Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758944Ab2HVJx0 (ORCPT ); Wed, 22 Aug 2012 05:53:26 -0400 Received: from mail-wi0-f178.google.com ([209.85.212.178]:45180 "EHLO mail-wi0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758875Ab2HVJxW (ORCPT ); Wed, 22 Aug 2012 05:53:22 -0400 Date: Wed, 22 Aug 2012 12:53:16 +0300 From: Ido Yariv To: Ingo Molnar Cc: linux-kernel@vger.kernel.org, Thomas Gleixner , Ingo Molnar , "H. Peter Anvin" , Shai Fultheim Subject: Re: [PATCH v2 2/2] x86/pat: Avoid contention on cpa_lock if possible Message-ID: <20120822095316.GA16110@WorkStation.localnet> References: <1344445299-22142-1-git-send-email-ido@wizery.com> <1344445299-22142-2-git-send-email-ido@wizery.com> <20120822084702.GA21764@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20120822084702.GA21764@gmail.com> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5684 Lines: 173 Hi Ingo, On Wed, Aug 22, 2012 at 10:47:02AM +0200, Ingo Molnar wrote: > > * Ido Yariv wrote: > > > vSMP Foundation does not require to serialize CPA by guaranteeing that > > the most recent TLB entry will always be used. > > > > To avoid needless contention on cpa_lock, do not lock/unlock it if it > > isn't necessary. > > > > Based on work by Shai Fultheim . > > > > Signed-off-by: Ido Yariv > > Acked-by: Shai Fultheim > > --- > > Changes from v1: > > - Use a synthetic CPUID bit and a use static_cpu_has() as suggested by > > H. Peter Avnin > > > > arch/x86/include/asm/cpufeature.h | 1 + > > arch/x86/kernel/vsmp_64.c | 10 ++++++++++ > > arch/x86/mm/pageattr.c | 30 +++++++++++++++++++++--------- > > 3 files changed, 32 insertions(+), 9 deletions(-) > > > > diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h > > index 6b7ee5f..92303a0 100644 > > --- a/arch/x86/include/asm/cpufeature.h > > +++ b/arch/x86/include/asm/cpufeature.h > > @@ -97,6 +97,7 @@ > > #define X86_FEATURE_EXTD_APICID (3*32+26) /* has extended APICID (8 bits) */ > > #define X86_FEATURE_AMD_DCM (3*32+27) /* multi-node processor */ > > #define X86_FEATURE_APERFMPERF (3*32+28) /* APERFMPERF */ > > +#define X86_FEATURE_NO_CPA_LOCK (3*32+29) /* Serializing cpa is not required */ > > Patch looks mostly good, but could we please use some more > hardware-ish name, instead of referring to a kernel lock? Sure thing. How about X86_FEATURE_TLB_RELIABLE? Thanks, Ido. >From fc3adcd1a9a86d65836d461bc22f54955ba3fc82 Mon Sep 17 00:00:00 2001 From: Ido Yariv Date: Mon, 11 Jun 2012 15:17:56 +0300 Subject: [PATCH v3] x86/pat: Avoid contention on cpa_lock if possible vSMP Foundation does not require to serialize CPA by guaranteeing that the most recent TLB entry will always be used. To avoid needless contention on cpa_lock, do not lock/unlock it if it isn't necessary. Based on work by Shai Fultheim . Signed-off-by: Ido Yariv Acked-by: Shai Fultheim --- arch/x86/include/asm/cpufeature.h | 1 + arch/x86/kernel/vsmp_64.c | 10 ++++++++++ arch/x86/mm/pageattr.c | 30 +++++++++++++++++++++--------- 3 files changed, 32 insertions(+), 9 deletions(-) diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h index 6b7ee5f..9b3d075 100644 --- a/arch/x86/include/asm/cpufeature.h +++ b/arch/x86/include/asm/cpufeature.h @@ -97,6 +97,7 @@ #define X86_FEATURE_EXTD_APICID (3*32+26) /* has extended APICID (8 bits) */ #define X86_FEATURE_AMD_DCM (3*32+27) /* multi-node processor */ #define X86_FEATURE_APERFMPERF (3*32+28) /* APERFMPERF */ +#define X86_FEATURE_TLB_RELIABLE (3*32+29) /* Serializing cpa is not required */ /* Intel-defined CPU features, CPUID level 0x00000001 (ecx), word 4 */ #define X86_FEATURE_XMM3 (4*32+ 0) /* "pni" SSE-3 */ diff --git a/arch/x86/kernel/vsmp_64.c b/arch/x86/kernel/vsmp_64.c index f655f2c..6e245d8 100644 --- a/arch/x86/kernel/vsmp_64.c +++ b/arch/x86/kernel/vsmp_64.c @@ -235,9 +235,19 @@ static void __init vsmp_platform_setup(void) set_vsmp_pv_ops(); } +static void __cpuinit vsmp_set_cpu_features(struct cpuinfo_x86 *c) +{ + /* + * vSMP guarantees that the most recent TLB entry will always be used, + * so we can avoid serializing cpa + */ + set_cpu_cap(c, X86_FEATURE_TLB_RELIABLE); +} + const __refconst struct hypervisor_x86 x86_hyper_vsmp = { .name = "ScaleMP vSMP Foundation", .detect = detect_vsmp_box, .init_platform = vsmp_platform_setup, + .set_cpu_features = vsmp_set_cpu_features, }; EXPORT_SYMBOL(x86_hyper_vsmp); diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c index a718e0d..bc4720e 100644 --- a/arch/x86/mm/pageattr.c +++ b/arch/x86/mm/pageattr.c @@ -46,7 +46,7 @@ struct cpa_data { * entries change the page attribute in parallel to some other cpu * splitting a large page entry along with changing the attribute. */ -static DEFINE_SPINLOCK(cpa_lock); +static DEFINE_SPINLOCK(_cpa_lock); #define CPA_FLUSHTLB 1 #define CPA_ARRAY 2 @@ -110,6 +110,22 @@ static inline unsigned long highmap_end_pfn(void) # define debug_pagealloc 0 #endif +static inline void cpa_lock(void) +{ + if (debug_pagealloc || static_cpu_has(X86_FEATURE_TLB_RELIABLE)) + return; + + spin_lock(&_cpa_lock); +} + +static inline void cpa_unlock(void) +{ + if (debug_pagealloc || static_cpu_has(X86_FEATURE_TLB_RELIABLE)) + return; + + spin_unlock(&_cpa_lock); +} + static inline int within(unsigned long addr, unsigned long start, unsigned long end) { @@ -509,11 +525,9 @@ static int split_large_page(pte_t *kpte, unsigned long address) pgprot_t ref_prot; struct page *base; - if (!debug_pagealloc) - spin_unlock(&cpa_lock); + cpa_unlock(); base = alloc_pages(GFP_KERNEL | __GFP_NOTRACK, 0); - if (!debug_pagealloc) - spin_lock(&cpa_lock); + cpa_lock(); if (!base) return -ENOMEM; @@ -801,11 +815,9 @@ static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias) if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY)) cpa->numpages = 1; - if (!debug_pagealloc) - spin_lock(&cpa_lock); + cpa_lock(); ret = __change_page_attr(cpa, checkalias); - if (!debug_pagealloc) - spin_unlock(&cpa_lock); + cpa_unlock(); if (ret) return ret; -- 1.7.7.6 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/