[lttng-dev] [PATCH v2 05/12] urcu/uatomic: Add CMM memory model

Paul E. McKenney paulmck at kernel.org
Wed Jun 21 19:28:48 EDT 2023


On Wed, Jun 07, 2023 at 02:53:52PM -0400, Olivier Dion wrote:
> Introducing the CMM memory model with the following new primitives:
> 
>   - uatomic_load(addr, memory_order)
> 
>   - uatomic_store(addr, value, memory_order)
>   - uatomic_and_mo(addr, mask, memory_order)
>   - uatomic_or_mo(addr, mask, memory_order)
>   - uatomic_add_mo(addr, value, memory_order)
>   - uatomic_sub_mo(addr, value, memory_order)
>   - uatomic_inc_mo(addr, memory_order)
>   - uatomic_dec_mo(addr, memory_order)
> 
>   - uatomic_add_return_mo(addr, value, memory_order)
>   - uatomic_sub_return_mo(addr, value, memory_order)
> 
>   - uatomic_xchg_mo(addr, value, memory_order)
> 
>   - uatomic_cmpxchg_mo(addr, old, new,
>                        memory_order_success,
>                        memory_order_failure)
> 
> The CMM memory model reflects the C11 memory model with an additional
> CMM_SEQ_CST_FENCE memory order. The memory order can be selected through
> the enum cmm_memorder.
> 
> * With Atomic Builtins
> 
> If configured with atomic builtins, the correspondence between the CMM
> memory model and the C11 memory model is a one to one at the exception
> of the CMM_SEQ_CST_FENCE memory order which implies the memory order
> CMM_SEQ_CST and a thread fence after the operation.
> 
> * Without Atomic Builtins
> 
> However, if not configured with atomic builtins, the following stipulate
> the memory model.
> 
> For load operations with uatomic_load(), the memory orders CMM_RELAXED,
> CMM_CONSUME, CMM_ACQUIRE, CMM_SEQ_CST and CMM_SEQ_CST_FENCE are
> allowed. A barrier may be inserted before and after the load from memory
> depending on the memory order:
> 
>   - CMM_RELAXED: No barrier
>   - CMM_CONSUME: Memory barrier after read
>   - CMM_ACQUIRE: Memory barrier after read
>   - CMM_SEQ_CST: Memory barriers before and after read
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after read
> 
> For store operations with uatomic_store(), the memory orders
> CMM_RELAXED, CMM_RELEASE, CMM_SEQ_CST and CMM_SEQ_CST_FENCE are
> allowed. A barrier may be inserted before and after the store to memory
> depending on the memory order:
> 
>   - CMM_RELAXED: No barrier
>   - CMM_RELEASE: Memory barrier before operation
>   - CMM_SEQ_CST: Memory barriers before and after operation
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after operation
> 
> For load/store operations with uatomic_and_mo(), uatomic_or_mo(),
> uatomic_add_mo(), uatomic_sub_mo(), uatomic_inc_mo(), uatomic_dec_mo(),
> uatomic_add_return_mo() and uatomic_sub_return_mo(), all memory orders
> are allowed. A barrier may be inserted before and after the operation
> depending on the memory order:
> 
>   - CMM_RELAXED: No barrier
>   - CMM_ACQUIRE: Memory barrier after operation
>   - CMM_CONSUME: Memory barrier after operation
>   - CMM_RELEASE: Memory barrier before operation
>   - CMM_ACQ_REL: Memory barriers before and after operation
>   - CMM_SEQ_CST: Memory barriers before and after operation
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after operation
> 
> For the exchange operation uatomic_xchg_mo(), any memory order is
> valid. A barrier may be inserted before and after the exchange to memory
> depending on the memory order:
> 
>   - CMM_RELAXED: No barrier
>   - CMM_ACQUIRE: Memory barrier after operation
>   - CMM_CONSUME: Memory barrier after operation
>   - CMM_RELEASE: Memory barrier before operation
>   - CMM_ACQ_REL: Memory barriers before and after operation
>   - CMM_SEQ_CST: Memory barriers before and after operation
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after operation
> 
> For the compare exchange operation uatomic_cmpxchg_mo(), the success
> memory order can be anything while the failure memory order cannot be
> CMM_RELEASE nor CMM_ACQ_REL and cannot be stronger than the success
> memory order. A barrier may be inserted before and after the store to
> memory depending on the memory orders:
> 
>  Success memory order:
> 
>   - CMM_RELAXED: No barrier
>   - CMM_ACQUIRE: Memory barrier after operation
>   - CMM_CONSUME: Memory barrier after operation
>   - CMM_RELEASE: Memory barrier before operation
>   - CMM_ACQ_REL: Memory barriers before and after operation
>   - CMM_SEQ_CST: Memory barriers before and after operation
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after operation
> 
>   Barriers after the operations are only emitted if the compare exchange
>   succeed.
> 
>  Failure memory order:
>   - CMM_RELAXED: No barrier
>   - CMM_ACQUIRE: Memory barrier after operation
>   - CMM_CONSUME: Memory barrier after operation
>   - CMM_SEQ_CST: Memory barriers before and after operation
>   - CMM_SEQ_CST_FENCE: Memory barriers before and after operation
> 
>   Barriers after the operations are only emitted if the compare exchange
>   failed.  Barriers before the operation are never emitted by this
>   memory order.
> 
> Change-Id: I213ba19c84e82a63083f00143a3142ffbdab1d52
> Co-authored-by: Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
> Signed-off-by: Olivier Dion <odion at efficios.com>
> ---
>  doc/uatomic-api.md                      |   3 +-
>  include/Makefile.am                     |   2 +
>  include/urcu/static/pointer.h           |  40 ++--
>  include/urcu/uatomic.h                  |  63 ++++++-
>  include/urcu/uatomic/builtins-generic.h | 170 +++++++++++++++++
>  include/urcu/uatomic/builtins.h         |  79 ++++++++
>  include/urcu/uatomic/generic.h          | 234 ++++++++++++++++++++++++
>  src/urcu-pointer.c                      |   9 +-
>  8 files changed, 565 insertions(+), 35 deletions(-)
>  create mode 100644 include/urcu/uatomic/builtins-generic.h
>  create mode 100644 include/urcu/uatomic/builtins.h
> 
> diff --git a/doc/uatomic-api.md b/doc/uatomic-api.md
> index 0962399..7341ee8 100644
> --- a/doc/uatomic-api.md
> +++ b/doc/uatomic-api.md
> @@ -52,7 +52,8 @@ An atomic read-modify-write operation that performs this
>  sequence of operations atomically: check if `addr` contains `old`.
>  If true, then replace the content of `addr` by `new`. Return the
>  value previously contained by `addr`. This function implies a full
> -memory barrier before and after the atomic operation.
> +memory barrier before and after the atomic operation. The second memory
> +barrier is only emitted if the operation succeeded.
>  
>  
>  ```c
> diff --git a/include/Makefile.am b/include/Makefile.am
> index ba1fe60..b20e56d 100644
> --- a/include/Makefile.am
> +++ b/include/Makefile.am
> @@ -63,6 +63,8 @@ nobase_include_HEADERS = \
>  	urcu/uatomic/alpha.h \
>  	urcu/uatomic_arch.h \
>  	urcu/uatomic/arm.h \
> +	urcu/uatomic/builtins.h \
> +	urcu/uatomic/builtins-generic.h \
>  	urcu/uatomic/gcc.h \
>  	urcu/uatomic/generic.h \
>  	urcu/uatomic.h \
> diff --git a/include/urcu/static/pointer.h b/include/urcu/static/pointer.h
> index 9e46a57..9da8657 100644
> --- a/include/urcu/static/pointer.h
> +++ b/include/urcu/static/pointer.h
> @@ -96,23 +96,8 @@ extern "C" {
>   * -Wincompatible-pointer-types errors.  Using the statement expression
>   * makes it an rvalue and gets rid of the const-ness.
