[PATCH] igb/ixgbe/e1000e: resolve tx multiqueue bug

[PATCH] igb/ixgbe/e1000e: resolve tx multiqueue bug

[Jeff Kirsher]

With the recent changes to tx mutiqueue, igb/ixgbe/e1000e was not calling
netif_tx_start_all_queues() before calling netif_tx_wake_all_queues().
This causes an issue during loading of the driver.

In addition, updated e1000e to use the updated tx mutliqueue api.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---

 drivers/net/e1000e/netdev.c    |   10 ++++++----
 drivers/net/igb/igb_main.c     |    2 ++
 drivers/net/ixgbe/ixgbe_main.c |    2 ++
 3 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
index be59533..869544b 100644
--- a/drivers/net/e1000e/netdev.c
+++ b/drivers/net/e1000e/netdev.c
@@ -2513,7 +2513,7 @@ void e1000e_down(struct e1000_adapter *adapter)
 	ew32(RCTL, rctl & ~E1000_RCTL_EN);
 	/* flush and sleep below */
 
-	netif_stop_queue(netdev);
+	netif_tx_stop_all_queues(netdev);
 
 	/* disable transmits in the hardware */
 	tctl = er32(TCTL);
@@ -2663,6 +2663,8 @@ static int e1000_open(struct net_device *netdev)
 
 	e1000_irq_enable(adapter);
 
+	netif_tx_start_all_queues(netdev);
+
 	/* fire a link status change interrupt to start the watchdog */
 	ew32(ICS, E1000_ICS_LSC);
 
@@ -3118,7 +3120,7 @@ static void e1000_watchdog_task(struct work_struct *work)
 			ew32(TCTL, tctl);
 
 			netif_carrier_on(netdev);
-			netif_wake_queue(netdev);
+			netif_tx_wake_all_queues(netdev);
 
 			if (!test_bit(__E1000_DOWN, &adapter->state))
 				mod_timer(&adapter->phy_info_timer,
@@ -3130,7 +3132,7 @@ static void e1000_watchdog_task(struct work_struct *work)
 			adapter->link_duplex = 0;
 			ndev_info(netdev, "Link is Down\n");
 			netif_carrier_off(netdev);
-			netif_stop_queue(netdev);
+			netif_tx_stop_all_queues(netdev);
 			if (!test_bit(__E1000_DOWN, &adapter->state))
 				mod_timer(&adapter->phy_info_timer,
 					  round_jiffies(jiffies + 2 * HZ));
@@ -4504,7 +4506,7 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
 
 	/* tell the stack to leave us alone until e1000_open() is called */
 	netif_carrier_off(netdev);
-	netif_stop_queue(netdev);
+	netif_tx_stop_all_queues(netdev);
 
 	strcpy(netdev->name, "eth%d");
 	err = register_netdev(netdev);
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 81bba69..1b7cb29 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1493,6 +1493,8 @@ static int igb_open(struct net_device *netdev)
 
 	igb_irq_enable(adapter);
 
+	netif_tx_start_all_queues(netdev);
+
 	/* Fire a link status change interrupt to start the watchdog. */
 	wr32(E1000_ICS, E1000_ICS_LSC);
 
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index efebd11..e5f3da8 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -2763,6 +2763,8 @@ static int ixgbe_open(struct net_device *netdev)
 	if (err)
 		goto err_up;
 
+	netif_tx_start_all_queues(netdev);
+
 	return 0;
 
 err_up:

Re: [PATCH] igb/ixgbe/e1000e: resolve tx multiqueue bug

[David Miller]

From: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Date: Fri, 18 Jul 2008 04:02:25 -0700

> With the recent changes to tx mutiqueue, igb/ixgbe/e1000e was not calling
> netif_tx_start_all_queues() before calling netif_tx_wake_all_queues().
> This causes an issue during loading of the driver.
> 
> In addition, updated e1000e to use the updated tx mutliqueue api.
> 
> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Thanks Jeff.  I've applied this to kill the BUG() triggers you were
seeing.

Background for everyone else:

1) These drivers have a watchdog handler that watches the PHY state
   and wakes the TX queue when the link comes up.

2) Since ->open() was not doing a start queue, we would do a wake
   queue while attached to the noop_qdisc

This causes the BUG check in netif_schedule() to trigger.

After some discussion with Jeff I've advised him that it's not
kosher for the driver to manage the TX queue wake state like
this in respose to link up and down events, but rather simply
maintaining the netif_carrier state is all that is needed.

When netif_carrier_ok() is false, we're attached to the noop_qdisc
which eats all packets and frees them up, never sending them down
to the device.

When netif_carrier_ok() is true, we attach to the configured normal
qdisc and packets begin flowing to the device again.

Jeff told me he will take a look more deeply into this.