diff mbox series

Use atomic_thread_fence

Message ID AM5PR0801MB16682C3A037C622C8D3C4CDE83969@AM5PR0801MB1668.eurprd08.prod.outlook.com
State New
Headers show
Series Use atomic_thread_fence | expand

Commit Message

Wilco Dijkstra July 28, 2022, 2:27 p.m. UTC 
Replace atomic barriers based on sync primitives with atomic_thread_fence.
Many uses appear suspect and in the future fixing these to use load_acquire
or store_release would be useful.

Passes regress on AArch64 and buildmanyglibc.

---

Comments

Carlos O'Donell Aug. 1, 2022, 1:24 p.m. UTC | #1 
On 7/28/22 10:27, Wilco Dijkstra via Libc-alpha wrote:
> Replace atomic barriers based on sync primitives with atomic_thread_fence.
> Many uses appear suspect and in the future fixing these to use load_acquire
> or store_release would be useful.
> 
> Passes regress on AArch64 and buildmanyglibc.

This came up in review today in the weekly meeting.

The pre-commit CI shows this doesn't apply:
https://patchwork.sourceware.org/project/glibc/patch/AM5PR0801MB16682C3A037C622C8D3C4CDE83969@AM5PR0801MB1668.eurprd08.prod.outlook.com/

How are you generating these patches?
 
