From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from e23smtp06.au.ibm.com (e23smtp06.au.ibm.com [202.81.31.148]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 319BD1A0008 for ; Wed, 4 Jun 2014 23:36:31 +1000 (EST) Received: from /spool/local by e23smtp06.au.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Wed, 4 Jun 2014 23:36:28 +1000 Received: from d23relay04.au.ibm.com (d23relay04.au.ibm.com [9.190.234.120]) by d23dlp01.au.ibm.com (Postfix) with ESMTP id 791D82CE8050 for ; Wed, 4 Jun 2014 23:36:26 +1000 (EST) Received: from d23av03.au.ibm.com (d23av03.au.ibm.com [9.190.234.97]) by d23relay04.au.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id s54DKTiY17235978 for ; Wed, 4 Jun 2014 23:20:29 +1000 Received: from d23av03.au.ibm.com (localhost [127.0.0.1]) by d23av03.au.ibm.com (8.14.4/8.14.4/NCO v10.0 AVout) with ESMTP id s54DaPOP008006 for ; Wed, 4 Jun 2014 23:36:25 +1000 From: "Aneesh Kumar K.V" To: Paul Mackerras Subject: Re: [PATCH] KVM: PPC: BOOK3S: PR: Fix PURR and SPURR emulation In-Reply-To: <87vbsh6ji4.fsf@linux.vnet.ibm.com> References: <1401797771-25606-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com> <20140604041547.GA32223@drongo> <87vbsh6ji4.fsf@linux.vnet.ibm.com> Date: Wed, 04 Jun 2014 19:06:22 +0530 Message-ID: <87sink7q5l.fsf@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain Cc: linuxppc-dev@lists.ozlabs.org, agraf@suse.de, kvm-ppc@vger.kernel.org, kvm@vger.kernel.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , "Aneesh Kumar K.V" writes: > Paul Mackerras writes: > >> On Tue, Jun 03, 2014 at 05:46:11PM +0530, Aneesh Kumar K.V wrote: >>> We use time base for PURR and SPURR emulation with PR KVM since we >>> are emulating a single threaded core. When using time base >>> we need to make sure that we don't accumulate time spent in the host >>> in PURR and SPURR value. >> >> Mostly looks good except for this... >> >>> @@ -170,6 +175,11 @@ void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu, >>> >>> out: >>> preempt_enable(); >>> + /* >>> + * Update purr and spurr using time base >>> + */ >>> + vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb; >>> + vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb; >> >> You need to do those updates before the "out:" label. Otherwise if >> this function gets called with !svcpu->in_use (which can happen if >> CONFIG_PREEMPT is enabled) we would do these updates a second time for >> one guest exit. The thing is that kvmppc_copy_from_svcpu() can get >> called from kvmppc_core_vcpu_put_pr() if the vcpu task gets preempted >> on the way out from the guest before we get to the regular call of >> kvmppc_copy_from_svcpu(). It would then get called again when the >> task gets to run, but this time it does nothing because svcpu->in_use >> is false. > > Looking at the code, since we enable MSR.EE early now, we might possibly > end up calling this function late in the guest exit path. That > implies, we may account that time (time spent after a preempt immediately > following a guest exit) in purr/spurr. I guess that amount of inaccuracy is > ok, because that is the best we could do here ? > May be not, we do call kvmppc_copy_to_svcpu from preempt notifier callbacks. That should ensure that we get the right value. I have sent patch -V2 taking care of the above review comment. -aneesh