[ltt-dev] [PATCH 07/11] add uatomic_gcc.h, use it for default definitions

[compudj@krystal.dyndns.org (Mathieu Desnoyers)]

* Paolo Bonzini (pbonzini at redhat.com) wrote:
> From most of the architectures, we can derive the implementation
> of xchg and add_return from cmpxchg:
> 
> - if cmpxchg is present, use it to implement xchg and add_return;

Not sure I understand this comment, and not sure it matches the patch.
x86 has xchg() which is faster than a cmpxchg-based fallback, and you
seem to leave the code as-is. Can you elaborate ?

> 
> - if it is not present, implement all three using __sync_* builtins
> 

Starting from which gcc versions does these __sync_* builtins work ?
(question applies for the builtin memory barrier too).

Thanks,

Mathieu

> The hunk in tests/test_uatomic.c is only needed for bisectability
> and will be removed later.
> 
> Signed-off-by: Paolo Bonzini <pbonzini at redhat.com>
> ---
>  Makefile.am                 |    2 +-
>  tests/test_uatomic.c        |    2 +
>  urcu/uatomic_arch_ppc.h     |   15 +---
>  urcu/uatomic_arch_s390.h    |   15 +---
>  urcu/uatomic_arch_sparc64.h |   15 +---
>  urcu/uatomic_arch_x86.h     |   26 ++----
>  urcu/uatomic_gcc.h          |  221 +++++++++++++++++++++++++++++++++++++++++++
>  7 files changed, 239 insertions(+), 57 deletions(-)
>  create mode 100644 urcu/uatomic_gcc.h
> 
> diff --git a/Makefile.am b/Makefile.am
> index 57e9299..d6ee789 100644
> --- a/Makefile.am
> +++ b/Makefile.am
> @@ -6,7 +6,7 @@ SUBDIRS = . tests
>  
>  include_HEADERS = urcu.h $(top_srcdir)/urcu-*.h
>  nobase_dist_include_HEADERS = urcu/compiler.h urcu/hlist.h urcu/list.h \
> -		urcu/rculist.h urcu/system.h urcu/urcu-futex.h
> +		urcu/rculist.h urcu/system.h urcu/urcu-futex.h urcu/uatomic_gcc.h
>  nobase_nodist_include_HEADERS = urcu/arch.h urcu/uatomic_arch.h urcu/config.h
>  
>  EXTRA_DIST = $(top_srcdir)/urcu/arch_*.h $(top_srcdir)/urcu/uatomic_arch_*.h \
> diff --git a/tests/test_uatomic.c b/tests/test_uatomic.c
> index 68cb6df..c0f36fe 100644
> --- a/tests/test_uatomic.c
> +++ b/tests/test_uatomic.c
> @@ -1,5 +1,7 @@
>  #include <stdio.h>
>  #include <assert.h>
> +
> +#define UATOMIC_NO_LINK_ERROR
>  #include <urcu/uatomic_arch.h>
>  
>  #if (defined(__i386__) || defined(__x86_64__))
> diff --git a/urcu/uatomic_arch_ppc.h b/urcu/uatomic_arch_ppc.h
> index 7106b99..c611c7a 100644
> --- a/urcu/uatomic_arch_ppc.h
> +++ b/urcu/uatomic_arch_ppc.h
> @@ -39,9 +39,6 @@ extern "C" {
>  
>  #define ILLEGAL_INSTR	".long	0xd00d00"
>  
> -#define uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
> -#define uatomic_read(addr)	LOAD_SHARED(*(addr))
> -
>  /*
>   * Using a isync as second barrier for exchange to provide acquire semantic.
