Revert "IPoIB: Use dedicated workqueues per interface"
authorRoland Dreier <roland@purestorage.com>
Fri, 30 Jan 2015 23:38:55 +0000 (15:38 -0800)
committerRoland Dreier <roland@purestorage.com>
Fri, 30 Jan 2015 23:38:55 +0000 (15:38 -0800)
This reverts commit 5141861cd5e17eac9676ff49c5abfafbea2b0e98.

The series of IPoIB bug fixes that went into 3.19-rc1 introduce
regressions, and after trying to sort things out, we decided to revert
to 3.18's IPoIB driver and get things right for 3.20.

Signed-off-by: Roland Dreier <roland@purestorage.com>
drivers/infiniband/ulp/ipoib/ipoib.h
drivers/infiniband/ulp/ipoib/ipoib_cm.c
drivers/infiniband/ulp/ipoib/ipoib_ib.c
drivers/infiniband/ulp/ipoib/ipoib_main.c
drivers/infiniband/ulp/ipoib/ipoib_multicast.c
drivers/infiniband/ulp/ipoib/ipoib_verbs.c

index 45fd10a72ec1cdcf1b2b60e6f788871755319ed5..f4c1b20b23b20d6daae8b4d96780dd77133288a6 100644 (file)
@@ -323,7 +323,6 @@ struct ipoib_dev_priv {
        struct list_head multicast_list;
        struct rb_root multicast_tree;
 
-       struct workqueue_struct *wq;
        struct delayed_work mcast_task;
        struct work_struct carrier_on_task;
        struct work_struct flush_light;
index 56959adb6c7da51ccbb6d20307247b7cb69ad55a..933efcea0d03f11b4da3967b8eedc137da21e08a 100644 (file)
@@ -474,7 +474,7 @@ static int ipoib_cm_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *even
        }
 
        spin_lock_irq(&priv->lock);
-       queue_delayed_work(priv->wq,
+       queue_delayed_work(ipoib_workqueue,
                           &priv->cm.stale_task, IPOIB_CM_RX_DELAY);
        /* Add this entry to passive ids list head, but do not re-add it
         * if IB_EVENT_QP_LAST_WQE_REACHED has moved it to flush list. */
@@ -576,7 +576,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
                        spin_lock_irqsave(&priv->lock, flags);
                        list_splice_init(&priv->cm.rx_drain_list, &priv->cm.rx_reap_list);
                        ipoib_cm_start_rx_drain(priv);
-                       queue_work(priv->wq, &priv->cm.rx_reap_task);
+                       queue_work(ipoib_workqueue, &priv->cm.rx_reap_task);
                        spin_unlock_irqrestore(&priv->lock, flags);
                } else
                        ipoib_warn(priv, "cm recv completion event with wrid %d (> %d)\n",
@@ -603,7 +603,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
                                spin_lock_irqsave(&priv->lock, flags);
                                list_move(&p->list, &priv->cm.rx_reap_list);
                                spin_unlock_irqrestore(&priv->lock, flags);
-                               queue_work(priv->wq, &priv->cm.rx_reap_task);
+                               queue_work(ipoib_workqueue, &priv->cm.rx_reap_task);
                        }
                        return;
                }
@@ -827,7 +827,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 
                if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) {
                        list_move(&tx->list, &priv->cm.reap_list);
-                       queue_work(priv->wq, &priv->cm.reap_task);
+                       queue_work(ipoib_workqueue, &priv->cm.reap_task);
                }
 
                clear_bit(IPOIB_FLAG_OPER_UP, &tx->flags);
@@ -1255,7 +1255,7 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 
                if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) {
                        list_move(&tx->list, &priv->cm.reap_list);
-                       queue_work(priv->wq, &priv->cm.reap_task);
+                       queue_work(ipoib_workqueue, &priv->cm.reap_task);
                }
 
                spin_unlock_irqrestore(&priv->lock, flags);
@@ -1284,7 +1284,7 @@ struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path
        tx->dev = dev;
        list_add(&tx->list, &priv->cm.start_list);
        set_bit(IPOIB_FLAG_INITIALIZED, &tx->flags);
-       queue_work(priv->wq, &priv->cm.start_task);
+       queue_work(ipoib_workqueue, &priv->cm.start_task);
        return tx;
 }
 
