From owner-freebsd-hardware Sat Aug 31 5:19: 8 2002 Delivered-To: freebsd-hardware@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id A5DF537B400 for ; Sat, 31 Aug 2002 05:19:06 -0700 (PDT) Received: from critter.freebsd.dk (critter.freebsd.dk [212.242.86.163]) by mx1.FreeBSD.org (Postfix) with ESMTP id DB90D43E42 for ; Sat, 31 Aug 2002 05:19:05 -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 g7VCFOd1040660; Sat, 31 Aug 2002 14:15:26 +0200 (CEST) (envelope-from phk@critter.freebsd.dk) To: Bruce Evans Cc: hardware@FreeBSD.ORG Subject: Re: PCI latency timer vs interrupt latency and ISA bus latency In-Reply-To: Your message of "Sat, 31 Aug 2002 22:03:18 +1000." <20020831215409.C5111-100000@gamplex.bde.org> Date: Sat, 31 Aug 2002 14:15:24 +0200 Message-ID: <40659.1030796124@critter.freebsd.dk> From: Poul-Henning Kamp Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org In message <20020831215409.C5111-100000@gamplex.bde.org>, Bruce Evans writes: >On Sat, 31 Aug 2002, Poul-Henning Kamp wrote: > >> In message <20020831160511.O3960-100000@gamplex.bde.org>, Bruce Evans writes: >> >> >Debugging of interrupt latency caused by critical_enter() showed that >> >DELAY(2) sometimes delayed for 170 or more usec for an Athlon1600 CPU >> >> DELAY(2) should not use i8254 when better alternatives exist. We may >> not want to deal with the TSC calibration issue, but both the ACPI timer >> and the APIC timer in the CPU would be good candidates. > >This is not the point here. DELAY() is just the messenger. What is >important is that bus accesses may be slowed down by a factor of 60 or >more for at least 3 accesses in succession, and that a using a simple >spinlock to provide exclusive access to the device and semi-exclusive >access to the CPU works surprisingly badly -- it prevents the CPU from >doing anything useful for 60 times longer than expected. I agree that we have a point in PCI latencies and stuff, an I know that NetBSD has some code to fix up stuff if it's wrong. But I think it is an equally valid point that DELAY() is in dire need for a brain-lobotomy. -- 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-hardware" in the body of the message