From: Andi Kleen Subject: Re: Re: [RFC PATCH 18/35] Support gdt/idt/ldt handling =?utf-8?q?on=09Xen=2E?= Date: Wed, 22 Mar 2006 18:36:23 +0100 Message-ID: <200603221836.24295.ak@suse.de> References: <20060322063040.960068000@sorel.sous-sol.org> <200603221530.51644.ak@suse.de> <44218E9C.3060102@vmware.com> Mime-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <44218E9C.3060102@vmware.com> Content-Disposition: inline List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Zachary Amsden Cc: Chris Wright , virtualization@lists.osdl.org, xen-devel@lists.xensource.com, linux-kernel@vger.kernel.org, Ian Pratt List-Id: xen-devel@lists.xenproject.org On Wednesday 22 March 2006 18:51, Zachary Amsden wrote: > > Yes, trapping works fine. Even LLDT is infrequent. Not when you use old style LinuxThreads which use the LDT for TLS. > No. First, you have to create a special #GP handler for the general > protection fault. [... etc ...] Sure but Xen already has the infrastructure for all of this and last time I checked it was approaching and exceeding the size of the main core kernel so a bit more of instruction emulation probably wouldn't do too much harm. In general I think any x86 hypervisor that attempts to work on current platforms needs instruction emulation because it is the only way to virtualize IO devices. If this was supposed to be a interface for lots of hypervisors then maybe, but so far it seems to only cover Xen and possibly some other bloatware ones. That said I don't feel very strongly about emulating these instructions or not as long as they can do that without too much code duplication. The current patch are still a bit too excessive on the duplication front. -Andi