1 files changed, 231 insertions, 50 deletions

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index dee05a353dbd..77683909ca3d 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -60,6 +60,7 @@
 #include <asm/hvcall.h>
 #include <linux/atomic.h>
 #include <asm/vio.h>
+#include <asm/xive.h>
 #include <asm/iommu.h>
 #include <linux/uaccess.h>
 #include <asm/firmware.h>
@@ -1429,6 +1430,15 @@ static int __ibmvnic_open(struct net_device *netdev)
 		return rc;
 	}
 
+	adapter->tx_queues_active = true;
+
+	/* Since queues were stopped until now, there shouldn't be any
+	 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
+	 * don't need the synchronize_rcu()? Leaving it for consistency
+	 * with setting ->tx_queues_active = false.
+	 */
+	synchronize_rcu();
+
 	netif_tx_start_all_queues(netdev);
 
 	if (prev_state == VNIC_CLOSED) {
@@ -1603,6 +1613,14 @@ static void ibmvnic_cleanup(struct net_device *netdev)
 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 
 	/* ensure that transmissions are stopped if called by do_reset */
+
+	adapter->tx_queues_active = false;
+
+	/* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
+	 * update so they don't restart a queue after we stop it below.
+	 */
+	synchronize_rcu();
+
 	if (test_bit(0, &adapter->resetting))
 		netif_tx_disable(netdev);
 	else
@@ -1842,14 +1860,21 @@ static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
 		tx_buff->skb = NULL;
 		adapter->netdev->stats.tx_dropped++;
 	}
+
 	ind_bufp->index = 0;
+
 	if (atomic_sub_return(entries, &tx_scrq->used) <=
 	    (adapter->req_tx_entries_per_subcrq / 2) &&
-	    __netif_subqueue_stopped(adapter->netdev, queue_num) &&
-	    !test_bit(0, &adapter->resetting)) {
-		netif_wake_subqueue(adapter->netdev, queue_num);
-		netdev_dbg(adapter->netdev, "Started queue %d\n",
-			   queue_num);
+	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
+		rcu_read_lock();
+
+		if (adapter->tx_queues_active) {
+			netif_wake_subqueue(adapter->netdev, queue_num);
+			netdev_dbg(adapter->netdev, "Started queue %d\n",
+				   queue_num);
+		}
+
+		rcu_read_unlock();
 	}
 }
 
@@ -1904,11 +1929,12 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
 	int index = 0;
 	u8 proto = 0;
 
-	tx_scrq = adapter->tx_scrq[queue_num];
-	txq = netdev_get_tx_queue(netdev, queue_num);
-	ind_bufp = &tx_scrq->ind_buf;
-
-	if (test_bit(0, &adapter->resetting)) {
+	/* If a reset is in progress, drop the packet since
+	 * the scrqs may get torn down. Otherwise use the
+	 * rcu to ensure reset waits for us to complete.
+	 */
+	rcu_read_lock();
+	if (!adapter->tx_queues_active) {
 		dev_kfree_skb_any(skb);
 
 		tx_send_failed++;
@@ -1917,6 +1943,10 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
 		goto out;
 	}
 
+	tx_scrq = adapter->tx_scrq[queue_num];
+	txq = netdev_get_tx_queue(netdev, queue_num);
+	ind_bufp = &tx_scrq->ind_buf;
+
 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
 		tx_dropped++;
 		tx_send_failed++;
@@ -1924,6 +1954,7 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
 		goto out;
 	}
+
 	if (skb_is_gso(skb))
 		tx_pool = &adapter->tso_pool[queue_num];
 	else
@@ -2078,6 +2109,7 @@ tx_err:
 		netif_carrier_off(netdev);
 	}
 out:
+	rcu_read_unlock();
 	netdev->stats.tx_dropped += tx_dropped;
 	netdev->stats.tx_bytes += tx_bytes;
 	netdev->stats.tx_packets += tx_packets;
@@ -2213,6 +2245,19 @@ static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
 }
 
