From 098bbd9ed847849834ff1f21766d23ea240c5bf0 Mon Sep 17 00:00:00 2001
From: Andrew Jones <drjones@redhat.com>
Date: Mon, 1 Aug 2016 14:27:09 +0200
Subject: RHEL-only: hw/char/pl011: fix SBSA reset
RH-Author: Andrew Jones <drjones@redhat.com>
Message-id: <1470061629-6395-1-git-send-email-drjones@redhat.com>
Patchwork-id: 71697
O-Subject: [AArch64 RHEL-7.3 qemu-kvm-rhev PATCH] RHEL-only: hw/char/pl011: fix SBSA reset
Bugzilla: 1266048
RH-Acked-by: Auger Eric <eric.auger@redhat.com>
RH-Acked-by: Laszlo Ersek <lersek@redhat.com>
RH-Acked-by: Wei Huang <wei@redhat.com>
When booting Linux with an SBSA UART, e.g. when booting mach-virt
with ACPI, if the user types on the console during boot, then when
the login prompt appears she won't be able to log in. This is
because during boot the SBSA UART needs to be reset, but the SBSA
specification doesn't provide registers to enable/disable the FIFOs.
This patch observes a couple registers the SBSA UART does write to
in order to attempt to guess when a reset is needed, and then do it.
We risk losing some characters from the FIFO if the guess is wrong,
but the risk of that should be quite low.
Signed-off-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
(cherry picked from commit 15ee295534f654d6b6ba9499cdd380aa9c954920)
(cherry picked from commit 49be481336c227fdad2f7edc02fa088f3d88c9a2)
(cherry picked from commit 9fcede24f35378e2c9113440a692d2c96cc94865)
(cherry picked from commit b721e7ff2bdd42b0a786b8630c1ba8b1d3560da3)
(cherry picked from commit 88f3eb70f909b03ed18074aadd12f32f28ad8437)
---
hw/char/pl011.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/hw/char/pl011.c b/hw/char/pl011.c
index 2aa277f..23fe047 100644
--- a/hw/char/pl011.c
+++ b/hw/char/pl011.c
@@ -209,6 +209,18 @@ static void pl011_write(void *opaque, hwaddr offset,
pl011_update(s);
break;
case 17: /* UARTICR */
+ /*
+ * RHEL-only, fixes BZ1266048
+ *
+ * Look for the "signature" of a driver init or shutdown in
+ * order to know that we need to reset the SBSA UART. Yes,
+ * this is hacky, but as SBSA drivers aren't required to write
+ * UARTLCR_H or UARTCR, then we don't have much choice...
+ */
+ if (s->int_enabled == 0 && value == 0xffff) {
+ s->read_count = 0;
+ s->read_pos = 0;
+ }
s->int_level &= ~value;
pl011_update(s);
break;
--
1.8.3.1