Merge branch 'slub/lockless' into for-linus

Conflicts:
	include/linux/slub_def.h

arch/alpha/kernel/vmlinux.lds.S

 #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 @@ SECTIONS
 	__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 @@ SECTIONS
 	/* 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)

arch/arm/kernel/vmlinux.lds.S

@@ -78,7 +78,7 @@ SECTIONS
 #endif
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(32, PAGE_SIZE)
 
 #ifndef CONFIG_XIP_KERNEL
 	. = ALIGN(PAGE_SIZE);

arch/blackfin/kernel/vmlinux.lds.S

@@ -136,7 +136,7 @@ SECTIONS
 
 	. = ALIGN(16);
 	INIT_DATA_SECTION(16)
-	PERCPU(4)
+	PERCPU(32, 4)
 
 	.exit.data :
 	{

arch/cris/kernel/vmlinux.lds.S

@@ -102,7 +102,7 @@ SECTIONS
 #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

arch/frv/kernel/vmlinux.lds.S

@@ -37,7 +37,7 @@ SECTIONS
   _einittext = .;
 
   INIT_DATA_SECTION(8)
-  PERCPU(4096)
+  PERCPU(L1_CACHE_BYTES, 4096)
 
   . = ALIGN(PAGE_SIZE);
   __init_end = .;

arch/ia64/kernel/vmlinux.lds.S

@@ -198,7 +198,7 @@ SECTIONS {
 
 	/* 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

arch/m32r/kernel/vmlinux.lds.S

@@ -53,7 +53,7 @@ SECTIONS
   __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 */

arch/mips/kernel/vmlinux.lds.S

@@ -115,7 +115,7 @@ SECTIONS
 		EXIT_DATA
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(1 << CONFIG_MIPS_L1_CACHE_SHIFT, PAGE_SIZE)
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;
 	/* freed after init ends here */

arch/mn10300/kernel/vmlinux.lds.S

@@ -70,7 +70,7 @@ SECTIONS
 	.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 */

arch/parisc/kernel/vmlinux.lds.S

@@ -145,7 +145,7 @@ SECTIONS
 		EXIT_DATA
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;
 	/* freed after init ends here */

arch/powerpc/kernel/vmlinux.lds.S

@@ -160,7 +160,7 @@ SECTIONS
 		INIT_RAM_FS
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 
 	. = ALIGN(8);
 	.machine.desc : AT(ADDR(.machine.desc) - LOAD_OFFSET) {

arch/s390/kernel/vmlinux.lds.S

@@ -77,7 +77,7 @@ SECTIONS
 	. = ALIGN(PAGE_SIZE);
 	INIT_DATA_SECTION(0x100)
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(0x100, PAGE_SIZE)
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;		/* freed after init ends here */
 

arch/sh/kernel/vmlinux.lds.S

@@ -66,7 +66,7 @@ SECTIONS
 		__machvec_end = .;
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(L1_CACHE_BYTES, PAGE_SIZE)
 
 	/*
 	 * .exit.text is discarded at runtime, not link time, to deal with

arch/sparc/kernel/vmlinux.lds.S

@@ -108,7 +108,7 @@ SECTIONS
 		__sun4v_2insn_patch_end = .;
 	}
 
-	PERCPU(PAGE_SIZE)
+	PERCPU(SMP_CACHE_BYTES, PAGE_SIZE)
 
 	. = ALIGN(PAGE_SIZE);
 	__init_end = .;

arch/tile/kernel/vmlinux.lds.S

@@ -63,7 +63,7 @@ SECTIONS
     *(.init.page)
   } :data =0
   INIT_DATA_SECTION(16)
-  PERCPU(PAGE_SIZE)
+  PERCPU(L2_CACHE_BYTES, PAGE_SIZE)
   . = ALIGN(PAGE_SIZE);
   VMLINUX_SYMBOL(_einitdata) = .;
 

arch/um/include/asm/common.lds.S

@@ -42,7 +42,7 @@
 	INIT_SETUP(0)
   }
 
-  PERCPU(32)
+  PERCPU(32, 32)
 	
   .initcall.init : {
 	INIT_CALLS

arch/x86/include/asm/percpu.h

@@ -451,6 +451,26 @@ do {									\
 #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 @@ do {									\
 #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 */

arch/x86/kernel/vmlinux.lds.S

@@ -230,7 +230,7 @@ SECTIONS
 	 * 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)
@@ -305,7 +305,7 @@ SECTIONS
 	}
 
