From mboxrd@z Thu Jan 1 00:00:00 1970 From: Vaidyanathan Srinivasan Subject: Re: [PATCH] cpuidle: powernv/pseries: Decrease the snooze residency Date: Wed, 3 Jun 2015 21:05:57 +0530 Message-ID: <20150603153557.GA5983@dirshya.in.ibm.com> References: <1432902728-31476-1-git-send-email-shilpa.bhat@linux.vnet.ibm.com> <55686DE5.6010001@linux.vnet.ibm.com> <20150530060141.GA19134@dirshya.in.ibm.com> <1433036302.4526.15.camel@au1.ibm.com> Reply-To: svaidy@linux.vnet.ibm.com Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Return-path: Received: from e28smtp09.in.ibm.com ([122.248.162.9]:55831 "EHLO e28smtp09.in.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752992AbbFCPgP (ORCPT ); Wed, 3 Jun 2015 11:36:15 -0400 Received: from /spool/local by e28smtp09.in.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Wed, 3 Jun 2015 21:06:13 +0530 Content-Disposition: inline In-Reply-To: <1433036302.4526.15.camel@au1.ibm.com> Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: Benjamin Herrenschmidt Cc: Preeti U Murthy , linux-pm@vger.kernel.org, Daniel Lezcano , rjw@rjwysocki.net, linux-kernel@vger.kernel.org, anton@samba.org, Shilpasri G Bhat , linuxppc-dev@lists.ozlabs.org * Benjamin Herrenschmidt [2015-05-30 20:38:22]: > On Sat, 2015-05-30 at 11:31 +0530, Vaidyanathan Srinivasan wrote: > > In shared lpar case, spinning in guest context may potentially take > > away cycles from other lpars waiting to run on the same physical cpu. > > > > So the policy in shared lpar case is to let PowerVM hypervisor know > > immediately that the guest cpu is idle which will allow the hypervisor > > to use the cycles for other tasks/lpars. > > But that will have negative side effects under KVM no ? Yes, you have a good point. If one of the thread in the core goes to cede, it can still come back quickly since the KVM guest context is not switched yet. But in single threaded guest, this can force an unnecessary exit/context switch overhead. Now that we have fixed the snooze loop to be bounded and exit predictably, KVM guest should actually use snooze state to improve latency. I will test this scenario and enable snooze state for KVM guest. > Suresh mentioned something with his new directed interrupts code that we > had many cases where the interrupts ended up arriving shortly after we > exited to host for NAP'ing ... > > Snooze might fix it... Right. This scenario is worth experimenting and then introduce snooze loop for guest. --Vaidy