>   */
> -#ifdef __URCU_DEREFERENCE_USE_ATOMIC_CONSUME
> -# define _rcu_dereference(p) __extension__ ({						\
> -				__typeof__(__extension__ ({				\
> -					__typeof__(p) __attribute__((unused)) _________p0 = { 0 }; \
> -					_________p0;					\
> -				})) _________p1;					\
> -				__atomic_load(&(p), &_________p1, __ATOMIC_CONSUME);	\

There is talk of getting rid of memory_order_consume.  But for the moment,
it is what there is.  Another alternative is to use a volatile load,
similar to old-style CMM_LOAD_SHARED() or in-kernel READ_ONCE().

						Thanx, Paul

> -				(_________p1);						\
> -			})
> -#else
> -# define _rcu_dereference(p) __extension__ ({						\
> -				__typeof__(p) _________p1 = CMM_LOAD_SHARED(p);		\
> -				cmm_smp_read_barrier_depends();				\
> -				(_________p1);						\
> -			})
> -#endif
> -
> +# define _rcu_dereference(p)			\
> +	uatomic_load(&(p), CMM_CONSUME)
>  /**
>   * _rcu_cmpxchg_pointer - same as rcu_assign_pointer, but tests if the pointer
>   * is as expected by "old". If succeeds, returns the previous pointer to the
> @@ -131,8 +116,9 @@ extern "C" {
>  	({								\
>  		__typeof__(*p) _________pold = (old);			\
>  		__typeof__(*p) _________pnew = (_new);			\
> -		uatomic_cmpxchg(p, _________pold, _________pnew);	\
> -	})
> +		uatomic_cmpxchg_mo(p, _________pold, _________pnew,	\
> +				   CMM_SEQ_CST, CMM_SEQ_CST);		\
> +	});
>  
>  /**
>   * _rcu_xchg_pointer - same as rcu_assign_pointer, but returns the previous
> @@ -149,17 +135,17 @@ extern "C" {
>  	__extension__					\
>  	({						\
>  		__typeof__(*p) _________pv = (v);	\
> -		uatomic_xchg(p, _________pv);		\
> +		uatomic_xchg_mo(p, _________pv,		\
> +				CMM_SEQ_CST);		\
>  	})
>  
>  
> -#define _rcu_set_pointer(p, v)				\
> -	do {						\
> -		__typeof__(*p) _________pv = (v);	\
> -		if (!__builtin_constant_p(v) || 	\
> -		    ((v) != NULL))			\
> -			cmm_wmb();				\
> -		uatomic_set(p, _________pv);		\
> +#define _rcu_set_pointer(p, v)						\
> +	do {								\
> +		__typeof__(*p) _________pv = (v);			\
> +		uatomic_store(p, _________pv,				\
> +			__builtin_constant_p(v) && (v) == NULL ?	\
> +			CMM_RELAXED : CMM_RELEASE);			\
>  	} while (0)
>  
>  /**
> diff --git a/include/urcu/uatomic.h b/include/urcu/uatomic.h
> index 2fb5fd4..be857e1 100644
> --- a/include/urcu/uatomic.h
> +++ b/include/urcu/uatomic.h
> @@ -21,9 +21,70 @@
>  #ifndef _URCU_UATOMIC_H
>  #define _URCU_UATOMIC_H
>  
> +#include <assert.h>
> +
>  #include <urcu/arch.h>
> +#include <urcu/config.h>
>  
> -#if defined(URCU_ARCH_X86)
> +enum cmm_memorder {
> +	CMM_RELAXED = 0,
> +	CMM_CONSUME = 1,
> +	CMM_ACQUIRE = 2,
> +	CMM_RELEASE = 3,
> +	CMM_ACQ_REL = 4,
> +	CMM_SEQ_CST = 5,
> +	CMM_SEQ_CST_FENCE = 6,
> +};
> +
> +#ifdef CONFIG_RCU_USE_ATOMIC_BUILTINS
> +
> +/*
> + * Make sure that CMM_SEQ_CST_FENCE is not equivalent to other memory orders.
