Merge branch 'for-2.6.39' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu

* 'for-2.6.39' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu:
  percpu, x86: Add arch-specific this_cpu_cmpxchg_double() support
  percpu: Generic support for this_cpu_cmpxchg_double()
  alpha: use L1_CACHE_BYTES for cacheline size in the linker script
  percpu: align percpu readmostly subsection to cacheline

Fix up trivial conflict in arch/x86/kernel/vmlinux.lds.S due to the
percpu alignment having changed ("x86: Reduce back the alignment of the
per-CPU data section")
diff --git a/arch/alpha/kernel/vmlinux.lds.S b/arch/alpha/kernel/vmlinux.lds.S
index 003ef4c..433be2a 100644
--- a/arch/alpha/kernel/vmlinux.lds.S
+++ b/arch/alpha/kernel/vmlinux.lds.S
@@ -1,5 +1,6 @@
 #include <asm-generic/vmlinux.lds.h>
 #include <asm/thread_info.h>
+#include <asm/cache.h>
 #include <asm/page.h>
 
 OUTPUT_FORMAT("elf64-alpha")
@@ -38,7 +39,7 @@
 	__init_begin = ALIGN(PAGE_SIZE);
 	INIT_TEXT_SECTION(PAGE_SIZE)
 	INIT_DATA_SECTION(16)
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 	/* Align to THREAD_SIZE rather than PAGE_SIZE here so any padding page
 	   needed for the THREAD_SIZE aligned init_task gets freed after init */
 	. = ALIGN(THREAD_SIZE);
@@ -46,7 +47,7 @@
 	/* Freed after init ends here */
 
 	_data = .;
-	RW_DATA_SECTION(64, PAGE_SIZE, THREAD_SIZE)
+	RW_DATA_SECTION(L1_CACHE_BYTES, PAGE_SIZE, THREAD_SIZE)
 
 	.got : {
 		*(.got)
diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S
index 6146279..28fea9b 100644
--- a/arch/arm/kernel/vmlinux.lds.S
+++ b/arch/arm/kernel/vmlinux.lds.S
@@ -78,7 +78,7 @@
 #endif
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(32, PAGE_SIZE)
 
 #ifndef CONFIG_XIP_KERNEL
 	. = ALIGN(PAGE_SIZE);
diff --git a/arch/blackfin/kernel/vmlinux.lds.S b/arch/blackfin/kernel/vmlinux.lds.S
index 4122678..c40d07f 100644
--- a/arch/blackfin/kernel/vmlinux.lds.S
+++ b/arch/blackfin/kernel/vmlinux.lds.S
@@ -136,7 +136,7 @@
 
 	. = ALIGN(16);
 	INIT_DATA_SECTION(16)
-	PERCPU(4)
+	PERCPU(32, 4)
 
 	.exit.data :
 	{
diff --git a/arch/cris/kernel/vmlinux.lds.S b/arch/cris/kernel/vmlinux.lds.S
index c49be84..728bbd9 100644
--- a/arch/cris/kernel/vmlinux.lds.S
+++ b/arch/cris/kernel/vmlinux.lds.S
@@ -102,7 +102,7 @@
 #endif
 	__vmlinux_end = .;		/* Last address of the physical file. */
 #ifdef CONFIG_ETRAX_ARCH_V32
-	PERCPU(PAGE_SIZE)
+	PERCPU(32, PAGE_SIZE)
 
 	.init.ramfs : {
 		INIT_RAM_FS
diff --git a/arch/frv/kernel/vmlinux.lds.S b/arch/frv/kernel/vmlinux.lds.S
index 8b973f3..0daae8a 100644
--- a/arch/frv/kernel/vmlinux.lds.S
+++ b/arch/frv/kernel/vmlinux.lds.S
@@ -37,7 +37,7 @@
   _einittext = .;
 
   INIT_DATA_SECTION(8)
-  PERCPU(4096)
+  PERCPU(L1_CACHE_BYTES, 4096)
 
   . = ALIGN(PAGE_SIZE);
   __init_end = .;
diff --git a/arch/ia64/kernel/vmlinux.lds.S b/arch/ia64/kernel/vmlinux.lds.S
index 5a4d044..787de4a 100644
--- a/arch/ia64/kernel/vmlinux.lds.S
+++ b/arch/ia64/kernel/vmlinux.lds.S
@@ -198,7 +198,7 @@
 
 	/* Per-cpu data: */
 	. = ALIGN(PERCPU_PAGE_SIZE);
-	PERCPU_VADDR(PERCPU_ADDR, :percpu)
+	PERCPU_VADDR(SMP_CACHE_BYTES, PERCPU_ADDR, :percpu)
 	__phys_per_cpu_start = __per_cpu_load;
 	/*
 	 * ensure percpu data fits
diff --git a/arch/m32r/kernel/vmlinux.lds.S b/arch/m32r/kernel/vmlinux.lds.S
index 7da94ea..c194d64 100644
--- a/arch/m32r/kernel/vmlinux.lds.S
+++ b/arch/m32r/kernel/vmlinux.lds.S
@@ -53,7 +53,7 @@
   __init_begin = .;
   INIT_TEXT_SECTION(PAGE_SIZE)
   INIT_DATA_SECTION(16)
-  PERCPU(PAGE_SIZE)
+  PERCPU(32, PAGE_SIZE)
   . = ALIGN(PAGE_SIZE);
   __init_end = .;
   /* freed after init ends here */
diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index 570607b..832afbb 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -115,7 +115,7 @@
 		EXIT_DATA
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(1 << CONFIG_MIPS_L1_CACHE_SHIFT, PAGE_SIZE)
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;
 	/* freed after init ends here */
diff --git a/arch/mn10300/kernel/vmlinux.lds.S b/arch/mn10300/kernel/vmlinux.lds.S
index febbeee..968bcd2 100644
--- a/arch/mn10300/kernel/vmlinux.lds.S
+++ b/arch/mn10300/kernel/vmlinux.lds.S
@@ -70,7 +70,7 @@
 	.exit.text : { EXIT_TEXT; }
 	.exit.data : { EXIT_DATA; }
 
-  PERCPU(PAGE_SIZE)
+  PERCPU(32, 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
index d64a6bb..8f1e4ef 100644
--- a/arch/parisc/kernel/vmlinux.lds.S
+++ b/arch/parisc/kernel/vmlinux.lds.S
@@ -145,7 +145,7 @@
 		EXIT_DATA
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 	. = ALIGN(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
index 8a0deef..b9150f0 100644
--- a/arch/powerpc/kernel/vmlinux.lds.S
+++ b/arch/powerpc/kernel/vmlinux.lds.S
@@ -160,7 +160,7 @@
 		INIT_RAM_FS
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 
 	. = ALIGN(8);
 	.machine.desc : AT(ADDR(.machine.desc) - LOAD_OFFSET) {
diff --git a/arch/s390/kernel/vmlinux.lds.S b/arch/s390/kernel/vmlinux.lds.S
index a68ac10..1bc18cd 100644
--- a/arch/s390/kernel/vmlinux.lds.S
+++ b/arch/s390/kernel/vmlinux.lds.S
@@ -77,7 +77,7 @@
 	. = ALIGN(PAGE_SIZE);
 	INIT_DATA_SECTION(0x100)
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(0x100, PAGE_SIZE)
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;		/* freed after init ends here */
 
diff --git a/arch/sh/kernel/vmlinux.lds.S b/arch/sh/kernel/vmlinux.lds.S
index 7f8a709..af4d461 100644
--- a/arch/sh/kernel/vmlinux.lds.S
+++ b/arch/sh/kernel/vmlinux.lds.S
@@ -66,7 +66,7 @@
 		__machvec_end = .;
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 
 	/*
 	 * .exit.text is discarded at runtime, not link time, to deal with
diff --git a/arch/sparc/kernel/vmlinux.lds.S b/arch/sparc/kernel/vmlinux.lds.S
index 0c1e678..92b557a 100644
--- a/arch/sparc/kernel/vmlinux.lds.S
+++ b/arch/sparc/kernel/vmlinux.lds.S
@@ -108,7 +108,7 @@
 		__sun4v_2insn_patch_end = .;
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(SMP_CACHE_BYTES, PAGE_SIZE)
 
