[PATCH] Re: [PATCH] net/fec: infinite spin on sirq-net-tx on real-time

Hi
Am Montag 06 Februar 2012, 14:39:59 schrieb Eric Dumazet:

Le lundi 06 février 2012 à 08:25 -0500, Steven Rostedt a écrit :

quoted

On Mon, 2012-02-06 at 12:03 +0100, Hector Palacios wrote:

quoted

quoted

quoted

@@ -530,6 +531,7 @@ fec_stop(struct net_device *ndev)

  	udelay(10);
  	writel(fep->phy_speed, fep->hwp + FEC_MII_SPEED);
  	writel(FEC_DEFAULT_IMASK, fep->hwp + FEC_IMASK);

+	fep->link = 0;

Why not call netif_stop_queue(dev) here ?

I'm no network driver expert so I'll leave it up to others to comment.
I just forward ported a patch I came across in Freescale's BSP which
solves the problem in mainline and in RT.

Hector,

Eric's suggestion may also work. Could your revert this patch and add
the netif_stop_queue(dev) there, and see if it fixes the problems in
both mainline and -rt?

Not sure it will be enough to call netif_stop_queue(dev) in fec_stop()

We probably need to start the device with its tx queue stopped, then
later when device is really ready wakeup the queue.

I am talking about 3.0-rt33 kernel since this is the one i can test the best.

Ok i just found out that removing:

        if (!fep->link) {
                /* Link is down or autonegotiation is in progress. */
                netif_stop_queue(ndev);
                return NETDEV_TX_BUSY;
        }
does not seem to work. Also netif_stop_queue seems to be mandantory to get the driver working.
If this condition in the hotpath is removed i see the ksoftirq high cpuload problem again.
I am not sure if the patch for lines 470 if netif_stop_queue(ndev); is really nessary.
I also found an double cleanup in my earlier patch. So heres the next iteration of the
fec ksoftirq fix for preempt rt.

Tim

diff --git a/drivers/net/fec.c b/drivers/net/fec.c
index 885d8ba..140cb13 100644
--- a/drivers/net/fec.c
+++ b/drivers/net/fec.c

@@ -244,6 +244,7 @@ fec_enet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 
        if (!fep->link) {
                /* Link is down or autonegotiation is in progress. */
+               netif_stop_queue(ndev);
                return NETDEV_TX_BUSY;
        }
 

@@ -470,6 +471,9 @@ fec_stop(struct net_device *ndev)
        udelay(10);
        writel(fep->phy_speed, fep->hwp + FEC_MII_SPEED);
        writel(FEC_DEFAULT_IMASK, fep->hwp + FEC_IMASK);
+
+       netif_stop_queue(ndev);
+       fep->link = 0;
 }
 
 

@@ -787,6 +791,7 @@ static void fec_enet_adjust_link(struct net_device *ndev)
        if (phy_dev->link) {
                if (fep->full_duplex != phy_dev->duplex) {
                        fec_restart(ndev, phy_dev->duplex);
+                       netif_wake_queue(ndev);
                        status_change = 1;
                }
        }

@@ -1118,6 +1123,10 @@ static int fec_enet_alloc_buffers(struct net_device *ndev)
        bdp = fep->tx_bd_base;
        for (i = 0; i < TX_RING_SIZE; i++) {
                fep->tx_bounce[i] = kmalloc(FEC_ENET_TX_FRSIZE, GFP_KERNEL);
+               if(!fep->tx_bounce[i])  {
+                       fec_enet_free_buffers(ndev);
+                       return -ENOMEM;
+               }
 
                bdp->cbd_sc = 0;
                bdp->cbd_bufaddr = 0;