> + */
> +# ifdef static_assert
> +static_assert(CMM_RELAXED == __ATOMIC_RELAXED, "");
> +static_assert(CMM_CONSUME == __ATOMIC_CONSUME, "");
> +static_assert(CMM_ACQUIRE == __ATOMIC_ACQUIRE, "");
> +static_assert(CMM_RELEASE == __ATOMIC_RELEASE, "");
> +static_assert(CMM_ACQ_REL == __ATOMIC_ACQ_REL, "");
> +static_assert(CMM_SEQ_CST == __ATOMIC_SEQ_CST, "");
> +# endif
> +
> +/*
> + * This is not part of the public API. It it used internally to implement the
> + * CMM_SEQ_CST_FENCE memory order.
> + *
> + * NOTE: Using switch here instead of if statement to avoid -Wduplicated-cond
> + * warning when memory order is conditionally determined.
> + */
> +static inline void cmm_seq_cst_fence_after_atomic(enum cmm_memorder mo)
> +{
> +	switch (mo) {
> +	case CMM_SEQ_CST_FENCE:
> +		cmm_smp_mb();
> +		break;
> +	default:
> +		break;
> +	}
> +}
> +
> +#endif
> +
> +/*
> + * This is not part of the public API. It is used internally to convert from the
> + * CMM memory model to the C11 memory model.
> + */
> +static inline int cmm_to_c11(int mo)
> +{
> +	if (mo == CMM_SEQ_CST_FENCE) {
> +		return CMM_SEQ_CST;
> +	}
> +	return mo;
> +}
> +
> +#if defined(CONFIG_RCU_USE_ATOMIC_BUILTINS)
> +#include <urcu/uatomic/builtins.h>
> +#elif defined(URCU_ARCH_X86)
>  #include <urcu/uatomic/x86.h>
>  #elif defined(URCU_ARCH_PPC)
>  #include <urcu/uatomic/ppc.h>
> diff --git a/include/urcu/uatomic/builtins-generic.h b/include/urcu/uatomic/builtins-generic.h
> new file mode 100644
> index 0000000..673e888
> --- /dev/null
> +++ b/include/urcu/uatomic/builtins-generic.h
> @@ -0,0 +1,170 @@
> +/*
> + * urcu/uatomic/builtins-generic.h
> + *
> + * Copyright (c) 2023 Olivier Dion <odion at efficios.com>
> + *
> + * This library is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * This library is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with this library; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +#ifndef _URCU_UATOMIC_BUILTINS_GENERIC_H
> +#define _URCU_UATOMIC_BUILTINS_GENERIC_H
> +
> +#include <urcu/system.h>
> +
> +#define uatomic_store(addr, v, mo)				\
> +	__extension__						\
> +	({							\
> +		__atomic_store_n(addr, v, cmm_to_c11(mo));	\
> +		cmm_seq_cst_fence_after_atomic(mo);		\
> +	})
> +
> +#define uatomic_set(addr, v)			\
> +	uatomic_store(addr, v, CMM_RELAXED)
> +
> +#define uatomic_load(addr, mo)						\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _value = __atomic_load_n(addr,	\
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +									\
> +		_value;							\
> +	})
> +
> +#define uatomic_read(addr)			\
> +	uatomic_load(addr, CMM_RELAXED)
> +
> +#define uatomic_cmpxchg_mo(addr, old, new, mos, mof)			\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _old = (__typeof__(*(addr)))old;	\
> +									\
> +		if (__atomic_compare_exchange_n(addr, &_old, new, 0,	\
> +							cmm_to_c11(mos), \
> +							cmm_to_c11(mof))) { \
> +			cmm_seq_cst_fence_after_atomic(mos);		\
> +		} else {						\
> +			cmm_seq_cst_fence_after_atomic(mof);		\
> +		}							\
> +		_old;							\
> +	})
> +
> +#define uatomic_cmpxchg(addr, old, new)					\
> +	uatomic_cmpxchg_mo(addr, old, new, CMM_SEQ_CST_FENCE, CMM_RELAXED)
> +
> +#define uatomic_xchg_mo(addr, v, mo)					\
> +	__extension__							\
> +	({								\
> +		__typeof__((*addr)) _old = __atomic_exchange_n(addr, v,	\
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +		_old;							\
> +	})
> +
> +#define uatomic_xchg(addr, v)						\
> +	uatomic_xchg_mo(addr, v, CMM_SEQ_CST_FENCE)
> +
> +#define uatomic_add_return_mo(addr, v, mo)				\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _old = __atomic_add_fetch(addr, v,	\
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +		_old;							\
> +	})
> +
> +#define uatomic_add_return(addr, v)					\
> +	uatomic_add_return_mo(addr, v, CMM_SEQ_CST_FENCE)
> +
> +#define uatomic_sub_return_mo(addr, v, mo)				\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _old = __atomic_sub_fetch(addr, v,	\
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +		_old;							\
> +	})
> +
> +#define uatomic_sub_return(addr, v)					\
> +	uatomic_sub_return_mo(addr, v, CMM_SEQ_CST_FENCE)
> +
> +#define uatomic_and_mo(addr, mask, mo)					\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _old = __atomic_and_fetch(addr, mask, \
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +		_old;							\
> +	})
> +
> +#define uatomic_and(addr, mask)				\
> +	(void) uatomic_and_mo(addr, mask, CMM_SEQ_CST)
> +
> +#define uatomic_or_mo(addr, mask, mo)					\
> +	__extension__							\
> +	({								\
> +		__typeof__(*(addr)) _old = __atomic_or_fetch(addr, mask, \
> +							cmm_to_c11(mo)); \
> +		cmm_seq_cst_fence_after_atomic(mo);			\
> +		_old;							\
> +	})
> +
> +
> +#define uatomic_or(addr, mask)				\
> +	(void) uatomic_or_mo(addr, mask, CMM_RELAXED)
> +
> +#define uatomic_add_mo(addr, v, mo)			\
> +	(void) uatomic_add_return_mo(addr, v, mo)
> +
> +#define uatomic_add(addr, v)				\
> +	(void) uatomic_add_mo(addr, v, CMM_RELAXED)
> +
> +#define uatomic_sub_mo(addr, v, mo)			\
> +	(void) uatomic_sub_return_mo(addr, v, mo)
> +
> +#define uatomic_sub(addr, v)				\
> +	(void) uatomic_sub_mo(addr, v, CMM_RELAXED)
> +
> +#define uatomic_inc_mo(addr, mo)		\
> +	(void) uatomic_add_mo(addr, 1, mo)
> +
> +#define uatomic_inc(addr)				\
> +	(void) uatomic_inc_mo(addr, CMM_RELAXED)
> +
> +#define uatomic_dec_mo(addr, mo)		\
> +	(void) uatomic_sub_mo(addr, 1, mo)
> +
> +#define uatomic_dec(addr)				\
> +	(void) uatomic_dec_mo(addr, CMM_RELAXED)
> +
> +#define cmm_smp_mb__before_uatomic_and() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_and()  cmm_smp_mb()
> +
> +#define cmm_smp_mb__before_uatomic_or() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_or()  cmm_smp_mb()
> +
> +#define cmm_smp_mb__before_uatomic_add() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_add()  cmm_smp_mb()
> +
> +#define cmm_smp_mb__before_uatomic_sub() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_sub()  cmm_smp_mb()
> +
> +#define cmm_smp_mb__before_uatomic_inc() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_inc() cmm_smp_mb()
> +
> +#define cmm_smp_mb__before_uatomic_dec() cmm_smp_mb()
> +#define cmm_smp_mb__after_uatomic_dec() cmm_smp_mb()
> +
> +#endif /* _URCU_UATOMIC_BUILTINS_X86_H */
> diff --git a/include/urcu/uatomic/builtins.h b/include/urcu/uatomic/builtins.h
> new file mode 100644
> index 0000000..82e98f8
> --- /dev/null
> +++ b/include/urcu/uatomic/builtins.h
> @@ -0,0 +1,79 @@
> +/*
> + * urcu/uatomic/builtins.h
> + *
> + * Copyright (c) 2023 Olivier Dion <odion at efficios.com>
> + *
> + * This library is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * This library is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with this library; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +#ifndef _URCU_UATOMIC_BUILTINS_H
> +#define _URCU_UATOMIC_BUILTINS_H
> +
> +#include <urcu/arch.h>
> +
> +#if defined(__has_builtin)
> +# if !__has_builtin(__atomic_store_n)
> +#  error "Toolchain does not support __atomic_store_n."
