[src/trunk]: src/sys/arch/aarch64/aarch64 Add a note about completion vs orde...

[skrll <skrll%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/72a031ba582a
branches:  trunk
changeset: 979122:72a031ba582a
user:      skrll <skrll%NetBSD.org@localhost>
date:      Mon Dec 14 19:32:29 2020 +0000

description:
Add a note about completion vs ordering barrier as well.

diffstat:

 sys/arch/aarch64/aarch64/bus_space.c |  8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)

diffs (29 lines):

diff -r 43048c1c6bce -r 72a031ba582a sys/arch/aarch64/aarch64/bus_space.c
--- a/sys/arch/aarch64/aarch64/bus_space.c      Mon Dec 14 19:25:28 2020 +0000
+++ b/sys/arch/aarch64/aarch64/bus_space.c      Mon Dec 14 19:32:29 2020 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: bus_space.c,v 1.14 2020/12/14 19:25:28 skrll Exp $ */
+/* $NetBSD: bus_space.c,v 1.15 2020/12/14 19:32:29 skrll Exp $ */
 
 /*
  * Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -27,7 +27,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(1, "$NetBSD: bus_space.c,v 1.14 2020/12/14 19:25:28 skrll Exp $");
+__KERNEL_RCSID(1, "$NetBSD: bus_space.c,v 1.15 2020/12/14 19:32:29 skrll Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -638,6 +638,10 @@
         * consulting the page tables), so just issue the barrier
         * unconditionally.  Chances are either it's necessary or the
         * cost is small in comparison to device register I/O.
+        *
+        * The bus_space(9) man page is not clear whether barriers
+        * should enforce ordering or completion. To be safe, use dsb
+        * (ensure completion) here instead of dmb (ordering).
         */
        switch (flags) {
        case BUS_SPACE_BARRIER_READ: