myri10ge: rework parity error check and cleanup
authorJon Mason <mason@myri.com>
Mon, 27 Jun 2011 17:57:28 +0000 (17:57 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 29 Jun 2011 13:02:04 +0000 (06:02 -0700)
Clean up watchdog reset code:
 - move code that checks for stuck slice to a common routine
 - unless there is a confirmed h/w fault, verify that a stuck
   slice is still stuck in the watchdog worker; if the slice is no
   longer stuck, abort the reset.
 - this removes an egregious 2000ms pause in the watchdog worker that
   was a diagnostic aid (to look for spurious resets) the snuck into
   production code.

v3 includes corrections from Joe Perches

Signed-off-by: Jon Mason <mason@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/myri10ge/myri10ge.c

index 0f0f83d50ddc4002e0e9c86dbc52f45502d32375..ca0345795fa6da03ee78853ed7f08409ba0f3fd9 100644 (file)
@@ -193,6 +193,7 @@ struct myri10ge_slice_state {
        int watchdog_tx_done;
        int watchdog_tx_req;
        int watchdog_rx_done;
+       int stuck;
 #ifdef CONFIG_MYRI10GE_DCA
        int cached_dca_tag;
        int cpu;
@@ -3442,6 +3443,42 @@ static u32 myri10ge_read_reboot(struct myri10ge_priv *mgp)
        return reboot;
 }
 
+static void
+myri10ge_check_slice(struct myri10ge_slice_state *ss, int *reset_needed,
+                    int *busy_slice_cnt, u32 rx_pause_cnt)
+{
+       struct myri10ge_priv *mgp = ss->mgp;
+       int slice = ss - mgp->ss;
+
+       if (ss->tx.req != ss->tx.done &&
+           ss->tx.done == ss->watchdog_tx_done &&
+           ss->watchdog_tx_req != ss->watchdog_tx_done) {
+               /* nic seems like it might be stuck.. */
+               if (rx_pause_cnt != mgp->watchdog_pause) {
+                       if (net_ratelimit())
+                               netdev_warn(mgp->dev, "slice %d: TX paused, "
+                                           "check link partner\n", slice);
+               } else {
+                       netdev_warn(mgp->dev,
+                                   "slice %d: TX stuck %d %d %d %d %d %d\n",
+                                   slice, ss->tx.queue_active, ss->tx.req,
+                                   ss->tx.done, ss->tx.pkt_start,
+                                   ss->tx.pkt_done,
+                                   (int)ntohl(mgp->ss[slice].fw_stats->
+                                              send_done_count));
+                       *reset_needed = 1;
+                       ss->stuck = 1;
+               }
+       }
+       if (ss->watchdog_tx_done != ss->tx.done ||
+           ss->watchdog_rx_done != ss->rx_done.cnt) {
+               *busy_slice_cnt += 1;
+       }
+       ss->watchdog_tx_done = ss->tx.done;
+       ss->watchdog_tx_req = ss->tx.req;
+       ss->watchdog_rx_done = ss->rx_done.cnt;
+}
+
 /*
  * This watchdog is used to check whether the board has suffered
  * from a parity error and needs to be recovered.
@@ -3450,10 +3487,12 @@ static void myri10ge_watchdog(struct work_struct *work)
 {
        struct myri10ge_priv *mgp =
            container_of(work, struct myri10ge_priv, watchdog_work);
-       struct myri10ge_tx_buf *tx;
-       u32 reboot;
+       struct myri10ge_slice_state *ss;
+       u32 reboot, rx_pause_cnt;
        int status, rebooted;
        int i;
+       int reset_needed = 0;
+       int busy_slice_cnt = 0;
        u16 cmd, vendor;
 
        mgp->watchdog_resets++;
@@ -3465,8 +3504,7 @@ static void myri10ge_watchdog(struct work_struct *work)
                 * For now, just report it */
                reboot = myri10ge_read_reboot(mgp);
                netdev_err(mgp->dev, "NIC rebooted (0x%x),%s resetting\n",
-                          reboot,
-                          myri10ge_reset_recover ? "" : " not");
+                          reboot, myri10ge_reset_recover ? "" : " not");
                if (myri10ge_reset_recover == 0)
                        return;
                rtnl_lock();
@@ -3498,23 +3536,24 @@ static void myri10ge_watchdog(struct work_struct *work)
                                return;
                        }
                }
-               /* Perhaps it is a software error.  Try to reset */
-
-               netdev_err(mgp->dev, "device timeout, resetting\n");
+               /* Perhaps it is a software error. See if stuck slice
+                * has recovered, reset if not */
+               rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
                for (i = 0; i < mgp->num_slices; i++) {
-                       tx = &mgp->ss[i].tx;
-                       netdev_err(mgp->dev, "(%d): %d %d %d %d %d %d\n",
-                                  i, tx->queue_active, tx->req,
-                                  tx->done, tx->pkt_start, tx->pkt_done,
-                                  (int)ntohl(mgp->ss[i].fw_stats->
-                                             send_done_count));
-                       msleep(2000);
-                       netdev_info(mgp->dev, "(%d): %d %d %d %d %d %d\n",
-                                   i, tx->queue_active, tx->req,
-                                   tx->done, tx->pkt_start, tx->pkt_done,
-                                   (int)ntohl(mgp->ss[i].fw_stats->
-                                              send_done_count));
+                       ss = mgp->ss;
+                       if (ss->stuck) {
+                               myri10ge_check_slice(ss, &reset_needed,
+                                                    &busy_slice_cnt,
+                                                    rx_pause_cnt);
+                               ss->stuck = 0;
+                       }
                }
+               if (!reset_needed) {
+                       netdev_dbg(mgp->dev, "not resetting\n");
+                       return;
+               }
+
+               netdev_err(mgp->dev, "device timeout, resetting\n");
        }
 
        if (!rebooted) {
@@ -3567,27 +3606,8 @@ static void myri10ge_watchdog_timer(unsigned long arg)
                            myri10ge_fill_thresh)
                                ss->rx_big.watchdog_needed = 0;
                }
-
-               if (ss->tx.req != ss->tx.done &&
-                   ss->tx.done == ss->watchdog_tx_done &&
-                   ss->watchdog_tx_req != ss->watchdog_tx_done) {
-                       /* nic seems like it might be stuck.. */
-                       if (rx_pause_cnt != mgp->watchdog_pause) {
-                               if (net_ratelimit())
-                                       netdev_err(mgp->dev, "slice %d: TX paused, check link partner\n",
-                                                  i);
-                       } else {
-                               netdev_warn(mgp->dev, "slice %d stuck:", i);
-                               reset_needed = 1;
-                       }
-               }
-               if (ss->watchdog_tx_done != ss->tx.done ||
-                   ss->watchdog_rx_done != ss->rx_done.cnt) {
-                       busy_slice_cnt++;
-               }
-               ss->watchdog_tx_done = ss->tx.done;
-               ss->watchdog_tx_req = ss->tx.req;
-               ss->watchdog_rx_done = ss->rx_done.cnt;
+               myri10ge_check_slice(ss, &reset_needed, &busy_slice_cnt,
+                                    rx_pause_cnt);
        }
        /* if we've sent or received no traffic, poll the NIC to
         * ensure it is still there.  Otherwise, we risk not noticing