From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 18444 invoked by alias); 5 Nov 2004 00:40:49 -0000 Mailing-List: contact libc-hacker-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-hacker-owner@sources.redhat.com Received: (qmail 18428 invoked from network); 5 Nov 2004 00:40:49 -0000 Received: from unknown (HELO palrel11.hp.com) (156.153.255.246) by sourceware.org with SMTP; 5 Nov 2004 00:40:49 -0000 Received: from hplms2.hpl.hp.com (hplms2.hpl.hp.com [15.0.152.33]) by palrel11.hp.com (Postfix) with ESMTP id CF63F49D7; Thu, 4 Nov 2004 16:40:48 -0800 (PST) Received: from napali.hpl.hp.com (napali.hpl.hp.com [15.4.89.123]) by hplms2.hpl.hp.com (8.13.1/8.13.1/HPL-PA Hub) with ESMTP id iA50ekLK002358; Thu, 4 Nov 2004 16:40:47 -0800 (PST) Received: from napali.hpl.hp.com (napali [127.0.0.1]) by napali.hpl.hp.com (8.13.1/8.13.1/Debian-16) with ESMTP id iA50eks6008835; Thu, 4 Nov 2004 16:40:46 -0800 Received: (from davidm@localhost) by napali.hpl.hp.com (8.13.1/8.13.1/Submit) id iA50ekF7008832; Thu, 4 Nov 2004 16:40:46 -0800 From: David Mosberger MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <16778.52238.419635.428364@napali.hpl.hp.com> Date: Fri, 05 Nov 2004 00:40:00 -0000 To: Roland McGrath Cc: davidm@hpl.hp.com, libc-hacker@sources.redhat.com Subject: Re: second thoughts on using dl_iterate_phdr() for cache-validation In-Reply-To: <200411042353.iA4NrtN1029576@magilla.sf.frob.com> References: <16778.48691.899903.792304@napali.hpl.hp.com> <200411042353.iA4NrtN1029576@magilla.sf.frob.com> Reply-To: davidm@hpl.hp.com X-URL: http://www.hpl.hp.com/personal/David_Mosberger/ X-SW-Source: 2004-11/txt/msg00009.txt.bz2 >>>>> On Thu, 4 Nov 2004 15:53:55 -0800, Roland McGrath said: Roland> No new ABI is going to go in very soon. We have frozen the Roland> GLIBC_2.3.4 ABI already. That's unfortunate. Is there a mechanism to queue this patch so it doesn't get lost again when 2.3.5 is opened up? Roland> unsigned long long int is not an atomic type on all Roland> platforms, so your code as written is not acceptable without Roland> adding locking for machines where it's not. The incrementing is always done under protection of a lock. The reading is not, but on those machines where reading an "unsigned long long int" isn't atomic, the effect is no worse than when using "unsigned int". And on those machines where it is atomic, "unsigned long long int" pretty much guarantees that the counter will never overflow. Do you still think the patch is unacceptable in this regard? Roland> An interface to return the address of such variables is Roland> highly dubious, since it constrains the implementation and Roland> does so permanently for the future. OK, I see your point. Thanks, --david