Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitf998948

Browse files
committed
Impose a full barrier in generic-xlc.h atomics functions.
pg_atomic_compare_exchange_*_impl() were providing only the semantics ofan acquire barrier. Buildfarm members hornet and mandrill revealed thisdeficit beginning with commit008608b.While we have no report of symptoms in 9.5, we can't rule out thepossibility of certain compilers, hardware, or extension code relying onthese functions' specified barrier semantics. Back-patch to 9.5, wherecommitb64d92f introduced atomics.Reviewed by Andres Freund.
1 parent46e006d commitf998948

File tree

1 file changed

+22
-4
lines changed

1 file changed

+22
-4
lines changed

‎src/include/port/atomics/generic-xlc.h

Lines changed: 22 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -46,12 +46,22 @@ static inline bool
4646
pg_atomic_compare_exchange_u32_impl(volatilepg_atomic_uint32*ptr,
4747
uint32*expected,uint32newval)
4848
{
49+
boolret;
50+
51+
/*
52+
* atomics.h specifies sequential consistency ("full barrier semantics")
53+
* for this interface. Since "lwsync" provides acquire/release
54+
* consistency only, do not use it here. GCC atomics observe the same
55+
* restriction; see its rs6000_pre_atomic_barrier().
56+
*/
57+
__asm__ __volatile__ ("sync \n" :::"memory");
58+
4959
/*
5060
* XXX: __compare_and_swap is defined to take signed parameters, but that
5161
* shouldn't matter since we don't perform any arithmetic operations.
5262
*/
53-
boolret=__compare_and_swap((volatileint*)&ptr->value,
54-
(int*)expected, (int)newval);
63+
ret=__compare_and_swap((volatileint*)&ptr->value,
64+
(int*)expected, (int)newval);
5565

5666
/*
5767
* xlc's documentation tells us:
@@ -69,6 +79,10 @@ pg_atomic_compare_exchange_u32_impl(volatile pg_atomic_uint32 *ptr,
6979
staticinlineuint32
7080
pg_atomic_fetch_add_u32_impl(volatilepg_atomic_uint32*ptr,int32add_)
7181
{
82+
/*
83+
* __fetch_and_add() emits a leading "sync" and trailing "isync", thereby
84+
* providing sequential consistency. This is undocumented.
85+
*/
7286
return__fetch_and_add((volatileint*)&ptr->value,add_);
7387
}
7488

@@ -79,8 +93,12 @@ static inline bool
7993
pg_atomic_compare_exchange_u64_impl(volatilepg_atomic_uint64*ptr,
8094
uint64*expected,uint64newval)
8195
{
82-
boolret=__compare_and_swaplp((volatilelong*)&ptr->value,
83-
(long*)expected, (long)newval);
96+
boolret;
97+
98+
__asm__ __volatile__ ("sync \n" :::"memory");
99+
100+
ret=__compare_and_swaplp((volatilelong*)&ptr->value,
101+
(long*)expected, (long)newval);
84102

85103
__isync();
86104

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp