1 files changed, 45 insertions, 0 deletions
diff --git a/meta-raptor/meta-asus/recipes-core/fan-ctrl/fan-ctrl/ast2050-timer-workaround.sh b/meta-raptor/meta-asus/recipes-core/fan-ctrl/fan-ctrl/ast2050-timer-workaround.sh
new file mode 100755
index 0000000..d8631dd
--- /dev/null
+++ b/meta-raptor/meta-asus/recipes-core/fan-ctrl/fan-ctrl/ast2050-timer-workaround.sh
@@ -0,0 +1,45 @@
+#/bin/bash
+#
+# Copyright 2017 Raptor Engineering, LLC
+#
+# Work around glitchy timer0 on the AST2050
+# For an unknown reason, sometimes when timer0
+# expires it does not fire an IRQ.  This in turn
+# causes Linux to hang waiting for the timer to
+# expire, causing all nanosleep calls to stall
+# indefinitely.  On a stock OpenBMC system, this
+# manifests as a spontanous reboot every few hours
+# of uptime as the watchdog timer kicks in.
+#
+# This script works around the problem by watching
+# timer0 and seeing if it is stuck at 0.  If so, it
+# gives the afflicted timer a good kick in the rear
+# to get it going again, which in turn un-stalls all
+# of the queued nanosleep calls.
+#
+# Note that due to what this does and how it operates,
+# using sleep in this script is a VERY BAD IDEA!
+
+function daemon_process {
+	while [[ 1 == 1 ]]; do
+		TIMER_COUNT=$(devmem 0x1e782000 32)
+		if [[ "$TIMER_COUNT" == "0x00000000" ]]; then
+			TIMER_COUNT=$(devmem 0x1e782000 32)
+			if [[ "$TIMER_COUNT" == "0x00000000" ]]; then
+				echo "Kicking timer back into operation!"
+				devmem 0x1e782004 32 0x00000001
+			fi
+		fi
+
+		# Busy wait at highest niceness
+		COUNTER=1000
+		until [[ $COUNTER -lt 1 ]]; do
+			let COUNTER-=1
+		done
+	done
+}
+
+renice -n 19 $$
+
+daemon_process < /dev/null &
+disown