Handle bus errors on thermal I2C bus

1) Properly report I2C errors on TMP006 as error, not device-not-powered.

2) Treat clock timeout and bus-busy I2C status as error (previously ignored).

3) If clock timeout or bus-busy, reset I2C master for that bus to clear the
error.

These should help with systems where the thermal I2C bus gets into a
weird state on suspend/resume.

BUG=chrome-os-partner:16262
BRANCH=link

TEST=boot system; 'battery' and 'temps' should give good info
Then run snanda's suspend_stress_test for a while and repeat.

Change-Id: I534be8236a4d6de82575fe6d33a68502ce0a3a95
Original-Change-Id: Iec5d6bbd357d2e5eb3dc3d361c829f353e996ab6
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://gerrit.chromium.org/gerrit/38444
Reviewed-on: https://gerrit.chromium.org/gerrit/38659
Reviewed-by: Yung-Chieh Lo <yjlou@chromium.org>

2 files changed, 43 insertions, 5 deletions

diff --git a/chip/lm4/i2c.c b/chip/lm4/i2c.c
index 7f0fa19c95..4a978e4a39 100644
--- a/chip/lm4/i2c.c
+++ b/chip/lm4/i2c.c
@@ -21,6 +21,7 @@
 
 #define NUM_PORTS 6  /* Number of physical ports */
 
+/* Flags for writes to MCS */
 #define LM4_I2C_MCS_RUN   (1 << 0)
 #define LM4_I2C_MCS_START (1 << 1)
 #define LM4_I2C_MCS_STOP  (1 << 2)
@@ -28,6 +29,16 @@
 #define LM4_I2C_MCS_HS    (1 << 4)
 #define LM4_I2C_MCS_QCMD  (1 << 5)
 
+/* Flags for reads from MCS */
+#define LM4_I2C_MCS_BUSY   (1 << 0)
+#define LM4_I2C_MCS_ERROR  (1 << 1)
+#define LM4_I2C_MCS_ADRACK (1 << 2)
+#define LM4_I2C_MCS_DATACK (1 << 3)
+#define LM4_I2C_MCS_ARBLST (1 << 4)
+#define LM4_I2C_MCS_IDLE   (1 << 5)
+#define LM4_I2C_MCS_BUSBSY (1 << 6)
+#define LM4_I2C_MCS_CLKTO  (1 << 7)
+
 #define START 1
 #define STOP  1
 #define NO_START 0
@@ -49,7 +60,7 @@ static int wait_idle(int port)
 	int event = 0;
 
 	i = LM4_I2C_MCS(port);
-	while (i & 0x01) {
+	while (i & LM4_I2C_MCS_BUSY) {
 		/* Port is busy, so wait for the interrupt */
 		task_waiting_on_port[port] = task_get_current();
 		LM4_I2C_MIMR(port) = 0x03;
@@ -81,7 +92,7 @@ static int wait_idle(int port)
 	task_set_event(task_get_current(), event, 0);
 
 	/* Check for errors */
-	if (i & 0x02)
+	if (i & LM4_I2C_MCS_ERROR)
 		return EC_ERROR_UNKNOWN;
 
 	return EC_SUCCESS;
@@ -110,6 +121,17 @@ static int i2c_xfer(int port, int slave_addr, const uint8_t *out, int out_size,
 	if (out_size == 0 && in_size == 0)
 		return EC_SUCCESS;
 
+	if (!started && (LM4_I2C_MCS(port) &
+			 (LM4_I2C_MCS_CLKTO | LM4_I2C_MCS_BUSBSY))) {
+		/*
+		 * Previous clock timeout or bus-busy.  Bounce the master to
+		 * clear these error states.
+		 */
+		LM4_I2C_MCR(port) = 0;
+		usleep(100000);
+		LM4_I2C_MCR(port) = 0x10;
+	}
+
 	if (out) {
 		LM4_I2C_MSA(port) = slave_addr & 0xff;
 		for (i = 0; i < out_size; i++) {
@@ -175,6 +197,11 @@ static int i2c_xfer(int port, int slave_addr, const uint8_t *out, int out_size,
 		}
 	}
 
+	/* Check for error conditions */
+	if (LM4_I2C_MCS(port) & (LM4_I2C_MCS_CLKTO | LM4_I2C_MCS_BUSBSY |
+				 LM4_I2C_MCS_ERROR))
+		return EC_ERROR_UNKNOWN;
+
 	return EC_SUCCESS;
 }
 
diff --git a/common/tmp006.c b/common/tmp006.c
index b4a1a8b967..3cda55ac1a 100644
--- a/common/tmp006.c
+++ b/common/tmp006.c
@@ -169,7 +169,10 @@ static int tmp006_poll_sensor(int sensor_id)
 	 */
 	if (tdata->fail && (FAIL_POWER | FAIL_INIT)) {
 		rv = i2c_read16(TMP006_PORT(addr), TMP006_REG(addr), 0x02, &v);
-		if (!(v & 0x80)) {
+		if (rv) {
+			tdata->fail |= FAIL_I2C;
+			return EC_ERROR_UNKNOWN;
+		} else if (!(v & 0x80)) {
 			tdata->fail |= FAIL_NOT_READY;
 			return EC_ERROR_UNKNOWN;
 		}
@@ -222,8 +225,16 @@ int tmp006_get_val(int idx, int *temp_ptr)
 	int tidx = idx >> 1;
 	const struct tmp006_data_t *tdata = tmp006_data + tidx;
 
-	if (tdata->fail & FAIL_POWER)
-		return EC_ERROR_NOT_POWERED;
+	if (tdata->fail & FAIL_POWER) {
+		/*
+		 * Sensor isn't powered, or hasn't successfully provided data
+		 * since being powered.  Keep reporting not-powered until
+		 * we get good data (which will clear FAIL_POWER) or there is
+		 * an I2C error.
+		 */
+		return (tdata->fail & FAIL_I2C) ? EC_ERROR_UNKNOWN :
+			EC_ERROR_NOT_POWERED;
+	}
 
 	/* Check the low bit to determine which temperature to read. */
 	if ((idx & 0x1) == 0)