> ---
> 
> diff --git a/crypt/crypt_util.c b/crypt/crypt_util.c
> index be925e3484e65d2180e07915f5d91b47f6b96393..a8c2b26ed13c27804a9465f2a54caa09aaec3814 100644
> --- a/crypt/crypt_util.c
> +++ b/crypt/crypt_util.c
> @@ -453,14 +453,14 @@ __init_des_r (struct crypt_data * __restrict __data)
>  	  efp[comes_from_word][word_value][o_long] |= mask2;
>        }
>      }
> -    atomic_write_barrier ();
> +    atomic_thread_fence_release ();
>      small_tables_initialized = 1;
>  #ifdef __GNU_LIBRARY__
>  small_tables_done:
>      __libc_lock_unlock(_ufc_tables_lock);
>  #endif
>    } else
> -    atomic_read_barrier ();
> +    atomic_thread_fence_acquire ();
>  
>    /*
>     * Create the sb tables:
> diff --git a/elf/dl-deps.c b/elf/dl-deps.c
> index 06005a0cc8686cc7e63cd8e1b1e7deda01fe6688..11b3fda5fdeb3830d3d5a0031084b43847444e04 100644
> --- a/elf/dl-deps.c
> +++ b/elf/dl-deps.c
> @@ -430,7 +430,7 @@ _dl_map_object_deps (struct link_map *map,
>  	  memcpy (&l_initfini[1], needed, nneeded * sizeof needed[0]);
>  	  memcpy (&l_initfini[nneeded + 1], l_initfini,
>  		  nneeded * sizeof needed[0]);
> -	  atomic_write_barrier ();
> +	  atomic_thread_fence_release ();
>  	  l->l_initfini = l_initfini;
>  	  l->l_free_initfini = 1;
>  	}
> @@ -555,12 +555,12 @@ _dl_map_object_deps (struct link_map *map,
>  
>    /* Terminate the list of dependencies.  */
>    l_initfini[nlist] = NULL;
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>    map->l_initfini = l_initfini;
>    map->l_free_initfini = 1;
>    if (l_reldeps != NULL)
>      {
> -      atomic_write_barrier ();
> +      atomic_thread_fence_release ();
>        void *old_l_reldeps = map->l_reldeps;
>        map->l_reldeps = l_reldeps;
>        _dl_scope_free (old_l_reldeps);
> diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
> index 02c63a7062b2be0f37a412160fdb2b3468cc70cf..894d3e7db198a2a08940e9a1d82c345b0e0343a0 100644
> --- a/elf/dl-lookup.c
> +++ b/elf/dl-lookup.c
> @@ -695,7 +695,7 @@ marking %s [%lu] as NODELETE due to memory allocation failure\n",
>  			l_reldepsact * sizeof (struct link_map *));
>  	      newp->list[l_reldepsact] = map;
>  	      newp->act = l_reldepsact + 1;
> -	      atomic_write_barrier ();
> +	      atomic_thread_fence_release ();
>  	      void *old = undef_map->l_reldeps;
>  	      undef_map->l_reldeps = newp;
>  	      undef_map->l_reldepsmax = max;
> @@ -706,7 +706,7 @@ marking %s [%lu] as NODELETE due to memory allocation failure\n",
>        else
>  	{
>  	  undef_map->l_reldeps->list[l_reldepsact] = map;
> -	  atomic_write_barrier ();
> +	  atomic_thread_fence_release ();
>  	  undef_map->l_reldeps->act = l_reldepsact + 1;
>  	}
>  
> diff --git a/elf/dl-open.c b/elf/dl-open.c
> index a23e65926bcfe797f06f8b4175f65040f4547a05..ba77f4b774cae69c382bceb6599936664203ef05 100644
> --- a/elf/dl-open.c
> +++ b/elf/dl-open.c
> @@ -202,7 +202,7 @@ add_to_global_update (struct link_map *new)
>    assert (added <= ns->_ns_global_scope_pending_adds);
>    ns->_ns_global_scope_pending_adds -= added;
>  
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>    ns->_ns_main_searchlist->r_nlist = new_nlist;
>  }
>  
> @@ -342,7 +342,7 @@ update_scopes (struct link_map *new)
>  	     might use the new last element and then use the garbage
>  	     at offset IDX+1.  */
>  	  imap->l_scope[cnt + 1] = NULL;
> -	  atomic_write_barrier ();
> +	  atomic_thread_fence_release ();
>  	  imap->l_scope[cnt] = &new->l_searchlist;
>  
>  	  from_scope = cnt;
> diff --git a/include/atomic.h b/include/atomic.h
> index 8eb56362ba18eb4836070930d5f2e769fb6a0a1e..dfe60ddb27fed1a06fb4967968e7d5d64de2f9c3 100644
> --- a/include/atomic.h
> +++ b/include/atomic.h
> @@ -104,21 +104,6 @@
>  #endif
>  
>  
> -#ifndef atomic_full_barrier
> -# define atomic_full_barrier() __sync_synchronize()
> -#endif
> -
> -
> -#ifndef atomic_read_barrier
> -# define atomic_read_barrier() atomic_full_barrier ()
> -#endif
> -
> -
> -#ifndef atomic_write_barrier
> -# define atomic_write_barrier() atomic_full_barrier ()
> -#endif
> -
> -
>  /* This is equal to 1 iff the architecture supports 64b atomic operations.  */
>  #ifndef __HAVE_64B_ATOMICS
>  #error Unable to determine if 64-bit atomics are present.
> diff --git a/include/list.h b/include/list.h
> index 7bea2c50a3759c0c8640971eff1e80874e3b543f..31a8a93fa2491b38c368b9e07dd65c473e4eb19f 100644
> --- a/include/list.h
> +++ b/include/list.h
> @@ -43,7 +43,7 @@ list_add (list_t *newp, list_t *head)
>    newp->next = head->next;
>    newp->prev = head;
>    head->next->prev = newp;
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>    head->next = newp;
>  }
>  
> diff --git a/malloc/arena.c b/malloc/arena.c
> index 7c74a18381a4be5fe6bcb94b38a62dbfa6b674f4..3ef52ade3aeaaba6bb5e5d4e70024202e0dc162b 100644
> --- a/malloc/arena.c
> +++ b/malloc/arena.c
> @@ -111,7 +111,7 @@ static mstate free_list;
>     malloc_state objects.
>  
>     Read access to the next member is supposed to synchronize with the
> -   atomic_write_barrier and the write to the next member in
> +   atomic_thread_fence_release and the write to the next member in
>     _int_new_arena.  This suffers from data races; see the FIXME
>     comments in _int_new_arena and reused_arena.
>  
> @@ -778,7 +778,7 @@ _int_new_arena (size_t size)
>    /* FIXME: The barrier is an attempt to synchronize with read access
>       in reused_arena, which does not acquire list_lock while
>       traversing the list.  */
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>    main_arena.next = a;
>  
>    __libc_lock_unlock (list_lock);
> diff --git a/manual/llio.texi b/manual/llio.texi
> index 1b801ee817db2935d8866894be23ffa516690ca3..eb8711a0f8e28281baf68206b7891f16f662de0b 100644
> --- a/manual/llio.texi
> +++ b/manual/llio.texi
> @@ -2543,14 +2543,14 @@ aiocb64}, since the LFS transparently replaces the old interface.
>  @c    deallocate_stack @asulock @ascuheap @aculock @acsmem
>  @c     lll_lock (state_cache_lock) @asulock @aculock
>  @c     stack_list_del ok
> -@c      atomic_write_barrier ok
> +@c      atomic_thread_fence_release ok
>  @c      list_del ok
> -@c      atomic_write_barrier ok
> +@c      atomic_thread_fence_release ok
>  @c     queue_stack @ascuheap @acsmem
>  @c      stack_list_add ok
> -@c       atomic_write_barrier ok
> +@c       atomic_thread_fence_release ok
>  @c       list_add ok
> -@c       atomic_write_barrier ok
> +@c       atomic_thread_fence_release ok
>  @c      free_stacks @ascuheap @acsmem
>  @c       list_for_each_prev_safe ok
>  @c       list_entry ok
> diff --git a/manual/memory.texi b/manual/memory.texi
> index 110e736a64c667988f4ca2fe92deb409225a4a88..2dfd09ea4aace004067e2e1d51c9e1292d1f7452 100644
> --- a/manual/memory.texi
> +++ b/manual/memory.texi
> @@ -395,7 +395,7 @@ this function is in @file{stdlib.h}.
>  @c     mutex_init ok
>  @c     mutex_lock (just-created mutex) ok, returns locked
>  @c     mutex_lock (list_lock) dup @asulock @aculock
> -@c     atomic_write_barrier ok
> +@c     atomic_thread_fence_release ok
>  @c     mutex_unlock (list_lock) @aculock
>  @c    atomic_fetch_add_relaxed ok
>  @c    reused_arena @asulock @aculock
> diff --git a/manual/startup.texi b/manual/startup.texi
> index 9bf24123f562f75ba27a4770c69147e003b94755..4c7c2976a92d772909203bca028347e8d798b2d7 100644
> --- a/manual/startup.texi
> +++ b/manual/startup.texi
> @@ -947,7 +947,7 @@ using @code{atexit} or @code{on_exit}.
>  @c     __libc_lock_lock @asulock @aculock
>  @c     calloc dup @ascuheap @acsmem
>  @c     __libc_lock_unlock @aculock
> -@c    atomic_write_barrier dup ok
> +@c    atomic_thread_fence_release dup ok
>  The @code{atexit} function registers the function @var{function} to be
>  called at normal program termination.  The @var{function} is called with
>  no arguments.
> @@ -961,7 +961,7 @@ the function cannot be registered.
>  @safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{} @asulock{}}@acunsafe{@aculock{} @acsmem{}}}
>  @c on_exit @ascuheap @asulock @aculock @acsmem
>  @c  new_exitfn dup @ascuheap @asulock @aculock @acsmem
> -@c  atomic_write_barrier dup ok
> +@c  atomic_thread_fence_release dup ok
>  This function is a somewhat more powerful variant of @code{atexit}.  It
>  accepts two arguments, a function @var{function} and an arbitrary
>  pointer @var{arg}.  At normal program termination, the @var{function} is
> diff --git a/nptl/nptl-stack.c b/nptl/nptl-stack.c
> index 20ce78eddbf100833d453d7032f63bc2ba8f01c7..7c04e7faaae5c15bf5ad98c32935bcba4849c1c1 100644
> --- a/nptl/nptl-stack.c
> +++ b/nptl/nptl-stack.c
> @@ -27,11 +27,11 @@ __nptl_stack_list_del (list_t *elem)
>  {
>    GL (dl_in_flight_stack) = (uintptr_t) elem;
>  
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>  
>    list_del (elem);
>  
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>  
>    GL (dl_in_flight_stack) = 0;
>  }
> @@ -42,11 +42,11 @@ __nptl_stack_list_add (list_t *elem, list_t *list)
>  {
>    GL (dl_in_flight_stack) = (uintptr_t) elem | 1;
>  
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>  
>    list_add (elem, list);
>  
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>  
>    GL (dl_in_flight_stack) = 0;
>  }
> diff --git a/nptl/pthread_mutex_setprioceiling.c b/nptl/pthread_mutex_setprioceiling.c
> index 2d71a750c8981e8ca271c265031887e2c510583a..b574a77250664bbb1487c932b9a32a9dee415072 100644
> --- a/nptl/pthread_mutex_setprioceiling.c
> +++ b/nptl/pthread_mutex_setprioceiling.c
> @@ -113,7 +113,7 @@ __pthread_mutex_setprioceiling (pthread_mutex_t *mutex, int prioceiling,
>      newlock = (mutex->__data.__lock & ~PTHREAD_MUTEX_PRIO_CEILING_MASK);
>    mutex->__data.__lock = newlock
>  			 | (prioceiling << PTHREAD_MUTEX_PRIO_CEILING_SHIFT);
> -  atomic_full_barrier ();
> +  atomic_thread_fence_seq_cst ();
>  
>    futex_wake ((unsigned int *)&mutex->__data.__lock, INT_MAX,
>  	      PTHREAD_MUTEX_PSHARED (mutex));
> diff --git a/nptl/sem_post.c b/nptl/sem_post.c
> index 7ec21e92eb4c71d7f17764e96bc7603837f7522d..d4e37cb0888cb1004881e608b82147900bc420a5 100644
> --- a/nptl/sem_post.c
> +++ b/nptl/sem_post.c
> @@ -90,7 +90,7 @@ __old_sem_post (sem_t *sem)
>  
>    /* We must need to synchronize with consumers of this token, so the atomic
>       increment must have release MO semantics.  */
> -  atomic_write_barrier ();
> +  atomic_thread_fence_release ();
>    atomic_fetch_add_release (futex, 1);
>    /* We always have to assume it is a shared semaphore.  */
>    futex_wake (futex, 1, LLL_SHARED);
> diff --git a/stdlib/msort.c b/stdlib/msort.c
> index cbe9a4a8fdb38113a4c18976c9f297be103d458f..e2f1eca94ad7e9005145c376b0de3dbd1ca14f18 100644
> --- a/stdlib/msort.c
> +++ b/stdlib/msort.c
> @@ -197,7 +197,7 @@ __qsort_r (void *b, size_t n, size_t s, __compar_d_fn_t cmp, void *arg)
>  	  phys_pages /= 4;
>  
>  	  /* Make sure phys_pages is written to memory.  */
> -	  atomic_write_barrier ();
> +	  atomic_thread_fence_release ();
>  
>  	  pagesize = __sysconf (_SC_PAGESIZE);
>  	}
> diff --git a/sysdeps/aarch64/nptl/tls.h b/sysdeps/aarch64/nptl/tls.h
> index 08aa2eff891b7be32243e9955d998892807c7b2e..0e5b4ece6a118b4b066bd2fd024dc85e978cc786 100644
> --- a/sysdeps/aarch64/nptl/tls.h
> +++ b/sysdeps/aarch64/nptl/tls.h
> @@ -108,7 +108,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/alpha/atomic-machine.h b/sysdeps/alpha/atomic-machine.h
> index f384a2bf0b3376cf240dc25d501e1d64a94bffe1..7fbe5b87eebf323d38fe1349b02aa56fe199cab3 100644
> --- a/sysdeps/alpha/atomic-machine.h
> +++ b/sysdeps/alpha/atomic-machine.h
> @@ -21,7 +21,3 @@
>  
>  /* XXX Is this actually correct?  */
>  #define ATOMIC_EXCHANGE_USES_CAS 1
> -
> -#define atomic_full_barrier()	__asm ("mb" : : : "memory");
> -#define atomic_read_barrier()	__asm ("mb" : : : "memory");
> -#define atomic_write_barrier()	__asm ("wmb" : : : "memory");
> diff --git a/sysdeps/alpha/nptl/tls.h b/sysdeps/alpha/nptl/tls.h
> index 8f5b69ad3b1b0c557fa1bae55278547572a374cc..914dba422c50e4531d22eb459b41c8b958a75263 100644
> --- a/sysdeps/alpha/nptl/tls.h
> +++ b/sysdeps/alpha/nptl/tls.h
> @@ -105,7 +105,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/arc/nptl/tls.h b/sysdeps/arc/nptl/tls.h
> index 7fc6602b236fa2455f8de4a0540442ae85d27c98..b2749f81d2980502043f507bf7c81da48f17aa9f 100644
> --- a/sysdeps/arc/nptl/tls.h
> +++ b/sysdeps/arc/nptl/tls.h
> @@ -113,7 +113,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/arm/nptl/tls.h b/sysdeps/arm/nptl/tls.h
> index 7657ca3dccc2d929c71236d42fc060a4b4902e2b..b1389ba034966aff17692f2b6d0e7b04a0baf9a0 100644
> --- a/sysdeps/arm/nptl/tls.h
> +++ b/sysdeps/arm/nptl/tls.h
> @@ -99,7 +99,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/csky/nptl/tls.h b/sysdeps/csky/nptl/tls.h
> index 58d6ab0fb2ae90de50cffd5b4a98426c6a793050..ac54606c3c0e28c1c8d57a6475d9249ca3566abe 100644
> --- a/sysdeps/csky/nptl/tls.h
> +++ b/sysdeps/csky/nptl/tls.h
> @@ -128,7 +128,7 @@ typedef struct
>    do									      \
>      {									      \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	      \
> -      atomic_write_barrier ();						      \
> +      atomic_thread_fence_release ();					      \
>      }									      \
>    while (0)
>  
> diff --git a/sysdeps/hppa/dl-fptr.c b/sysdeps/hppa/dl-fptr.c
> index 40bf5cd3b306315d8eeb6bdba2b2b46b1ea5059e..0562467d6f79f76b78b2cf169fdd059a993296d3 100644
> --- a/sysdeps/hppa/dl-fptr.c
> +++ b/sysdeps/hppa/dl-fptr.c
> @@ -369,7 +369,7 @@ _dl_lookup_address (const void *address)
>  
>    /* First load the relocation offset.  */
>    reloc_arg = (ElfW(Word)) desc[1];
> -  atomic_full_barrier();
> +  atomic_thread_fence_seq_cst ();
>  
>    /* Then load first word of candidate descriptor.  It should be a pointer
>       with word alignment and point to memory that can be read.  */
> diff --git a/sysdeps/hppa/dl-machine.h b/sysdeps/hppa/dl-machine.h
> index c865713be1e3f8e0430bbb35c8db7ebe3e7a6abf..61635ca9115e1fa77305eaa3cc4ab5bf9bb91d7b 100644
> --- a/sysdeps/hppa/dl-machine.h
> +++ b/sysdeps/hppa/dl-machine.h
> @@ -136,7 +136,7 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t t,
>        /* Need to ensure that the gp is visible before the code
>           entry point is updated */
>        rfdesc[1] = value.gp;
> -      atomic_full_barrier();
> +      atomic_thread_fence_seq_cst ();
>        rfdesc[0] = value.ip;
>      }
>    else
> diff --git a/sysdeps/hppa/nptl/tls.h b/sysdeps/hppa/nptl/tls.h
> index e6b0bd5c7182b497aaf0d2bb08f62551a223c403..5b2495637b8aec9df3c8b3d47a2b46aa632c84da 100644
> --- a/sysdeps/hppa/nptl/tls.h
> +++ b/sysdeps/hppa/nptl/tls.h
> @@ -133,7 +133,7 @@ static inline void __set_cr27(struct pthread *cr27)
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/htl/pt-once.c b/sysdeps/htl/pt-once.c
> index b85b196645958fc7f47b08b39e91077b82817cdc..55db6c3d7176a9ca6cf2d0caccc1cf01aeb5ea2f 100644
> --- a/sysdeps/htl/pt-once.c
> +++ b/sysdeps/htl/pt-once.c
> @@ -33,7 +33,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
>  {
>    ASSERT_TYPE_SIZE (pthread_once_t, __SIZEOF_PTHREAD_ONCE_T);
>  
> -  atomic_full_barrier ();
> +  atomic_thread_fence_seq_cst ();
>    if (once_control->__run == 0)
>      {
>        __pthread_spin_wait (&once_control->__lock);
> @@ -44,7 +44,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
>  	  init_routine ();
>  	  pthread_cleanup_pop (0);
>  
> -	  atomic_full_barrier ();
> +	  atomic_thread_fence_seq_cst ();
>  	  once_control->__run = 1;
>  	}
>  
> diff --git a/sysdeps/ia64/nptl/tls.h b/sysdeps/ia64/nptl/tls.h
> index d2411b3c1ac29733c0bb3683d83388e2e0e8e277..7709e644ee04ebc935dc659806481eebcb4129f0 100644
> --- a/sysdeps/ia64/nptl/tls.h
> +++ b/sysdeps/ia64/nptl/tls.h
> @@ -157,7 +157,7 @@ register struct pthread *__thread_self __asm__("r13");
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/m68k/nptl/tls.h b/sysdeps/m68k/nptl/tls.h
> index 742e1b6767d99fa6011ac1d207264c7b82e53787..dfba7a568016b8e10dac6c21d65c785eaab12a09 100644
> --- a/sysdeps/m68k/nptl/tls.h
> +++ b/sysdeps/m68k/nptl/tls.h
> @@ -132,7 +132,7 @@ extern void * __m68k_read_tp (void);
>    do									\
>      {									\
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
> -      atomic_write_barrier ();						\
> +      atomic_thread_fence_release ();					\
>      }									\
>    while (0)
>  
> diff --git a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
> index 71f789cbdfa20b79f9e3cef5a2523d4e243b4f19..feb9085af9499b5d954a1bf4f04d9cbcb804e8cf 100644
> --- a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
> +++ b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
> @@ -26,7 +26,7 @@
>  int
>  __pthread_mutex_destroy (pthread_mutex_t *mtxp)
>  {
> -  atomic_read_barrier ();
> +  atomic_thread_fence_acquire ();
>    if (*(volatile unsigned int *) &mtxp->__lock != 0)
>      return EBUSY;
>  
> diff --git a/sysdeps/mach/hurd/htl/pt-mutex.h b/sysdeps/mach/hurd/htl/pt-mutex.h
> index 4021e72a6e8d15316336296ff732a4e7fd1acdff..ebdf8a5fbde0f755c7625cd38ad185ee0c977b5f 100644
> --- a/sysdeps/mach/hurd/htl/pt-mutex.h
> +++ b/sysdeps/mach/hurd/htl/pt-mutex.h
> @@ -54,7 +54,7 @@
>            if (ret == EOWNERDEAD)   \
>              {   \
>                mtxp->__lock = mtxp->__lock | LLL_DEAD_OWNER;   \
> -              atomic_write_barrier ();   \
> +              atomic_thread_fence_release ();   \
>              }   \
>          }   \
>      }   \
> diff --git a/sysdeps/microblaze/nptl/tls.h b/sysdeps/microblaze/nptl/tls.h
> index 588fd1c5d63ee4e6a1b284cc19e216b6730a2091..30e5d628be8b78cf9c7b8e9386ab1b2355819f4a 100644
> --- a/sysdeps/microblaze/nptl/tls.h
> +++ b/sysdeps/microblaze/nptl/tls.h
> @@ -110,7 +110,7 @@ typedef struct
>    do                                                                        \
>      {                                                                       \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;            \
> -      atomic_write_barrier ();                                              \
> +      atomic_thread_fence_release ();                                       \
>      }                                                                       \
>    while (0)
>  
> diff --git a/sysdeps/mips/nptl/tls.h b/sysdeps/mips/nptl/tls.h
> index 2aa7cb4bb8d0b5a31889aa33d5751104ff1e4f45..e4c5d2a876db7943b38daa270f4681d17b441c58 100644
> --- a/sysdeps/mips/nptl/tls.h
> +++ b/sysdeps/mips/nptl/tls.h
> @@ -160,7 +160,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/nios2/nptl/tls.h b/sysdeps/nios2/nptl/tls.h
> index cb231e2a4bbfa52495c4c017a7d3e1c6dd7937ca..50dbbef119af30112136a9bef187c79037d0849d 100644
> --- a/sysdeps/nios2/nptl/tls.h
> +++ b/sysdeps/nios2/nptl/tls.h
> @@ -140,7 +140,7 @@ register struct pthread *__thread_self __asm__("r23");
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/or1k/nptl/tls.h b/sysdeps/or1k/nptl/tls.h
> index e82f444738de222c0e4866d5a2ab8191ce99ddc9..886c017be12d06ee63198a36917c59e24be77a60 100644
> --- a/sysdeps/or1k/nptl/tls.h
> +++ b/sysdeps/or1k/nptl/tls.h
> @@ -175,7 +175,7 @@ register tcbhead_t *__thread_self __asm__("r10");
>    do									\
>      {									\
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
> -      atomic_write_barrier ();						\
> +      atomic_thread_fence_release ();					\
>      }									\
>    while (0)
>  
> diff --git a/sysdeps/powerpc/nptl/tls.h b/sysdeps/powerpc/nptl/tls.h
> index e62a96238aa95c79ac1f749b4dbf03985b6e15d4..c8d233a7347f609b4cdbffb5dafa2f55e18ac18e 100644
> --- a/sysdeps/powerpc/nptl/tls.h
> +++ b/sysdeps/powerpc/nptl/tls.h
> @@ -224,7 +224,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/powerpc/powerpc32/atomic-machine.h b/sysdeps/powerpc/powerpc32/atomic-machine.h
> index f72d4be13709e38006255d236efb0e94f3976e68..6a2aae8bdb34281144e6810924377a6a62857d15 100644
> --- a/sysdeps/powerpc/powerpc32/atomic-machine.h
> +++ b/sysdeps/powerpc/powerpc32/atomic-machine.h
> @@ -36,24 +36,3 @@
>  
>  #define __HAVE_64B_ATOMICS 0
>  #define ATOMIC_EXCHANGE_USES_CAS 1
> -
> -#ifdef _ARCH_PWR4
> -/*
> - * Newer powerpc64 processors support the new "light weight" sync (lwsync)
> - * So if the build is using -mcpu=[power4,power5,power5+,970] we can
> - * safely use lwsync.
> - */
> -# define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
> -/*
> - * "light weight" sync can also be used for the release barrier.
> - */
> -# define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
> -#else
> -/*
> - * Older powerpc32 processors don't support the new "light weight"
> - * sync (lwsync).  So the only safe option is to use normal sync
> - * for all powerpc32 applications.
> - */
> -# define atomic_read_barrier()	__asm ("sync" ::: "memory")
> -# define atomic_write_barrier()	__asm ("sync" ::: "memory")
> -#endif
> diff --git a/sysdeps/powerpc/powerpc64/atomic-machine.h b/sysdeps/powerpc/powerpc64/atomic-machine.h
> index fcb1592be9ad6a3981f56c513deac2f5f8ac5bb7..2932f889c5bc6d0fa49d5ad36875b50c27ad07e9 100644
> --- a/sysdeps/powerpc/powerpc64/atomic-machine.h
> +++ b/sysdeps/powerpc/powerpc64/atomic-machine.h
> @@ -36,12 +36,3 @@
>  
>  #define __HAVE_64B_ATOMICS 1
>  #define ATOMIC_EXCHANGE_USES_CAS 1
> -
> -/*
> - * All powerpc64 processors support the new "light weight"  sync (lwsync).
> - */
> -#define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
> -/*
> - * "light weight" sync can also be used for the release barrier.
> - */
> -#define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
> diff --git a/sysdeps/riscv/nptl/tls.h b/sysdeps/riscv/nptl/tls.h
> index 700c2f51899b0385d7ebaa4810c84de4fa6f2b45..020a986ceee89e1feb8f76c51f224a8faea71bbb 100644
> --- a/sysdeps/riscv/nptl/tls.h
> +++ b/sysdeps/riscv/nptl/tls.h
> @@ -123,7 +123,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/s390/nptl/tls.h b/sysdeps/s390/nptl/tls.h
> index 98d7870148ce6bc1d6397b1465dfabe96f7280b2..f1664d9ade6fd562db38c3dddd3fa6237a47faea 100644
> --- a/sysdeps/s390/nptl/tls.h
> +++ b/sysdeps/s390/nptl/tls.h
> @@ -167,7 +167,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/sh/nptl/tls.h b/sysdeps/sh/nptl/tls.h
> index 1530489a6ce4286bc5146e6cd83e3b463b965467..00ae1b998b9e0b1e6c347c4be4e99a90e530e924 100644
> --- a/sysdeps/sh/nptl/tls.h
> +++ b/sysdeps/sh/nptl/tls.h
> @@ -139,7 +139,7 @@ typedef struct
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/sparc/atomic-machine.h b/sysdeps/sparc/atomic-machine.h
> index a7042f1ee546b9f238153cb923409d42eb45cc03..1f0eb0a9b1171c06dc19dc21c4fe7de94adc4bce 100644
> --- a/sysdeps/sparc/atomic-machine.h
> +++ b/sysdeps/sparc/atomic-machine.h
> @@ -29,13 +29,6 @@
>  #define ATOMIC_EXCHANGE_USES_CAS     __HAVE_64B_ATOMICS
>  
>  #ifdef __sparc_v9__
> -# define atomic_full_barrier() \
> -  __asm __volatile ("membar #LoadLoad | #LoadStore"			      \
> -		    " | #StoreLoad | #StoreStore" : : : "memory")
> -# define atomic_read_barrier() \
> -  __asm __volatile ("membar #LoadLoad | #LoadStore" : : : "memory")
> -# define atomic_write_barrier() \
> -  __asm __volatile ("membar #LoadStore | #StoreStore" : : : "memory")
>  
>  extern void __cpu_relax (void);
>  # define atomic_spin_nop() __cpu_relax ()
> diff --git a/sysdeps/sparc/nptl/tls.h b/sysdeps/sparc/nptl/tls.h
> index 95a69cb8249dc79c3a063637a21d976d2660c48f..bc7ada0d3fe66751506e1cf5516f14ec2c205af8 100644
> --- a/sysdeps/sparc/nptl/tls.h
> +++ b/sysdeps/sparc/nptl/tls.h
> @@ -140,7 +140,7 @@ register struct pthread *__thread_self __asm__("%g7");
>    do									     \
>      {									     \
>        THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
> -      atomic_write_barrier ();						     \
> +      atomic_thread_fence_release ();					     \
>      }									     \
>    while (0)
>  
> diff --git a/sysdeps/unix/sysv/linux/arm/atomic-machine.h b/sysdeps/unix/sysv/linux/arm/atomic-machine.h
> deleted file mode 100644
> index 20068c72f359442769f8d49e11f7e771c922ef0b..0000000000000000000000000000000000000000
> --- a/sysdeps/unix/sysv/linux/arm/atomic-machine.h
> +++ /dev/null
> @@ -1,115 +0,0 @@
> -/* Atomic operations.  ARM/Linux version.
> -   Copyright (C) 2002-2022 Free Software Foundation, Inc.
> -   This file is part of the GNU C Library.
> -
> -   The GNU C 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.
> -
> -   The GNU C 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 the GNU C Library.  If not, see
> -   <https://www.gnu.org/licenses/>.  */
> -
> -#include <stdint.h>
> -
> -/* If the compiler doesn't provide a primitive, we'll use this macro
> -   to get assistance from the kernel.  */
> -#ifdef __thumb2__
> -# define __arm_assisted_full_barrier() \
> -     __asm__ __volatile__						      \
> -	     ("movw\tip, #0x0fa0\n\t"					      \
> -	      "movt\tip, #0xffff\n\t"					      \
> -	      "blx\tip"							      \
> -	      : : : "ip", "lr", "cc", "memory");
> -#else
> -# define __arm_assisted_full_barrier() \
> -     __asm__ __volatile__						      \
> -	     ("mov\tip, #0xffff0fff\n\t"				      \
> -	      "mov\tlr, pc\n\t"						      \
> -	      "add\tpc, ip, #(0xffff0fa0 - 0xffff0fff)"			      \
> -	      : : : "ip", "lr", "cc", "memory");
> -#endif
> -
> -/* Atomic compare and exchange.  This sequence relies on the kernel to
> -   provide a compare and exchange operation which is atomic on the
> -   current architecture, either via cleverness on pre-ARMv6 or via
> -   ldrex / strex on ARMv6.
> -
> -   It doesn't matter what register is used for a_oldval2, but we must
> -   specify one to work around GCC PR rtl-optimization/21223.  Otherwise
> -   it may cause a_oldval or a_tmp to be moved to a different register.
> -
> -   We use the union trick rather than simply using __typeof (...) in the
> -   declarations of A_OLDVAL et al because when NEWVAL or OLDVAL is of the
> -   form *PTR and PTR has a 'volatile ... *' type, then __typeof (*PTR) has
> -   a 'volatile ...' type and this triggers -Wvolatile-register-var to
> -   complain about 'register volatile ... asm ("reg")'.
> -
> -   We use the same union trick in the declaration of A_PTR because when
> -   MEM is of the from *PTR and PTR has a 'const ... *' type, then __typeof
> -   (*PTR) has a 'const ...' type and this enables the compiler to substitute
> -   the variable with its initializer in asm statements, which may cause the
> -   corresponding operand to appear in a different register.  */
> -#ifdef __thumb2__
> -/* Thumb-2 has ldrex/strex.  However it does not have barrier instructions,
> -   so we still need to use the kernel helper.  */
> -# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
> -  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
> -     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
> -     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
> -     register uint32_t a_oldval asm ("r0");				      \
> -     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
> -     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
> -     register uint32_t a_tmp asm ("r3");				      \
> -     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
> -     __asm__ __volatile__						      \
> -	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
> -	      "cmp\t%[tmp], %[old2]\n\t"				      \
> -	      "bne\t1f\n\t"						      \
> -	      "mov\t%[old], %[old2]\n\t"				      \
> -	      "movw\t%[tmp], #0x0fc0\n\t"				      \
> -	      "movt\t%[tmp], #0xffff\n\t"				      \
> -	      "blx\t%[tmp]\n\t"						      \
> -	      "bcc\t0b\n\t"						      \
> -	      "mov\t%[tmp], %[old2]\n\t"				      \
> -	      "1:"							      \
> -	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
> -	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
> -		[old2] "r" (a_oldval2)					      \
> -	      : "ip", "lr", "cc", "memory");				      \
> -     (__typeof (oldval)) a_tmp; })
> -#else
> -# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
> -  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
> -     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
> -     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
> -     register uint32_t a_oldval asm ("r0");				      \
> -     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
> -     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
> -     register uint32_t a_tmp asm ("r3");				      \
> -     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
> -     __asm__ __volatile__						      \
> -	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
> -	      "cmp\t%[tmp], %[old2]\n\t"				      \
> -	      "bne\t1f\n\t"						      \
> -	      "mov\t%[old], %[old2]\n\t"				      \
> -	      "mov\t%[tmp], #0xffff0fff\n\t"				      \
> -	      "mov\tlr, pc\n\t"						      \
> -	      "add\tpc, %[tmp], #(0xffff0fc0 - 0xffff0fff)\n\t"		      \
> -	      "bcc\t0b\n\t"						      \
> -	      "mov\t%[tmp], %[old2]\n\t"				      \
> -	      "1:"							      \
> -	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
> -	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
> -		[old2] "r" (a_oldval2)					      \
> -	      : "ip", "lr", "cc", "memory");				      \
> -     (__typeof (oldval)) a_tmp; })
> -#endif
> -
> -#include <sysdeps/arm/atomic-machine.h>
> diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
> index 6f83fb2965bd162f0f76e0e3586472ade39af607..02e54847a42bfbc93ae1e07b7e32965be644daba 100644
> --- a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
> +++ b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
> @@ -25,7 +25,4 @@
>  /* XXX Is this actually correct?  */
>  #define ATOMIC_EXCHANGE_USES_CAS 1
>  
> -# define atomic_full_barrier()				\
> -  (INTERNAL_SYSCALL_CALL (atomic_barrier), (void) 0)
> -
>  #endif
> diff --git a/sysdeps/x86/atomic-machine.h b/sysdeps/x86/atomic-machine.h
> index b9be51c52d8cbef2a95a62192c8ef7011e7f2c12..98541a2d06ff5e4aa8c789ab7405215097471971 100644
> --- a/sysdeps/x86/atomic-machine.h
> +++ b/sysdeps/x86/atomic-machine.h
> @@ -32,9 +32,6 @@
>  #endif
>  #define ATOMIC_EXCHANGE_USES_CAS	0
>  
> -#define atomic_read_barrier() __asm ("" ::: "memory")
> -#define atomic_write_barrier() __asm ("" ::: "memory")
> -
>  #define atomic_spin_nop() __asm ("pause")
>  
>  #endif /* atomic-machine.h */
> 
>
Wilco Dijkstra Aug. 5, 2022, 5:09 p.m. UTC | #2 
Hi Carlos,