>   * According to uatomic_ops/sysdeps/gcc/powerpc.h, the documentation is "fairly
> @@ -217,18 +214,10 @@ unsigned long _uatomic_add_return(void *addr, unsigned long val,
>  						  (unsigned long)(v),	\
>  						  sizeof(*(addr))))
>  
> -/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */
> -
> -#define uatomic_sub_return(addr, v)	uatomic_add_return((addr), -(v))
> -
> -#define uatomic_add(addr, v)		(void)uatomic_add_return((addr), (v))
> -#define uatomic_sub(addr, v)		(void)uatomic_sub_return((addr), (v))
> -
> -#define uatomic_inc(addr)		uatomic_add((addr), 1)
> -#define uatomic_dec(addr)		uatomic_add((addr), -1)
> -
>  #ifdef __cplusplus 
>  }
>  #endif
>  
> +#include <urcu/uatomic_gcc.h>
> +
>  #endif /* _URCU_ARCH_UATOMIC_PPC_H */
> diff --git a/urcu/uatomic_arch_s390.h b/urcu/uatomic_arch_s390.h
> index 6247601..9d4b74e 100644
> --- a/urcu/uatomic_arch_s390.h
> +++ b/urcu/uatomic_arch_s390.h
> @@ -70,9 +70,6 @@ struct __uatomic_dummy {
>  };
>  #define __hp(x)	((struct __uatomic_dummy *)(x))
>  
> -#define uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
> -#define uatomic_read(addr)	LOAD_SHARED(*(addr))
> -
>  /* xchg */
>  
>  static inline __attribute__((always_inline))
> @@ -200,18 +197,10 @@ unsigned long _uatomic_add_return(void *addr, unsigned long val, int len)
>  						  (unsigned long)(v),	\
>  						  sizeof(*(addr))))
>  
> -/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */
> -
> -#define uatomic_sub_return(addr, v)	uatomic_add_return((addr), -(v))
> -
> -#define uatomic_add(addr, v)		(void)uatomic_add_return((addr), (v))
> -#define uatomic_sub(addr, v)		(void)uatomic_sub_return((addr), (v))
> -
> -#define uatomic_inc(addr)		uatomic_add((addr), 1)
> -#define uatomic_dec(addr)		uatomic_add((addr), -1)
> -
>  #ifdef __cplusplus 
>  }
>  #endif
>  
> +#include <urcu/uatomic_gcc.h>
> +
>  #endif /* _URCU_UATOMIC_ARCH_S390_H */
> diff --git a/urcu/uatomic_arch_sparc64.h b/urcu/uatomic_arch_sparc64.h
> index e984986..d21d73c 100644
> --- a/urcu/uatomic_arch_sparc64.h
> +++ b/urcu/uatomic_arch_sparc64.h
> @@ -31,9 +31,6 @@ extern "C" {
>  #define BITS_PER_LONG	(__SIZEOF_LONG__ * 8)
>  #endif
>  
> -#define uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
> -#define uatomic_read(addr)	LOAD_SHARED(*(addr))
> -
>  /* cmpxchg */
>  
>  static inline __attribute__((always_inline))
> @@ -161,18 +158,10 @@ unsigned long _uatomic_add_return(void *addr, unsigned long val, int len)
>  						  (unsigned long)(v),	\
>  						  sizeof(*(addr))))
>  
> -/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */
> -
> -#define uatomic_sub_return(addr, v)	uatomic_add_return((addr), -(v))
> -
> -#define uatomic_add(addr, v)		(void)uatomic_add_return((addr), (v))
> -#define uatomic_sub(addr, v)		(void)uatomic_sub_return((addr), (v))
> -
> -#define uatomic_inc(addr)		uatomic_add((addr), 1)
> -#define uatomic_dec(addr)		uatomic_add((addr), -1)
> -
>  #ifdef __cplusplus 
>  }
>  #endif
>  
> +#include <urcu/uatomic_gcc.h>
> +
>  #endif /* _URCU_ARCH_UATOMIC_PPC_H */
> diff --git a/urcu/uatomic_arch_x86.h b/urcu/uatomic_arch_x86.h
> index 34f5f87..0aadbd5 100644
> --- a/urcu/uatomic_arch_x86.h
> +++ b/urcu/uatomic_arch_x86.h
> @@ -41,7 +41,11 @@ struct __uatomic_dummy {
>  #define __hp(x)	((struct __uatomic_dummy *)(x))
>  
>  #define _uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
> +
> +#if 0
> +/* Read is atomic even in compat mode */
>  #define _uatomic_read(addr)	LOAD_SHARED(*(addr))
> +#endif
>  
>  /* cmpxchg */
>  
