From mboxrd@z Thu Jan 1 00:00:00 1970 From: "KOSAKI Motohiro" Subject: Re: [sample] mem_notify v6: usage example Date: Tue, 12 Feb 2008 03:37:10 +0900 Message-ID: <2f11576a0802111037j4fe75e80l695f5a401ec93a7a@mail.gmail.com> References: <2f11576a0802090755n123c9b7dh26e0af6a2fef28af@mail.gmail.com> <2f11576a0802090846t7655e988pb1b712696cad1098@mail.gmail.com> <20080211181526.GC3029@webber.adilger.int> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: "Jon Masters" , "linux-mm@kvack.org" , "linux-kernel@vger.kernel.org" , "Marcelo Tosatti" , "Daniel Spang" , "Rik van Riel" , "Andrew Morton" , "Alan Cox" , "linux-fsdevel@vger.kernel.org" , "Pavel Machek" , "Al Boldi" , "Zan Lynx" To: "Andreas Dilger" Return-path: Received: from py-out-1112.google.com ([64.233.166.178]:8859 "EHLO py-out-1112.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750901AbYBKShM (ORCPT ); Mon, 11 Feb 2008 13:37:12 -0500 Received: by py-out-1112.google.com with SMTP id u52so6948657pyb.10 for ; Mon, 11 Feb 2008 10:37:11 -0800 (PST) In-Reply-To: <20080211181526.GC3029@webber.adilger.int> Content-Disposition: inline Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Hi Andreas, Thank you very good comment. > Having such notification handled by glibc to free up unused malloc (or > any heap allocations) would be very useful, because even if a program > does "free" there is no guarantee the memory is returned to the kernel. Yes, no guarantee. but current glibc-malloc very frequently return memory to kernel. glibc default behavior 1. over 1M memory: return memory just free(3) called. (you can change threshold by MALLOC_MMAP_MAX_ environment) 2. more lower: return memory when exist continuous 128k at heap tail. (you can change threashold by MALLOC_TRIM_THRESHOLD_ environment) if you know very memory consumption by already freed memory situation, please tell me situation detail and consumption memory size. > I think that having a generic reservation framework is too complex, but > hiding the details of /dev/mem_notify from applications is desirable. > A simple wrapper (possibly part of glibc) to return the poll fd, or set > up the signal is enough. Agreed. if large consumption situation exist, I'm behind you. Thanks!