lkml.org 
[lkml]   [2012]   [Aug]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH v2 2/2] x86/pat: Avoid contention on cpa_lock if possible
Hi Ingo,

On Wed, Aug 22, 2012 at 10:47:02AM +0200, Ingo Molnar wrote:
>
> * Ido Yariv <ido@wizery.com> 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 <shai@scalemp.com>.
> >
> > Signed-off-by: Ido Yariv <ido@wizery.com>
> > Acked-by: Shai Fultheim <shai@scalemp.com>
> > ---
> > 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 <ido@wizery.com>
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 <shai@scalemp.com>.

Signed-off-by: Ido Yariv <ido@wizery.com>
Acked-by: Shai Fultheim <shai@scalemp.com>
---
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


\
 
 \ /
  Last update: 2012-08-22 12:41    [W:0.071 / U:0.420 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site