]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/commitdiff
net/mlx5e: Add tx timeout support for mlx5e tx reporter
authorEran Ben Elisha <eranbe@mellanox.com>
Thu, 7 Feb 2019 09:36:41 +0000 (11:36 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 7 Feb 2019 18:34:29 +0000 (10:34 -0800)
With this patch, ndo_tx_timeout callback will be redirected to the tx
reporter in order to detect a tx timeout error and report it to the
devlink health. (The watchdog detects tx timeouts, but the driver verify
the issue still exists before launching any recover method).

In addition, recover from tx timeout in case of lost interrupt was added
to the tx reporter recover method. The tx timeout recover from lost
interrupt is not a new feature in the driver, this patch re-organize the
functionality and move it to the tx reporter recovery flow.

tx timeout example:
(with auto_recover set to false, if set to true, the manual recover and
diagnose sections are irrelevant)

$cat /sys/kernel/debug/tracing/trace
...
devlink_health_report: bus_name=pci dev_name=0000:00:09.0
driver_name=mlx5_core reporter_name=tx: TX timeout on queue: 0, SQ: 0x8a,
CQ: 0x35, SQ Cons: 0x2 SQ Prod: 0x2, usecs since last trans: 14912000

$devlink health show
pci/0000:00:09.0:
  name tx
    state healthy #err 1 #recover 0 last_dump_ts N/A
    parameters:
      grace_period 500 auto_recover false

$devlink health diagnose pci/0000:00:09.0 reporter tx -j -p
{
    "SQs": [ {
            "sqn": 138,
            "HW state": 1,
            "stopped": true
        },{
            "sqn": 142,
            "HW state": 1,
            "stopped": false
        } ]
}

$devlink health diagnose pci/0000:00:09.0 reporter tx
SQs:
  sqn: 138 HW state: 1 stopped: true
  sqn: 142 HW state: 1 stopped: false

$devlink health recover pci/0000:00:09 reporter tx
$devlink health show
pci/0000:00:09.0:
  name tx
    state healthy #err 1 #recover 1 last_dump_ts N/A
    parameters:
      grace_period 500 auto_recover false

Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx5/core/en/reporter.h
drivers/net/ethernet/mellanox/mlx5/core/en/reporter_tx.c
drivers/net/ethernet/mellanox/mlx5/core/en_main.c

index 1c90059db5f8042f59ee16b37e0f97b44c391136..e78e92753d738933cb76527dabe61f8c12bd324d 100644 (file)
@@ -10,5 +10,6 @@
 int mlx5e_tx_reporter_create(struct mlx5e_priv *priv);
 void mlx5e_tx_reporter_destroy(struct mlx5e_priv *priv);
 void mlx5e_tx_reporter_err_cqe(struct mlx5e_txqsq *sq);
+int mlx5e_tx_reporter_timeout(struct mlx5e_txqsq *sq);
 
 #endif
index 35568e4820de4bca620392c2bd510ea40cee9640..0aebfb377cf04edd6b8fab5e6e1de43fd8be3c0f 100644 (file)
@@ -126,6 +126,44 @@ void mlx5e_tx_reporter_err_cqe(struct mlx5e_txqsq *sq)
                              &err_ctx);
 }
 
