From mboxrd@z Thu Jan  1 00:00:00 1970
From: Marcelo Tosatti <mtosatti@redhat.com>
Subject: Re: [PATCH v5] KVM: VMX: Execute WBINVD to keep data consistency
 with assigned devices
Date: Tue, 29 Jun 2010 10:35:16 -0300
Message-ID: <20100629133516.GC28951@amt.cnet>
References: <4C286CCE.10309@redhat.com>
 <1277781419-13227-1-git-send-email-sheng@linux.intel.com>
 <20100629132541.GB27338@amt.cnet>
 <4C29F503.7020400@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Sheng Yang <sheng@linux.intel.com>,
	Jan Kiszka <jan.kiszka@siemens.com>,
	Joerg Roedel <joerg.roedel@amd.com>, kvm@vger.kernel.org,
	"Yaozu (Eddie) Dong" <eddie.dong@intel.com>
To: Avi Kivity <avi@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:56705 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1754857Ab0F2NgT (ORCPT <rfc822;kvm@vger.kernel.org>);
	Tue, 29 Jun 2010 09:36:19 -0400
Content-Disposition: inline
In-Reply-To: <4C29F503.7020400@redhat.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Tue, Jun 29, 2010 at 04:28:35PM +0300, Avi Kivity wrote:
> On 06/29/2010 04:25 PM, Marcelo Tosatti wrote:
> >
> >>+			smp_call_function_single(vcpu->cpu,
> >>+					wbinvd_ipi, NULL, 1);
> >>+	}
> >>+
> >>  	kvm_x86_ops->vcpu_load(vcpu, cpu);
> >>  	if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
> >>  		unsigned long khz = cpufreq_quick_get(cpu);
> >>@@ -3650,6 +3670,21 @@ int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
> >>  	return X86EMUL_CONTINUE;
> >>  }
> >>
> >>+int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
> >>+{
> >>+	if (!need_emulate_wbinvd(vcpu))
> >>+		return X86EMUL_CONTINUE;
> >>+
> >>+	if (kvm_x86_ops->has_wbinvd_exit()) {
> >>+		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
> >>+				wbinvd_ipi, NULL, 1);
> >work_on_cpu() loop instead of smp_call_function_many(), to avoid executing
> >wbinvd with interrupts disabled.
> 
> Why?  wbinvd is not interruptible.

Right. But still, smp_call_function_many() is going to busy-spin until
the target CPUs finish their work, while work_on_cpu() will schedule.

Also the IPI request has to handled immediately, bypassing the
scheduler.