 #if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
-	PERCPU(THREAD_SIZE)
+	PERCPU(INTERNODE_CACHE_BYTES, THREAD_SIZE)
 #endif
 
 	. = ALIGN(PAGE_SIZE);

arch/x86/lib/Makefile

@@ -42,4 +42,5 @@ else
         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

arch/x86/lib/cmpxchg16b_emu.S

+/*
+ *	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)

arch/xtensa/kernel/vmlinux.lds.S

@@ -155,7 +155,7 @@ SECTIONS
     INIT_RAM_FS
   }
 
-  PERCPU(PAGE_SIZE)
+  PERCPU(XCHAL_ICACHE_LINESIZE, PAGE_SIZE)
 
   /* We need this dummy segment here */
 

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 = .;
@@ -683,13 +683,18 @@
 
 /**
  * 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
@@ -700,7 +705,7 @@
  * 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) {			\
@@ -708,7 +713,9 @@
 		*(.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) = .;			\
@@ -717,18 +724,18 @@
 
 /**
  * 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) = .;			\
@@ -736,7 +743,9 @@
 		*(.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) = .;			\

include/linux/percpu.h

@@ -255,6 +255,30 @@ extern void __bad_size_call_parameter(void);
 	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));					\
@@ -500,6 +524,45 @@ do {									\
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
 #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
@@ -703,6 +766,39 @@ do {									\
 	__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 @@ do {									\
 	__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 */

include/linux/slub_def.h

@@ -35,7 +35,10 @@ enum stat_item {
 	NR_SLUB_STAT_ITEMS };
 
 struct kmem_cache_cpu {
-	void **freelist;	/* Pointer to first free per cpu object */
+	void **freelist;	/* Pointer to next available object */
+#ifdef CONFIG_CMPXCHG_LOCAL
+	unsigned long tid;	/* Globally unique transaction id */
+#endif
 	struct page *page;	/* The slab from which we are allocating */
 	int node;		/* The node of the page (or -1 for debug) */
 #ifdef CONFIG_SLUB_STATS
@@ -70,6 +73,7 @@ struct kmem_cache {
 	struct kmem_cache_cpu __percpu *cpu_slab;
 	/* Used for retriving partial slabs etc */
 	unsigned long flags;
+	unsigned long min_partial;
 	int size;		/* The size of an object including meta data */
 	int objsize;		/* The size of an object without meta data */
 	int offset;		/* Free pointer offset. */
@@ -84,7 +88,6 @@ struct kmem_cache {
 	int inuse;		/* Offset to metadata */
 	int align;		/* Alignment */
 	int reserved;		/* Reserved bytes at the end of slabs */
-	unsigned long min_partial;
 	const char *name;	/* Name (only for display!) */
 	struct list_head list;	/* List of slab caches */
 #ifdef CONFIG_SYSFS

mm/slub.c

@@ -836,14 +836,24 @@ static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void
 static inline void slab_free_hook(struct kmem_cache *s, void *x)
 {
 	kmemleak_free_recursive(x, s->flags);
-}
 
-static inline void slab_free_hook_irq(struct kmem_cache *s, void *object)
-{
-	kmemcheck_slab_free(s, object, s->objsize);
-	debug_check_no_locks_freed(object, s->objsize);
+	/*
+	 * Trouble is that we may no longer disable interupts in the fast path
+	 * So in order to make the debug calls that expect irqs to be
+	 * disabled we need to disable interrupts temporarily.
+	 */
+#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
+	{
+		unsigned long flags;
+
+		local_irq_save(flags);
+		kmemcheck_slab_free(s, x, s->objsize);
+		debug_check_no_locks_freed(x, s->objsize);
 		if (!(s->flags & SLAB_DEBUG_OBJECTS))
-		debug_check_no_obj_freed(object, s->objsize);
+			debug_check_no_obj_freed(x, s->objsize);
+		local_irq_restore(flags);
+	}
+#endif
 }
 
 /*
@@ -1130,9 +1140,6 @@ static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
 
 static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
 
-static inline void slab_free_hook_irq(struct kmem_cache *s,
-		void *object) {}
-
 #endif /* CONFIG_SLUB_DEBUG */
 
 /*
@@ -1533,6 +1540,77 @@ static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
 	}
 }
 