> +# endif
> +# if !__has_builtin(__atomic_load_n)
> +#  error "Toolchain does not support __atomic_load_n."
> +# endif
> +# if !__has_builtin(__atomic_exchange_n)
> +#  error "Toolchain does not support __atomic_exchange_n."
> +# endif
> +# if !__has_builtin(__atomic_compare_exchange_n)
> +#  error "Toolchain does not support __atomic_compare_exchange_n."
> +# endif
> +# if !__has_builtin(__atomic_add_fetch)
> +#  error "Toolchain does not support __atomic_add_fetch."
> +# endif
> +# if !__has_builtin(__atomic_sub_fetch)
> +#  error "Toolchain does not support __atomic_sub_fetch."
> +# endif
> +# if !__has_builtin(__atomic_or_fetch)
> +#  error "Toolchain does not support __atomic_or_fetch."
> +# endif
> +# if !__has_builtin(__atomic_thread_fence)
> +#  error "Toolchain does not support __atomic_thread_fence."
> +# endif
> +# if !__has_builtin(__atomic_signal_fence)
> +#  error "Toolchain does not support __atomic_signal_fence."
> +# endif
> +#elif defined(__GNUC__)
> +# define GCC_VERSION (__GNUC__       * 10000 + \
> +		       __GNUC_MINOR__ * 100   + \
> +		       __GNUC_PATCHLEVEL__)
> +# if  GCC_VERSION < 40700
> +#  error "GCC version is too old. Version must be 4.7 or greater"
> +# endif
> +# undef  GCC_VERSION
> +#else
> +# error "Toolchain is not supported."
> +#endif
> +
> +#if defined(__GNUC__)
> +# define UATOMIC_HAS_ATOMIC_BYTE  __GCC_ATOMIC_CHAR_LOCK_FREE
> +# define UATOMIC_HAS_ATOMIC_SHORT __GCC_ATOMIC_SHORT_LOCK_FREE
> +#elif defined(__clang__)
> +# define UATOMIC_HAS_ATOMIC_BYTE  __CLANG_ATOMIC_CHAR_LOCK_FREE
> +# define UATOMIC_HAS_ATOMIC_SHORT __CLANG_ATOMIC_SHORT_LOCK_FREE
> +#else
> +/* #  define UATOMIC_HAS_ATOMIC_BYTE  */
> +/* #  define UATOMIC_HAS_ATOMIC_SHORT */
> +#endif
> +
> +#include <urcu/uatomic/builtins-generic.h>
> +
> +#endif	/* _URCU_UATOMIC_BUILTINS_H */
> diff --git a/include/urcu/uatomic/generic.h b/include/urcu/uatomic/generic.h
> index e31a19b..6b9c153 100644
> --- a/include/urcu/uatomic/generic.h
> +++ b/include/urcu/uatomic/generic.h
> @@ -33,10 +33,244 @@ extern "C" {
>  #define uatomic_set(addr, v)	((void) CMM_STORE_SHARED(*(addr), (v)))
>  #endif
>  
> +extern void abort(void);
> +
> +#define uatomic_load_store_return_op(op, addr, v, mo)			\
> +	__extension__							\
> +	({								\
> +									\
> +		switch (mo) {						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_CONSUME:					\
> +		case CMM_RELAXED:					\
> +			break;						\
> +		case CMM_RELEASE:					\
> +		case CMM_ACQ_REL:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +									\
