From patchwork Wed Jun 22 14:35:11 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xiao Guangrong X-Patchwork-Id: 905482 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by demeter2.kernel.org (8.14.4/8.14.4) with ESMTP id p5MEXeH3009443 for ; Wed, 22 Jun 2011 14:33:40 GMT Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932178Ab1FVOdO (ORCPT ); Wed, 22 Jun 2011 10:33:14 -0400 Received: from cn.fujitsu.com ([222.73.24.84]:60618 "EHLO song.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1758296Ab1FVOdN (ORCPT ); Wed, 22 Jun 2011 10:33:13 -0400 Received: from tang.cn.fujitsu.com (tang.cn.fujitsu.com [10.167.250.3]) by song.cn.fujitsu.com (Postfix) with ESMTP id EFB4F170080; Wed, 22 Jun 2011 22:33:10 +0800 (CST) Received: from mailserver.fnst.cn.fujitsu.com (tang.cn.fujitsu.com [127.0.0.1]) by tang.cn.fujitsu.com (8.14.3/8.13.1) with ESMTP id p5MEXAgP001937; Wed, 22 Jun 2011 22:33:10 +0800 Received: from localhost.localdomain ([10.167.225.99]) by mailserver.fnst.cn.fujitsu.com (Lotus Domino Release 8.5.1FP4) with ESMTP id 2011062222324618-642053 ; Wed, 22 Jun 2011 22:32:46 +0800 Message-ID: <4E01FD9F.9000908@cn.fujitsu.com> Date: Wed, 22 Jun 2011 22:35:11 +0800 From: Xiao Guangrong User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.17) Gecko/20110428 Fedora/3.1.10-1.fc15 Thunderbird/3.1.10 MIME-Version: 1.0 To: Avi Kivity CC: Marcelo Tosatti , LKML , KVM Subject: [PATCH 18/22] KVM: MMU: do not need atomicly to set/clear spte References: <4E01FBC9.3020009@cn.fujitsu.com> In-Reply-To: <4E01FBC9.3020009@cn.fujitsu.com> X-MIMETrack: Itemize by SMTP Server on mailserver/fnst(Release 8.5.1FP4|July 25, 2010) at 2011-06-22 22:32:46, Serialize by Router on mailserver/fnst(Release 8.5.1FP4|July 25, 2010) at 2011-06-22 22:32:46, Serialize complete at 2011-06-22 22:32:46 Sender: kvm-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: kvm@vger.kernel.org X-Greylist: IP, sender and recipient auto-whitelisted, not delayed by milter-greylist-4.2.6 (demeter2.kernel.org [140.211.167.43]); Wed, 22 Jun 2011 14:33:40 +0000 (UTC) Now, the spte is just from nonprsent to present or present to nonprsent, so we can use some trick to set/clear spte non-atomicly as linux kernel does Signed-off-by: Xiao Guangrong --- arch/x86/kvm/mmu.c | 82 +++++++++++++++++++++++++++++++++++++++++++-------- 1 files changed, 69 insertions(+), 13 deletions(-) diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c index 339e1a3..0ba94e9 100644 --- a/arch/x86/kvm/mmu.c +++ b/arch/x86/kvm/mmu.c @@ -259,26 +259,82 @@ static gfn_t pse36_gfn_delta(u32 gpte) return (gpte & PT32_DIR_PSE36_MASK) << shift; } +#ifdef CONFIG_X86_64 static void __set_spte(u64 *sptep, u64 spte) { - set_64bit(sptep, spte); + *sptep = spte; } -static u64 __xchg_spte(u64 *sptep, u64 new_spte) +static void __update_clear_spte_fast(u64 *sptep, u64 spte) { -#ifdef CONFIG_X86_64 - return xchg(sptep, new_spte); + *sptep = spte; +} + +static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) +{ + return xchg(sptep, spte); +} #else - u64 old_spte; +union split_spte { + struct { + u32 spte_low; + u32 spte_high; + }; + u64 spte; +}; - do { - old_spte = *sptep; - } while (cmpxchg64(sptep, old_spte, new_spte) != old_spte); +static void __set_spte(u64 *sptep, u64 spte) +{ + union split_spte *ssptep, sspte; - return old_spte; -#endif + ssptep = (union split_spte *)sptep; + sspte = (union split_spte)spte; + + ssptep->spte_high = sspte.spte_high; + + /* + * If we map the spte from nonpresent to present, We should store + * the high bits firstly, then set present bit, so cpu can not + * fetch this spte while we are setting the spte. + */ + smp_wmb(); + + ssptep->spte_low = sspte.spte_low; } +static void __update_clear_spte_fast(u64 *sptep, u64 spte) +{ + union split_spte *ssptep, sspte; + + ssptep = (union split_spte *)sptep; + sspte = (union split_spte)spte; + + ssptep->spte_low = sspte.spte_low; + + /* + * If we map the spte from present to nonpresent, we should clear + * present bit firstly to avoid vcpu fetch the old high bits. + */ + smp_wmb(); + + ssptep->spte_high = sspte.spte_high; +} + +static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) +{ + union split_spte *ssptep, sspte, orig; + + ssptep = (union split_spte *)sptep; + sspte = (union split_spte)spte; + + /* xchg acts as a barrier before the setting of the high bits */ + orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); + orig.spte_high = ssptep->spte_high = sspte.spte_high; + + return orig.spte; +} +#endif + static bool spte_is_bit_cleared(u64 old_spte, u64 new_spte, u64 bit_mask) { return (old_spte & bit_mask) && !(new_spte & bit_mask); @@ -319,9 +375,9 @@ static u64 spte_get_and_update_clear(u64 *sptep, u64 new_spte) u64 old_spte = *sptep; if (!spte_has_volatile_bits(old_spte, new_spte)) - __set_spte(sptep, new_spte); + __update_clear_spte_fast(sptep, new_spte); else - old_spte = __xchg_spte(sptep, new_spte); + old_spte = __update_clear_spte_slow(sptep, new_spte); return old_spte; } @@ -379,7 +435,7 @@ static int mmu_spte_clear_track_bits(u64 *sptep) */ static void mmu_spte_clear_no_track(u64 *sptep) { - __set_spte(sptep, 0ull); + __update_clear_spte_fast(sptep, 0ull); } static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,