+#ifdef CONFIG_CMPXCHG_LOCAL
+#ifdef CONFIG_PREEMPT
+/*
+ * Calculate the next globally unique transaction for disambiguiation
+ * during cmpxchg. The transactions start with the cpu number and are then
+ * incremented by CONFIG_NR_CPUS.
+ */
+#define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
+#else
+/*
+ * No preemption supported therefore also no need to check for
+ * different cpus.
+ */
+#define TID_STEP 1
+#endif
+
+static inline unsigned long next_tid(unsigned long tid)
+{
+	return tid + TID_STEP;
+}
+
+static inline unsigned int tid_to_cpu(unsigned long tid)
+{
+	return tid % TID_STEP;
+}
+
+static inline unsigned long tid_to_event(unsigned long tid)
+{
+	return tid / TID_STEP;
+}
+
+static inline unsigned int init_tid(int cpu)
+{
+	return cpu;
+}
+
+static inline void note_cmpxchg_failure(const char *n,
+		const struct kmem_cache *s, unsigned long tid)
+{
+#ifdef SLUB_DEBUG_CMPXCHG
+	unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
+
+	printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
+
+#ifdef CONFIG_PREEMPT
+	if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
+		printk("due to cpu change %d -> %d\n",
+			tid_to_cpu(tid), tid_to_cpu(actual_tid));
+	else
+#endif
+	if (tid_to_event(tid) != tid_to_event(actual_tid))
+		printk("due to cpu running other code. Event %ld->%ld\n",
+			tid_to_event(tid), tid_to_event(actual_tid));
+	else
+		printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
+			actual_tid, tid, next_tid(tid));
+#endif
+}
+
+#endif
+
+void init_kmem_cache_cpus(struct kmem_cache *s)
+{
+#if defined(CONFIG_CMPXCHG_LOCAL) && defined(CONFIG_PREEMPT)
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
+#endif
+
+}
 /*
  * Remove the cpu slab
  */
@@ -1564,6 +1642,9 @@ static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
 		page->inuse--;
 	}
 	c->page = NULL;
+#ifdef CONFIG_CMPXCHG_LOCAL
+	c->tid = next_tid(c->tid);
+#endif
 	unfreeze_slab(s, page, tail);
 }
 
@@ -1698,6 +1779,19 @@ static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 {
 	void **object;
 	struct page *new;
+#ifdef CONFIG_CMPXCHG_LOCAL
+	unsigned long flags;
+
+	local_irq_save(flags);
+#ifdef CONFIG_PREEMPT
+	/*
+	 * We may have been preempted and rescheduled on a different
+	 * cpu before disabling interrupts. Need to reload cpu area
+	 * pointer.
+	 */
+	c = this_cpu_ptr(s->cpu_slab);
+#endif
+#endif
 
 	/* We handle __GFP_ZERO in the caller */
 	gfpflags &= ~__GFP_ZERO;
@@ -1724,6 +1818,10 @@ static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 	c->node = page_to_nid(c->page);
 unlock_out:
 	slab_unlock(c->page);
+#ifdef CONFIG_CMPXCHG_LOCAL
+	c->tid = next_tid(c->tid);
+	local_irq_restore(flags);
+#endif
 	stat(s, ALLOC_SLOWPATH);
 	return object;
 
@@ -1785,23 +1883,76 @@ static __always_inline void *slab_alloc(struct kmem_cache *s,
 {
 	void **object;
 	struct kmem_cache_cpu *c;
+#ifdef CONFIG_CMPXCHG_LOCAL
+	unsigned long tid;
+#else
 	unsigned long flags;
+#endif
 
 	if (slab_pre_alloc_hook(s, gfpflags))
 		return NULL;
 
+#ifndef CONFIG_CMPXCHG_LOCAL
 	local_irq_save(flags);
+#else
+redo:
+#endif
+
+	/*
+	 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
+	 * enabled. We may switch back and forth between cpus while
+	 * reading from one cpu area. That does not matter as long
+	 * as we end up on the original cpu again when doing the cmpxchg.
+	 */
 	c = __this_cpu_ptr(s->cpu_slab);
+
+#ifdef CONFIG_CMPXCHG_LOCAL
+	/*
+	 * The transaction ids are globally unique per cpu and per operation on
+	 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
+	 * occurs on the right processor and that there was no operation on the
+	 * linked list in between.
+	 */
+	tid = c->tid;
+	barrier();
+#endif
+
 	object = c->freelist;
 	if (unlikely(!object || !node_match(c, node)))
 
 		object = __slab_alloc(s, gfpflags, node, addr, c);
 
