mei: use hbm idle state to prevent spurious resets

When reset is caused by hbm protocol mismatch or timeout
we might end up in an endless reset loop and hbm protocol
will never sync

Cc: <stable@vger.kernel.org>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/misc/mei/hbm.c

@@ -126,6 +126,17 @@ static bool is_treat_specially_client(struct mei_cl *cl,
 	return false;
 }
 
+/**
+ * mei_hbm_idle - set hbm to idle state
+ *
+ * @dev: the device structure
+ */
+void mei_hbm_idle(struct mei_device *dev)
+{
+	dev->init_clients_timer = 0;
+	dev->hbm_state = MEI_HBM_IDLE;
+}
+
 int mei_hbm_start_wait(struct mei_device *dev)
 {
 	int ret;
@@ -583,6 +594,14 @@ int mei_hbm_dispatch(struct mei_device *dev, struct mei_msg_hdr *hdr)
 	mei_read_slots(dev, dev->rd_msg_buf, hdr->length);
 	mei_msg = (struct mei_bus_message *)dev->rd_msg_buf;
 
+	/* ignore spurious message and prevent reset nesting
+	 * hbm is put to idle during system reset
+	 */
+	if (dev->hbm_state == MEI_HBM_IDLE) {
+		dev_dbg(&dev->pdev->dev, "hbm: state is idle ignore spurious messages\n");
+		return 0;
+	}
+
 	switch (mei_msg->hbm_cmd) {
 	case HOST_START_RES_CMD:
 		dev_dbg(&dev->pdev->dev, "hbm: start: response message received.\n");

drivers/misc/mei/hbm.h

@@ -49,6 +49,7 @@ static inline void mei_hbm_hdr(struct mei_msg_hdr *hdr, size_t length)
 	hdr->reserved = 0;
 }
 
+void mei_hbm_idle(struct mei_device *dev);
 int mei_hbm_start_req(struct mei_device *dev);
 int mei_hbm_start_wait(struct mei_device *dev);
 int mei_hbm_cl_flow_control_req(struct mei_device *dev, struct mei_cl *cl);

drivers/misc/mei/init.c

@@ -129,14 +129,19 @@ void mei_reset(struct mei_device *dev, int interrupts_enabled)
 		dev_warn(&dev->pdev->dev, "unexpected reset: dev_state = %s\n",
 			 mei_dev_state_str(dev->dev_state));
 
+	/* we're already in reset, cancel the init timer
+	 * if the reset was called due the hbm protocol error
+	 * we need to call it before hw start
+	 * so the hbm watchdog won't kick in
+	 */
+	mei_hbm_idle(dev);
+
 	ret = mei_hw_reset(dev, interrupts_enabled);
 	if (ret) {
 		dev_err(&dev->pdev->dev, "hw reset failed disabling the device\n");
 		interrupts_enabled = false;
-		dev->dev_state = MEI_DEV_DISABLED;
 	}
 
-	dev->hbm_state = MEI_HBM_IDLE;
 
 	if (dev->dev_state != MEI_DEV_INITIALIZING &&
 	    dev->dev_state != MEI_DEV_POWER_UP) {
@@ -160,8 +165,6 @@ void mei_reset(struct mei_device *dev, int interrupts_enabled)
 		memset(&dev->wr_ext_msg, 0, sizeof(dev->wr_ext_msg));
 	}
 
-	/* we're already in reset, cancel the init timer */
-	dev->init_clients_timer = 0;
 
 	dev->me_clients_num = 0;
 	dev->rd_msg_hdr = 0;
@@ -169,6 +172,7 @@ void mei_reset(struct mei_device *dev, int interrupts_enabled)
 
 	if (!interrupts_enabled) {
 		dev_dbg(&dev->pdev->dev, "intr not enabled end of reset\n");
+		dev->dev_state = MEI_DEV_DISABLED;
 		return;
 	}
 

drivers/misc/mei/interrupt.c

@@ -537,7 +537,6 @@ EXPORT_SYMBOL_GPL(mei_irq_write_handler);
  *
  * @work: pointer to the work_struct structure
  *
- * NOTE: This function is called by timer interrupt work
  */
 void mei_timer(struct work_struct *work)
 {
@@ -552,18 +551,24 @@ void mei_timer(struct work_struct *work)
 
 
 	mutex_lock(&dev->device_lock);
-	if (dev->dev_state != MEI_DEV_ENABLED) {
-		if (dev->dev_state == MEI_DEV_INIT_CLIENTS) {
+
+	/* Catch interrupt stalls during HBM init handshake */
+	if (dev->dev_state == MEI_DEV_INIT_CLIENTS &&
+	    dev->hbm_state != MEI_HBM_IDLE) {
+
 		if (dev->init_clients_timer) {
 			if (--dev->init_clients_timer == 0) {
-					dev_err(&dev->pdev->dev, "reset: init clients timeout hbm_state = %d.\n",
+				dev_err(&dev->pdev->dev, "timer: init clients timeout hbm_state = %d.\n",
 					dev->hbm_state);
 				mei_reset(dev, 1);
+				goto out;
 			}
 		}
 	}
+
+	if (dev->dev_state != MEI_DEV_ENABLED)
 		goto out;
-	}
+
 	/*** connect/disconnect timeouts ***/
 	list_for_each_entry_safe(cl_pos, cl_next, &dev->file_list, link) {
 		if (cl_pos->timer_count) {