[net-next v2 10/15] e1000e: Fix Hardware Unit Hang

[Jeff Kirsher <jeffrey.t.kirsher@intel.com>]

From: David Ertman <davidx.m.ertman@intel.com>

The check for pending Tx work when link is lost was mistakenly moved to be
done only when link is first detected to be lost.  It turns out there is a
small window of opportunity for additional Tx work to get queued up shortly
after link is dropped.

Move the check back to the place it was before in the watchdog task.  Put in
additional debug information for other reset paths and a final catch-all for
false hangs in the scheduled function that prints out the hardware hang
message.

Signed-off-by: Dave Ertman <davidx.m.ertman@intel.com>
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
 drivers/net/ethernet/intel/e1000e/netdev.c | 32 +++++++++++++++++++++---------
 1 file changed, 23 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
index eafad41..d577723 100644
--- a/drivers/net/ethernet/intel/e1000e/netdev.c
+++ b/drivers/net/ethernet/intel/e1000e/netdev.c
@@ -1090,8 +1090,14 @@ static void e1000_print_hw_hang(struct work_struct *work)
 		adapter->tx_hang_recheck = true;
 		return;
 	}
-	/* Real hang detected */
 	adapter->tx_hang_recheck = false;
+
+	if (er32(TDH(0)) == er32(TDT(0))) {
+		e_dbg("false hang detected, ignoring\n");
+		return;
+	}
+
+	/* Real hang detected */
 	netif_stop_queue(netdev);
 
 	e1e_rphy(hw, MII_BMSR, &phy_status);
@@ -1121,6 +1127,8 @@ static void e1000_print_hw_hang(struct work_struct *work)
 	      eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
 	      phy_status, phy_1000t_status, phy_ext_status, pci_status);
 
+	e1000e_dump(adapter);
+
 	/* Suggest workaround for known h/w issue */
 	if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
 		e_err("Try turning off Tx pause (flow control) via ethtool\n");
@@ -4798,6 +4806,7 @@ static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
 
 	if (adapter->phy_hang_count > 1) {
 		adapter->phy_hang_count = 0;
+		e_dbg("PHY appears hung - resetting\n");
 		schedule_work(&adapter->reset_task);
 	}
 }
@@ -4956,15 +4965,11 @@ static void e1000_watchdog_task(struct work_struct *work)
 				mod_timer(&adapter->phy_info_timer,
 					  round_jiffies(jiffies + 2 * HZ));
 
-			/* The link is lost so the controller stops DMA.
-			 * If there is queued Tx work that cannot be done
-			 * or if on an 8000ES2LAN which requires a Rx packet
-			 * buffer work-around on link down event, reset the
-			 * controller to flush the Tx/Rx packet buffers.
-			 * (Do the reset outside of interrupt context).
+			/* 8000ES2LAN requires a Rx packet buffer work-around
+			 * on link down event; reset the controller to flush
+			 * the Rx packet buffer.
 			 */
-			if ((adapter->flags & FLAG_RX_NEEDS_RESTART) ||
-			    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
+			if (adapter->flags & FLAG_RX_NEEDS_RESTART)
 				adapter->flags |= FLAG_RESTART_NOW;
 			else
 				pm_schedule_suspend(netdev->dev.parent,
@@ -4987,6 +4992,15 @@ link_up:
 	adapter->gotc_old = adapter->stats.gotc;
 	spin_unlock(&adapter->stats64_lock);
 
+	/* If the link is lost the controller stops DMA, but
+	 * if there is queued Tx work it cannot be done.  So
+	 * reset the controller to flush the Tx packet buffers.
+	 */
+	if (!netif_carrier_ok(netdev) &&
+	    (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
+		adapter->flags |= FLAG_RESTART_NOW;
+
+	/* If reset is necessary, do it outside of interrupt context. */
 	if (adapter->flags & FLAG_RESTART_NOW) {
 		schedule_work(&adapter->reset_task);
 		/* return immediately since reset is imminent */
-- 
1.8.3.1