From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751606Ab1GYPFj (ORCPT ); Mon, 25 Jul 2011 11:05:39 -0400 Received: from mx1.redhat.com ([209.132.183.28]:22883 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751350Ab1GYPFg (ORCPT ); Mon, 25 Jul 2011 11:05:36 -0400 Message-ID: <4E2D862C.10204@redhat.com> Date: Mon, 25 Jul 2011 18:05:16 +0300 From: Avi Kivity User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.18) Gecko/20110621 Fedora/3.1.11-1.fc15 Thunderbird/3.1.11 MIME-Version: 1.0 To: Joerg Roedel CC: Alexander Graf , Pekka Enberg , Ingo Molnar , Jan Kiszka , torvalds@linux-foundation.org, akpm@linux-foundation.org, linux-kernel@vger.kernel.org, kvm@vger.kernel.org, gorcunov@gmail.com, levinsasha928@gmail.com, asias.hejun@gmail.com, prasadjoshi124@gmail.com Subject: Re: [GIT PULL] Native Linux KVM tool for 3.1 References: <4E2CA6DE.4040900@web.de> <20110725075305.GA32294@elte.hu> <0EAA5203-D598-4CBA-B8D2-AB371A7689A9@suse.de> <4E2D2AE8.7050100@redhat.com> <20110725131022.GJ24072@8bytes.org> In-Reply-To: <20110725131022.GJ24072@8bytes.org> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 07/25/2011 04:10 PM, Joerg Roedel wrote: > > That actually makes a ton of sense. One immediate win would be that > > klibc can be tuned to the kernel it ships with (the dynamic loader will > > pick the correct object), so less #ifdef trees. Another would be to > > make klibc the formal kernel interface, which allows us to reimplement > > an older interface in terms of the one that supercedes it. > > A libc in the kernel tree would indeed make a lot of sense and I am all > for it :) It would also help to get new interfaces to the users faster. > But it should probably not be the primary exported interface. This would > just move the compatability problem from the kernel into user-space. It's a better place for it, IMO. For example, if we dump xnotify in favour of ynotify, we can implement xnotify using ynotify in klibc, and so avoid increasing the kernel's footprint. As another example, there's an OS that implements its 32-bit ABI in userspace. (of course that only works for new interfaces; the old ones have to be implemented in the kernel unless we thunk them somehow) -- error compiling committee.c: too many arguments to function