From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail.lixom.net (lixom.net [66.141.50.11]) by ozlabs.org (Postfix) with ESMTP id 57B4667B32 for ; Wed, 19 Jul 2006 08:39:44 +1000 (EST) Date: Tue, 18 Jul 2006 17:38:54 -0500 To: Mike Kravetz Subject: Re: [PATCH 0/3] powerpc: Instrument Hypervisor Calls Message-ID: <20060718223854.GC5238@pb15.lixom.net> References: <20060718204723.GA6104@w-mikek2.ibm.com> <20060718212913.GB3091@w-mikek2.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20060718212913.GB3091@w-mikek2.ibm.com> From: Olof Johansson Cc: linuxppc-dev@ozlabs.org, Paul Mackerras List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Tue, Jul 18, 2006 at 02:29:13PM -0700, Mike Kravetz wrote: > 88 1046 108106 109322 [...] > 104 1 1040 1061 Heh. Shouldn't PURR/TB be <= 1 at all times? :-) I bet this is because they're not sampled at exactly the same time, etc. Maybe recoding it to assembly so you can do mftb and mfspr right after each other, and do the arithmetics/stores later could help. Inline asm might be sufficient. -Olof