diff --git a/src/pl/plperl/SPI.xs b/src/pl/plperl/SPI.xs
index 6b8dcf62990..0447c50df19 100644
--- a/src/pl/plperl/SPI.xs
+++ b/src/pl/plperl/SPI.xs
@@ -41,7 +41,7 @@ do_plperl_return_next(SV *sv)
 		FlushErrorState();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 	}
 	PG_END_TRY();
 }
diff --git a/src/pl/plperl/Util.xs b/src/pl/plperl/Util.xs
index b2e0dfcf75d..8c3c47fec9f 100644
--- a/src/pl/plperl/Util.xs
+++ b/src/pl/plperl/Util.xs
@@ -58,7 +58,7 @@ do_util_elog(int level, SV *msg)
 			pfree(cmsg);
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 	}
 	PG_END_TRY();
 }
diff --git a/src/pl/plperl/expected/plperl_elog.out b/src/pl/plperl/expected/plperl_elog.out
index c447fa22cbc..d2ac1eb71d4 100644
--- a/src/pl/plperl/expected/plperl_elog.out
+++ b/src/pl/plperl/expected/plperl_elog.out
@@ -104,3 +104,16 @@ PL/Perl function "indirect_die_caller"
                    2
 (1 row)
 
+-- Test non-ASCII error messages
+--
+-- Note: this test case is known to fail if the database encoding is
+-- EUC_CN, EUC_JP, EUC_KR, or EUC_TW, for lack of any equivalent to
+-- U+00A0 (no-break space) in those encodings.  However, testing with
+-- plain ASCII data would be rather useless, so we must live with that.
+SET client_encoding TO UTF8;
+create or replace function error_with_nbsp() returns void language plperl as $$
+  elog(ERROR, "this message contains a no-break space");
+$$;
+select error_with_nbsp();
+ERROR:  this message contains a no-break space at line 2.
+CONTEXT:  PL/Perl function "error_with_nbsp"
diff --git a/src/pl/plperl/expected/plperl_elog_1.out b/src/pl/plperl/expected/plperl_elog_1.out
index 0932fde2668..136e8210991 100644
--- a/src/pl/plperl/expected/plperl_elog_1.out
+++ b/src/pl/plperl/expected/plperl_elog_1.out
@@ -104,3 +104,16 @@ PL/Perl function "indirect_die_caller"
                    2
 (1 row)
 
+-- Test non-ASCII error messages
+--
+-- Note: this test case is known to fail if the database encoding is
+-- EUC_CN, EUC_JP, EUC_KR, or EUC_TW, for lack of any equivalent to
+-- U+00A0 (no-break space) in those encodings.  However, testing with
+-- plain ASCII data would be rather useless, so we must live with that.
+SET client_encoding TO UTF8;
+create or replace function error_with_nbsp() returns void language plperl as $$
+  elog(ERROR, "this message contains a no-break space");
+$$;
+select error_with_nbsp();
+ERROR:  this message contains a no-break space at line 2.
+CONTEXT:  PL/Perl function "error_with_nbsp"
diff --git a/src/pl/plperl/plperl.c b/src/pl/plperl/plperl.c
index d6ec1c4baf4..27e7022b11e 100644
--- a/src/pl/plperl/plperl.c
+++ b/src/pl/plperl/plperl.c
@@ -2876,7 +2876,7 @@ plperl_spi_exec(char *query, int limit)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
@@ -3109,7 +3109,7 @@ plperl_spi_query(char *query)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
@@ -3195,7 +3195,7 @@ plperl_spi_fetchrow(char *cursor)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
@@ -3370,7 +3370,7 @@ plperl_spi_prepare(char *query, int argc, SV **argv)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
@@ -3511,7 +3511,7 @@ plperl_spi_exec_prepared(char *query, HV *attr, int argc, SV **argv)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
@@ -3640,7 +3640,7 @@ plperl_spi_query_prepared(char *query, int argc, SV **argv)
 		SPI_restore_connection();
 
 		/* Punt the error to Perl */
-		croak("%s", edata->message);
+		croak_cstr(edata->message);
 
 		/* Can't get here, but keep compiler quiet */
 		return NULL;
diff --git a/src/pl/plperl/plperl_helpers.h b/src/pl/plperl/plperl_helpers.h
index ed99194ed1e..eb08cd51518 100644
--- a/src/pl/plperl/plperl_helpers.h
+++ b/src/pl/plperl/plperl_helpers.h
@@ -132,4 +132,42 @@ cstr2sv(const char *str)
 	return sv;
 }
 
+/*
+ * croak() with specified message, which is given in the database encoding.
+ *
+ * Ideally we'd just write croak("%s", str), but plain croak() does not play
+ * nice with non-ASCII data.  In modern Perl versions we can call cstr2sv()
+ * and pass the result to croak_sv(); in versions that don't have croak_sv(),
+ * we have to work harder.
+ */
+static inline void
+croak_cstr(const char *str)
+{
+#ifdef croak_sv
+	/* Use sv_2mortal() to be sure the transient SV gets freed */
+	croak_sv(sv_2mortal(cstr2sv(str)));
+#else
+
+	/*
+	 * The older way to do this is to assign a UTF8-marked value to ERRSV and
+	 * then call croak(NULL).  But if we leave it to croak() to append the
+	 * error location, it does so too late (only after popping the stack) in
+	 * some Perl versions.  Hence, use mess() to create an SV with the error
+	 * location info already appended.
+	 */
+	SV		   *errsv = get_sv("@", GV_ADD);
+	char	   *utf8_str = utf_e2u(str);
+	SV		   *ssv;
+
+	ssv = mess("%s", utf8_str);
+	SvUTF8_on(ssv);
+
+	pfree(utf8_str);
+
+	sv_setsv(errsv, ssv);
+
+	croak(NULL);
+#endif   /* croak_sv */
+}
+
 #endif   /* PL_PERL_HELPERS_H */
diff --git a/src/pl/plperl/sql/plperl_elog.sql b/src/pl/plperl/sql/plperl_elog.sql
index 032fd8b8ba7..9ea1350069b 100644
--- a/src/pl/plperl/sql/plperl_elog.sql
+++ b/src/pl/plperl/sql/plperl_elog.sql
@@ -76,3 +76,18 @@ return $a + $b;
 $$;
 
 select indirect_die_caller();
+
+-- Test non-ASCII error messages
+--
+-- Note: this test case is known to fail if the database encoding is
+-- EUC_CN, EUC_JP, EUC_KR, or EUC_TW, for lack of any equivalent to
+-- U+00A0 (no-break space) in those encodings.  However, testing with
+-- plain ASCII data would be rather useless, so we must live with that.
+
+SET client_encoding TO UTF8;
+
+create or replace function error_with_nbsp() returns void language plperl as $$
+  elog(ERROR, "this message contains a no-break space");
+$$;
+
+select error_with_nbsp();