> @@ -168,7 +172,7 @@ unsigned long __uatomic_exchange(void *addr, unsigned long val, int len)
>  	((__typeof__(*(addr))) __uatomic_exchange((addr), (unsigned long)(v), \
>  						sizeof(*(addr))))
>  
> -/* uatomic_add_return, uatomic_sub_return */
> +/* uatomic_add_return */
>  
>  static inline __attribute__((always_inline))
>  unsigned long __uatomic_add_return(void *addr, unsigned long val,
> @@ -233,9 +237,7 @@ unsigned long __uatomic_add_return(void *addr, unsigned long val,
>  						  (unsigned long)(v),	\
>  						  sizeof(*(addr))))
>  
> -#define _uatomic_sub_return(addr, v)	_uatomic_add_return((addr), -(v))
> -
> -/* uatomic_add, uatomic_sub */
> +/* uatomic_add */
>  
>  static inline __attribute__((always_inline))
>  void __uatomic_add(void *addr, unsigned long val, int len)
> @@ -289,8 +291,6 @@ void __uatomic_add(void *addr, unsigned long val, int len)
>  #define _uatomic_add(addr, v)						   \
>  	(__uatomic_add((addr), (unsigned long)(v), sizeof(*(addr))))
>  
> -#define _uatomic_sub(addr, v)	_uatomic_add((addr), -(v))
> -
>  
>  /* uatomic_inc */
>  
> @@ -441,24 +441,17 @@ extern unsigned long _compat_uatomic_xchg(void *addr,
>  						(unsigned long)(v), 	       \
>  						sizeof(*(addr))))
>  
> -#define compat_uatomic_sub_return(addr, v)				       \
> -		compat_uatomic_add_return((addr), -(v))
>  #define compat_uatomic_add(addr, v)					       \
>  		((void)compat_uatomic_add_return((addr), (v)))
> -#define compat_uatomic_sub(addr, v)					       \
> -		((void)compat_uatomic_sub_return((addr), (v)))
>  #define compat_uatomic_inc(addr)					       \
>  		(compat_uatomic_add((addr), 1))
>  #define compat_uatomic_dec(addr)					       \
> -		(compat_uatomic_sub((addr), 1))
> +		(compat_uatomic_add((addr), -1))
>  
>  #else
>  #define UATOMIC_COMPAT(insn)	(_uatomic_##insn)
>  #endif
>  
> -/* Read is atomic even in compat mode */
> -#define uatomic_read(addr)	_uatomic_read(addr)
> -
>  #define uatomic_set(addr, v)			\
>  		UATOMIC_COMPAT(set(addr, v))
>  #define uatomic_cmpxchg(addr, old, _new)	\
> @@ -467,10 +460,7 @@ extern unsigned long _compat_uatomic_xchg(void *addr,
>  		UATOMIC_COMPAT(xchg(addr, v))
>  #define uatomic_add_return(addr, v)		\
>  		UATOMIC_COMPAT(add_return(addr, v))
> -#define uatomic_sub_return(addr, v)		\
> -		UATOMIC_COMPAT(sub_return(addr, v))
>  #define uatomic_add(addr, v)	UATOMIC_COMPAT(add(addr, v))
> -#define uatomic_sub(addr, v)	UATOMIC_COMPAT(sub(addr, v))
>  #define uatomic_inc(addr)	UATOMIC_COMPAT(inc(addr))
>  #define uatomic_dec(addr)	UATOMIC_COMPAT(dec(addr))
>  
> @@ -478,4 +468,6 @@ extern unsigned long _compat_uatomic_xchg(void *addr,
>  }
>  #endif
>  
> +#include <urcu/uatomic_gcc.h>
> +
>  #endif /* _URCU_ARCH_UATOMIC_X86_H */
> diff --git a/urcu/uatomic_gcc.h b/urcu/uatomic_gcc.h
> new file mode 100644
> index 0000000..28f6d71
> --- /dev/null
> +++ b/urcu/uatomic_gcc.h
> @@ -0,0 +1,223 @@
> +#ifndef _URCU_UATOMIC_GCC_H
> +#define _URCU_UATOMIC_GCC_H
> +
> +/*
> + * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
> + * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
> + * Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
> + * Copyright (c) 2009      Mathieu Desnoyers
> + * Copyright (c) 2010      Paolo Bonzini
> + *
> + * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
> + * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
> + *
> + * Permission is hereby granted to use or copy this program
> + * for any purpose,  provided the above notices are retained on all copies.
> + * Permission to modify the code and to distribute modified code is granted,
> + * provided the above notices are retained, and a notice that the code was
> + * modified is included with the above copyright notice.
> + *
> + * Code inspired from libuatomic_ops-1.2, inherited in part from the
> + * Boehm-Demers-Weiser conservative garbage collector.
> + */
> +
> +#include <urcu/compiler.h>
> +#include <urcu/system.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +#ifndef BITS_PER_LONG
> +#define BITS_PER_LONG	(__SIZEOF_LONG__ * 8)
> +#endif
> +
> +#ifndef uatomic_set
> +#define uatomic_set(addr, v)	STORE_SHARED(*(addr), (v))
> +#endif
> +
> +#ifndef uatomic_read
> +#define uatomic_read(addr)	LOAD_SHARED(*(addr))
> +#endif
> +
> +#if !defined __OPTIMIZE__  || defined UATOMIC_NO_LINK_ERROR
> +static inline __attribute__((always_inline))
> +void _uatomic_link_error()
> +{
> +#ifdef ILLEGAL_INSTR
> +	/* generate an illegal instruction. Cannot catch this with linker tricks
> +	 * when optimizations are disabled. */
> +	__asm__ __volatile__(ILLEGAL_INSTR);
> +#else
> +	__builtin_trap ();
> +#endif
> +}
> +#else
> +extern void _uatomic_link_error ();
> +#endif
> +
> +/* cmpxchg */
> +
> +#ifndef uatomic_cmpxchg
> +static inline __attribute__((always_inline))
> +unsigned long _uatomic_cmpxchg(void *addr, unsigned long old,
> +			      unsigned long _new, int len)
> +{
> +	switch (len) {
> +	case 4:
> +		return __sync_val_compare_and_swap_4(addr, old, _new);
> +#if (BITS_PER_LONG == 64)
> +	case 8:
> +		return __sync_val_compare_and_swap_8(addr, old, _new);
> +#endif
> +	}
> +	_uatomic_link_error();
> +	return 0;
> +}
> +
> +
> +#define uatomic_cmpxchg(addr, old, _new)				    \
> +	((__typeof__(*(addr))) _uatomic_cmpxchg((addr), (unsigned long)(old),\
> +						(unsigned long)(_new), 	    \
> +						sizeof(*(addr))))
> +
> +
> +/* uatomic_add_return */
> +
> +#ifndef uatomic_add_return
> +static inline __attribute__((always_inline))
> +unsigned long _uatomic_add_return(void *addr, unsigned long val,
> +				 int len)
> +{
> +	switch (len) {
> +	case 4:
> +		return __sync_add_and_fetch_4(addr, val);
> +#if (BITS_PER_LONG == 64)
> +	case 8:
> +		return __sync_add_and_fetch_8(addr, val);
> +#endif
> +	}
> +	_uatomic_link_error();
> +	return 0;
> +}
> +
> +
> +#define uatomic_add_return(addr, v)					\
> +	((__typeof__(*(addr))) _uatomic_add_return((addr),		\
> +						  (unsigned long)(v),	\
> +						  sizeof(*(addr))))
> +#endif
> +
> +#else
> +
> +#ifndef uatomic_add_return
> +/* uatomic_add_return */
> +
> +static inline __attribute__((always_inline))
