[src/trunk]: src/sys/arch/sandpoint At least KuroBox Classic and HG seem to h...

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

details:   https://anonhg.NetBSD.org/src/rev/92e33af164c1
branches:  trunk
changeset: 953928:92e33af164c1
user:      rin <rin%NetBSD.org@localhost>
date:      Thu Mar 25 05:35:50 2021 +0000

description:
At least KuroBox Classic and HG seem to have buggy UART, for which
ETXRDY interrupts are sometimes (once per a few hours?) lost; TX stalls
until next RX arrives.

Since interrupt-based I/O works just fine for other devices, I believe
this problem is due to UART and not to PIC or its driver.

In order to work around this bug, enable COM_HW_BROKEN_ETXRDY bit, as
well as carry out polling once per 10 sec or 1k tics (arbitrary, but
1 min seems too long).

XXX
Apply this for other MPC824x-based boards?

diffstat:

 sys/arch/sandpoint/conf/KUROBOX         |   5 +++--
 sys/arch/sandpoint/sandpoint/com_eumb.c |  17 +++++++++++++++--
 2 files changed, 18 insertions(+), 4 deletions(-)

diffs (57 lines):

diff -r 7d9b89cfa379 -r 92e33af164c1 sys/arch/sandpoint/conf/KUROBOX
--- a/sys/arch/sandpoint/conf/KUROBOX   Thu Mar 25 05:34:49 2021 +0000
+++ b/sys/arch/sandpoint/conf/KUROBOX   Thu Mar 25 05:35:50 2021 +0000
@@ -1,4 +1,4 @@
-#      $NetBSD: KUROBOX,v 1.8 2020/01/03 03:44:42 thorpej Exp $
+#      $NetBSD: KUROBOX,v 1.9 2021/03/25 05:35:50 rin Exp $
 #
 # KuroBox/LinkStation support
 #
@@ -7,7 +7,8 @@
 
 no com0
 no satmgr0
-com0   at eumb? unit 1                 # console at 0x4600
+com0   at eumb? unit 1 flags 0x1       # console at 0x4600
+                                       # XXX flags 0x1 to work around H/W bug
 #com1  at eumb? unit 0
 satmgr0 at eumb? unit 0                        # satmgr at 0x4500
 
diff -r 7d9b89cfa379 -r 92e33af164c1 sys/arch/sandpoint/sandpoint/com_eumb.c
--- a/sys/arch/sandpoint/sandpoint/com_eumb.c   Thu Mar 25 05:34:49 2021 +0000
+++ b/sys/arch/sandpoint/sandpoint/com_eumb.c   Thu Mar 25 05:35:50 2021 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: com_eumb.c,v 1.11 2021/03/17 14:50:11 rin Exp $ */
+/* $NetBSD: com_eumb.c,v 1.12 2021/03/25 05:35:50 rin Exp $ */
 
 /*-
  * Copyright (c) 2007 The NetBSD Foundation, Inc.
@@ -30,7 +30,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: com_eumb.c,v 1.11 2021/03/17 14:50:11 rin Exp $");
+__KERNEL_RCSID(0, "$NetBSD: com_eumb.c,v 1.12 2021/03/25 05:35:50 rin Exp $");
 
 #include <sys/param.h>
 #include <sys/device.h>
@@ -99,6 +99,19 @@
        sc->sc_frequency = 4 * ticks_per_sec;
        epicirq = (eaa->eumb_unit == 1) ? 25 : 24;
 
+       /*
+        * XXX
+        * At least KuroBox Classic and HG seem to have a hardware bug, by
+        * which ETXRDY interrupt is sometimes, say, once per few hours,
+        * lost; output stalls until next input arrives. In order to work
+        * around this problem, push TX queue manually by low-rate.
+        */
+#define        COM_EUMB_FLAG_BROKEN_ETXRDY     __BIT(0)
+       if (device_cfdata(self)->cf_flags & COM_EUMB_FLAG_BROKEN_ETXRDY) {
+               SET(sc->sc_hwflags, COM_HW_BROKEN_ETXRDY);
+               sc->sc_poll_ticks = 10 * hz;    /* once per 10 sec */
+       }
+
        com_attach_subr(sc);
 
        intr_establish_xname(epicirq + I8259_ICU, IST_LEVEL, IPL_SERIAL,