From owner-freebsd-arch Sun Oct 28 12: 2:24 2001 Delivered-To: freebsd-arch@freebsd.org Received: from netau1.alcanet.com.au (ntp.alcanet.com.au [203.62.196.27]) by hub.freebsd.org (Postfix) with ESMTP id 8152037B401 for ; Sun, 28 Oct 2001 12:02:19 -0800 (PST) Received: from mfg1.cim.alcatel.com.au (mfg1.cim.alcatel.com.au [139.188.23.1]) by netau1.alcanet.com.au (8.9.3 (PHNE_22672)/8.9.3) with ESMTP id GAA24471; Mon, 29 Oct 2001 06:49:40 +1100 (EDT) Received: from gsmx07.alcatel.com.au by cim.alcatel.com.au (PMDF V5.2-32 #37640) with ESMTP id <01KA2BKB5NGWVMP4S3@cim.alcatel.com.au>; Mon, 29 Oct 2001 06:49:39 +1000 Received: (from jeremyp@localhost) by gsmx07.alcatel.com.au (8.11.1/8.11.1) id f9SJnbA90739; Mon, 29 Oct 2001 06:49:38 +1100 (EST envelope-from jeremyp) Content-return: prohibited Date: Mon, 29 Oct 2001 06:49:37 +1100 From: Peter Jeremy Subject: Re: 64 bit times revisited.. In-reply-to: <200110260020.f9Q0KQR63759@khavrinen.lcs.mit.edu>; from wollman@khavrinen.lcs.mit.edu on Thu, Oct 25, 2001 at 08:20:26PM -0400 To: Garrett Wollman Cc: arch@FreeBSD.ORG Mail-Followup-To: Garrett Wollman , arch@FreeBSD.ORG Message-id: <20011029064937.C75481@gsmx07.alcatel.com.au> MIME-version: 1.0 Content-type: text/plain; charset=us-ascii Content-disposition: inline User-Agent: Mutt/1.2.5i References: <20011025233602.587C63808@overcee.netplex.com.au> <200110260020.f9Q0KQR63759@khavrinen.lcs.mit.edu> Sender: owner-freebsd-arch@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG On Thu, Oct 25, 2001 at 08:20:26PM -0400, Garrett Wollman wrote: >V) Do what the Large File Summit did for off_t: define 32- and 64-bit >versions of every interface that uses a time_t, and let a preprocessor >macro switch between them. How many syscalls would this impact? My guess is that it would be a significant number. Also, this approach can easily lead to exponential growth in the number of syscalls: Consider a hypothetical system that has two off_t sizes and two time_t sizes - it will need 4 stat() syscalls (and maybe others). It's also relatively difficult to recover old syscalls. (A number of wrapper functions around a single syscall() would be more manageable). > On new platforms (regardless of native word size), use the 64-bit >versions. I agree about using a 64-bit time_t for new 64-bit (or larger) architectures. I'm less certain about using 64-bit time_t on any new 32-bit architectures - supporting double-width operation is usually fairly expensive. > Meanwhile, start deploying a new inode format with 64-bit times. This is a different bikeshed - we need to extend the inode timestamps at both ends - we need more resolution as well as more range. (The 1-second granularity has been a problem for a few years already). It's unclear how to easily achieve this (since C doesn't have native support for fixed-point binary types and bitfields are only supported for int (and shorter) types). Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-arch" in the body of the message