define new percpu interface for shared data

per cpu data section contains two types of data. One set which is exclusively accessed by the local cpu and the other set which is per cpu, but also shared by remote cpus. In the current kernel, these two sets are not clearely separated out. This can potentially cause the same data cacheline shared between the two sets of data, which will result in unnecessary bouncing of the cacheline between cpus. One way to fix the problem is to cacheline align the remotely accessed per cpu data, both at the beginning and at the end. Because of the padding at both ends, this will likely cause some memory wastage and also the interface to achieve this is not clean. This patch: Moves the remotely accessed per cpu data (which is currently marked as ____cacheline_aligned_in_smp) into a different section, where all the data elements are cacheline aligned. And as such, this differentiates the local only data and remotely accessed data cleanly. Signed-off-by: Fenghua Yu <[email protected]> Acked-by: Suresh Siddha <[email protected]> Cc: Rusty Russell <[email protected]> Cc: Christoph Lameter <[email protected]> Cc: <[email protected]> Cc: "Luck, Tony" <[email protected]> Cc: Andi Kleen <[email protected]> Signed-off-by: Andrew Morton <[email protected]> Signed-off-by: Linus Torvalds <[email protected]>
chewitt · Jul 19, 2007 · 5fb7dc3 · 5fb7dc3
1 parent 3d7e338
commit 5fb7dc3
Show file tree

Hide file tree

Showing 26 changed files with 84 additions and 53 deletions.
diff --git a/arch/alpha/kernel/vmlinux.lds.S b/arch/alpha/kernel/vmlinux.lds.S
@@ -69,10 +69,7 @@ SECTIONS
   . = ALIGN(8);
   SECURITY_INIT
 
-  . = ALIGN(8192);
-  __per_cpu_start = .;
-  .data.percpu : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(8192)
 
   . = ALIGN(2*8192);
   __init_end = .;

diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S
@@ -66,6 +66,7 @@ SECTIONS
 		. = ALIGN(4096);
 		__per_cpu_start = .;
 			*(.data.percpu)
+			*(.data.percpu.shared_aligned)
 		__per_cpu_end = .;
 #ifndef CONFIG_XIP_KERNEL
 		__init_begin = _stext;

diff --git a/arch/cris/arch-v32/vmlinux.lds.S b/arch/cris/arch-v32/vmlinux.lds.S
@@ -91,10 +91,7 @@ SECTIONS
 	}
 	SECURITY_INIT
 
-	. =  ALIGN (8192);
-	__per_cpu_start = .;
-	.data.percpu  : { *(.data.percpu) }
-	__per_cpu_end = .;
+	PERCPU(8192)
 
 #ifdef CONFIG_BLK_DEV_INITRD
 	.init.ramfs : {

diff --git a/arch/frv/kernel/vmlinux.lds.S b/arch/frv/kernel/vmlinux.lds.S
@@ -57,10 +57,7 @@ SECTIONS
   __alt_instructions_end = .;
  .altinstr_replacement : { *(.altinstr_replacement) }
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
 
 #ifdef CONFIG_BLK_DEV_INITRD
   . = ALIGN(4096);

diff --git a/arch/i386/kernel/vmlinux.lds.S b/arch/i386/kernel/vmlinux.lds.S
@@ -181,6 +181,7 @@ SECTIONS
   .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) {
 	__per_cpu_start = .;
 	*(.data.percpu)
+	*(.data.percpu.shared_aligned)
 	__per_cpu_end = .;
   }
   . = ALIGN(4096);

diff --git a/arch/ia64/kernel/vmlinux.lds.S b/arch/ia64/kernel/vmlinux.lds.S
@@ -206,6 +206,7 @@ SECTIONS
 	{
 		__per_cpu_start = .;
 		*(.data.percpu)
+		*(.data.percpu.shared_aligned)
 		__per_cpu_end = .;
 	}
   . = __phys_per_cpu_start + PERCPU_PAGE_SIZE;	/* ensure percpu data fits

diff --git a/arch/m32r/kernel/vmlinux.lds.S b/arch/m32r/kernel/vmlinux.lds.S
@@ -110,10 +110,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
   . = ALIGN(4096);
   __init_end = .;
   /* freed after init ends here */

diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
@@ -119,10 +119,7 @@ SECTIONS
   .init.ramfs : { *(.init.ramfs) }
   __initramfs_end = .;
 #endif
-  . = ALIGN(_PAGE_SIZE);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(_PAGE_SIZE)
   . = ALIGN(_PAGE_SIZE);
   __init_end = .;
   /* freed after init ends here */

diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
@@ -181,10 +181,9 @@ SECTIONS
   .init.ramfs : { *(.init.ramfs) }
   __initramfs_end = .;
 #endif
-  . = ALIGN(ASM_PAGE_SIZE);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+
+  PERCPU(ASM_PAGE_SIZE)
+
   . = ALIGN(ASM_PAGE_SIZE);
   __init_end = .;
   /* freed after init ends here */

diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S
@@ -144,6 +144,7 @@ SECTIONS
 	.data.percpu : {
 		__per_cpu_start = .;
 		*(.data.percpu)
+		*(.data.percpu.shared_aligned)
 		__per_cpu_end = .;
 	}
 

diff --git a/arch/ppc/kernel/vmlinux.lds.S b/arch/ppc/kernel/vmlinux.lds.S
@@ -130,10 +130,7 @@ SECTIONS
   __ftr_fixup : { *(__ftr_fixup) }
   __stop___ftr_fixup = .;
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
 
 #ifdef CONFIG_BLK_DEV_INITRD
   . = ALIGN(4096);

diff --git a/arch/s390/kernel/vmlinux.lds.S b/arch/s390/kernel/vmlinux.lds.S
@@ -107,10 +107,7 @@ SECTIONS
   . = ALIGN(2);
   __initramfs_end = .;
 #endif
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
   . = ALIGN(4096);
   __init_end = .;
   /* freed after init ends here */

diff --git a/arch/sh/kernel/vmlinux.lds.S b/arch/sh/kernel/vmlinux.lds.S
@@ -60,10 +60,7 @@ SECTIONS
   . = ALIGN(PAGE_SIZE);
   __nosave_end = .;
 
-  . = ALIGN(PAGE_SIZE);
-  __per_cpu_start = .;
-  .data.percpu : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(PAGE_SIZE)
   .data.cacheline_aligned : { *(.data.cacheline_aligned) }
 
   _edata = .;			/* End of data section */

diff --git a/arch/sh64/kernel/vmlinux.lds.S b/arch/sh64/kernel/vmlinux.lds.S
@@ -87,7 +87,10 @@ SECTIONS
 
   . = ALIGN(PAGE_SIZE);
   __per_cpu_start = .;
-  .data.percpu : C_PHYS(.data.percpu) { *(.data.percpu) }
+  .data.percpu : C_PHYS(.data.percpu) {
+	*(.data.percpu)
+	*(.data.percpu.shared_aligned)
+  }
   __per_cpu_end = . ;
   .data.cacheline_aligned : C_PHYS(.data.cacheline_aligned) { *(.data.cacheline_aligned) }
 

diff --git a/arch/sparc/kernel/vmlinux.lds.S b/arch/sparc/kernel/vmlinux.lds.S
@@ -65,10 +65,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
   . = ALIGN(4096);
   __init_end = .;
   . = ALIGN(32);

diff --git a/arch/sparc64/kernel/vmlinux.lds.S b/arch/sparc64/kernel/vmlinux.lds.S
@@ -90,10 +90,8 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(PAGE_SIZE);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(PAGE_SIZE)
+
   . = ALIGN(PAGE_SIZE);
   __init_end = .;
   __bss_start = .;

diff --git a/arch/x86_64/kernel/vmlinux.lds.S b/arch/x86_64/kernel/vmlinux.lds.S
@@ -194,10 +194,8 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
+
   . = ALIGN(4096);
   __init_end = .;
 

diff --git a/arch/xtensa/kernel/vmlinux.lds.S b/arch/xtensa/kernel/vmlinux.lds.S
@@ -190,10 +190,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(4096);
-  __per_cpu_start = .;
-  .data.percpu  : { *(.data.percpu) }
-  __per_cpu_end = .;
+  PERCPU(4096)
 
 
   /* We need this dummy segment here */

diff --git a/include/asm-generic/percpu.h b/include/asm-generic/percpu.h
@@ -14,6 +14,11 @@ extern unsigned long __per_cpu_offset[NR_CPUS];
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_aligned_in_smp
+
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*({				\
 	extern int simple_identifier_##var(void);	\
@@ -34,6 +39,9 @@ do {								\
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+    DEFINE_PER_CPU(type, name)
+
 #define per_cpu(var, cpu)			(*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)			per_cpu__##var
 #define __raw_get_cpu_var(var)			per_cpu__##var