> +		__typeof__((*addr)) _value = op(addr, v);		\
> +									\
> +		switch (mo) {						\
> +		case CMM_CONSUME:					\
> +			cmm_smp_read_barrier_depends();			\
> +			break;						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_ACQ_REL:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		case CMM_RELAXED:					\
> +		case CMM_RELEASE:					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +		_value;							\
> +	})
> +
> +#define uatomic_load_store_op(op, addr, v, mo)				\
> +	({								\
> +		switch (mo) {						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_CONSUME:					\
> +		case CMM_RELAXED:					\
> +			break;						\
> +		case CMM_RELEASE:					\
> +		case CMM_ACQ_REL:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +									\
> +		op(addr, v);						\
> +									\
> +		switch (mo) {						\
> +		case CMM_CONSUME:					\
> +			cmm_smp_read_barrier_depends();			\
> +			break;						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_ACQ_REL:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		case CMM_RELAXED:					\
> +		case CMM_RELEASE:					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +	})
> +
> +#define uatomic_store(addr, v, mo)			\
> +	({						\
> +		switch (mo) {				\
> +		case CMM_RELAXED:			\
> +			break;				\
> +		case CMM_RELEASE:			\
> +		case CMM_SEQ_CST:			\
> +		case CMM_SEQ_CST_FENCE:			\
> +			cmm_smp_mb();			\
> +			break;				\
> +		default:				\
> +			abort();			\
> +		}					\
> +							\
> +		uatomic_set(addr, v);			\
> +							\
> +		switch (mo) {				\
> +		case CMM_RELAXED:			\
> +		case CMM_RELEASE:			\
> +			break;				\
> +		case CMM_SEQ_CST:			\
> +		case CMM_SEQ_CST_FENCE:			\
> +			cmm_smp_mb();			\
> +			break;				\
> +		default:				\
> +			abort();			\
> +		}					\
> +	})
> +
> +#define uatomic_and_mo(addr, v, mo)				\
> +	uatomic_load_store_op(uatomic_and, addr, v, mo)
> +
> +#define uatomic_or_mo(addr, v, mo)				\
> +	uatomic_load_store_op(uatomic_or, addr, v, mo)
> +
> +#define uatomic_add_mo(addr, v, mo)				\
> +	uatomic_load_store_op(uatomic_add, addr, v, mo)
> +
> +#define uatomic_sub_mo(addr, v, mo)				\
> +	uatomic_load_store_op(uatomic_sub, addr, v, mo)
> +
> +#define uatomic_inc_mo(addr, mo)				\
> +	uatomic_load_store_op(uatomic_add, addr, 1, mo)
> +
> +#define uatomic_dec_mo(addr, mo)				\
> +	uatomic_load_store_op(uatomic_add, addr, -1, mo)
> +/*
> + * NOTE: We can not just do switch (_value == (old) ? mos : mof) otherwise the
> + * compiler emit a -Wduplicated-cond warning.