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;
diff --git a/arch/tile/kernel/vmlinux.lds.S b/arch/tile/kernel/vmlinux.lds.S
index 25fdc0c..c6ce378 100644
--- a/arch/tile/kernel/vmlinux.lds.S
+++ b/arch/tile/kernel/vmlinux.lds.S
@@ -63,7 +63,7 @@
     *(.init.page)
   } :data =0
   INIT_DATA_SECTION(16)
-  PERCPU(PAGE_SIZE)
+  PERCPU(L2_CACHE_BYTES, PAGE_SIZE)
   . = ALIGN(PAGE_SIZE);
   VMLINUX_SYMBOL(_einitdata) = .;
 
diff --git a/arch/um/include/asm/common.lds.S b/arch/um/include/asm/common.lds.S
index ac55b9e..34bede8 100644
--- a/arch/um/include/asm/common.lds.S
+++ b/arch/um/include/asm/common.lds.S
@@ -42,7 +42,7 @@
 	INIT_SETUP(0)
   }
 
-  PERCPU(32)
+  PERCPU(32, 32)
 	
   .initcall.init : {
 	INIT_CALLS
diff --git a/arch/x86/include/asm/percpu.h b/arch/x86/include/asm/percpu.h
index 7e17295..a09e1f0 100644
--- a/arch/x86/include/asm/percpu.h
+++ b/arch/x86/include/asm/percpu.h
@@ -451,6 +451,26 @@
 #define irqsafe_cpu_cmpxchg_4(pcp, oval, nval)	percpu_cmpxchg_op(pcp, oval, nval)
 #endif /* !CONFIG_M386 */
 
+#ifdef CONFIG_X86_CMPXCHG64
+#define percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)			\
+({									\
+	char __ret;							\
+	typeof(o1) __o1 = o1;						\
+	typeof(o1) __n1 = n1;						\
+	typeof(o2) __o2 = o2;						\
+	typeof(o2) __n2 = n2;						\
+	typeof(o2) __dummy = n2;					\
+	asm volatile("cmpxchg8b "__percpu_arg(1)"\n\tsetz %0\n\t"	\
+		    : "=a"(__ret), "=m" (pcp1), "=d"(__dummy)		\
+		    :  "b"(__n1), "c"(__n2), "a"(__o1), "d"(__o2));	\
+	__ret;								\
+})
+
+#define __this_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)		percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
+#define this_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)		percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
+#define irqsafe_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)	percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
+#endif /* CONFIG_X86_CMPXCHG64 */
+
 /*
  * Per cpu atomic 64 bit operations are only available under 64 bit.
  * 32 bit must fall back to generic operations.
@@ -480,6 +500,34 @@
 #define irqsafe_cpu_xor_8(pcp, val)	percpu_to_op("xor", (pcp), val)
 #define irqsafe_cpu_xchg_8(pcp, nval)	percpu_xchg_op(pcp, nval)
 #define irqsafe_cpu_cmpxchg_8(pcp, oval, nval)	percpu_cmpxchg_op(pcp, oval, nval)
+
+/*
+ * Pretty complex macro to generate cmpxchg16 instruction.  The instruction
+ * is not supported on early AMD64 processors so we must be able to emulate
+ * it in software.  The address used in the cmpxchg16 instruction must be
+ * aligned to a 16 byte boundary.
+ */
+#define percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)			\
+({									\
+	char __ret;							\
+	typeof(o1) __o1 = o1;						\
+	typeof(o1) __n1 = n1;						\
+	typeof(o2) __o2 = o2;						\
+	typeof(o2) __n2 = n2;						\
+	typeof(o2) __dummy;						\
+	alternative_io("call this_cpu_cmpxchg16b_emu\n\t" P6_NOP4,	\
+		       "cmpxchg16b %%gs:(%%rsi)\n\tsetz %0\n\t",	\
+		       X86_FEATURE_CX16,				\
+		       ASM_OUTPUT2("=a"(__ret), "=d"(__dummy)),		\
+		       "S" (&pcp1), "b"(__n1), "c"(__n2),		\
+		       "a"(__o1), "d"(__o2));				\
+	__ret;								\
+})
+
+#define __this_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)		percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
+#define this_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)		percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
+#define irqsafe_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)	percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
+
 #endif
 
