From mboxrd@z Thu Jan 1 00:00:00 1970 From: Weidong Han Subject: Re: [PATCH 0/3 v2] XSAVE/XRSTOR fixes and enhancements Date: Wed, 01 Sep 2010 09:53:44 +0800 Message-ID: <4C7DB228.30709@intel.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Keir Fraser Cc: Xen-devel , Jan Beulich List-Id: xen-devel@lists.xenproject.org Keir Fraser wrote: > On 31/08/2010 15:52, "Han, Weidong" wrote: > > >> Change logs from v1 -> v2: >> Due to not guarantee backward compatibility, drop the guest save/restore patch >> here. Will re-implement it later. In addition, split the original fix frozen >> states patch into XSAVE/XRSTOR cleanup patch and fix frozen state patch. >> >> Patch 1/3: XSAVE/XRSTOR: some cleanups >> Replace xfeature_low and xfeature_high with a u64 variable xfeature_mask. >> In structure hvm_vcpu, rename xfeature_mask to xcr0 >> Provide EDX:EAX with all bits set to 1 for XSAVE and XRSTOR as spec >> recommends. >> >> Patch 2/3: Fix frozen states >> If a guest sets a state and dirties the state, but later temporarily clears >> the state, and at this time if this vcpu is scheduled out, then other vcpus >> may corrupt the state before the vcpu is scheduled in again, thus the state >> cannot be restored correctly. To solve this issue, this patch save/restore all >> states unconditionally on vcpu context switch. >> > > Performance overhead of this fix? Is there no other lazy save technique that > can work? > I think the cost of set_xcr0 which just changes some bits in XCR0 register should be little. I don't have any optimization for it now. > >> Patch 3/3. Enable guest AVX >> This patch enables Intel(R) Advanced Vector Extension (AVX) for guest. >> > > If we enable this but don't implement save/restore then don't guests lose > state across s/r with unpredictable results? > Yes. As I said in another email, actually it already breaks hvm guests save/restore on platforms which supports XSAVE/XRSTOR. Regards, Weidong > > > >