From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kiszka Subject: Re: [PATCH] KVM: VMX: Do not overwrite vcpu->srcu_idx in vmx_vcpu_reset Date: Thu, 14 Mar 2013 20:20:26 +0100 Message-ID: <514222FA.9060706@siemens.com> References: <5141E41B.8080804@siemens.com> <20130314150004.GX11223@redhat.com> <20130314191438.GA9009@amt.cnet> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Gleb Natapov , kvm , Paolo Bonzini To: Marcelo Tosatti Return-path: Received: from david.siemens.de ([192.35.17.14]:16999 "EHLO david.siemens.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751446Ab3CNTUd (ORCPT ); Thu, 14 Mar 2013 15:20:33 -0400 In-Reply-To: <20130314191438.GA9009@amt.cnet> Sender: kvm-owner@vger.kernel.org List-ID: On 2013-03-14 20:14, Marcelo Tosatti wrote: > On Thu, Mar 14, 2013 at 05:00:04PM +0200, Gleb Natapov wrote: >> On Thu, Mar 14, 2013 at 03:52:11PM +0100, Jan Kiszka wrote: >>> vmx_vcpu_reset may now be called while already holding the srcu lock, so >>> we may overwrite what was already saved there. Also, we lock and unlock >>> in the same context, thus there was no need to save to the vcpu anyway. >>> >>> Signed-off-by: Jan Kiszka >>> --- >>> >>> Marcelo just suggested this as the simplest fix for the issue caused by >>> the INIT/SIPI patch. Avoiding srcu lock for TSS handling might still be >>> possible but more tricky. >>> >>> arch/x86/kvm/vmx.c | 5 +++-- >>> 1 files changed, 3 insertions(+), 2 deletions(-) >>> >>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c >>> index 958ac3a..be5b1dc 100644 >>> --- a/arch/x86/kvm/vmx.c >>> +++ b/arch/x86/kvm/vmx.c >>> @@ -4117,6 +4117,7 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu) >>> { >>> struct vcpu_vmx *vmx = to_vmx(vcpu); >>> u64 msr; >>> + int idx; >>> >>> vmx->rmode.vm86_active = 0; >>> >>> @@ -4190,9 +4191,9 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu) >>> vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid); >>> >>> vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET; >>> - vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); >>> + idx = srcu_read_lock(&vcpu->kvm->srcu); >>> vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */ >>> - srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); >>> + srcu_read_unlock(&vcpu->kvm->srcu, idx); >> vmx_set_cr0() does: >> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); >> vmx_set_tss_addr(vcpu->kvm, 0xfeffd000); >> vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); >> So with this change the sequence will be: >> >> vcpu->srcu_idx = srcu_read_lock() >> idx = srcu_read_lock(&vcpu->kvm->srcu); >> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); >> vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); >> srcu_read_unlock(&vcpu->kvm->srcu, idx); >> srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); >> >> Not sure this is valid. > > The lock/unlocks must be paired. Did you find out more than what Paolo reported? > > Pass parameters around to make that happen? Or we save/restore srcu_idx when overwriting. Jan -- Siemens AG, Corporate Technology, CT RTC ITP SDP-DE Corporate Competence Center Embedded Linux