From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934064AbYDWBga (ORCPT ); Tue, 22 Apr 2008 21:36:30 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1758827AbYDWBgU (ORCPT ); Tue, 22 Apr 2008 21:36:20 -0400 Received: from ns2.suse.de ([195.135.220.15]:57543 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758590AbYDWBgS (ORCPT ); Tue, 22 Apr 2008 21:36:18 -0400 Date: Wed, 23 Apr 2008 03:36:17 +0200 From: Nick Piggin To: Linus Torvalds Cc: Ingo Molnar , Jens Axboe , linux-arch@vger.kernel.org, Linux Kernel Mailing List , peterz@infradead.org, sam@ravnborg.org Subject: Re: [PATCH 2/11] x86: convert to generic helpers for IPI function calls Message-ID: <20080423013617.GB23192@wotan.suse.de> References: <1208890227-24808-1-git-send-email-jens.axboe@oracle.com> <1208890227-24808-3-git-send-email-jens.axboe@oracle.com> <20080422191213.GA6370@elte.hu> <20080422192601.GB12588@elte.hu> <20080423011153.GB17572@wotan.suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.9i Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Apr 22, 2008 at 06:22:17PM -0700, Linus Torvalds wrote: > > > On Wed, 23 Apr 2008, Nick Piggin wrote: > > > > The "too expensive to check the shared queue" is one aspect of it. The > > shared queue need not have events *for us* (at least, unless Jens has > > changed the implementation a bit) but it can still have events that we > > would need to check through. > > > > I don't think deadlock is a problem (any more than with multiple vectors). > > I'm happy with it either way, but as pointed out by Ralf, for the MIPS > version of the patch there aren't even other vectors available. Yep. > So sharing the vector is going to be a requirement on other architectures, > maybe we should look at whether we can make it generic and peform well > regardless? I don't think it will perform terribly _poorly_ by sharing the vector. If you have a look at how serialised the old code is... If there is a noticable slowdown anywhere, we could I suppose create a 2nd variant for small numbers of CPUs, or shared vector case, but I'd say the existing code should be fine I think.