From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [RFC PATCH 0/5] Utilizing VMX preemption for timer virtualization Date: Fri, 20 May 2016 16:49:21 -0400 (EDT) Message-ID: <1863826576.16229326.1463777361204.JavaMail.zimbra@redhat.com> References: <1463708703-19208-1-git-send-email-yunhong.jiang@linux.intel.com> <20160520181830.GA28312@amt.cnet> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Cc: Yunhong Jiang , kvm@vger.kernel.org, rkrcmar@redhat.com To: Marcelo Tosatti Return-path: Received: from mx5-phx2.redhat.com ([209.132.183.37]:56872 "EHLO mx5-phx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750995AbcETUtX (ORCPT ); Fri, 20 May 2016 16:49:23 -0400 In-Reply-To: <20160520181830.GA28312@amt.cnet> Sender: kvm-owner@vger.kernel.org List-ID: > On Thu, May 19, 2016 at 06:44:58PM -0700, Yunhong Jiang wrote: > > The VMX-preemption timer is a feature on VMX, it counts down, from the > > value loaded by VM entry, in VMX nonroot operation. When the timer > > counts down to zero, it stops counting down and a VM exit occurs. > > > > The VMX preemption timer for tsc deadline timer virtualization. The > > VMX preemption timer is armed when the vCPU is running, and a VMExit > > will happen if the virtual TSC deadline timer expires. > > > > When the vCPU thread is scheduled out, the tsc deadline timer > > virtualization will be switched to use the current solution, i.e. use > > the timer for it. It's switched back to VMX preemption timer when the > > vCPU thread is scheduled int. > > > > This solution replace the complex OS's hrtimer system, and also the > > host timer interrupt handling cost, with a preemption_timer VMexit. It > > fits well for some NFV usage scenario, when the vCPU is bound to a > > pCPU and the pCPU is isolated, or some similar scenarioes. > > > > However, it possibly has impact if the vCPU thread is scheduled in/out > > very frequently, because it switches from/to the hrtimer emulation a > > lot. A module parameter is provided to turn it on or off. > > > > Signed-off-by: Yunhong Jiang > > Hi Yunhong Jiang, > > This adds cost to the VM-exit and VM-entry paths (additional > instructions and i-cache pressure). Also it adds cost to > kvm_sched_out. For now my review limited itself to making the code nicer without touching the overall design too much. I'm confident that we can reduce it to a dozen instructions on vmentry and only pay the cost of hrtimer_start on failed HLT polls (enabling the hrtimer only before going to sleep). Assuming that device interrupts are delivered while the guest is running or during a successful HLT polls, that would be very rare. > What is the benefit the switch from external interrupt to VMX preemption > timer brings? hrtimer_start/hrtimer_cancel do show up in profiles for dynticks guests. Since they touch a red-black tree, it shouldn't be hard to outperform them. Paolo