From owner-freebsd-current Fri Aug 23 23:43:50 2002 Delivered-To: freebsd-current@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 5AAB537B400 for ; Fri, 23 Aug 2002 23:43:44 -0700 (PDT) Received: from critter.freebsd.dk (critter.freebsd.dk [212.242.86.163]) by mx1.FreeBSD.org (Postfix) with ESMTP id 892C443E75 for ; Fri, 23 Aug 2002 23:43:43 -0700 (PDT) (envelope-from phk@critter.freebsd.dk) Received: from critter.freebsd.dk (localhost [127.0.0.1]) by critter.freebsd.dk (8.12.3/8.12.2) with ESMTP id g7O6eHoc037249; Sat, 24 Aug 2002 08:40:17 +0200 (CEST) (envelope-from phk@critter.freebsd.dk) To: John Polstra Cc: current@freebsd.org Subject: Re: how to compute the skew between TSC in SMP systems ? In-Reply-To: Your message of "Fri, 23 Aug 2002 14:54:59 PDT." <200208232154.g7NLsxsi088258@vashon.polstra.com> Date: Sat, 24 Aug 2002 08:40:16 +0200 Message-ID: <37248.1030171216@critter.freebsd.dk> From: Poul-Henning Kamp Sender: owner-freebsd-current@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG In message <200208232154.g7NLsxsi088258@vashon.polstra.com>, John Polstra write s: >In article <30560.1030138677@critter.freebsd.dk>, >Poul-Henning Kamp wrote: >> In message <200208232135.g7NLZZqx088123@vashon.polstra.com>, John Polstra write >> s: >> >> Still, the feature has come in handy when debugging certain >> >multiprocessor situations where I really needed to know the relative >> >ordering of events taking place on both CPUs. >> >> Right, you might get away with that in tightly controlled circumstances, >> but it would probably be far smarter to use the counter in the >> IOAPIC (exists on all SMP) > >You mean the timer in the local APIC, don't you? I don't recall >that the IOAPIC has one, and if it did it would probably be somewhat >expensive to read it. Right, I remembered wrong: the messed up and put no timer in the ioapic, so for SMP forget about the APIC. >> or the ACPI counter even in such cases. > >The trouble with both of those counters is that they don't give you >the 1-CPU-cycle resolution that you get with the TSC. For some kinds >of performance evaluation you really want to be able to count CPU >cycles. I generally find BB profiling much more attractive in such cases since it tells you (basic block of) instruction by (bbo) instruction how many times something was executed. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message