 /* This is not atomic against other CPUs -- CPU preemption needs to be off */
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
index 0381e1f..101c323 100644
--- a/arch/x86/kernel/vmlinux.lds.S
+++ b/arch/x86/kernel/vmlinux.lds.S
@@ -231,7 +231,7 @@
 	 * output PHDR, so the next output section - .init.text - should
 	 * start another segment - init.
 	 */
-	PERCPU_VADDR(0, :percpu)
+	PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
 #endif
 
 	INIT_TEXT_SECTION(PAGE_SIZE)
@@ -306,7 +306,7 @@
 	}
 
 #if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
-	PERCPU(PAGE_SIZE)
+	PERCPU(INTERNODE_CACHE_BYTES, PAGE_SIZE)
 #endif
 
 	. = ALIGN(PAGE_SIZE);
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index e10cf07..f2479f1 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -42,4 +42,5 @@
         lib-y += memmove_64.o memset_64.o
         lib-y += copy_user_64.o rwlock_64.o copy_user_nocache_64.o
 	lib-$(CONFIG_RWSEM_XCHGADD_ALGORITHM) += rwsem_64.o
+	lib-y += cmpxchg16b_emu.o
 endif
diff --git a/arch/x86/lib/cmpxchg16b_emu.S b/arch/x86/lib/cmpxchg16b_emu.S
new file mode 100644
index 0000000..3e8b08a
--- /dev/null
+++ b/arch/x86/lib/cmpxchg16b_emu.S
@@ -0,0 +1,59 @@
+/*
+ *	This program is free software; you can redistribute it and/or
+ *	modify it under the terms of the GNU General Public License
+ *	as published by the Free Software Foundation; version 2
+ *	of the License.
+ *
+ */
+#include <linux/linkage.h>
+#include <asm/alternative-asm.h>
+#include <asm/frame.h>
+#include <asm/dwarf2.h>
+
+.text
+
+/*
+ * Inputs:
+ * %rsi : memory location to compare
+ * %rax : low 64 bits of old value
+ * %rdx : high 64 bits of old value
+ * %rbx : low 64 bits of new value
+ * %rcx : high 64 bits of new value
+ * %al  : Operation successful
+ */
+ENTRY(this_cpu_cmpxchg16b_emu)
+CFI_STARTPROC
+
+#
+# Emulate 'cmpxchg16b %gs:(%rsi)' except we return the result in %al not
+# via the ZF.  Caller will access %al to get result.
+#
+# Note that this is only useful for a cpuops operation.  Meaning that we
+# do *not* have a fully atomic operation but just an operation that is
+# *atomic* on a single cpu (as provided by the this_cpu_xx class of
+# macros).
+#
+this_cpu_cmpxchg16b_emu:
+	pushf
+	cli
+
+	cmpq %gs:(%rsi), %rax
+	jne not_same
+	cmpq %gs:8(%rsi), %rdx
+	jne not_same
+
+	movq %rbx, %gs:(%rsi)
+	movq %rcx, %gs:8(%rsi)
+
+	popf
+	mov $1, %al
+	ret
+
+ not_same:
+	popf
+	xor %al,%al
+	ret
+
+CFI_ENDPROC
+
+ENDPROC(this_cpu_cmpxchg16b_emu)
diff --git a/arch/xtensa/kernel/vmlinux.lds.S b/arch/xtensa/kernel/vmlinux.lds.S
index 9b52615..a282006 100644
--- a/arch/xtensa/kernel/vmlinux.lds.S
+++ b/arch/xtensa/kernel/vmlinux.lds.S
@@ -155,7 +155,7 @@
     INIT_RAM_FS
   }
 
-  PERCPU(PAGE_SIZE)
+  PERCPU(XCHAL_ICACHE_LINESIZE, PAGE_SIZE)
 
   /* We need this dummy segment here */
 
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 906c3ce..32c45e5 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -15,7 +15,7 @@
  *	HEAD_TEXT_SECTION
  *	INIT_TEXT_SECTION(PAGE_SIZE)
  *	INIT_DATA_SECTION(...)
- *	PERCPU(PAGE_SIZE)
+ *	PERCPU(CACHELINE_SIZE, PAGE_SIZE)
  *	__init_end = .;
  *
  *	_stext = .;
@@ -689,13 +689,18 @@
 