 	else {
+#ifdef CONFIG_CMPXCHG_LOCAL
+		/*
+		 * The cmpxchg will only match if there was no additonal
+		 * operation and if we are on the right processor.
+		 *
+		 * The cmpxchg does the following atomically (without lock semantics!)
+		 * 1. Relocate first pointer to the current per cpu area.
+		 * 2. Verify that tid and freelist have not been changed
+		 * 3. If they were not changed replace tid and freelist
+		 *
+		 * Since this is without lock semantics the protection is only against
+		 * code executing on this cpu *not* from access by other cpus.
+		 */
+		if (unlikely(!this_cpu_cmpxchg_double(
+				s->cpu_slab->freelist, s->cpu_slab->tid,
+				object, tid,
+				get_freepointer(s, object), next_tid(tid)))) {
+
+			note_cmpxchg_failure("slab_alloc", s, tid);
+			goto redo;
+		}
+#else
 		c->freelist = get_freepointer(s, object);
+#endif
 		stat(s, ALLOC_FASTPATH);
 	}
+
+#ifndef CONFIG_CMPXCHG_LOCAL
 	local_irq_restore(flags);
+#endif
 
 	if (unlikely(gfpflags & __GFP_ZERO) && object)
 		memset(object, 0, s->objsize);
@@ -1879,9 +2030,13 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
 {
 	void *prior;
 	void **object = (void *)x;
+#ifdef CONFIG_CMPXCHG_LOCAL
+	unsigned long flags;
 
-	stat(s, FREE_SLOWPATH);
+	local_irq_save(flags);
+#endif
 	slab_lock(page);
+	stat(s, FREE_SLOWPATH);
 
 	if (kmem_cache_debug(s))
 		goto debug;
@@ -1911,6 +2066,9 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
 
 out_unlock:
 	slab_unlock(page);
+#ifdef CONFIG_CMPXCHG_LOCAL
+	local_irq_restore(flags);
+#endif
 	return;
 
 slab_empty:
@@ -1922,6 +2080,9 @@ static void __slab_free(struct kmem_cache *s, struct page *page,
 		stat(s, FREE_REMOVE_PARTIAL);
 	}
 	slab_unlock(page);
+#ifdef CONFIG_CMPXCHG_LOCAL
+	local_irq_restore(flags);
+#endif
 	stat(s, FREE_SLAB);
 	discard_slab(s, page);
 	return;
@@ -1948,23 +2109,56 @@ static __always_inline void slab_free(struct kmem_cache *s,
 {
 	void **object = (void *)x;
 	struct kmem_cache_cpu *c;
+#ifdef CONFIG_CMPXCHG_LOCAL
+	unsigned long tid;
+#else
 	unsigned long flags;
+#endif
 
 	slab_free_hook(s, x);
 
+#ifndef CONFIG_CMPXCHG_LOCAL
 	local_irq_save(flags);
+
+#else
+redo:
+#endif
+
+	/*
+	 * Determine the currently cpus per cpu slab.
+	 * The cpu may change afterward. However that does not matter since
+	 * data is retrieved via this pointer. If we are on the same cpu
+	 * during the cmpxchg then the free will succedd.
+	 */
 	c = __this_cpu_ptr(s->cpu_slab);
 
-	slab_free_hook_irq(s, x);
+#ifdef CONFIG_CMPXCHG_LOCAL
+	tid = c->tid;
+	barrier();
+#endif
 
 	if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
 		set_freepointer(s, object, c->freelist);
+
+#ifdef CONFIG_CMPXCHG_LOCAL
+		if (unlikely(!this_cpu_cmpxchg_double(
+				s->cpu_slab->freelist, s->cpu_slab->tid,
+				c->freelist, tid,
+				object, next_tid(tid)))) {
+
+			note_cmpxchg_failure("slab_free", s, tid);
+			goto redo;
+		}
+#else
 		c->freelist = object;
+#endif
 		stat(s, FREE_FASTPATH);
 	} else
 		__slab_free(s, page, x, addr);
 
+#ifndef CONFIG_CMPXCHG_LOCAL
 	local_irq_restore(flags);
+#endif
 }
 
 void kmem_cache_free(struct kmem_cache *s, void *x)
@@ -2156,9 +2350,23 @@ static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
 			SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
 
+#ifdef CONFIG_CMPXCHG_LOCAL
+	/*
+	 * Must align to double word boundary for the double cmpxchg instructions
+	 * to work.
+	 */
+	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
+#else
+	/* Regular alignment is sufficient */
 	s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
+#endif
 
-	return s->cpu_slab != NULL;
+	if (!s->cpu_slab)
+		return 0;
+
+	init_kmem_cache_cpus(s);
+
+	return 1;
 }
 
 static struct kmem_cache *kmem_cache_node;