Here is a rebased version that should build without needing other atomic
patches to be commited before it. But now it will clash with the other patches.

Cheers,
Wilco


v2: rebased to latest GLIBC, remove ancient atomic_full_barrier definitions

Replace atomic barriers based on sync primitives with atomic_thread_fence.
Many uses appear suspect and in the future fixing these to use load_acquire
or store_release would be useful.

---
diff --git a/crypt/crypt_util.c b/crypt/crypt_util.c
index be925e3484e65d2180e07915f5d91b47f6b96393..a8c2b26ed13c27804a9465f2a54caa09aaec3814 100644
--- a/crypt/crypt_util.c
+++ b/crypt/crypt_util.c
@@ -453,14 +453,14 @@ __init_des_r (struct crypt_data * __restrict __data)
 	  efp[comes_from_word][word_value][o_long] |= mask2;
       }
     }
-    atomic_write_barrier ();
+    atomic_thread_fence_release ();
     small_tables_initialized = 1;
 #ifdef __GNU_LIBRARY__
 small_tables_done:
     __libc_lock_unlock(_ufc_tables_lock);
 #endif
   } else
-    atomic_read_barrier ();
+    atomic_thread_fence_acquire ();
 
   /*
    * Create the sb tables:
diff --git a/elf/dl-deps.c b/elf/dl-deps.c
index 06005a0cc8686cc7e63cd8e1b1e7deda01fe6688..11b3fda5fdeb3830d3d5a0031084b43847444e04 100644
--- a/elf/dl-deps.c
+++ b/elf/dl-deps.c
@@ -430,7 +430,7 @@ _dl_map_object_deps (struct link_map *map,
 	  memcpy (&l_initfini[1], needed, nneeded * sizeof needed[0]);
 	  memcpy (&l_initfini[nneeded + 1], l_initfini,
 		  nneeded * sizeof needed[0]);
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  l->l_initfini = l_initfini;
 	  l->l_free_initfini = 1;
 	}
@@ -555,12 +555,12 @@ _dl_map_object_deps (struct link_map *map,
 
   /* Terminate the list of dependencies.  */
   l_initfini[nlist] = NULL;
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   map->l_initfini = l_initfini;
   map->l_free_initfini = 1;
   if (l_reldeps != NULL)
     {
-      atomic_write_barrier ();
+      atomic_thread_fence_release ();
       void *old_l_reldeps = map->l_reldeps;
       map->l_reldeps = l_reldeps;
       _dl_scope_free (old_l_reldeps);
diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
index 4c86dc694e0eebb23421e08103ee7f3c0adaadf9..68c66f61bc57da2e4e8736bf171f59b3e1b9052f 100644
--- a/elf/dl-lookup.c
+++ b/elf/dl-lookup.c
@@ -532,7 +532,7 @@ add_dependency (struct link_map *undef_map, struct link_map *map, int flags)
     = atomic_forced_read (undef_map->l_reldeps);
 
   /* Make sure l_reldeps is read before l_initfini.  */
-  atomic_read_barrier ();
+  atomic_thread_fence_acquire ();
 
   /* Determine whether UNDEF_MAP already has a reference to MAP.  First
      look in the normal dependencies.  */
@@ -697,7 +697,7 @@ marking %s [%lu] as NODELETE due to memory allocation failure\n",
 			l_reldepsact * sizeof (struct link_map *));
 	      newp->list[l_reldepsact] = map;
 	      newp->act = l_reldepsact + 1;
-	      atomic_write_barrier ();
+	      atomic_thread_fence_release ();
 	      void *old = undef_map->l_reldeps;
 	      undef_map->l_reldeps = newp;
 	      undef_map->l_reldepsmax = max;
@@ -708,7 +708,7 @@ marking %s [%lu] as NODELETE due to memory allocation failure\n",
       else
 	{
 	  undef_map->l_reldeps->list[l_reldepsact] = map;
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  undef_map->l_reldeps->act = l_reldepsact + 1;
 	}
 
diff --git a/elf/dl-open.c b/elf/dl-open.c
index a23e65926bcfe797f06f8b4175f65040f4547a05..ba77f4b774cae69c382bceb6599936664203ef05 100644
--- a/elf/dl-open.c
+++ b/elf/dl-open.c
@@ -202,7 +202,7 @@ add_to_global_update (struct link_map *new)
   assert (added <= ns->_ns_global_scope_pending_adds);
   ns->_ns_global_scope_pending_adds -= added;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   ns->_ns_main_searchlist->r_nlist = new_nlist;
 }
 
@@ -342,7 +342,7 @@ update_scopes (struct link_map *new)
 	     might use the new last element and then use the garbage
 	     at offset IDX+1.  */
 	  imap->l_scope[cnt + 1] = NULL;
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  imap->l_scope[cnt] = &new->l_searchlist;
 
 	  from_scope = cnt;
diff --git a/include/atomic.h b/include/atomic.h
index 2cb52c9cfd894308b97b97a04dd574b2287bf1b2..27038aee6559bf1fb6ca277a1dbdc1d41776b813 100644
--- a/include/atomic.h
+++ b/include/atomic.h
@@ -504,20 +504,6 @@
      __atg19_old; })
 #endif
 
-#ifndef atomic_full_barrier
-# define atomic_full_barrier() __asm ("" ::: "memory")
-#endif
-
-
-#ifndef atomic_read_barrier
-# define atomic_read_barrier() atomic_full_barrier ()
-#endif
-
-
-#ifndef atomic_write_barrier
-# define atomic_write_barrier() atomic_full_barrier ()
-#endif
-
 
 #ifndef atomic_forced_read
 # define atomic_forced_read(x) \
@@ -655,15 +641,12 @@ void __atomic_link_error (void);
 /* By default, we assume that read, write, and full barriers are equivalent
    to acquire, release, and seq_cst barriers.  Archs for which this does not
    hold have to provide custom definitions of the fences.  */
-# ifndef atomic_thread_fence_acquire
-#  define atomic_thread_fence_acquire() atomic_read_barrier ()
-# endif
-# ifndef atomic_thread_fence_release
-#  define atomic_thread_fence_release() atomic_write_barrier ()
-# endif
-# ifndef atomic_thread_fence_seq_cst
-#  define atomic_thread_fence_seq_cst() atomic_full_barrier ()
-# endif
+# define atomic_thread_fence_acquire() \
+  __atomic_thread_fence (__ATOMIC_ACQUIRE)
+# define atomic_thread_fence_release() \
+  __atomic_thread_fence (__ATOMIC_RELEASE)
+# define atomic_thread_fence_seq_cst() \
+  __atomic_thread_fence (__ATOMIC_SEQ_CST)
 
 # ifndef atomic_load_relaxed
 #  define atomic_load_relaxed(mem) \
diff --git a/include/list.h b/include/list.h
index 7bea2c50a3759c0c8640971eff1e80874e3b543f..31a8a93fa2491b38c368b9e07dd65c473e4eb19f 100644
--- a/include/list.h
+++ b/include/list.h
@@ -43,7 +43,7 @@ list_add (list_t *newp, list_t *head)
   newp->next = head->next;
   newp->prev = head;
   head->next->prev = newp;
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   head->next = newp;
 }
 
diff --git a/malloc/arena.c b/malloc/arena.c
index defd25c8a6850188824c1e51f41845ace11e1060..7b0c3076efb664d8daa61992a04f4f0853ff5fe1 100644
--- a/malloc/arena.c
+++ b/malloc/arena.c
@@ -111,7 +111,7 @@ static mstate free_list;
    malloc_state objects.
 
