From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1030296AbXCAV2Q (ORCPT ); Thu, 1 Mar 2007 16:28:16 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1030297AbXCAV2Q (ORCPT ); Thu, 1 Mar 2007 16:28:16 -0500 Received: from smtp.osdl.org ([65.172.181.24]:42212 "EHLO smtp.osdl.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1030296AbXCAV2P (ORCPT ); Thu, 1 Mar 2007 16:28:15 -0500 Date: Thu, 1 Mar 2007 13:27:22 -0800 (PST) From: Linus Torvalds To: Ingo Molnar cc: Davide Libenzi , Ulrich Drepper , Linux Kernel Mailing List , Arjan van de Ven , Christoph Hellwig , Andrew Morton , Alan Cox , Zach Brown , Evgeniy Polyakov , "David S. Miller" , Suparna Bhattacharya , Jens Axboe , Thomas Gleixner Subject: Re: [patch 00/13] Syslets, "Threadlets", generic AIO support, v3 In-Reply-To: <20070228231228.GA12328@elte.hu> Message-ID: References: <20070221211355.GA7302@elte.hu> <20070221233111.GB5895@elte.hu> <45DCD9E5.2010106@redhat.com> <20070222074044.GA4158@elte.hu> <20070228094522.GA17716@elte.hu> <20070228231228.GA12328@elte.hu> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 1 Mar 2007, Ingo Molnar wrote: > > wrt. one-shot syscalls, the user-space stack footprint would still > probably be there, because even async contexts that only do single-shot > processing need to drop out of kernel mode to handle signals. Why? The easiest thing to do with signals is to just not pick them up. If the signal was to that *particular* threadlet (ie a "cancel"), then we just want to kill the threadlet. And if the signal was to the thread group, there is no reason why the threadlet should pick it up. In neither case is there *any* reason to handle the signal in the threadlet, afaik. And having to have a stack allocation for each threadlet certainly means that you complicate things a lot. Suddenly you have allocations that can't just go away. Again, I'm pointing to the problems I already pointed out with the allocations of the atom structures - quite often you do *not* want to keep track of anything specific for completion time, and that means that you MUST NOT have to de-allocate anythign either. Again, think aio_read(). With the *exact* current binary interface. PLEASE. If you cannot emulate that with threadlets, then threadlets are *pointless*. On eof the major reasons for the whole exercise was to get rid of the special code in fs/aio.c. So I repeat: if you cannot do that, and remain binary compatible, don't even bother. Linus