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
@@ -245,3 +245,11 @@
   	*(.initcall7.init)						\
   	*(.initcall7s.init)
 
+#define PERCPU(align)							\
+	. = ALIGN(align);						\
+	__per_cpu_start = .;						\
+	.data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) {		\
+		*(.data.percpu)						\
+		*(.data.percpu.shared_aligned)				\
+	}								\
+	__per_cpu_end = .;
diff --git a/include/asm-i386/percpu.h b/include/asm-i386/percpu.h
@@ -54,6 +54,11 @@ extern unsigned long __per_cpu_offset[];
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_aligned_in_smp
+
 /* We can use this directly for local CPU (faster). */
 DECLARE_PER_CPU(unsigned long, this_cpu_off);
 

diff --git a/include/asm-ia64/percpu.h b/include/asm-ia64/percpu.h
@@ -29,6 +29,16 @@
 	__attribute__((__section__(".data.percpu")))		\
 	__SMALL_ADDR_AREA __typeof__(type) per_cpu__##name
 
+#ifdef CONFIG_SMP
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
+	__attribute__((__section__(".data.percpu.shared_aligned")))	\
+	__SMALL_ADDR_AREA __typeof__(type) per_cpu__##name		\
+	____cacheline_aligned_in_smp
+#else
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+	DEFINE_PER_CPU(type, name)
+#endif
+
 /*
  * Pretty much a literal copy of asm-generic/percpu.h, except that percpu_modcopy() is an
  * external routine, to avoid include-hell.

diff --git a/include/asm-powerpc/percpu.h b/include/asm-powerpc/percpu.h
@@ -20,6 +20,11 @@
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_aligned_in_smp
+
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset(cpu)))
 #define __get_cpu_var(var) (*RELOC_HIDE(&per_cpu__##var, __my_cpu_offset()))
@@ -40,6 +45,8 @@ extern void setup_per_cpu_areas(void);
 
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+    DEFINE_PER_CPU(type, name)
 
 #define per_cpu(var, cpu)			(*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)			per_cpu__##var

diff --git a/include/asm-s390/percpu.h b/include/asm-s390/percpu.h
@@ -41,6 +41,11 @@ extern unsigned long __per_cpu_offset[NR_CPUS];
     __attribute__((__section__(".data.percpu"))) \
     __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_aligned_in_smp
+
 #define __get_cpu_var(var) __reloc_hide(var,S390_lowcore.percpu_offset)
 #define __raw_get_cpu_var(var) __reloc_hide(var,S390_lowcore.percpu_offset)
 #define per_cpu(var,cpu) __reloc_hide(var,__per_cpu_offset[cpu])
@@ -59,6 +64,8 @@ do {								\
 
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+    DEFINE_PER_CPU(type, name)
 
 #define __get_cpu_var(var) __reloc_hide(var,0)
 #define __raw_get_cpu_var(var) __reloc_hide(var,0)

diff --git a/include/asm-sparc64/percpu.h b/include/asm-sparc64/percpu.h
@@ -18,6 +18,11 @@ extern unsigned long __per_cpu_shift;
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_aligned_in_smp
+
 register unsigned long __local_per_cpu_offset asm("g5");
 
 /* var is in discarded region: offset to particular copy we want */
@@ -38,6 +43,8 @@ do {								\
 #define real_setup_per_cpu_areas()		do { } while (0)
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+    DEFINE_PER_CPU(type, name)
 
 #define per_cpu(var, cpu)			(*((void)cpu, &per_cpu__##var))
 #define __get_cpu_var(var)			per_cpu__##var

diff --git a/include/asm-x86_64/percpu.h b/include/asm-x86_64/percpu.h
@@ -20,6 +20,11 @@
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)		\
+    __attribute__((__section__(".data.percpu.shared_aligned"))) \
+    __typeof__(type) per_cpu__##name				\
+    ____cacheline_internodealigned_in_smp
+
 /* var is in discarded region: offset to particular copy we want */
 #define per_cpu(var, cpu) (*({				\
 	extern int simple_identifier_##var(void);	\
@@ -46,6 +51,8 @@ extern void setup_per_cpu_areas(void);
 
 #define DEFINE_PER_CPU(type, name) \
     __typeof__(type) per_cpu__##name
+#define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
+    DEFINE_PER_CPU(type, name)
 
 #define per_cpu(var, cpu)			(*((void)(cpu), &per_cpu__##var))
 #define __get_cpu_var(var)			per_cpu__##var