    Read access to the next member is supposed to synchronize with the
-   atomic_write_barrier and the write to the next member in
+   atomic_thread_fence_release and the write to the next member in
    _int_new_arena.  This suffers from data races; see the FIXME
    comments in _int_new_arena and reused_arena.
 
@@ -773,7 +773,7 @@ _int_new_arena (size_t size)
   /* FIXME: The barrier is an attempt to synchronize with read access
      in reused_arena, which does not acquire list_lock while
      traversing the list.  */
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   main_arena.next = a;
 
   __libc_lock_unlock (list_lock);
diff --git a/manual/llio.texi b/manual/llio.texi
index e55b02d8bd839827a352c751f11ad18114c32ed3..9d37d8dd731a942bd968bec41499a5213dc09470 100644
--- a/manual/llio.texi
+++ b/manual/llio.texi
@@ -2543,14 +2543,14 @@ aiocb64}, since the LFS transparently replaces the old interface.
 @c    deallocate_stack @asulock @ascuheap @aculock @acsmem
 @c     lll_lock (state_cache_lock) @asulock @aculock
 @c     stack_list_del ok
-@c      atomic_write_barrier ok
+@c      atomic_thread_fence_release ok
 @c      list_del ok
-@c      atomic_write_barrier ok
+@c      atomic_thread_fence_release ok
 @c     queue_stack @ascuheap @acsmem
 @c      stack_list_add ok
-@c       atomic_write_barrier ok
+@c       atomic_thread_fence_release ok
 @c       list_add ok
-@c       atomic_write_barrier ok
+@c       atomic_thread_fence_release ok
 @c      free_stacks @ascuheap @acsmem
 @c       list_for_each_prev_safe ok
 @c       list_entry ok
diff --git a/manual/memory.texi b/manual/memory.texi
index 23a039c57e60c81787252d935e3b309fd8290902..e3f1e7cf33e426cb888738f5614e6be11af49ec2 100644
--- a/manual/memory.texi
+++ b/manual/memory.texi
@@ -395,7 +395,7 @@ this function is in @file{stdlib.h}.
 @c     mutex_init ok
 @c     mutex_lock (just-created mutex) ok, returns locked
 @c     mutex_lock (list_lock) dup @asulock @aculock
-@c     atomic_write_barrier ok
+@c     atomic_thread_fence_release ok
 @c     mutex_unlock (list_lock) @aculock
 @c    catomic_decrement ok
 @c    reused_arena @asulock @aculock
diff --git a/manual/startup.texi b/manual/startup.texi
index 9bf24123f562f75ba27a4770c69147e003b94755..4c7c2976a92d772909203bca028347e8d798b2d7 100644
--- a/manual/startup.texi
+++ b/manual/startup.texi
@@ -947,7 +947,7 @@ using @code{atexit} or @code{on_exit}.
 @c     __libc_lock_lock @asulock @aculock
 @c     calloc dup @ascuheap @acsmem
 @c     __libc_lock_unlock @aculock
-@c    atomic_write_barrier dup ok
+@c    atomic_thread_fence_release dup ok
 The @code{atexit} function registers the function @var{function} to be
 called at normal program termination.  The @var{function} is called with
 no arguments.
@@ -961,7 +961,7 @@ the function cannot be registered.
 @safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{} @asulock{}}@acunsafe{@aculock{} @acsmem{}}}
 @c on_exit @ascuheap @asulock @aculock @acsmem
 @c  new_exitfn dup @ascuheap @asulock @aculock @acsmem
-@c  atomic_write_barrier dup ok
+@c  atomic_thread_fence_release dup ok
 This function is a somewhat more powerful variant of @code{atexit}.  It
 accepts two arguments, a function @var{function} and an arbitrary
 pointer @var{arg}.  At normal program termination, the @var{function} is
diff --git a/nptl/nptl-stack.c b/nptl/nptl-stack.c
index 20ce78eddbf100833d453d7032f63bc2ba8f01c7..7c04e7faaae5c15bf5ad98c32935bcba4849c1c1 100644
--- a/nptl/nptl-stack.c
+++ b/nptl/nptl-stack.c
@@ -27,11 +27,11 @@ __nptl_stack_list_del (list_t *elem)
 {
   GL (dl_in_flight_stack) = (uintptr_t) elem;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   list_del (elem);
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   GL (dl_in_flight_stack) = 0;
 }
@@ -42,11 +42,11 @@ __nptl_stack_list_add (list_t *elem, list_t *list)
 {
   GL (dl_in_flight_stack) = (uintptr_t) elem | 1;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   list_add (elem, list);
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   GL (dl_in_flight_stack) = 0;
 }
diff --git a/nptl/pthread_mutex_setprioceiling.c b/nptl/pthread_mutex_setprioceiling.c
index 2d71a750c8981e8ca271c265031887e2c510583a..b574a77250664bbb1487c932b9a32a9dee415072 100644
--- a/nptl/pthread_mutex_setprioceiling.c
+++ b/nptl/pthread_mutex_setprioceiling.c
@@ -113,7 +113,7 @@ __pthread_mutex_setprioceiling (pthread_mutex_t *mutex, int prioceiling,
     newlock = (mutex->__data.__lock & ~PTHREAD_MUTEX_PRIO_CEILING_MASK);
   mutex->__data.__lock = newlock
 			 | (prioceiling << PTHREAD_MUTEX_PRIO_CEILING_SHIFT);
-  atomic_full_barrier ();
+  atomic_thread_fence_seq_cst ();
 
   futex_wake ((unsigned int *)&mutex->__data.__lock, INT_MAX,
 	      PTHREAD_MUTEX_PSHARED (mutex));
diff --git a/nptl/sem_post.c b/nptl/sem_post.c
index 9e5741753a741034db96cfff8b8978908fafc1f6..5e5753bee43943d669d10bddc1ae36b700915dc5 100644
--- a/nptl/sem_post.c
+++ b/nptl/sem_post.c
@@ -90,7 +90,7 @@ __old_sem_post (sem_t *sem)
 
   /* We must need to synchronize with consumers of this token, so the atomic
      increment must have release MO semantics.  */
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   (void) atomic_increment_val (futex);
   /* We always have to assume it is a shared semaphore.  */
   futex_wake (futex, 1, LLL_SHARED);
diff --git a/stdlib/msort.c b/stdlib/msort.c
index cbe9a4a8fdb38113a4c18976c9f297be103d458f..e2f1eca94ad7e9005145c376b0de3dbd1ca14f18 100644
--- a/stdlib/msort.c
+++ b/stdlib/msort.c
@@ -197,7 +197,7 @@ __qsort_r (void *b, size_t n, size_t s, __compar_d_fn_t cmp, void *arg)
 	  phys_pages /= 4;
 
 	  /* Make sure phys_pages is written to memory.  */
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 
 	  pagesize = __sysconf (_SC_PAGESIZE);
 	}
diff --git a/sysdeps/aarch64/atomic-machine.h b/sysdeps/aarch64/atomic-machine.h
index 52b3fb2047cbf2518644b1595ed9a11cddcc89d2..dccd7a6bffd1ce2222a920c01ca4622aa0eec0e3 100644
--- a/sysdeps/aarch64/atomic-machine.h
+++ b/sysdeps/aarch64/atomic-machine.h
@@ -147,7 +147,4 @@
   __atomic_val_bysize (__arch_exchange_and_add, int, mem, value,	\
 		       __ATOMIC_RELEASE)
 
-/* Barrier macro. */
-#define atomic_full_barrier() __sync_synchronize()
-
 #endif
diff --git a/sysdeps/aarch64/nptl/tls.h b/sysdeps/aarch64/nptl/tls.h
index 8d62b31e23d53ff2789bb32a828eac2254434679..da6a4e0d63a17d4c71a4b97619e80bd6287b7f30 100644
--- a/sysdeps/aarch64/nptl/tls.h
+++ b/sysdeps/aarch64/nptl/tls.h
@@ -108,7 +108,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/alpha/atomic-machine.h b/sysdeps/alpha/atomic-machine.h
index df28b90261aa485125d951bc1abf76602730bfd0..6e7e1b21ea3d947648311bfc35833fe475dbb5ed 100644
--- a/sysdeps/alpha/atomic-machine.h
+++ b/sysdeps/alpha/atomic-machine.h
@@ -333,7 +333,3 @@
 #define atomic_bit_test_set(mem, bit)
 
 */
-
-#define atomic_full_barrier()	__asm ("mb" : : : "memory");
-#define atomic_read_barrier()	__asm ("mb" : : : "memory");
-#define atomic_write_barrier()	__asm ("wmb" : : : "memory");
diff --git a/sysdeps/alpha/nptl/tls.h b/sysdeps/alpha/nptl/tls.h
index ddf7de0705b37be20dbd7c981ff79e36659e0439..ac6c1514672b83677128761df992bcfae11a9b2a 100644
--- a/sysdeps/alpha/nptl/tls.h
+++ b/sysdeps/alpha/nptl/tls.h
@@ -105,7 +105,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/arc/atomic-machine.h b/sysdeps/arc/atomic-machine.h
index 3d17f7899083c859efd645878e0f0596da32a073..f155bafca3ee277f70af030f11a8e15b1c19bf9f 100644
--- a/sysdeps/arc/atomic-machine.h
+++ b/sysdeps/arc/atomic-machine.h
@@ -52,6 +52,4 @@
   __atomic_val_bysize (__arch_compare_and_exchange_val, int,		\
 		       mem, new, old, __ATOMIC_ACQUIRE)
 
-#define atomic_full_barrier()  ({ asm volatile ("dmb 3":::"memory"); })
-
 #endif /* _ARC_BITS_ATOMIC_H */
diff --git a/sysdeps/arc/nptl/tls.h b/sysdeps/arc/nptl/tls.h
index 15adfc94918dda2364ab5321d3a6ce5693ba4ad4..a9f1350ea6fcb1d86ab75207f0833a11f457af30 100644
--- a/sysdeps/arc/nptl/tls.h
+++ b/sysdeps/arc/nptl/tls.h
@@ -113,7 +113,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/arm/atomic-machine.h b/sysdeps/arm/atomic-machine.h
index 952404379748e6dc5dee1da7731fb5c6faab4e57..d5258a0cb16da045357fd5404e752a229af248ac 100644
--- a/sysdeps/arm/atomic-machine.h
+++ b/sysdeps/arm/atomic-machine.h
@@ -22,12 +22,6 @@
 
 void __arm_link_error (void);
 
-#ifdef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4
-# define atomic_full_barrier() __sync_synchronize ()
-#else
-# define atomic_full_barrier() __arm_assisted_full_barrier ()
-#endif
-
 /* An OS-specific atomic-machine.h file will define this macro if
    the OS can provide something.  If not, we'll fail to build
    with a compiler that doesn't supply the operation.  */
diff --git a/sysdeps/arm/nptl/tls.h b/sysdeps/arm/nptl/tls.h
index 02d9be202d77b9ec871a52be5a4e49feffc300b1..28249c31f72200b2e7c06b6598a5a6076584b09e 100644
--- a/sysdeps/arm/nptl/tls.h
+++ b/sysdeps/arm/nptl/tls.h
@@ -99,7 +99,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/csky/nptl/tls.h b/sysdeps/csky/nptl/tls.h
index cd135d546444e2e8f8ce0d53d5a16e2b13709bb9..da4efbe6d51b6d09e4f91ce63c5c5d2a9fa102c3 100644
--- a/sysdeps/csky/nptl/tls.h
+++ b/sysdeps/csky/nptl/tls.h
@@ -128,7 +128,7 @@ typedef struct
   do									      \
     {									      \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	      \
-      atomic_write_barrier ();						      \
+      atomic_thread_fence_release ();					      \
     }									      \
   while (0)
 
diff --git a/sysdeps/generic/malloc-machine.h b/sysdeps/generic/malloc-machine.h
index 001a8e7e606c584dabacc9cbf6713f137bb9b4a7..ebd6983ecc14b5b314f457fc1766a9f86561d32f 100644
--- a/sysdeps/generic/malloc-machine.h
+++ b/sysdeps/generic/malloc-machine.h
@@ -22,18 +22,6 @@
 
 #include <atomic.h>
 
-#ifndef atomic_full_barrier
-# define atomic_full_barrier() __asm ("" ::: "memory")
-#endif
-
-#ifndef atomic_read_barrier
-# define atomic_read_barrier() atomic_full_barrier ()
-#endif
-
-#ifndef atomic_write_barrier
-# define atomic_write_barrier() atomic_full_barrier ()
-#endif
-
 #ifndef DEFAULT_TOP_PAD
 # define DEFAULT_TOP_PAD 131072
 #endif
diff --git a/sysdeps/hppa/dl-fptr.c b/sysdeps/hppa/dl-fptr.c
index 9ed21602d6155d4b960278f8d1fac4ffa885b9d5..97882a8106a719642b2778d3ca8a69ef202cce2d 100644
--- a/sysdeps/hppa/dl-fptr.c
+++ b/sysdeps/hppa/dl-fptr.c
@@ -371,7 +371,7 @@ _dl_lookup_address (const void *address)
 
   /* First load the relocation offset.  */
   reloc_arg = (ElfW(Word)) desc[1];
-  atomic_full_barrier();
+  atomic_thread_fence_seq_cst ();
 
