lkml.org 
[lkml]   [2010]   [Sep]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH UPDATED 1/2] x86, percpu: Optimize this_cpu_ptr
Allow arches to implement __this_cpu_ptr, and provide an x86 version.

Before:
movq $foo, %rax
movq %gs:this_cpu_off, %rdx
addq %rdx, %rax

After:
movq $foo, %rax
addq %gs:this_cpu_off, %rax

The benefit is doing it in one less instruction and not clobbering
a temporary register.

tj: beefed up the comment a bit and renamed in-macro temp variable to
match neighboring macros.

Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
---
Applied both patches to the following branch. I updated the first
patch slightly.

git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu.git for-next

Thank you.

arch/x86/include/asm/percpu.h | 14 ++++++++++++++
include/asm-generic/percpu.h | 9 +++++++--
2 files changed, 21 insertions(+), 2 deletions(-)

Index: percpu/arch/x86/include/asm/percpu.h
===================================================================
--- percpu.orig/arch/x86/include/asm/percpu.h
+++ percpu/arch/x86/include/asm/percpu.h
@@ -47,6 +47,20 @@
#ifdef CONFIG_SMP
#define __percpu_arg(x) "%%"__stringify(__percpu_seg)":%P" #x
#define __my_cpu_offset percpu_read(this_cpu_off)
+
+/*
+ * Compared to the generic __my_cpu_offset version, the following
+ * saves one instruction and avoids clobbering a temp register.
+ */
+#define __this_cpu_ptr(ptr) \
+({ \
+ typeof(ptr) tcp_ptr__ = (ptr); \
+ __verify_pcpu_ptr(ptr); \
+ asm volatile("add " __percpu_arg(1) ", %0" \
+ : "+r" (tcp_ptr__) \
+ : "m" (this_cpu_off)); \
+ tcp_ptr__; \
+})
#else
#define __percpu_arg(x) "%P" #x
#endif
Index: percpu/include/asm-generic/percpu.h
===================================================================
--- percpu.orig/include/asm-generic/percpu.h
+++ percpu/include/asm-generic/percpu.h
@@ -60,9 +60,14 @@ extern unsigned long __per_cpu_offset[NR
#define __raw_get_cpu_var(var) \
(*SHIFT_PERCPU_PTR(&(var), __my_cpu_offset))

-#define this_cpu_ptr(ptr) SHIFT_PERCPU_PTR(ptr, my_cpu_offset)
+#ifndef __this_cpu_ptr
#define __this_cpu_ptr(ptr) SHIFT_PERCPU_PTR(ptr, __my_cpu_offset)
-
+#endif
+#ifdef CONFIG_DEBUG_PREEMPT
+#define this_cpu_ptr(ptr) SHIFT_PERCPU_PTR(ptr, my_cpu_offset)
+#else
+#define this_cpu_ptr(ptr) __this_cpu_ptr(ptr)
+#endif

#ifdef CONFIG_HAVE_SETUP_PER_CPU_AREA
extern void setup_per_cpu_areas(void);

\
 
 \ /
  Last update: 2010-09-09 18:21    [W:0.044 / U:1.512 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site