index bd7a70b..089247c 100644 > --- a/arch/x86/kvm/x86.c > +++ b/arch/x86/kvm/x86.c > @@ -5463,6 +5463,39 @@ int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port) > } > EXPORT_SYMBOL_GPL(kvm_fast_pio_out); > > +static int complete_fast_pio(struct kvm_vcpu *vcpu) > +{ > + u32 new_rax = kvm_register_read(vcpu, VCPU_REGS_RAX); u64. > + > + if (!vcpu->arch.pio.count) > + return 0; > + if (vcpu->arch.pio.count * vcpu->arch.pio.size > 8) > + return 0; sizeof(new_rax). (safer and easier to understand) Both should never happen in KVM code, BUG_ON(). > + > + memcpy(&new_rax, vcpu->arch.pio_data, > + vcpu->arch.pio.count * vcpu->arch.pio.size); Use emulator_pio_in_emulated() here, for code sharing. (We want to trace the read here too; it could be better to split the path from emulator_pio_in_emulated() first.) > + kvm_register_write(vcpu, VCPU_REGS_RAX, new_rax); > + > + vcpu->arch.pio.count = 0; > + return 1; > +} > + > +int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port) > +{ > + unsigned long val; > + int ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, > + port, &val, 1); > + > + if (ret) { > + kvm_register_write(vcpu, VCPU_REGS_RAX, val); > + vcpu->arch.pio.count = 0; (emulator_pio_in_emulated() sets count to zero if it returns true.) > + } else > + vcpu->arch.complete_userspace_io = complete_fast_pio; > + > + return ret; > +} > +EXPORT_SYMBOL_GPL(kvm_fast_pio_in); > + > static void tsc_bad(void *info) > { > __this_cpu_write(cpu_tsc_khz, 0); >