   /* Then load first word of candidate descriptor.  It should be a pointer
      with word alignment and point to memory that can be read.  */
diff --git a/sysdeps/hppa/dl-machine.h b/sysdeps/hppa/dl-machine.h
index c865713be1e3f8e0430bbb35c8db7ebe3e7a6abf..61635ca9115e1fa77305eaa3cc4ab5bf9bb91d7b 100644
--- a/sysdeps/hppa/dl-machine.h
+++ b/sysdeps/hppa/dl-machine.h
@@ -136,7 +136,7 @@ elf_machine_fixup_plt (struct link_map *map, lookup_t t,
       /* Need to ensure that the gp is visible before the code
          entry point is updated */
       rfdesc[1] = value.gp;
-      atomic_full_barrier();
+      atomic_thread_fence_seq_cst ();
       rfdesc[0] = value.ip;
     }
   else
diff --git a/sysdeps/hppa/nptl/tls.h b/sysdeps/hppa/nptl/tls.h
index 204960d524246af3e8be40ecad20ffb567f2dc6d..c502b1a0fd9c02bcf86b8703c99d16e6cc8374eb 100644
--- a/sysdeps/hppa/nptl/tls.h
+++ b/sysdeps/hppa/nptl/tls.h
@@ -133,7 +133,7 @@ static inline void __set_cr27(struct pthread *cr27)
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/htl/pt-once.c b/sysdeps/htl/pt-once.c
index b85b196645958fc7f47b08b39e91077b82817cdc..55db6c3d7176a9ca6cf2d0caccc1cf01aeb5ea2f 100644
--- a/sysdeps/htl/pt-once.c
+++ b/sysdeps/htl/pt-once.c
@@ -33,7 +33,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 {
   ASSERT_TYPE_SIZE (pthread_once_t, __SIZEOF_PTHREAD_ONCE_T);
 
-  atomic_full_barrier ();
+  atomic_thread_fence_seq_cst ();
   if (once_control->__run == 0)
     {
       __pthread_spin_wait (&once_control->__lock);
@@ -44,7 +44,7 @@ __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 	  init_routine ();
 	  pthread_cleanup_pop (0);
 
-	  atomic_full_barrier ();
+	  atomic_thread_fence_seq_cst ();
 	  once_control->__run = 1;
 	}
 
diff --git a/sysdeps/ia64/atomic-machine.h b/sysdeps/ia64/atomic-machine.h
index 71afcfe0d0031a6ceae5879a2b3b19ed2ee2f110..9fef5306e2bfff85b0363da195c96a0542f35940 100644
--- a/sysdeps/ia64/atomic-machine.h
+++ b/sysdeps/ia64/atomic-machine.h
@@ -93,4 +93,3 @@
      while (__builtin_expect (__val != __oldval, 0));			      \
      __oldval & __mask; })
 
-#define atomic_full_barrier() __sync_synchronize ()
diff --git a/sysdeps/ia64/nptl/tls.h b/sysdeps/ia64/nptl/tls.h
index 8ccedb73e610095337b8a082ee0742866ac4e2f8..2f43a5511c2e2a9b0e1ca107a8c800ed80d54399 100644
--- a/sysdeps/ia64/nptl/tls.h
+++ b/sysdeps/ia64/nptl/tls.h
@@ -157,7 +157,7 @@ register struct pthread *__thread_self __asm__("r13");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/loongarch/nptl/tls.h b/sysdeps/loongarch/nptl/tls.h
index 3b271c4b60f220c0b916583324a629dae28a4727..a4828dc25bf71f4d0c874936a61672eaa8a1cf88 100644
--- a/sysdeps/loongarch/nptl/tls.h
+++ b/sysdeps/loongarch/nptl/tls.h
@@ -129,7 +129,7 @@ typedef struct
   do \
     { \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED; \
-      atomic_write_barrier (); \
+      atomic_thread_fence_release (); \
     } \
   while (0)
 
diff --git a/sysdeps/m68k/nptl/tls.h b/sysdeps/m68k/nptl/tls.h
index 30d58e36e9552d1776fffe071653c78a8012408b..76b10a2e44225650fb39d897505f62df7d9b7533 100644
--- a/sysdeps/m68k/nptl/tls.h
+++ b/sysdeps/m68k/nptl/tls.h
@@ -132,7 +132,7 @@ extern void * __m68k_read_tp (void);
   do									\
     {									\
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
-      atomic_write_barrier ();						\
+      atomic_thread_fence_release ();					\
     }									\
   while (0)
 
diff --git a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
index 71f789cbdfa20b79f9e3cef5a2523d4e243b4f19..feb9085af9499b5d954a1bf4f04d9cbcb804e8cf 100644
--- a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
+++ b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
@@ -26,7 +26,7 @@
 int
 __pthread_mutex_destroy (pthread_mutex_t *mtxp)
 {
-  atomic_read_barrier ();
+  atomic_thread_fence_acquire ();
   if (*(volatile unsigned int *) &mtxp->__lock != 0)
     return EBUSY;
 
diff --git a/sysdeps/mach/hurd/htl/pt-mutex.h b/sysdeps/mach/hurd/htl/pt-mutex.h
index 4021e72a6e8d15316336296ff732a4e7fd1acdff..ebdf8a5fbde0f755c7625cd38ad185ee0c977b5f 100644
--- a/sysdeps/mach/hurd/htl/pt-mutex.h
+++ b/sysdeps/mach/hurd/htl/pt-mutex.h
@@ -54,7 +54,7 @@
           if (ret == EOWNERDEAD)   \
             {   \
               mtxp->__lock = mtxp->__lock | LLL_DEAD_OWNER;   \
-              atomic_write_barrier ();   \
+              atomic_thread_fence_release ();   \
             }   \
         }   \
     }   \
diff --git a/sysdeps/microblaze/nptl/tls.h b/sysdeps/microblaze/nptl/tls.h
index 40a6acd71e710c35f2b8ef2495fdeb34906f9517..ee043c67ced211a7751aa94be79756b4751f452c 100644
--- a/sysdeps/microblaze/nptl/tls.h
+++ b/sysdeps/microblaze/nptl/tls.h
@@ -110,7 +110,7 @@ typedef struct
   do                                                                        \
     {                                                                       \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;            \
-      atomic_write_barrier ();                                              \
+      atomic_thread_fence_release ();                                       \
     }                                                                       \
   while (0)
 
diff --git a/sysdeps/mips/atomic-machine.h b/sysdeps/mips/atomic-machine.h
index 88805ee30b11d8af96e296b8794a8a0d2cedfeb7..11fa155988e1c65ea7e8b2d185c12c18477f0004 100644
--- a/sysdeps/mips/atomic-machine.h
+++ b/sysdeps/mips/atomic-machine.h
@@ -167,18 +167,4 @@
   __atomic_val_bysize (__arch_exchange_and_add, int, mem, value,	\
 		       __ATOMIC_RELEASE)
 
-/* TODO: More atomic operations could be implemented efficiently; only the
-   basic requirements are done.  */
-
-#ifdef __mips16
-# define atomic_full_barrier() __sync_synchronize ()
-
-#else /* !__mips16 */
-# define atomic_full_barrier() \
-  __asm__ __volatile__ (".set push\n\t"					      \
-			MIPS_PUSH_MIPS2					      \
-			MIPS_SYNC_STR "\n\t"				      \
-			".set pop" : : : "memory")
-#endif /* !__mips16 */
-
 #endif /* atomic-machine.h */
diff --git a/sysdeps/mips/nptl/tls.h b/sysdeps/mips/nptl/tls.h
index 03a5b24abd5d1d6254418e79cb231b03cee72d99..ea52828ffcd7c79d3aa09466227ea24b503e3538 100644
--- a/sysdeps/mips/nptl/tls.h
+++ b/sysdeps/mips/nptl/tls.h
@@ -160,7 +160,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/nios2/nptl/tls.h b/sysdeps/nios2/nptl/tls.h
index 18080275ef7458c80c7826fc0fa0f6a4fadac320..065628120c5198b8ea9a08dbe27205a9ee3c86a5 100644
--- a/sysdeps/nios2/nptl/tls.h
+++ b/sysdeps/nios2/nptl/tls.h
@@ -140,7 +140,7 @@ register struct pthread *__thread_self __asm__("r23");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/or1k/atomic-machine.h b/sysdeps/or1k/atomic-machine.h
index 0d27298d70037a771de9abb1033cbfcb48cdb1b8..4744928e33fc8e6b55b23ddb6965e6deadf23979 100644
--- a/sysdeps/or1k/atomic-machine.h
+++ b/sysdeps/or1k/atomic-machine.h
@@ -66,6 +66,4 @@
   __atomic_val_bysize (__arch_compare_and_exchange_val, int,		\
 		       mem, new, old, __ATOMIC_ACQUIRE)
 
-#define atomic_full_barrier() ({ asm volatile ("l.msync" ::: "memory"); })
-
 #endif /* atomic-machine.h */
diff --git a/sysdeps/or1k/nptl/tls.h b/sysdeps/or1k/nptl/tls.h
index c6ffe62c3f8c1ae330e6402640294cbc18714883..6b4fb2c15275689b3f584cb773222bb73aba5faa 100644
--- a/sysdeps/or1k/nptl/tls.h
+++ b/sysdeps/or1k/nptl/tls.h
@@ -175,7 +175,7 @@ register tcbhead_t *__thread_self __asm__("r10");
   do									\
     {									\
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
-      atomic_write_barrier ();						\
+      atomic_thread_fence_release ();					\
     }									\
   while (0)
 
diff --git a/sysdeps/powerpc/atomic-machine.h b/sysdeps/powerpc/atomic-machine.h
index 853bd8a75eaa429feb4a707069a8f5859943d103..aed1c72a451ad773d91bf9828ee19e211076bf10 100644
--- a/sysdeps/powerpc/atomic-machine.h
+++ b/sysdeps/powerpc/atomic-machine.h
@@ -47,8 +47,6 @@
 # define MUTEX_HINT_REL
 #endif
 
-#define atomic_full_barrier()	__asm ("sync" ::: "memory")
-
 #define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval)	      \
   ({									      \
       __typeof (*(mem)) __tmp;						      \
diff --git a/sysdeps/powerpc/nptl/tls.h b/sysdeps/powerpc/nptl/tls.h
index 22b00752350196ad7e6d0fbdcafdc271f0af5bed..08e9d921f47a3aa0ee5185865ff0113a71391245 100644
--- a/sysdeps/powerpc/nptl/tls.h
+++ b/sysdeps/powerpc/nptl/tls.h
@@ -224,7 +224,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/powerpc/powerpc32/atomic-machine.h b/sysdeps/powerpc/powerpc32/atomic-machine.h
index 33ff2aa071b93c7497e6c8ded8e0fbc33ce41722..ba37e3da042a0c21e22db7995d353047f0289325 100644
--- a/sysdeps/powerpc/powerpc32/atomic-machine.h
+++ b/sysdeps/powerpc/powerpc32/atomic-machine.h
@@ -95,25 +95,7 @@
     ({ abort (); (*mem)--; })
 
 #ifdef _ARCH_PWR4
-/*
- * Newer powerpc64 processors support the new "light weight" sync (lwsync)
- * So if the build is using -mcpu=[power4,power5,power5+,970] we can
- * safely use lwsync.
- */
-# define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
-/*
- * "light weight" sync can also be used for the release barrier.
- */
 # define __ARCH_REL_INSTR	"lwsync"
-# define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
-#else
-/*
- * Older powerpc32 processors don't support the new "light weight"
- * sync (lwsync).  So the only safe option is to use normal sync
- * for all powerpc32 applications.
- */
-# define atomic_read_barrier()	__asm ("sync" ::: "memory")
-# define atomic_write_barrier()	__asm ("sync" ::: "memory")
 #endif
 
 /*
diff --git a/sysdeps/powerpc/powerpc64/atomic-machine.h b/sysdeps/powerpc/powerpc64/atomic-machine.h
index cbaf01162a277519bdcf730c33d4e5526c3dcb39..8693388bd49154ac32aa99d986eebf214e418a51 100644
--- a/sysdeps/powerpc/powerpc64/atomic-machine.h
+++ b/sysdeps/powerpc/powerpc64/atomic-machine.h
@@ -222,15 +222,7 @@
      __val;								      \
   })
 
-/*
- * All powerpc64 processors support the new "light weight"  sync (lwsync).
- */
-#define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
-/*
- * "light weight" sync can also be used for the release barrier.
- */
 #define __ARCH_REL_INSTR	"lwsync"
-#define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
 
 /*
  * Include the rest of the atomic ops macros which are common to both
diff --git a/sysdeps/riscv/nptl/tls.h b/sysdeps/riscv/nptl/tls.h
index 4e586f348adf6556151c850e983bbcaa42a91b0f..7d4df591d571eba294fa7af8b14ae73639939593 100644
--- a/sysdeps/riscv/nptl/tls.h
+++ b/sysdeps/riscv/nptl/tls.h
@@ -123,7 +123,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/s390/nptl/tls.h b/sysdeps/s390/nptl/tls.h
index ff210ffeb25a140bb19dbf54ab299350dc261ac7..a4cad2d3a9733cce6d16787bfb6172add464c9e6 100644
--- a/sysdeps/s390/nptl/tls.h
+++ b/sysdeps/s390/nptl/tls.h
@@ -167,7 +167,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/sh/nptl/tls.h b/sysdeps/sh/nptl/tls.h
index 76591ab6ef6c12b73dda0feb18ab5adfbf53c28e..556ca89588f6750b464338647049bdb602d5be03 100644
--- a/sysdeps/sh/nptl/tls.h
+++ b/sysdeps/sh/nptl/tls.h
@@ -139,7 +139,7 @@ typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/sparc/atomic-machine.h b/sysdeps/sparc/atomic-machine.h
index 653c2035f76bbf8cd5ef31463807f199528b417f..aefe481401ee4a1786c3781984e927e4b8bc58b0 100644
--- a/sysdeps/sparc/atomic-machine.h
+++ b/sysdeps/sparc/atomic-machine.h
@@ -53,13 +53,6 @@
   })
 