 /**
  * PERCPU_VADDR - define output section for percpu area
+ * @cacheline: cacheline size
  * @vaddr: explicit base address (optional)
  * @phdr: destination PHDR (optional)
  *
- * Macro which expands to output section for percpu area.  If @vaddr
- * is not blank, it specifies explicit base address and all percpu
- * symbols will be offset from the given address.  If blank, @vaddr
- * always equals @laddr + LOAD_OFFSET.
+ * Macro which expands to output section for percpu area.
+ *
+ * @cacheline is used to align subsections to avoid false cacheline
+ * sharing between subsections for different purposes.
+ *
+ * If @vaddr is not blank, it specifies explicit base address and all
+ * percpu symbols will be offset from the given address.  If blank,
+ * @vaddr always equals @laddr + LOAD_OFFSET.
  *
  * @phdr defines the output PHDR to use if not blank.  Be warned that
  * output PHDR is sticky.  If @phdr is specified, the next output
@@ -706,7 +711,7 @@
  * If there is no need to put the percpu section at a predetermined
  * address, use PERCPU().
  */
-#define PERCPU_VADDR(vaddr, phdr)					\
+#define PERCPU_VADDR(cacheline, vaddr, phdr)				\
 	VMLINUX_SYMBOL(__per_cpu_load) = .;				\
 	.data..percpu vaddr : AT(VMLINUX_SYMBOL(__per_cpu_load)		\
 				- LOAD_OFFSET) {			\
@@ -714,7 +719,9 @@
 		*(.data..percpu..first)					\
 		. = ALIGN(PAGE_SIZE);					\
 		*(.data..percpu..page_aligned)				\
+		. = ALIGN(cacheline);					\
 		*(.data..percpu..readmostly)				\
+		. = ALIGN(cacheline);					\
 		*(.data..percpu)					\
 		*(.data..percpu..shared_aligned)			\
 		VMLINUX_SYMBOL(__per_cpu_end) = .;			\
@@ -723,18 +730,18 @@
 
 /**
  * PERCPU - define output section for percpu area, simple version
+ * @cacheline: cacheline size
  * @align: required alignment
  *
- * Align to @align and outputs output section for percpu area.  This
- * macro doesn't maniuplate @vaddr or @phdr and __per_cpu_load and
+ * Align to @align and outputs output section for percpu area.  This macro
+ * doesn't manipulate @vaddr or @phdr and __per_cpu_load and
  * __per_cpu_start will be identical.
  *
- * This macro is equivalent to ALIGN(align); PERCPU_VADDR( , ) except
- * that __per_cpu_load is defined as a relative symbol against
- * .data..percpu which is required for relocatable x86_32
- * configuration.
+ * This macro is equivalent to ALIGN(@align); PERCPU_VADDR(@cacheline,,)
+ * except that __per_cpu_load is defined as a relative symbol against
+ * .data..percpu which is required for relocatable x86_32 configuration.
  */
-#define PERCPU(align)							\
+#define PERCPU(cacheline, align)					\
 	. = ALIGN(align);						\
 	.data..percpu	: AT(ADDR(.data..percpu) - LOAD_OFFSET) {	\
 		VMLINUX_SYMBOL(__per_cpu_load) = .;			\
@@ -742,7 +749,9 @@
 		*(.data..percpu..first)					\
 		. = ALIGN(PAGE_SIZE);					\
 		*(.data..percpu..page_aligned)				\
+		. = ALIGN(cacheline);					\
 		*(.data..percpu..readmostly)				\
+		. = ALIGN(cacheline);					\
 		*(.data..percpu)					\
 		*(.data..percpu..shared_aligned)			\
 		VMLINUX_SYMBOL(__per_cpu_end) = .;			\
diff --git a/include/linux/percpu.h b/include/linux/percpu.h
index 27c3c6f..3a5c444 100644
--- a/include/linux/percpu.h
+++ b/include/linux/percpu.h
@@ -255,6 +255,30 @@
 	pscr2_ret__;							\
 })
 
+/*
+ * Special handling for cmpxchg_double.  cmpxchg_double is passed two
+ * percpu variables.  The first has to be aligned to a double word
+ * boundary and the second has to follow directly thereafter.
+ */
+#define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...)		\
+({									\
+	bool pdcrb_ret__;						\
+	__verify_pcpu_ptr(&pcp1);					\
+	BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2));			\
+	VM_BUG_ON((unsigned long)(&pcp1) % (2 * sizeof(pcp1)));		\
+	VM_BUG_ON((unsigned long)(&pcp2) !=				\
+		  (unsigned long)(&pcp1) + sizeof(pcp1));		\
+	switch(sizeof(pcp1)) {						\
+	case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break;	\
+	case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break;	\
+	case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break;	\
+	case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break;	\
+	default:							\
+		__bad_size_call_parameter(); break;			\
+	}								\
+	pdcrb_ret__;							\
+})
+
 #define __pcpu_size_call(stem, variable, ...)				\
 do {									\
 	__verify_pcpu_ptr(&(variable));					\
@@ -501,6 +525,45 @@
 #endif
 
 /*
+ * cmpxchg_double replaces two adjacent scalars at once.  The first
+ * two parameters are per cpu variables which have to be of the same
+ * size.  A truth value is returned to indicate success or failure
+ * (since a double register result is difficult to handle).  There is
+ * very limited hardware support for these operations, so only certain
+ * sizes may work.
+ */
+#define _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+({									\
+	int ret__;							\
+	preempt_disable();						\
+	ret__ = __this_cpu_generic_cmpxchg_double(pcp1, pcp2,		\
+			oval1, oval2, nval1, nval2);			\
+	preempt_enable();						\
+	ret__;								\
+})
+
+#ifndef this_cpu_cmpxchg_double
+# ifndef this_cpu_cmpxchg_double_1
+#  define this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef this_cpu_cmpxchg_double_2
+#  define this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef this_cpu_cmpxchg_double_4
+#  define this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef this_cpu_cmpxchg_double_8
+#  define this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
+#endif
+
+/*
  * Generic percpu operations that do not require preemption handling.
  * Either we do not care about races or the caller has the
  * responsibility of handling preemptions issues. Arch code can still
@@ -703,6 +766,39 @@
 	__pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval)
 #endif
 
+#define __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+({									\
+	int __ret = 0;							\
+	if (__this_cpu_read(pcp1) == (oval1) &&				\
+			 __this_cpu_read(pcp2)  == (oval2)) {		\
+		__this_cpu_write(pcp1, (nval1));			\
+		__this_cpu_write(pcp2, (nval2));			\
+		__ret = 1;						\
+	}								\
+	(__ret);							\
+})
+
+#ifndef __this_cpu_cmpxchg_double
+# ifndef __this_cpu_cmpxchg_double_1
+#  define __this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef __this_cpu_cmpxchg_double_2
+#  define __this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef __this_cpu_cmpxchg_double_4
+#  define __this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef __this_cpu_cmpxchg_double_8
+#  define __this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
+#endif
+
 /*
  * IRQ safe versions of the per cpu RMW operations. Note that these operations
  * are *not* safe against modification of the same variable from another
@@ -823,4 +919,36 @@
 	__pcpu_size_call_return2(irqsafe_cpu_cmpxchg_, (pcp), oval, nval)
 #endif
 
+#define irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+({									\
+	int ret__;							\
+	unsigned long flags;						\
+	local_irq_save(flags);						\
+	ret__ = __this_cpu_generic_cmpxchg_double(pcp1, pcp2,		\
+			oval1, oval2, nval1, nval2);			\
+	local_irq_restore(flags);					\
+	ret__;								\
+})
+
+#ifndef irqsafe_cpu_cmpxchg_double
+# ifndef irqsafe_cpu_cmpxchg_double_1
+#  define irqsafe_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef irqsafe_cpu_cmpxchg_double_2
+#  define irqsafe_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef irqsafe_cpu_cmpxchg_double_4
+#  define irqsafe_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# ifndef irqsafe_cpu_cmpxchg_double_8
+#  define irqsafe_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
+# endif
+# define irqsafe_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__pcpu_double_call_return_int(irqsafe_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
+#endif
+
 #endif /* __LINUX_PERCPU_H */