@@ -1295,7 +1295,7 @@ void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
        if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) {
                spin_lock_irqsave(&priv->lock, flags);
                list_move(&tx->list, &priv->cm.reap_list);
-               queue_work(priv->wq, &priv->cm.reap_task);
+               queue_work(ipoib_workqueue, &priv->cm.reap_task);
                ipoib_dbg(priv, "Reap connection for gid %pI6\n",
                          tx->neigh->daddr + 4);
                tx->neigh = NULL;
@@ -1417,7 +1417,7 @@ void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
 
        skb_queue_tail(&priv->cm.skb_queue, skb);
        if (e)
-               queue_work(priv->wq, &priv->cm.skb_task);
+               queue_work(ipoib_workqueue, &priv->cm.skb_task);
 }
 
 static void ipoib_cm_rx_reap(struct work_struct *work)
@@ -1450,7 +1450,7 @@ static void ipoib_cm_stale_task(struct work_struct *work)
        }
 
        if (!list_empty(&priv->cm.passive_ids))
-               queue_delayed_work(priv->wq,
+               queue_delayed_work(ipoib_workqueue,
                                   &priv->cm.stale_task, IPOIB_CM_RX_DELAY);
        spin_unlock_irq(&priv->lock);
 }
index bfd17d41b5f2b52db679be3ea786e8d7e13f36d0..72626c3481749b962fe96b79722d7c8e9c99c585 100644 (file)
@@ -655,7 +655,7 @@ void ipoib_reap_ah(struct work_struct *work)
        __ipoib_reap_ah(dev);
 
        if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
-               queue_delayed_work(priv->wq, &priv->ah_reap_task,
+               queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
                                   round_jiffies_relative(HZ));
 }
 
@@ -696,7 +696,7 @@ int ipoib_ib_dev_open(struct net_device *dev, int flush)
        }
 
        clear_bit(IPOIB_STOP_REAPER, &priv->flags);
-       queue_delayed_work(priv->wq, &priv->ah_reap_task,
+       queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
                           round_jiffies_relative(HZ));
 
        if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
@@ -881,7 +881,7 @@ timeout:
        set_bit(IPOIB_STOP_REAPER, &priv->flags);
        cancel_delayed_work(&priv->ah_reap_task);
        if (flush)
-               flush_workqueue(priv->wq);
+               flush_workqueue(ipoib_workqueue);
 
        begin = jiffies;
 
index 42e5c278f4892c249cec5830882792c9277ec149..2cf81ef51412d7d5d47e4756f96d7a588bd08835 100644 (file)
@@ -839,7 +839,7 @@ static void ipoib_set_mcast_list(struct net_device *dev)
                return;
        }
 
-       queue_work(priv->wq, &priv->restart_task);
+       queue_work(ipoib_workqueue, &priv->restart_task);
 }
 
 static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
@@ -954,7 +954,7 @@ static void ipoib_reap_neigh(struct work_struct *work)
        __ipoib_reap_neigh(priv);
 
        if (!test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
-               queue_delayed_work(priv->wq, &priv->neigh_reap_task,
+               queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task,
                                   arp_tbl.gc_interval);
 }
 
@@ -1133,7 +1133,7 @@ static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
 
        /* start garbage collection */
        clear_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
-       queue_delayed_work(priv->wq, &priv->neigh_reap_task,
+       queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task,
                           arp_tbl.gc_interval);
 
        return 0;
@@ -1293,7 +1293,7 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
        return 0;
 
 out_dev_uninit:
-       ipoib_ib_dev_cleanup(dev);
+       ipoib_ib_dev_cleanup();
 
 out_tx_ring_cleanup:
        vfree(priv->tx_ring);
@@ -1646,7 +1646,7 @@ register_failed:
        /* Stop GC if started before flush */
        set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
        cancel_delayed_work(&priv->neigh_reap_task);
-       flush_workqueue(priv->wq);
+       flush_workqueue(ipoib_workqueue);
 
 event_failed:
        ipoib_dev_cleanup(priv->dev);
@@ -1717,7 +1717,7 @@ static void ipoib_remove_one(struct ib_device *device)
                /* Stop GC */
                set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
                cancel_delayed_work(&priv->neigh_reap_task);
-               flush_workqueue(priv->wq);
+               flush_workqueue(ipoib_workqueue);
 
                unregister_netdev(priv->dev);
                free_netdev(priv->dev);
@@ -1758,13 +1758,8 @@ static int __init ipoib_init_module(void)
         * unregister_netdev() and linkwatch_event take the rtnl lock,
         * so flush_scheduled_work() can deadlock during device
         * removal.
-        *
-        * In addition, bringing one device up and another down at the
-        * same time can deadlock a single workqueue, so we have this
-        * global fallback workqueue, but we also attempt to open a
-        * per device workqueue each time we bring an interface up
         */
-       ipoib_workqueue = create_singlethread_workqueue("ipoib_flush");
+       ipoib_workqueue = create_singlethread_workqueue("ipoib");
        if (!ipoib_workqueue) {
                ret = -ENOMEM;
                goto err_fs;
index 845f910eb21401a9b38fc31e7ab005c74034c517..41325960e4e0410ab91c0f982e9a654fecfc0f1a 100644 (file)
@@ -388,7 +388,7 @@ void ipoib_mcast_carrier_on_task(struct work_struct *work)
         * the workqueue while holding the rtnl lock, so loop
         * on trylock until either we get the lock or we see
         * FLAG_ADMIN_UP go away as that signals that we are bailing
-        * and can safely ignore the carrier on work.
+        * and can safely ignore the carrier on work
         */
        while (!rtnl_trylock()) {
                if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
@@ -432,14 +432,15 @@ static int ipoib_mcast_join_complete(int status,
        if (!status) {
                mcast->backoff = 1;
                if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
-                       queue_delayed_work(priv->wq, &priv->mcast_task, 0);
+                       queue_delayed_work(ipoib_workqueue,
+                                          &priv->mcast_task, 0);
 