 #ifdef __sparc_v9__
-# define atomic_full_barrier() \
-  __asm __volatile ("membar #LoadLoad | #LoadStore"			      \
-		    " | #StoreLoad | #StoreStore" : : : "memory")
-# define atomic_read_barrier() \
-  __asm __volatile ("membar #LoadLoad | #LoadStore" : : : "memory")
-# define atomic_write_barrier() \
-  __asm __volatile ("membar #LoadStore | #StoreStore" : : : "memory")
 
 extern void __cpu_relax (void);
 # define atomic_spin_nop() __cpu_relax ()
diff --git a/sysdeps/sparc/nptl/tls.h b/sysdeps/sparc/nptl/tls.h
index d1e2bb4ad192cc7c825fd62b03bec45bc537aaf6..555d22fdf4d703b6de1c705fb3e47109f0bdd459 100644
--- a/sysdeps/sparc/nptl/tls.h
+++ b/sysdeps/sparc/nptl/tls.h
@@ -140,7 +140,7 @@ register struct pthread *__thread_self __asm__("%g7");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/unix/sysv/linux/arm/atomic-machine.h b/sysdeps/unix/sysv/linux/arm/atomic-machine.h
deleted file mode 100644
index 20068c72f359442769f8d49e11f7e771c922ef0b..0000000000000000000000000000000000000000
--- a/sysdeps/unix/sysv/linux/arm/atomic-machine.h
+++ /dev/null
@@ -1,115 +0,0 @@
-/* Atomic operations.  ARM/Linux version.
-   Copyright (C) 2002-2022 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C 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.
-
-   The GNU C 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 the GNU C Library.  If not, see
-   <https://www.gnu.org/licenses/>.  */
-
-#include <stdint.h>
-
-/* If the compiler doesn't provide a primitive, we'll use this macro
-   to get assistance from the kernel.  */
-#ifdef __thumb2__
-# define __arm_assisted_full_barrier() \
-     __asm__ __volatile__						      \
-	     ("movw\tip, #0x0fa0\n\t"					      \
-	      "movt\tip, #0xffff\n\t"					      \
-	      "blx\tip"							      \
-	      : : : "ip", "lr", "cc", "memory");
-#else
-# define __arm_assisted_full_barrier() \
-     __asm__ __volatile__						      \
-	     ("mov\tip, #0xffff0fff\n\t"				      \
-	      "mov\tlr, pc\n\t"						      \
-	      "add\tpc, ip, #(0xffff0fa0 - 0xffff0fff)"			      \
-	      : : : "ip", "lr", "cc", "memory");
-#endif
-
-/* Atomic compare and exchange.  This sequence relies on the kernel to
-   provide a compare and exchange operation which is atomic on the
-   current architecture, either via cleverness on pre-ARMv6 or via
-   ldrex / strex on ARMv6.
-
-   It doesn't matter what register is used for a_oldval2, but we must
-   specify one to work around GCC PR rtl-optimization/21223.  Otherwise
-   it may cause a_oldval or a_tmp to be moved to a different register.
-
-   We use the union trick rather than simply using __typeof (...) in the
-   declarations of A_OLDVAL et al because when NEWVAL or OLDVAL is of the
-   form *PTR and PTR has a 'volatile ... *' type, then __typeof (*PTR) has
-   a 'volatile ...' type and this triggers -Wvolatile-register-var to
-   complain about 'register volatile ... asm ("reg")'.
-
-   We use the same union trick in the declaration of A_PTR because when
-   MEM is of the from *PTR and PTR has a 'const ... *' type, then __typeof
-   (*PTR) has a 'const ...' type and this enables the compiler to substitute
-   the variable with its initializer in asm statements, which may cause the
-   corresponding operand to appear in a different register.  */
-#ifdef __thumb2__
-/* Thumb-2 has ldrex/strex.  However it does not have barrier instructions,
-   so we still need to use the kernel helper.  */
-# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
-     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
-     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
-     register uint32_t a_oldval asm ("r0");				      \
-     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
-     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
-     register uint32_t a_tmp asm ("r3");				      \
-     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
-     __asm__ __volatile__						      \
-	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
-	      "cmp\t%[tmp], %[old2]\n\t"				      \
-	      "bne\t1f\n\t"						      \
-	      "mov\t%[old], %[old2]\n\t"				      \
-	      "movw\t%[tmp], #0x0fc0\n\t"				      \
-	      "movt\t%[tmp], #0xffff\n\t"				      \
-	      "blx\t%[tmp]\n\t"						      \
-	      "bcc\t0b\n\t"						      \
-	      "mov\t%[tmp], %[old2]\n\t"				      \
-	      "1:"							      \
-	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
-	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
-		[old2] "r" (a_oldval2)					      \
-	      : "ip", "lr", "cc", "memory");				      \
-     (__typeof (oldval)) a_tmp; })
-#else
-# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
-     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
-     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
-     register uint32_t a_oldval asm ("r0");				      \
-     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
-     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
-     register uint32_t a_tmp asm ("r3");				      \
-     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
-     __asm__ __volatile__						      \
-	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
-	      "cmp\t%[tmp], %[old2]\n\t"				      \
-	      "bne\t1f\n\t"						      \
-	      "mov\t%[old], %[old2]\n\t"				      \
-	      "mov\t%[tmp], #0xffff0fff\n\t"				      \
-	      "mov\tlr, pc\n\t"						      \
-	      "add\tpc, %[tmp], #(0xffff0fc0 - 0xffff0fff)\n\t"		      \
-	      "bcc\t0b\n\t"						      \
-	      "mov\t%[tmp], %[old2]\n\t"				      \
-	      "1:"							      \
-	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
-	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
-		[old2] "r" (a_oldval2)					      \
-	      : "ip", "lr", "cc", "memory");				      \
-     (__typeof (oldval)) a_tmp; })
-#endif
-
-#include <sysdeps/arm/atomic-machine.h>
diff --git a/sysdeps/unix/sysv/linux/hppa/atomic-machine.h b/sysdeps/unix/sysv/linux/hppa/atomic-machine.h
index 393a056ece1add048f574f720cfdc71015964efa..37604c2963e09da41404324276324d71bb8db583 100644
--- a/sysdeps/unix/sysv/linux/hppa/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/hppa/atomic-machine.h
@@ -18,8 +18,6 @@
 #ifndef _ATOMIC_MACHINE_H
 #define _ATOMIC_MACHINE_H	1
 
-#define atomic_full_barrier() __sync_synchronize ()
-
 #define __HAVE_64B_ATOMICS 0
 #define USE_ATOMIC_COMPILER_BUILTINS 0
 
diff --git a/sysdeps/unix/sysv/linux/loongarch/atomic-machine.h b/sysdeps/unix/sysv/linux/loongarch/atomic-machine.h
index d1b8f1c11b6cfa9e5b2c0b2c18813d34e93eba22..016af7b4fe00a07566aeb7d13d18a3463f384b65 100644
--- a/sysdeps/unix/sysv/linux/loongarch/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/loongarch/atomic-machine.h
@@ -19,8 +19,6 @@
 #ifndef _LINUX_LOONGARCH_BITS_ATOMIC_H
 #define _LINUX_LOONGARCH_BITS_ATOMIC_H 1
 
-#define atomic_full_barrier() __sync_synchronize ()
-
 #define __HAVE_64B_ATOMICS (__loongarch_grlen >= 64)
 #define USE_ATOMIC_COMPILER_BUILTINS 1
 #define ATOMIC_EXCHANGE_USES_CAS 0
diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
index 67467fe9d6de19060a0c2f53031a9c9af4dea102..17921c3020ccfd3fae32a86708aff68ab4b82036 100644
--- a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
@@ -48,7 +48,4 @@
     (__typeof (oldval)) _d0;						\
   })
 
-# define atomic_full_barrier()				\
-  (INTERNAL_SYSCALL_CALL (atomic_barrier), (void) 0)
-
 #endif
diff --git a/sysdeps/unix/sysv/linux/nios2/atomic-machine.h b/sysdeps/unix/sysv/linux/nios2/atomic-machine.h
index 951aa463797a5acb2f7360e79d8495edf9343130..d832d0c7a9fb6cc38cbd4e7241fd6b0abf289633 100644
--- a/sysdeps/unix/sysv/linux/nios2/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/nios2/atomic-machine.h
@@ -78,6 +78,4 @@
      r2;								\
   })
 
-#define atomic_full_barrier()  ({ asm volatile ("sync"); })
-
 #endif /* _NIOS2_ATOMIC_MACHINE_H */
diff --git a/sysdeps/unix/sysv/linux/riscv/atomic-machine.h b/sysdeps/unix/sysv/linux/riscv/atomic-machine.h
index 9ae89e0ef12ad28319755ac51260908779b9579f..9cee9d7d996f8bbbca6ee2b70b367beb717630ca 100644
--- a/sysdeps/unix/sysv/linux/riscv/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/riscv/atomic-machine.h
@@ -19,8 +19,6 @@
 #ifndef _LINUX_RISCV_BITS_ATOMIC_H
 #define _LINUX_RISCV_BITS_ATOMIC_H 1
 
-#define atomic_full_barrier() __sync_synchronize ()
-
 #ifdef __riscv_atomic
 
 # define __HAVE_64B_ATOMICS (__riscv_xlen >= 64)
diff --git a/sysdeps/x86/atomic-machine.h b/sysdeps/x86/atomic-machine.h
index f24f1c71ed718c601c71decc1ee0c4b49fdf32f8..84ba01c9fe05cb2cb6b4ca5784ec804f8f3fbe05 100644
--- a/sysdeps/x86/atomic-machine.h
+++ b/sysdeps/x86/atomic-machine.h
@@ -531,13 +531,6 @@
 
 #define catomic_or(mem, mask) __arch_or_body (__arch_cprefix, mem, mask)
 
-/* We don't use mfence because it is supposedly slower due to having to
-   provide stronger guarantees (e.g., regarding self-modifying code).  */
-#define atomic_full_barrier() \
-    __asm __volatile (LOCK_PREFIX "orl $0, (%%" SP_REG ")" ::: "memory")
-#define atomic_read_barrier() __asm ("" ::: "memory")
-#define atomic_write_barrier() __asm ("" ::: "memory")
-
 #define atomic_spin_nop() __asm ("pause")
 
 #endif /* atomic-machine.h */
diff mbox series

Patch

diff --git a/crypt/crypt_util.c b/crypt/crypt_util.c
index be925e3484e65d2180e07915f5d91b47f6b96393..a8c2b26ed13c27804a9465f2a54caa09aaec3814 100644
--- a/crypt/crypt_util.c
+++ b/crypt/crypt_util.c
@@ -453,14 +453,14 @@  __init_des_r (struct crypt_data * __restrict __data)
 	  efp[comes_from_word][word_value][o_long] |= mask2;
       }
     }
-    atomic_write_barrier ();
+    atomic_thread_fence_release ();
     small_tables_initialized = 1;
 #ifdef __GNU_LIBRARY__
 small_tables_done:
     __libc_lock_unlock(_ufc_tables_lock);
 #endif
   } else
-    atomic_read_barrier ();
+    atomic_thread_fence_acquire ();
 
   /*
    * Create the sb tables:
diff --git a/elf/dl-deps.c b/elf/dl-deps.c
index 06005a0cc8686cc7e63cd8e1b1e7deda01fe6688..11b3fda5fdeb3830d3d5a0031084b43847444e04 100644
--- a/elf/dl-deps.c
+++ b/elf/dl-deps.c
@@ -430,7 +430,7 @@  _dl_map_object_deps (struct link_map *map,
 	  memcpy (&l_initfini[1], needed, nneeded * sizeof needed[0]);
 	  memcpy (&l_initfini[nneeded + 1], l_initfini,
 		  nneeded * sizeof needed[0]);
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  l->l_initfini = l_initfini;
 	  l->l_free_initfini = 1;
 	}
@@ -555,12 +555,12 @@  _dl_map_object_deps (struct link_map *map,
 
   /* Terminate the list of dependencies.  */
   l_initfini[nlist] = NULL;
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   map->l_initfini = l_initfini;
   map->l_free_initfini = 1;
   if (l_reldeps != NULL)
     {
-      atomic_write_barrier ();
+      atomic_thread_fence_release ();
       void *old_l_reldeps = map->l_reldeps;
       map->l_reldeps = l_reldeps;
       _dl_scope_free (old_l_reldeps);
diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
index 02c63a7062b2be0f37a412160fdb2b3468cc70cf..894d3e7db198a2a08940e9a1d82c345b0e0343a0 100644
--- a/elf/dl-lookup.c
+++ b/elf/dl-lookup.c
@@ -695,7 +695,7 @@  marking %s [%lu] as NODELETE due to memory allocation failure\n",
 			l_reldepsact * sizeof (struct link_map *));
 	      newp->list[l_reldepsact] = map;
 	      newp->act = l_reldepsact + 1;
-	      atomic_write_barrier ();
+	      atomic_thread_fence_release ();
 	      void *old = undef_map->l_reldeps;
 	      undef_map->l_reldeps = newp;
 	      undef_map->l_reldepsmax = max;
@@ -706,7 +706,7 @@  marking %s [%lu] as NODELETE due to memory allocation failure\n",
       else
 	{
 	  undef_map->l_reldeps->list[l_reldepsact] = map;
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  undef_map->l_reldeps->act = l_reldepsact + 1;
 	}
 
diff --git a/elf/dl-open.c b/elf/dl-open.c
index a23e65926bcfe797f06f8b4175f65040f4547a05..ba77f4b774cae69c382bceb6599936664203ef05 100644
--- a/elf/dl-open.c
+++ b/elf/dl-open.c
@@ -202,7 +202,7 @@  add_to_global_update (struct link_map *new)
   assert (added <= ns->_ns_global_scope_pending_adds);
   ns->_ns_global_scope_pending_adds -= added;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   ns->_ns_main_searchlist->r_nlist = new_nlist;
 }
 
@@ -342,7 +342,7 @@  update_scopes (struct link_map *new)
 	     might use the new last element and then use the garbage
 	     at offset IDX+1.  */
 	  imap->l_scope[cnt + 1] = NULL;
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 	  imap->l_scope[cnt] = &new->l_searchlist;
 
 	  from_scope = cnt;
diff --git a/include/atomic.h b/include/atomic.h
index 8eb56362ba18eb4836070930d5f2e769fb6a0a1e..dfe60ddb27fed1a06fb4967968e7d5d64de2f9c3 100644
--- a/include/atomic.h
+++ b/include/atomic.h
@@ -104,21 +104,6 @@ 
 #endif
 
 
-#ifndef atomic_full_barrier
-# define atomic_full_barrier() __sync_synchronize()
-#endif
-
-
-#ifndef atomic_read_barrier
-# define atomic_read_barrier() atomic_full_barrier ()
-#endif
-
-
-#ifndef atomic_write_barrier
-# define atomic_write_barrier() atomic_full_barrier ()
-#endif
-
-
 /* This is equal to 1 iff the architecture supports 64b atomic operations.  */
 #ifndef __HAVE_64B_ATOMICS
 #error Unable to determine if 64-bit atomics are present.
diff --git a/include/list.h b/include/list.h
index 7bea2c50a3759c0c8640971eff1e80874e3b543f..31a8a93fa2491b38c368b9e07dd65c473e4eb19f 100644
--- a/include/list.h
+++ b/include/list.h
@@ -43,7 +43,7 @@  list_add (list_t *newp, list_t *head)
   newp->next = head->next;
   newp->prev = head;
   head->next->prev = newp;
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   head->next = newp;
 }
 
diff --git a/malloc/arena.c b/malloc/arena.c
index 7c74a18381a4be5fe6bcb94b38a62dbfa6b674f4..3ef52ade3aeaaba6bb5e5d4e70024202e0dc162b 100644
--- a/malloc/arena.c
+++ b/malloc/arena.c
@@ -111,7 +111,7 @@  static mstate free_list;
    malloc_state objects.
 