> + */
> +#define uatomic_cmpxchg_mo(addr, old, new, mos, mof)			\
> +	__extension__							\
> +	({								\
> +		switch (mos) {						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_CONSUME:					\
> +		case CMM_RELAXED:					\
> +			break;						\
> +		case CMM_RELEASE:					\
> +		case CMM_ACQ_REL:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +									\
> +		__typeof__(*(addr)) _value = uatomic_cmpxchg(addr, old,	\
> +							new);		\
> +									\
> +		if (_value == (old)) {					\
> +			switch (mos) {					\
> +			case CMM_CONSUME:				\
> +				cmm_smp_read_barrier_depends();		\
> +				break;					\
> +			case CMM_ACQUIRE:				\
> +			case CMM_ACQ_REL:				\
> +			case CMM_SEQ_CST:				\
> +			case CMM_SEQ_CST_FENCE:				\
> +				cmm_smp_mb();				\
> +				break;					\
> +			case CMM_RELAXED:				\
> +			case CMM_RELEASE:				\
> +				break;					\
> +			default:					\
> +				abort();				\
> +			}						\
> +		} else {						\
> +			switch (mof) {					\
> +			case CMM_CONSUME:				\
> +				cmm_smp_read_barrier_depends();		\
> +				break;					\
> +			case CMM_ACQUIRE:				\
> +			case CMM_ACQ_REL:				\
> +			case CMM_SEQ_CST:				\
> +			case CMM_SEQ_CST_FENCE:				\
> +				cmm_smp_mb();				\
> +				break;					\
> +			case CMM_RELAXED:				\
> +			case CMM_RELEASE:				\
> +				break;					\
> +			default:					\
> +				abort();				\
> +			}						\
> +		}							\
> +		_value;							\
> +	})
> +
> +#define uatomic_xchg_mo(addr, v, mo)				\
> +	uatomic_load_store_return_op(uatomic_xchg, addr, v, mo)
> +
> +#define uatomic_add_return_mo(addr, v, mo)				\
> +	uatomic_load_store_return_op(uatomic_add_return, addr, v)
> +
> +#define uatomic_sub_return_mo(addr, v, mo)				\
> +	uatomic_load_store_return_op(uatomic_sub_return, addr, v)
> +
> +
>  #ifndef uatomic_read
>  #define uatomic_read(addr)	CMM_LOAD_SHARED(*(addr))
>  #endif
>  
> +#define uatomic_load(addr, mo)						\
> +	__extension__							\
> +	({								\
> +		switch (mo) {						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_CONSUME:					\
> +		case CMM_RELAXED:					\
> +			break;						\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +									\
> +		__typeof__(*(addr)) _rcu_value = uatomic_read(addr);	\
> +									\
> +		switch (mo) {						\
> +		case CMM_RELAXED:					\
> +			break;						\
> +		case CMM_CONSUME:					\
> +			cmm_smp_read_barrier_depends();			\
> +			break;						\
> +		case CMM_ACQUIRE:					\
> +		case CMM_SEQ_CST:					\
> +		case CMM_SEQ_CST_FENCE:					\
> +			cmm_smp_mb();					\
> +			break;						\
> +		default:						\
> +			abort();					\
> +		}							\
> +									\
> +		_rcu_value;						\
> +	})
> +
>  #if !defined __OPTIMIZE__  || defined UATOMIC_NO_LINK_ERROR
>  #ifdef ILLEGAL_INSTR
>  static inline __attribute__((always_inline))
> diff --git a/src/urcu-pointer.c b/src/urcu-pointer.c
> index d0854ac..cea8aeb 100644
> --- a/src/urcu-pointer.c
> +++ b/src/urcu-pointer.c
> @@ -39,19 +39,16 @@ void *rcu_dereference_sym(void *p)
>  
>  void *rcu_set_pointer_sym(void **p, void *v)
>  {
> -	cmm_wmb();
> -	uatomic_set(p, v);
> +	uatomic_store(p, v, CMM_RELEASE);
>  	return v;
>  }
>  
>  void *rcu_xchg_pointer_sym(void **p, void *v)
>  {
> -	cmm_wmb();
> -	return uatomic_xchg(p, v);
> +	return uatomic_xchg_mo(p, v, CMM_SEQ_CST);
>  }
>  
>  void *rcu_cmpxchg_pointer_sym(void **p, void *old, void *_new)
>  {
> -	cmm_wmb();
> -	return uatomic_cmpxchg(p, old, _new);
> +	return uatomic_cmpxchg_mo(p, old, _new, CMM_SEQ_CST, CMM_SEQ_CST);
>  }
> -- 
> 2.40.1
> 


More information about the lttng-dev mailing list