                /*
-                * Defer carrier on work to priv->wq to avoid a
+                * Defer carrier on work to ipoib_workqueue to avoid a
                 * deadlock on rtnl_lock here.
                 */
                if (mcast == priv->broadcast)
-                       queue_work(priv->wq, &priv->carrier_on_task);
+                       queue_work(ipoib_workqueue, &priv->carrier_on_task);
        } else {
                if (mcast->logcount++ < 20) {
                        if (status == -ETIMEDOUT || status == -EAGAIN) {
@@ -464,7 +465,7 @@ out:
        if (status == -ENETRESET)
                status = 0;
        if (status && test_bit(IPOIB_MCAST_RUN, &priv->flags))
-               queue_delayed_work(priv->wq, &priv->mcast_task,
+               queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
                                   mcast->backoff * HZ);
        spin_unlock_irq(&priv->lock);
        mutex_unlock(&mcast_mutex);
@@ -534,7 +535,8 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
                        mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
 
                if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
-                       queue_delayed_work(priv->wq, &priv->mcast_task,
+                       queue_delayed_work(ipoib_workqueue,
+                                          &priv->mcast_task,
                                           mcast->backoff * HZ);
        }
        mutex_unlock(&mcast_mutex);
@@ -574,8 +576,8 @@ void ipoib_mcast_join_task(struct work_struct *work)
                        ipoib_warn(priv, "failed to allocate broadcast group\n");
                        mutex_lock(&mcast_mutex);
                        if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
-                               queue_delayed_work(priv->wq, &priv->mcast_task,
-                                                  HZ);
+                               queue_delayed_work(ipoib_workqueue,
+                                                  &priv->mcast_task, HZ);
                        mutex_unlock(&mcast_mutex);
                        return;
                }
@@ -642,7 +644,7 @@ int ipoib_mcast_start_thread(struct net_device *dev)
 
        mutex_lock(&mcast_mutex);
        if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
-               queue_delayed_work(priv->wq, &priv->mcast_task, 0);
+               queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
        mutex_unlock(&mcast_mutex);
 
        return 0;
@@ -660,7 +662,7 @@ int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
        mutex_unlock(&mcast_mutex);
 
        if (flush)
-               flush_workqueue(priv->wq);
+               flush_workqueue(ipoib_workqueue);
 
        return 0;
 }
@@ -727,7 +729,7 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
                __ipoib_mcast_add(dev, mcast);
                list_add_tail(&mcast->list, &priv->multicast_list);
                if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
-                       queue_delayed_work(priv->wq, &priv->mcast_task, 0);
+                       queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
        }
 
        if (!mcast->ah) {
@@ -942,7 +944,7 @@ void ipoib_mcast_restart_task(struct work_struct *work)
         * completes.  So do like the carrier on task and attempt to
         * take the rtnl lock, but if we can't before the ADMIN_UP flag
         * goes away, then just return and know that the remove list will
-        * get flushed later by mcast_stop_thread.
+        * get flushed later by mcast_dev_flush.
         */
        while (!rtnl_trylock()) {
                if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
index b72a753eb41dc3031608269c56434ed507b96f5f..c56d5d44c53b3f11725b6d6da220ea2c440fe496 100644 (file)
@@ -145,20 +145,10 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
        int ret, size;
        int i;
 
-       /*
-        * the various IPoIB tasks assume they will never race against
-        * themselves, so always use a single thread workqueue
-        */
-       priv->wq = create_singlethread_workqueue("ipoib_wq");
-       if (!priv->wq) {
-               printk(KERN_WARNING "ipoib: failed to allocate device WQ\n");
-               return -ENODEV;
-       }
-
        priv->pd = ib_alloc_pd(priv->ca);
        if (IS_ERR(priv->pd)) {
                printk(KERN_WARNING "%s: failed to allocate PD\n", ca->name);
-               goto out_free_wq;
+               return -ENODEV;
        }
 
        priv->mr = ib_get_dma_mr(priv->pd, IB_ACCESS_LOCAL_WRITE);
@@ -252,10 +242,6 @@ out_free_mr:
 
 out_free_pd:
        ib_dealloc_pd(priv->pd);
-
-out_free_wq:
-       destroy_workqueue(priv->wq);
-       priv->wq = NULL;
        return -ENODEV;
 }
 
@@ -284,12 +270,6 @@ void ipoib_transport_dev_cleanup(struct net_device *dev)
 
        if (ib_dealloc_pd(priv->pd))
                ipoib_warn(priv, "ib_dealloc_pd failed\n");
-
-       if (priv->wq) {
-               flush_workqueue(priv->wq);
-               destroy_workqueue(priv->wq);
-               priv->wq = NULL;
-       }
 }
 
 void ipoib_event(struct ib_event_handler *handler,