+static int mlx5e_tx_reporter_timeout_recover(struct mlx5e_txqsq *sq)
+{
+       struct mlx5_eq_comp *eq = sq->cq.mcq.eq;
+       u32 eqe_count;
+       int ret;
+
+       netdev_err(sq->channel->netdev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n",
+                  eq->core.eqn, eq->core.cons_index, eq->core.irqn);
+
+       eqe_count = mlx5_eq_poll_irq_disabled(eq);
+       ret = eqe_count ? true : false;
+       if (!eqe_count) {
+               clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
+               return ret;
+       }
+
+       netdev_err(sq->channel->netdev, "Recover %d eqes on EQ 0x%x\n",
+                  eqe_count, eq->core.eqn);
+       sq->channel->stats->eq_rearm++;
+       return ret;
+}
+
+int mlx5e_tx_reporter_timeout(struct mlx5e_txqsq *sq)
+{
+       char err_str[MLX5E_TX_REPORTER_PER_SQ_MAX_LEN];
+       struct mlx5e_tx_err_ctx err_ctx;
+
+       err_ctx.sq       = sq;
+       err_ctx.recover  = mlx5e_tx_reporter_timeout_recover;
+       sprintf(err_str,
+               "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x, usecs since last trans: %u\n",
+               sq->channel->ix, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc,
+               jiffies_to_usecs(jiffies - sq->txq->trans_start));
+
+       return devlink_health_report(sq->channel->priv->tx_reporter, err_str,
+                                    &err_ctx);
+}
+
 /* state lock cannot be grabbed within this function.
  * It can cause a dead lock or a read-after-free.
  */
index 93e1f037b01984a87b75cd5e8ab8e412164b6b1f..d81ebba7c04ee66f7bf09562bf4de5b4ea13fab5 100644 (file)
@@ -4116,31 +4116,13 @@ netdev_features_t mlx5e_features_check(struct sk_buff *skb,
        return features;
 }
 
-static bool mlx5e_tx_timeout_eq_recover(struct net_device *dev,
-                                       struct mlx5e_txqsq *sq)
-{
-       struct mlx5_eq_comp *eq = sq->cq.mcq.eq;
-       u32 eqe_count;
-
-       netdev_err(dev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n",
-                  eq->core.eqn, eq->core.cons_index, eq->core.irqn);
-
-       eqe_count = mlx5_eq_poll_irq_disabled(eq);
-       if (!eqe_count)
-               return false;
-
-       netdev_err(dev, "Recover %d eqes on EQ 0x%x\n", eqe_count, eq->core.eqn);
-       sq->channel->stats->eq_rearm++;
-       return true;
-}
-
 static void mlx5e_tx_timeout_work(struct work_struct *work)
 {
        struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
                                               tx_timeout_work);
-       struct net_device *dev = priv->netdev;
-       bool reopen_channels = false;
-       int i, err;
+       bool report_failed = false;
+       int err;
+       int i;
 
        rtnl_lock();
        mutex_lock(&priv->state_lock);
@@ -4149,31 +4131,22 @@ static void mlx5e_tx_timeout_work(struct work_struct *work)
                goto unlock;
 
        for (i = 0; i < priv->channels.num * priv->channels.params.num_tc; i++) {
-               struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, i);
+               struct netdev_queue *dev_queue =
+                       netdev_get_tx_queue(priv->netdev, i);
                struct mlx5e_txqsq *sq = priv->txq2sq[i];
 
                if (!netif_xmit_stopped(dev_queue))
                        continue;
 
-               netdev_err(dev,
-                          "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x, usecs since last trans: %u\n",
-                          i, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc,
-                          jiffies_to_usecs(jiffies - dev_queue->trans_start));
-
-               /* If we recover a lost interrupt, most likely TX timeout will
-                * be resolved, skip reopening channels
-                */
-               if (!mlx5e_tx_timeout_eq_recover(dev, sq)) {
-                       clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
-                       reopen_channels = true;
-               }
+               if (mlx5e_tx_reporter_timeout(sq))
+                       report_failed = true;
        }
 
-       if (!reopen_channels)
+       if (!report_failed)
                goto unlock;
 
-       mlx5e_close_locked(dev);
-       err = mlx5e_open_locked(dev);
+       mlx5e_close_locked(priv->netdev);
+       err = mlx5e_open_locked(priv->netdev);
        if (err)
                netdev_err(priv->netdev,
                           "mlx5e_open_locked failed recovering from a tx_timeout, err(%d).\n",
@@ -4189,6 +4162,12 @@ static void mlx5e_tx_timeout(struct net_device *dev)
        struct mlx5e_priv *priv = netdev_priv(dev);
 
        netdev_err(dev, "TX timeout detected\n");
+
+       if (IS_ERR_OR_NULL(priv->tx_reporter)) {
+               netdev_err_once(priv->netdev, "tx timeout will not be handled, no valid tx reporter\n");
+               return;
+       }
+
        queue_work(priv->wq, &priv->tx_timeout_work);
 }