From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 2/2] i387: split up into exported and internal interfaces Date: Tue, 28 Feb 2012 13:21:00 +0200 Message-ID: <4F4CB89C.4060500@redhat.com> References: <4F42FE08.5020309@zytor.com> <4F43DB69.9060509@zytor.com> <4F440945.1020904@zytor.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: "H. Peter Anvin" , Josh Boyer , Jongman Heo , Thomas Gleixner , Ingo Molnar , x86@kernel.org, Linux Kernel Mailing List , KVM list To: Linus Torvalds Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On 02/21/2012 11:41 PM, Linus Torvalds wrote: > Btw, I really don't like what arch/x86/kvm/ does with CR0.TS and the FP > state. I'm not at all sure that's all kosher. But I don't know the > code, so I just made sure that at no point did any of the semantics > change. > Can you elaborate on what you don't like in the kvm code (apart from "it does virtualiztion")? btw, some time ago I did some work to lazify fpu save (as opposed to just fpu restore) and abstract out the various users (user mode, kernel threads, irq context, guest mode, and signal handlers). This would allow you to run task A's user mode with task B's fpu loaded, have preemptible kernel fpu being, avoid fpu switching while handling signals, and run user mode with a guest fpu loaded or vice versa. However I abandoned the effort as too complex. Perhaps a more determined hacker can make more progress there. -- error compiling committee.c: too many arguments to function