    Read access to the next member is supposed to synchronize with the
-   atomic_write_barrier and the write to the next member in
+   atomic_thread_fence_release and the write to the next member in
    _int_new_arena.  This suffers from data races; see the FIXME
    comments in _int_new_arena and reused_arena.
 
@@ -778,7 +778,7 @@  _int_new_arena (size_t size)
   /* FIXME: The barrier is an attempt to synchronize with read access
      in reused_arena, which does not acquire list_lock while
      traversing the list.  */
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   main_arena.next = a;
 
   __libc_lock_unlock (list_lock);
diff --git a/manual/llio.texi b/manual/llio.texi
index 1b801ee817db2935d8866894be23ffa516690ca3..eb8711a0f8e28281baf68206b7891f16f662de0b 100644
--- a/manual/llio.texi
+++ b/manual/llio.texi
@@ -2543,14 +2543,14 @@  aiocb64}, since the LFS transparently replaces the old interface.
 @c    deallocate_stack @asulock @ascuheap @aculock @acsmem
 @c     lll_lock (state_cache_lock) @asulock @aculock
 @c     stack_list_del ok
-@c      atomic_write_barrier ok
+@c      atomic_thread_fence_release ok
 @c      list_del ok
-@c      atomic_write_barrier ok
+@c      atomic_thread_fence_release ok
 @c     queue_stack @ascuheap @acsmem
 @c      stack_list_add ok
-@c       atomic_write_barrier ok
+@c       atomic_thread_fence_release ok
 @c       list_add ok
-@c       atomic_write_barrier ok
+@c       atomic_thread_fence_release ok
 @c      free_stacks @ascuheap @acsmem
 @c       list_for_each_prev_safe ok
 @c       list_entry ok
diff --git a/manual/memory.texi b/manual/memory.texi
index 110e736a64c667988f4ca2fe92deb409225a4a88..2dfd09ea4aace004067e2e1d51c9e1292d1f7452 100644
--- a/manual/memory.texi
+++ b/manual/memory.texi
@@ -395,7 +395,7 @@  this function is in @file{stdlib.h}.
 @c     mutex_init ok
 @c     mutex_lock (just-created mutex) ok, returns locked
 @c     mutex_lock (list_lock) dup @asulock @aculock
-@c     atomic_write_barrier ok
+@c     atomic_thread_fence_release ok
 @c     mutex_unlock (list_lock) @aculock
 @c    atomic_fetch_add_relaxed ok
 @c    reused_arena @asulock @aculock
diff --git a/manual/startup.texi b/manual/startup.texi
index 9bf24123f562f75ba27a4770c69147e003b94755..4c7c2976a92d772909203bca028347e8d798b2d7 100644
--- a/manual/startup.texi
+++ b/manual/startup.texi
@@ -947,7 +947,7 @@  using @code{atexit} or @code{on_exit}.
 @c     __libc_lock_lock @asulock @aculock
 @c     calloc dup @ascuheap @acsmem
 @c     __libc_lock_unlock @aculock
-@c    atomic_write_barrier dup ok
+@c    atomic_thread_fence_release dup ok
 The @code{atexit} function registers the function @var{function} to be
 called at normal program termination.  The @var{function} is called with
 no arguments.
@@ -961,7 +961,7 @@  the function cannot be registered.
 @safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{} @asulock{}}@acunsafe{@aculock{} @acsmem{}}}
 @c on_exit @ascuheap @asulock @aculock @acsmem
 @c  new_exitfn dup @ascuheap @asulock @aculock @acsmem
-@c  atomic_write_barrier dup ok
+@c  atomic_thread_fence_release dup ok
 This function is a somewhat more powerful variant of @code{atexit}.  It
 accepts two arguments, a function @var{function} and an arbitrary
 pointer @var{arg}.  At normal program termination, the @var{function} is
diff --git a/nptl/nptl-stack.c b/nptl/nptl-stack.c
index 20ce78eddbf100833d453d7032f63bc2ba8f01c7..7c04e7faaae5c15bf5ad98c32935bcba4849c1c1 100644
--- a/nptl/nptl-stack.c
+++ b/nptl/nptl-stack.c
@@ -27,11 +27,11 @@  __nptl_stack_list_del (list_t *elem)
 {
   GL (dl_in_flight_stack) = (uintptr_t) elem;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   list_del (elem);
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   GL (dl_in_flight_stack) = 0;
 }
@@ -42,11 +42,11 @@  __nptl_stack_list_add (list_t *elem, list_t *list)
 {
   GL (dl_in_flight_stack) = (uintptr_t) elem | 1;
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   list_add (elem, list);
 
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
 
   GL (dl_in_flight_stack) = 0;
 }
diff --git a/nptl/pthread_mutex_setprioceiling.c b/nptl/pthread_mutex_setprioceiling.c
index 2d71a750c8981e8ca271c265031887e2c510583a..b574a77250664bbb1487c932b9a32a9dee415072 100644
--- a/nptl/pthread_mutex_setprioceiling.c
+++ b/nptl/pthread_mutex_setprioceiling.c
@@ -113,7 +113,7 @@  __pthread_mutex_setprioceiling (pthread_mutex_t *mutex, int prioceiling,
     newlock = (mutex->__data.__lock & ~PTHREAD_MUTEX_PRIO_CEILING_MASK);
   mutex->__data.__lock = newlock
 			 | (prioceiling << PTHREAD_MUTEX_PRIO_CEILING_SHIFT);
-  atomic_full_barrier ();
+  atomic_thread_fence_seq_cst ();
 
   futex_wake ((unsigned int *)&mutex->__data.__lock, INT_MAX,
 	      PTHREAD_MUTEX_PSHARED (mutex));
diff --git a/nptl/sem_post.c b/nptl/sem_post.c
index 7ec21e92eb4c71d7f17764e96bc7603837f7522d..d4e37cb0888cb1004881e608b82147900bc420a5 100644
--- a/nptl/sem_post.c
+++ b/nptl/sem_post.c
@@ -90,7 +90,7 @@  __old_sem_post (sem_t *sem)
 
   /* We must need to synchronize with consumers of this token, so the atomic
      increment must have release MO semantics.  */
-  atomic_write_barrier ();
+  atomic_thread_fence_release ();
   atomic_fetch_add_release (futex, 1);
   /* We always have to assume it is a shared semaphore.  */
   futex_wake (futex, 1, LLL_SHARED);
diff --git a/stdlib/msort.c b/stdlib/msort.c
index cbe9a4a8fdb38113a4c18976c9f297be103d458f..e2f1eca94ad7e9005145c376b0de3dbd1ca14f18 100644
--- a/stdlib/msort.c
+++ b/stdlib/msort.c
@@ -197,7 +197,7 @@  __qsort_r (void *b, size_t n, size_t s, __compar_d_fn_t cmp, void *arg)
 	  phys_pages /= 4;
 
 	  /* Make sure phys_pages is written to memory.  */
-	  atomic_write_barrier ();
+	  atomic_thread_fence_release ();
 
 	  pagesize = __sysconf (_SC_PAGESIZE);
 	}
diff --git a/sysdeps/aarch64/nptl/tls.h b/sysdeps/aarch64/nptl/tls.h
index 08aa2eff891b7be32243e9955d998892807c7b2e..0e5b4ece6a118b4b066bd2fd024dc85e978cc786 100644
--- a/sysdeps/aarch64/nptl/tls.h
+++ b/sysdeps/aarch64/nptl/tls.h
@@ -108,7 +108,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/alpha/atomic-machine.h b/sysdeps/alpha/atomic-machine.h
index f384a2bf0b3376cf240dc25d501e1d64a94bffe1..7fbe5b87eebf323d38fe1349b02aa56fe199cab3 100644
--- a/sysdeps/alpha/atomic-machine.h
+++ b/sysdeps/alpha/atomic-machine.h
@@ -21,7 +21,3 @@ 
 
 /* XXX Is this actually correct?  */
 #define ATOMIC_EXCHANGE_USES_CAS 1
-
-#define atomic_full_barrier()	__asm ("mb" : : : "memory");
-#define atomic_read_barrier()	__asm ("mb" : : : "memory");
-#define atomic_write_barrier()	__asm ("wmb" : : : "memory");
diff --git a/sysdeps/alpha/nptl/tls.h b/sysdeps/alpha/nptl/tls.h
index 8f5b69ad3b1b0c557fa1bae55278547572a374cc..914dba422c50e4531d22eb459b41c8b958a75263 100644
--- a/sysdeps/alpha/nptl/tls.h
+++ b/sysdeps/alpha/nptl/tls.h
@@ -105,7 +105,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/arc/nptl/tls.h b/sysdeps/arc/nptl/tls.h
index 7fc6602b236fa2455f8de4a0540442ae85d27c98..b2749f81d2980502043f507bf7c81da48f17aa9f 100644
--- a/sysdeps/arc/nptl/tls.h
+++ b/sysdeps/arc/nptl/tls.h
@@ -113,7 +113,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/arm/nptl/tls.h b/sysdeps/arm/nptl/tls.h
index 7657ca3dccc2d929c71236d42fc060a4b4902e2b..b1389ba034966aff17692f2b6d0e7b04a0baf9a0 100644
--- a/sysdeps/arm/nptl/tls.h
+++ b/sysdeps/arm/nptl/tls.h
@@ -99,7 +99,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/csky/nptl/tls.h b/sysdeps/csky/nptl/tls.h
index 58d6ab0fb2ae90de50cffd5b4a98426c6a793050..ac54606c3c0e28c1c8d57a6475d9249ca3566abe 100644
--- a/sysdeps/csky/nptl/tls.h
+++ b/sysdeps/csky/nptl/tls.h
@@ -128,7 +128,7 @@  typedef struct
   do									      \
     {									      \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	      \
-      atomic_write_barrier ();						      \
+      atomic_thread_fence_release ();					      \
     }									      \
   while (0)
 
diff --git a/sysdeps/hppa/dl-fptr.c b/sysdeps/hppa/dl-fptr.c
index 40bf5cd3b306315d8eeb6bdba2b2b46b1ea5059e..0562467d6f79f76b78b2cf169fdd059a993296d3 100644
--- a/sysdeps/hppa/dl-fptr.c
+++ b/sysdeps/hppa/dl-fptr.c
@@ -369,7 +369,7 @@  _dl_lookup_address (const void *address)
 
   /* First load the relocation offset.  */
   reloc_arg = (ElfW(Word)) desc[1];
-  atomic_full_barrier();
+  atomic_thread_fence_seq_cst ();
 
   /* Then load first word of candidate descriptor.  It should be a pointer
      with word alignment and point to memory that can be read.  */
diff --git a/sysdeps/hppa/dl-machine.h b/sysdeps/hppa/dl-machine.h
index c865713be1e3f8e0430bbb35c8db7ebe3e7a6abf..61635ca9115e1fa77305eaa3cc4ab5bf9bb91d7b 100644
--- a/sysdeps/hppa/dl-machine.h
+++ b/sysdeps/hppa/dl-machine.h
@@ -136,7 +136,7 @@  elf_machine_fixup_plt (struct link_map *map, lookup_t t,
       /* Need to ensure that the gp is visible before the code
          entry point is updated */
       rfdesc[1] = value.gp;
-      atomic_full_barrier();
+      atomic_thread_fence_seq_cst ();
       rfdesc[0] = value.ip;
     }
   else
diff --git a/sysdeps/hppa/nptl/tls.h b/sysdeps/hppa/nptl/tls.h
index e6b0bd5c7182b497aaf0d2bb08f62551a223c403..5b2495637b8aec9df3c8b3d47a2b46aa632c84da 100644
--- a/sysdeps/hppa/nptl/tls.h
+++ b/sysdeps/hppa/nptl/tls.h
@@ -133,7 +133,7 @@  static inline void __set_cr27(struct pthread *cr27)
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/htl/pt-once.c b/sysdeps/htl/pt-once.c
index b85b196645958fc7f47b08b39e91077b82817cdc..55db6c3d7176a9ca6cf2d0caccc1cf01aeb5ea2f 100644
--- a/sysdeps/htl/pt-once.c
+++ b/sysdeps/htl/pt-once.c
@@ -33,7 +33,7 @@  __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 {
   ASSERT_TYPE_SIZE (pthread_once_t, __SIZEOF_PTHREAD_ONCE_T);
 
-  atomic_full_barrier ();
+  atomic_thread_fence_seq_cst ();
   if (once_control->__run == 0)
     {
       __pthread_spin_wait (&once_control->__lock);
@@ -44,7 +44,7 @@  __pthread_once (pthread_once_t *once_control, void (*init_routine) (void))
 	  init_routine ();
 	  pthread_cleanup_pop (0);
 
