From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Jan Beulich" Subject: Re: [PATCH] x86: introduce specialized clear_page() Date: Wed, 20 Jun 2007 16:35:38 +0100 Message-ID: <4679656A.76E4.0078.0@novell.com> References: <4677C925.76E4.0078.0@novell.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: 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: xen-devel@lists.xensource.com, Keir Fraser List-Id: xen-devel@lists.xenproject.org >>> Keir Fraser 20.06.07 17:16 >>> >On 19/6/07 11:16, "Jan Beulich" wrote: > >> While the patch also adds an SSE version, this is is currently orphaned >> as I am not certain about the benefit of special casing idle VCPUs in >> a few places (during context switching), so that at least in that >> context using %xmmN registers would be possible without crashing and/or >> corrupting guest state. The benefit of adding such support could be to >> reduce scheduling latency when a VCPU is to transition out of idle, but >> is busy doing page cleaning. > >I measure no benefit from the XMM version (nor from upgrading the SSE2 >version on x86/64 to clear from %rax instead of %eax). I guess this is >because the move is non-temporal and it's actually quite easy to max the >memory bandwidth. Neither did I, except (obviously) on machines having SSE, but not SSE2 = (which was what I considered applying this for, with the additional need to tweak idle vcpu handling so that in that context [where FP context altering = doesn't matter, as long as the rest of the system's aware of it] it could then be = used). >So I stripped out the xmm version (it can't work anyway without some >kernel_fpu_begin/end kind of mechanism) and made the sse2 version common >across i386 and x86_64 since that was easily done and increases code >sharing. Fine by me - I tried to say this in the description. Jan