> +unsigned long _uatomic_add_return(void *addr, unsigned long val, int len)
> +{
> +	switch (len) {
> +	case 4:
> +	{
> +		unsigned int old, oldt;
> +
> +		oldt = uatomic_read((unsigned int *)addr);
> +		do {
> +			old = oldt;
> +			oldt = _uatomic_cmpxchg(addr, old, old + val, 4);
> +		} while (oldt != old);
> +
> +		return old + val;
> +	}
> +#if (BITS_PER_LONG == 64)
> +	case 8:
> +	{
> +		unsigned long old, oldt;
> +
> +		oldt = uatomic_read((unsigned long *)addr);
> +		do {
> +			old = oldt;
> +			oldt = _uatomic_cmpxchg(addr, old, old + val, 8);
> +		} while (oldt != old);
> +
> +		return old + val;
> +	}
> +#endif
> +	}
> +	_uatomic_link_error();
> +	return 0;
> +}
> +
> +#define uatomic_add_return(addr, v)					\
> +	((__typeof__(*(addr))) _uatomic_add_return((addr),		\
> +						  (unsigned long)(v),	\
> +						  sizeof(*(addr))))
> +#endif
> +
> +#endif
> +
> +#ifndef uatomic_xchg
> +/* xchg */
> +
> +static inline __attribute__((always_inline))
> +unsigned long _uatomic_exchange(void *addr, unsigned long val, int len)
> +{
> +	switch (len) {
> +	case 4:
> +	{
> +		unsigned int old, oldt;
> +
> +		oldt = uatomic_read((unsigned int *)addr);
> +		do {
> +			old = oldt;
> +			oldt = _uatomic_cmpxchg(addr, old, val, 4);
> +		} while (oldt != old);
> +
> +		return old;
> +	}
> +#if (BITS_PER_LONG == 64)
> +	case 8:
> +	{
> +		unsigned long old, oldt;
> +
> +		oldt = uatomic_read((unsigned long *)addr);
> +		do {
> +			old = oldt;
> +			oldt = _uatomic_cmpxchg(addr, old, val, 8);
> +		} while (oldt != old);
> +
> +		return old;
> +	}
> +#endif
> +	}
> +	_uatomic_link_error();
> +	return 0;
> +}
> +
> +#define uatomic_xchg(addr, v)						    \
> +	((__typeof__(*(addr))) _uatomic_exchange((addr), (unsigned long)(v), \
> +						sizeof(*(addr))))
> +#endif
> +
> +/* uatomic_sub_return, uatomic_add, uatomic_sub, uatomic_inc, uatomic_dec */
> +
> +#ifndef uatomic_add
> +#define uatomic_add(addr, v)		(void)uatomic_add_return((addr), (v))
> +#endif
> +
> +#define uatomic_sub_return(addr, v)	uatomic_add_return((addr), -(v))
> +#define uatomic_sub(addr, v)		uatomic_add((addr), -(v))
> +
> +#ifndef uatomic_inc
> +#define uatomic_inc(addr)		uatomic_add((addr), 1)
> +#endif
> +
> +#ifndef uatomic_dec
> +#define uatomic_dec(addr)		uatomic_add((addr), -1)
> +#endif
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif /* _URCU_UATOMIC_GCC_H */
> -- 
> 1.6.6
> 
> 
> 
> _______________________________________________
> ltt-dev mailing list
> ltt-dev at lists.casi.polymtl.ca
> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68