-	  atomic_full_barrier ();
+	  atomic_thread_fence_seq_cst ();
 	  once_control->__run = 1;
 	}
 
diff --git a/sysdeps/ia64/nptl/tls.h b/sysdeps/ia64/nptl/tls.h
index d2411b3c1ac29733c0bb3683d83388e2e0e8e277..7709e644ee04ebc935dc659806481eebcb4129f0 100644
--- a/sysdeps/ia64/nptl/tls.h
+++ b/sysdeps/ia64/nptl/tls.h
@@ -157,7 +157,7 @@  register struct pthread *__thread_self __asm__("r13");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/m68k/nptl/tls.h b/sysdeps/m68k/nptl/tls.h
index 742e1b6767d99fa6011ac1d207264c7b82e53787..dfba7a568016b8e10dac6c21d65c785eaab12a09 100644
--- a/sysdeps/m68k/nptl/tls.h
+++ b/sysdeps/m68k/nptl/tls.h
@@ -132,7 +132,7 @@  extern void * __m68k_read_tp (void);
   do									\
     {									\
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
-      atomic_write_barrier ();						\
+      atomic_thread_fence_release ();					\
     }									\
   while (0)
 
diff --git a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
index 71f789cbdfa20b79f9e3cef5a2523d4e243b4f19..feb9085af9499b5d954a1bf4f04d9cbcb804e8cf 100644
--- a/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
+++ b/sysdeps/mach/hurd/htl/pt-mutex-destroy.c
@@ -26,7 +26,7 @@ 
 int
 __pthread_mutex_destroy (pthread_mutex_t *mtxp)
 {
-  atomic_read_barrier ();
+  atomic_thread_fence_acquire ();
   if (*(volatile unsigned int *) &mtxp->__lock != 0)
     return EBUSY;
 
diff --git a/sysdeps/mach/hurd/htl/pt-mutex.h b/sysdeps/mach/hurd/htl/pt-mutex.h
index 4021e72a6e8d15316336296ff732a4e7fd1acdff..ebdf8a5fbde0f755c7625cd38ad185ee0c977b5f 100644
--- a/sysdeps/mach/hurd/htl/pt-mutex.h
+++ b/sysdeps/mach/hurd/htl/pt-mutex.h
@@ -54,7 +54,7 @@ 
           if (ret == EOWNERDEAD)   \
             {   \
               mtxp->__lock = mtxp->__lock | LLL_DEAD_OWNER;   \
-              atomic_write_barrier ();   \
+              atomic_thread_fence_release ();   \
             }   \
         }   \
     }   \
diff --git a/sysdeps/microblaze/nptl/tls.h b/sysdeps/microblaze/nptl/tls.h
index 588fd1c5d63ee4e6a1b284cc19e216b6730a2091..30e5d628be8b78cf9c7b8e9386ab1b2355819f4a 100644
--- a/sysdeps/microblaze/nptl/tls.h
+++ b/sysdeps/microblaze/nptl/tls.h
@@ -110,7 +110,7 @@  typedef struct
   do                                                                        \
     {                                                                       \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;            \
-      atomic_write_barrier ();                                              \
+      atomic_thread_fence_release ();                                       \
     }                                                                       \
   while (0)
 
diff --git a/sysdeps/mips/nptl/tls.h b/sysdeps/mips/nptl/tls.h
index 2aa7cb4bb8d0b5a31889aa33d5751104ff1e4f45..e4c5d2a876db7943b38daa270f4681d17b441c58 100644
--- a/sysdeps/mips/nptl/tls.h
+++ b/sysdeps/mips/nptl/tls.h
@@ -160,7 +160,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/nios2/nptl/tls.h b/sysdeps/nios2/nptl/tls.h
index cb231e2a4bbfa52495c4c017a7d3e1c6dd7937ca..50dbbef119af30112136a9bef187c79037d0849d 100644
--- a/sysdeps/nios2/nptl/tls.h
+++ b/sysdeps/nios2/nptl/tls.h
@@ -140,7 +140,7 @@  register struct pthread *__thread_self __asm__("r23");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/or1k/nptl/tls.h b/sysdeps/or1k/nptl/tls.h
index e82f444738de222c0e4866d5a2ab8191ce99ddc9..886c017be12d06ee63198a36917c59e24be77a60 100644
--- a/sysdeps/or1k/nptl/tls.h
+++ b/sysdeps/or1k/nptl/tls.h
@@ -175,7 +175,7 @@  register tcbhead_t *__thread_self __asm__("r10");
   do									\
     {									\
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	\
-      atomic_write_barrier ();						\
+      atomic_thread_fence_release ();					\
     }									\
   while (0)
 
diff --git a/sysdeps/powerpc/nptl/tls.h b/sysdeps/powerpc/nptl/tls.h
index e62a96238aa95c79ac1f749b4dbf03985b6e15d4..c8d233a7347f609b4cdbffb5dafa2f55e18ac18e 100644
--- a/sysdeps/powerpc/nptl/tls.h
+++ b/sysdeps/powerpc/nptl/tls.h
@@ -224,7 +224,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/powerpc/powerpc32/atomic-machine.h b/sysdeps/powerpc/powerpc32/atomic-machine.h
index f72d4be13709e38006255d236efb0e94f3976e68..6a2aae8bdb34281144e6810924377a6a62857d15 100644
--- a/sysdeps/powerpc/powerpc32/atomic-machine.h
+++ b/sysdeps/powerpc/powerpc32/atomic-machine.h
@@ -36,24 +36,3 @@ 
 
 #define __HAVE_64B_ATOMICS 0
 #define ATOMIC_EXCHANGE_USES_CAS 1
-
-#ifdef _ARCH_PWR4
-/*
- * Newer powerpc64 processors support the new "light weight" sync (lwsync)
- * So if the build is using -mcpu=[power4,power5,power5+,970] we can
- * safely use lwsync.
- */
-# define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
-/*
- * "light weight" sync can also be used for the release barrier.
- */
-# define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
-#else
-/*
- * Older powerpc32 processors don't support the new "light weight"
- * sync (lwsync).  So the only safe option is to use normal sync
- * for all powerpc32 applications.
- */
-# define atomic_read_barrier()	__asm ("sync" ::: "memory")
-# define atomic_write_barrier()	__asm ("sync" ::: "memory")
-#endif
diff --git a/sysdeps/powerpc/powerpc64/atomic-machine.h b/sysdeps/powerpc/powerpc64/atomic-machine.h
index fcb1592be9ad6a3981f56c513deac2f5f8ac5bb7..2932f889c5bc6d0fa49d5ad36875b50c27ad07e9 100644
--- a/sysdeps/powerpc/powerpc64/atomic-machine.h
+++ b/sysdeps/powerpc/powerpc64/atomic-machine.h
@@ -36,12 +36,3 @@ 
 
 #define __HAVE_64B_ATOMICS 1
 #define ATOMIC_EXCHANGE_USES_CAS 1
-
-/*
- * All powerpc64 processors support the new "light weight"  sync (lwsync).
- */
-#define atomic_read_barrier()	__asm ("lwsync" ::: "memory")
-/*
- * "light weight" sync can also be used for the release barrier.
- */
-#define atomic_write_barrier()	__asm ("lwsync" ::: "memory")
diff --git a/sysdeps/riscv/nptl/tls.h b/sysdeps/riscv/nptl/tls.h
index 700c2f51899b0385d7ebaa4810c84de4fa6f2b45..020a986ceee89e1feb8f76c51f224a8faea71bbb 100644
--- a/sysdeps/riscv/nptl/tls.h
+++ b/sysdeps/riscv/nptl/tls.h
@@ -123,7 +123,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/s390/nptl/tls.h b/sysdeps/s390/nptl/tls.h
index 98d7870148ce6bc1d6397b1465dfabe96f7280b2..f1664d9ade6fd562db38c3dddd3fa6237a47faea 100644
--- a/sysdeps/s390/nptl/tls.h
+++ b/sysdeps/s390/nptl/tls.h
@@ -167,7 +167,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/sh/nptl/tls.h b/sysdeps/sh/nptl/tls.h
index 1530489a6ce4286bc5146e6cd83e3b463b965467..00ae1b998b9e0b1e6c347c4be4e99a90e530e924 100644
--- a/sysdeps/sh/nptl/tls.h
+++ b/sysdeps/sh/nptl/tls.h
@@ -139,7 +139,7 @@  typedef struct
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/sparc/atomic-machine.h b/sysdeps/sparc/atomic-machine.h
index a7042f1ee546b9f238153cb923409d42eb45cc03..1f0eb0a9b1171c06dc19dc21c4fe7de94adc4bce 100644
--- a/sysdeps/sparc/atomic-machine.h
+++ b/sysdeps/sparc/atomic-machine.h
@@ -29,13 +29,6 @@ 
 #define ATOMIC_EXCHANGE_USES_CAS     __HAVE_64B_ATOMICS
 
 #ifdef __sparc_v9__
-# define atomic_full_barrier() \
-  __asm __volatile ("membar #LoadLoad | #LoadStore"			      \
-		    " | #StoreLoad | #StoreStore" : : : "memory")
-# define atomic_read_barrier() \
-  __asm __volatile ("membar #LoadLoad | #LoadStore" : : : "memory")
-# define atomic_write_barrier() \
-  __asm __volatile ("membar #LoadStore | #StoreStore" : : : "memory")
 
 extern void __cpu_relax (void);
 # define atomic_spin_nop() __cpu_relax ()
diff --git a/sysdeps/sparc/nptl/tls.h b/sysdeps/sparc/nptl/tls.h
index 95a69cb8249dc79c3a063637a21d976d2660c48f..bc7ada0d3fe66751506e1cf5516f14ec2c205af8 100644
--- a/sysdeps/sparc/nptl/tls.h
+++ b/sysdeps/sparc/nptl/tls.h
@@ -140,7 +140,7 @@  register struct pthread *__thread_self __asm__("%g7");
   do									     \
     {									     \
       THREAD_SELF->header.gscope_flag = THREAD_GSCOPE_FLAG_USED;	     \
-      atomic_write_barrier ();						     \
+      atomic_thread_fence_release ();					     \
     }									     \
   while (0)
 
diff --git a/sysdeps/unix/sysv/linux/arm/atomic-machine.h b/sysdeps/unix/sysv/linux/arm/atomic-machine.h
deleted file mode 100644
index 20068c72f359442769f8d49e11f7e771c922ef0b..0000000000000000000000000000000000000000
--- a/sysdeps/unix/sysv/linux/arm/atomic-machine.h
+++ /dev/null
@@ -1,115 +0,0 @@ 
-/* Atomic operations.  ARM/Linux version.
-   Copyright (C) 2002-2022 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C 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.
-
-   The GNU C 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 the GNU C Library.  If not, see
-   <https://www.gnu.org/licenses/>.  */
-
-#include <stdint.h>
-
-/* If the compiler doesn't provide a primitive, we'll use this macro
-   to get assistance from the kernel.  */
-#ifdef __thumb2__
-# define __arm_assisted_full_barrier() \
-     __asm__ __volatile__						      \
-	     ("movw\tip, #0x0fa0\n\t"					      \
-	      "movt\tip, #0xffff\n\t"					      \
-	      "blx\tip"							      \
-	      : : : "ip", "lr", "cc", "memory");
-#else
-# define __arm_assisted_full_barrier() \
-     __asm__ __volatile__						      \
-	     ("mov\tip, #0xffff0fff\n\t"				      \
-	      "mov\tlr, pc\n\t"						      \
-	      "add\tpc, ip, #(0xffff0fa0 - 0xffff0fff)"			      \
-	      : : : "ip", "lr", "cc", "memory");
-#endif
-
-/* Atomic compare and exchange.  This sequence relies on the kernel to
-   provide a compare and exchange operation which is atomic on the
-   current architecture, either via cleverness on pre-ARMv6 or via
-   ldrex / strex on ARMv6.
-
-   It doesn't matter what register is used for a_oldval2, but we must
-   specify one to work around GCC PR rtl-optimization/21223.  Otherwise
-   it may cause a_oldval or a_tmp to be moved to a different register.
-
-   We use the union trick rather than simply using __typeof (...) in the
-   declarations of A_OLDVAL et al because when NEWVAL or OLDVAL is of the
-   form *PTR and PTR has a 'volatile ... *' type, then __typeof (*PTR) has
-   a 'volatile ...' type and this triggers -Wvolatile-register-var to
-   complain about 'register volatile ... asm ("reg")'.
-
-   We use the same union trick in the declaration of A_PTR because when
-   MEM is of the from *PTR and PTR has a 'const ... *' type, then __typeof
-   (*PTR) has a 'const ...' type and this enables the compiler to substitute
-   the variable with its initializer in asm statements, which may cause the
-   corresponding operand to appear in a different register.  */
-#ifdef __thumb2__
-/* Thumb-2 has ldrex/strex.  However it does not have barrier instructions,
-   so we still need to use the kernel helper.  */
-# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
-     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
-     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
-     register uint32_t a_oldval asm ("r0");				      \
-     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
-     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
-     register uint32_t a_tmp asm ("r3");				      \
-     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
-     __asm__ __volatile__						      \
-	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
-	      "cmp\t%[tmp], %[old2]\n\t"				      \
-	      "bne\t1f\n\t"						      \
-	      "mov\t%[old], %[old2]\n\t"				      \
-	      "movw\t%[tmp], #0x0fc0\n\t"				      \
-	      "movt\t%[tmp], #0xffff\n\t"				      \
-	      "blx\t%[tmp]\n\t"						      \
-	      "bcc\t0b\n\t"						      \
-	      "mov\t%[tmp], %[old2]\n\t"				      \
-	      "1:"							      \
-	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
-	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
-		[old2] "r" (a_oldval2)					      \
-	      : "ip", "lr", "cc", "memory");				      \
-     (__typeof (oldval)) a_tmp; })
-#else
-# define __arm_assisted_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ union { __typeof (mem) a; uint32_t v; } mem_arg = { .a = (mem) };       \
-     union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
-     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
-     register uint32_t a_oldval asm ("r0");				      \
-     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
-     register uint32_t a_ptr asm ("r2") = mem_arg.v;			      \
-     register uint32_t a_tmp asm ("r3");				      \
-     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
-     __asm__ __volatile__						      \
-	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
-	      "cmp\t%[tmp], %[old2]\n\t"				      \
-	      "bne\t1f\n\t"						      \
-	      "mov\t%[old], %[old2]\n\t"				      \
-	      "mov\t%[tmp], #0xffff0fff\n\t"				      \
-	      "mov\tlr, pc\n\t"						      \
-	      "add\tpc, %[tmp], #(0xffff0fc0 - 0xffff0fff)\n\t"		      \
-	      "bcc\t0b\n\t"						      \
-	      "mov\t%[tmp], %[old2]\n\t"				      \
-	      "1:"							      \
-	      : [old] "=&r" (a_oldval), [tmp] "=&r" (a_tmp)		      \
-	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
-		[old2] "r" (a_oldval2)					      \
-	      : "ip", "lr", "cc", "memory");				      \
-     (__typeof (oldval)) a_tmp; })
-#endif
-
-#include <sysdeps/arm/atomic-machine.h>
diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
index 6f83fb2965bd162f0f76e0e3586472ade39af607..02e54847a42bfbc93ae1e07b7e32965be644daba 100644
--- a/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
+++ b/sysdeps/unix/sysv/linux/m68k/coldfire/atomic-machine.h
@@ -25,7 +25,4 @@ 
 /* XXX Is this actually correct?  */
 #define ATOMIC_EXCHANGE_USES_CAS 1
 
-# define atomic_full_barrier()				\
-  (INTERNAL_SYSCALL_CALL (atomic_barrier), (void) 0)
-
 #endif
diff --git a/sysdeps/x86/atomic-machine.h b/sysdeps/x86/atomic-machine.h
index b9be51c52d8cbef2a95a62192c8ef7011e7f2c12..98541a2d06ff5e4aa8c789ab7405215097471971 100644
--- a/sysdeps/x86/atomic-machine.h
+++ b/sysdeps/x86/atomic-machine.h
@@ -32,9 +32,6 @@ 
 #endif
 #define ATOMIC_EXCHANGE_USES_CAS	0
 
-#define atomic_read_barrier() __asm ("" ::: "memory")
-#define atomic_write_barrier() __asm ("" ::: "memory")
-
 #define atomic_spin_nop() __asm ("pause")
 
 #endif /* atomic-machine.h */