Call xlc __isync() after, not before, associated compare-and-swap.

Architecture reference material specifies this order, and s_lock.h
inline assembly agrees.  The former order failed to provide mutual
exclusion to lwlock.c and perhaps to other clients.  The two xlc
buildfarm members, hornet and mandrill, have failed sixteen times with
duplicate key errors involving pg_class_oid_index or pg_type_oid_index.
Back-patch to 9.5, where commit b64d92f1
introduced atomics.

Reviewed by Andres Freund and Tom Lane.

src/bin/pgbench/.gitignore

 /exprparse.c
 /exprscan.c
 /pgbench
+/tmp_check/

src/bin/pgbench/Makefile

@@ -40,3 +40,9 @@ clean distclean:
 
 maintainer-clean: distclean
 	rm -f exprparse.c exprscan.c
+
+check:
+	$(prove_check)
+
+installcheck:
+	$(prove_installcheck)

src/bin/pgbench/t/001_pgbench.pl

+use strict;
+use warnings;
+
+use PostgresNode;
+use TestLib;
+use Test::More tests => 3;
+
+# Test concurrent insertion into table with UNIQUE oid column.  DDL expects
+# GetNewOidWithIndex() to successfully avoid violating uniqueness for indexes
+# like pg_class_oid_index and pg_proc_oid_index.  This indirectly exercises
+# LWLock and spinlock concurrency.  This test makes a 5-MiB table.
+my $node = get_new_node('main');
+$node->init;
+$node->start;
+$node->psql('postgres',
+	    'CREATE UNLOGGED TABLE oid_tbl () WITH OIDS; '
+	  . 'ALTER TABLE oid_tbl ADD UNIQUE (oid);');
+my $script = $node->basedir . '/pgbench_script';
+append_to_file($script,
+	'INSERT INTO oid_tbl SELECT FROM generate_series(1,1000);');
+$node->command_like(
+	[   qw(pgbench --no-vacuum --client=5 --protocol=prepared
+		  --transactions=25 --file), $script ],
+	qr{processed: 125/125},
+	'concurrent OID generation');

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

@@ -40,19 +40,23 @@ static inline bool
 pg_atomic_compare_exchange_u32_impl(volatile pg_atomic_uint32 *ptr,
 									uint32 *expected, uint32 newval)
 {
+	/*
+	 * XXX: __compare_and_swap is defined to take signed parameters, but that
+	 * shouldn't matter since we don't perform any arithmetic operations.
+	 */
+	bool		ret = __compare_and_swap((volatile int*)&ptr->value,
+										 (int *)expected, (int)newval);
+
 	/*
 	 * xlc's documentation tells us:
 	 * "If __compare_and_swap is used as a locking primitive, insert a call to
 	 * the __isync built-in function at the start of any critical sections."
+	 *
+	 * The critical section begins immediately after __compare_and_swap().
 	 */
 	__isync();
 
-	/*
-	 * XXX: __compare_and_swap is defined to take signed parameters, but that
-	 * shouldn't matter since we don't perform any arithmetic operations.
-	 */
-	return __compare_and_swap((volatile int*)&ptr->value,
-							  (int *)expected, (int)newval);
+	return ret;
 }
 
 #define PG_HAVE_ATOMIC_FETCH_ADD_U32
@@ -69,10 +73,12 @@ static inline bool
 pg_atomic_compare_exchange_u64_impl(volatile pg_atomic_uint64 *ptr,
 									uint64 *expected, uint64 newval)
 {
+	bool		ret = __compare_and_swaplp((volatile long*)&ptr->value,
+										   (long *)expected, (long)newval);
+
 	__isync();
 
-	return __compare_and_swaplp((volatile long*)&ptr->value,
-								(long *)expected, (long)newval);;
+	return ret;
 }
 
 #define PG_HAVE_ATOMIC_FETCH_ADD_U64

src/tools/msvc/clean.bat

@@ -94,6 +94,7 @@ if exist src\bin\pg_basebackup\tmp_check rd /s /q src\bin\pg_basebackup\tmp_chec
 if exist src\bin\pg_config\tmp_check rd /s /q src\bin\pg_config\tmp_check
 if exist src\bin\pg_ctl\tmp_check rd /s /q src\bin\pg_ctl\tmp_check
 if exist src\bin\pg_rewind\tmp_check rd /s /q src\bin\pg_rewind\tmp_check
+if exist src\bin\pgbench\tmp_check rd /s /q src\bin\pgbench\tmp_check
 if exist src\bin\scripts\tmp_check rd /s /q src\bin\scripts\tmp_check
 
 REM Clean up datafiles built with contrib