 /*
+ * Initialize the init_done completion and return code values. We
+ * can get a transport event just after registering the CRQ and the
+ * tasklet will use this to communicate the transport event. To ensure
+ * we don't miss the notification/error, initialize these _before_
+ * regisering the CRQ.
+ */
+static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
+{
+	reinit_completion(&adapter->init_done);
+	adapter->init_done_rc = 0;
+}
+
+/*
  * do_reset returns zero if we are able to keep processing reset events, or
  * non-zero if we hit a fatal error and must halt.
  */
@@ -2318,6 +2363,8 @@ static int do_reset(struct ibmvnic_adapter *adapter,
 		 */
 		adapter->state = VNIC_PROBED;
 
+		reinit_init_done(adapter);
+
 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
 			rc = init_crq_queue(adapter);
 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
@@ -2461,7 +2508,8 @@ static int do_hard_reset(struct ibmvnic_adapter *adapter,
 	 */
 	adapter->state = VNIC_PROBED;
 
-	reinit_completion(&adapter->init_done);
+	reinit_init_done(adapter);
+
 	rc = init_crq_queue(adapter);
 	if (rc) {
 		netdev_err(adapter->netdev,
@@ -2602,23 +2650,82 @@ out:
 static void __ibmvnic_reset(struct work_struct *work)
 {
 	struct ibmvnic_adapter *adapter;
-	bool saved_state = false;
+	unsigned int timeout = 5000;
 	struct ibmvnic_rwi *tmprwi;
+	bool saved_state = false;
 	struct ibmvnic_rwi *rwi;
 	unsigned long flags;
-	u32 reset_state;
+	struct device *dev;
+	bool need_reset;
 	int num_fails = 0;
+	u32 reset_state;
 	int rc = 0;
 
 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
+		dev = &adapter->vdev->dev;
 
-	if (test_and_set_bit_lock(0, &adapter->resetting)) {
+	/* Wait for ibmvnic_probe() to complete. If probe is taking too long
+	 * or if another reset is in progress, defer work for now. If probe
+	 * eventually fails it will flush and terminate our work.
+	 *
+	 * Three possibilities here:
+	 * 1. Adpater being removed  - just return
+	 * 2. Timed out on probe or another reset in progress - delay the work
+	 * 3. Completed probe - perform any resets in queue
+	 */
+	if (adapter->state == VNIC_PROBING &&
+	    !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
+		dev_err(dev, "Reset thread timed out on probe");
 		queue_delayed_work(system_long_wq,
 				   &adapter->ibmvnic_delayed_reset,
 				   IBMVNIC_RESET_DELAY);
 		return;
 	}
 
+	/* adapter is done with probe (i.e state is never VNIC_PROBING now) */
+	if (adapter->state == VNIC_REMOVING)
+		return;
+
+	/* ->rwi_list is stable now (no one else is removing entries) */
+
+	/* ibmvnic_probe() may have purged the reset queue after we were
+	 * scheduled to process a reset so there maybe no resets to process.
+	 * Before setting the ->resetting bit though, we have to make sure
+	 * that there is infact a reset to process. Otherwise we may race
+	 * with ibmvnic_open() and end up leaving the vnic down:
+	 *
+	 *	__ibmvnic_reset()	    ibmvnic_open()
+	 *	-----------------	    --------------
+	 *
+	 *  set ->resetting bit
+	 *  				find ->resetting bit is set
+	 *  				set ->state to IBMVNIC_OPEN (i.e
+	 *  				assume reset will open device)
+	 *  				return
+	 *  find reset queue empty
+	 *  return
+	 *
+	 *  	Neither performed vnic login/open and vnic stays down
+	 *
+	 * If we hold the lock and conditionally set the bit, either we
+	 * or ibmvnic_open() will complete the open.
+	 */
+	need_reset = false;
+	spin_lock(&adapter->rwi_lock);
+	if (!list_empty(&adapter->rwi_list)) {
+		if (test_and_set_bit_lock(0, &adapter->resetting)) {
+			queue_delayed_work(system_long_wq,
+					   &adapter->ibmvnic_delayed_reset,
+					   IBMVNIC_RESET_DELAY);
+		} else {
+			need_reset = true;
+		}
+	}
+	spin_unlock(&adapter->rwi_lock);
+
+	if (!need_reset)
+		return;
+
 	rwi = get_next_rwi(adapter);
 	while (rwi) {
 		spin_lock_irqsave(&adapter->state_lock, flags);
@@ -2735,12 +2842,23 @@ static void __ibmvnic_delayed_reset(struct work_struct *work)
 	__ibmvnic_reset(&adapter->ibmvnic_reset);
 }
 
+static void flush_reset_queue(struct ibmvnic_adapter *adapter)
+{
+	struct list_head *entry, *tmp_entry;
+
+	if (!list_empty(&adapter->rwi_list)) {
+		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
+			list_del(entry);
+			kfree(list_entry(entry, struct ibmvnic_rwi, list));
+		}
+	}
+}
+
 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
 			 enum ibmvnic_reset_reason reason)
 {
-	struct list_head *entry, *tmp_entry;
-	struct ibmvnic_rwi *rwi, *tmp;
 	struct net_device *netdev = adapter->netdev;
+	struct ibmvnic_rwi *rwi, *tmp;
 	unsigned long flags;
 	int ret;
 
@@ -2759,13 +2877,6 @@ static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
 		goto err;
 	}
 
-	if (adapter->state == VNIC_PROBING) {
-		netdev_warn(netdev, "Adapter reset during probe\n");
-		adapter->init_done_rc = -EAGAIN;
-		ret = EAGAIN;
-		goto err;
-	}
-
 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
 		if (tmp->reset_reason == reason) {
 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
@@ -2783,10 +2894,9 @@ static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
 	/* if we just received a transport event,
 	 * flush reset queue and process this reset
 	 */
-	if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
-		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
-			list_del(entry);
-	}
+	if (adapter->force_reset_recovery)
+		flush_reset_queue(adapter);
+
 	rwi->reset_reason = reason;
 	list_add_tail(&rwi->list, &adapter->rwi_list);
 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
@@ -3562,6 +3672,30 @@ static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
 	return rc;
 }
 
+/* We can not use the IRQ chip EOI handler because that has the
+ * unintended effect of changing the interrupt priority.
+ */
+static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
+{
+	u64 val = 0xff000000 | scrq->hw_irq;
+	unsigned long rc;
+
+	rc = plpar_hcall_norets(H_EOI, val);
+	if (rc)
+		dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
+}
+
+/* Due to a firmware bug, the hypervisor can send an interrupt to a
+ * transmit or receive queue just prior to a partition migration.
+ * Force an EOI after migration.
+ */
+static void ibmvnic_clear_pending_interrupt(struct device *dev,
+					    struct ibmvnic_sub_crq_queue *scrq)
+{
+	if (!xive_enabled())
+		ibmvnic_xics_eoi(dev, scrq);
+}
+
 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
 			   struct ibmvnic_sub_crq_queue *scrq)
 {
@@ -3575,15 +3709,7 @@ static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
 
 	if (test_bit(0, &adapter->resetting) &&
 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
-		u64 val = (0xff000000) | scrq->hw_irq;
-
-		rc = plpar_hcall_norets(H_EOI, val);
-		/* H_EOI would fail with rc = H_FUNCTION when running
-		 * in XIVE mode which is expected, but not an error.
-		 */
-		if (rc && (rc != H_FUNCTION))
-			dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
-				val, rc);
+		ibmvnic_clear_pending_interrupt(dev, scrq);
 	}
 
 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
@@ -3654,9 +3780,15 @@ restart_loop:
 		    (adapter->req_tx_entries_per_subcrq / 2) &&
 		    __netif_subqueue_stopped(adapter->netdev,
 					     scrq->pool_index)) {
-			netif_wake_subqueue(adapter->netdev, scrq->pool_index);
-			netdev_dbg(adapter->netdev, "Started queue %d\n",
-				   scrq->pool_index);
+			rcu_read_lock();
+			if (adapter->tx_queues_active) {
+				netif_wake_subqueue(adapter->netdev,
+						    scrq->pool_index);
+				netdev_dbg(adapter->netdev,
+					   "Started queue %d\n",
+					   scrq->pool_index);
+			}
+			rcu_read_unlock();
 		}
 	}
 
@@ -5321,9 +5453,9 @@ static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
 			}
 
 			if (!completion_done(&adapter->init_done)) {
-				complete(&adapter->init_done);
 				if (!adapter->init_done_rc)
 					adapter->init_done_rc = -EAGAIN;
+				complete(&adapter->init_done);
 			}
 
 			break;
@@ -5346,6 +5478,13 @@ static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
 			adapter->fw_done_rc = -EIO;
 			complete(&adapter->fw_done);
 		}
+
+		/* if we got here during crq-init, retry crq-init */
+		if (!completion_done(&adapter->init_done)) {
+			adapter->init_done_rc = -EAGAIN;
+			complete(&adapter->init_done);
+		}
+
 		if (!completion_done(&adapter->stats_done))
 			complete(&adapter->stats_done);
 		if (test_bit(0, &adapter->resetting))
@@ -5662,10 +5801,6 @@ static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
 
 	adapter->from_passive_init = false;
 
-	if (reset)
-		reinit_completion(&adapter->init_done);
-
-	adapter->init_done_rc = 0;
 	rc = ibmvnic_send_crq_init(adapter);
 	if (rc) {
 		dev_err(dev, "Send crq init failed with error %d\n", rc);
@@ -5679,12 +5814,14 @@ static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
 
 	if (adapter->init_done_rc) {
 		release_crq_queue(adapter);
+		dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
 		return adapter->init_done_rc;
 	}
 
 	if (adapter->from_passive_init) {
 		adapter->state = VNIC_OPEN;
 		adapter->from_passive_init = false;
+		dev_err(dev, "CRQ-init failed, passive-init\n");
 		return -EINVAL;
 	}
 
@@ -5724,6 +5861,7 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
 	struct ibmvnic_adapter *adapter;
 	struct net_device *netdev;
 	unsigned char *mac_addr_p;
+	unsigned long flags;
 	bool init_success;
 	int rc;
 
@@ -5768,6 +5906,7 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
 	spin_lock_init(&adapter->rwi_lock);
 	spin_lock_init(&adapter->state_lock);
 	mutex_init(&adapter->fw_lock);
+	init_completion(&adapter->probe_done);
 	init_completion(&adapter->init_done);
 	init_completion(&adapter->fw_done);
 	init_completion(&adapter->reset_done);
@@ -5778,6 +5917,33 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
 
 	init_success = false;
 	do {
+		reinit_init_done(adapter);
+
+		/* clear any failovers we got in the previous pass
+		 * since we are reinitializing the CRQ
+		 */
+		adapter->failover_pending = false;
+
+		/* If we had already initialized CRQ, we may have one or
+		 * more resets queued already. Discard those and release
+		 * the CRQ before initializing the CRQ again.
+		 */
+		release_crq_queue(adapter);
+
+		/* Since we are still in PROBING state, __ibmvnic_reset()
+		 * will not access the ->rwi_list and since we released CRQ,
+		 * we won't get _new_ transport events. But there maybe an
+		 * ongoing ibmvnic_reset() call. So serialize access to
+		 * rwi_list. If we win the race, ibvmnic_reset() could add
+		 * a reset after we purged but thats ok - we just may end
+		 * up with an extra reset (i.e similar to having two or more
+		 * resets in the queue at once).
+		 * CHECK.
+		 */
+		spin_lock_irqsave(&adapter->rwi_lock, flags);
+		flush_reset_queue(adapter);
+		spin_unlock_irqrestore(&adapter->rwi_lock, flags);
+
 		rc = init_crq_queue(adapter);
 		if (rc) {
 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
@@ -5809,12 +5975,6 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
 		goto ibmvnic_dev_file_err;
 
 	netif_carrier_off(netdev);
-	rc = register_netdev(netdev);
-	if (rc) {
-		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
-		goto ibmvnic_register_fail;
-	}
-	dev_info(&dev->dev, "ibmvnic registered\n");
 
 	if (init_success) {
 		adapter->state = VNIC_PROBED;
@@ -5827,6 +5987,16 @@ static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
 
 	adapter->wait_for_reset = false;
 	adapter->last_reset_time = jiffies;
+
+	rc = register_netdev(netdev);
+	if (rc) {
+		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
+		goto ibmvnic_register_fail;
+	}
+	dev_info(&dev->dev, "ibmvnic registered\n");
+
+	complete(&adapter->probe_done);
+
 	return 0;
 
 ibmvnic_register_fail:
@@ -5841,6 +6011,17 @@ ibmvnic_stats_fail:
 ibmvnic_init_fail:
 	release_sub_crqs(adapter, 1);
 	release_crq_queue(adapter);
+
+	/* cleanup worker thread after releasing CRQ so we don't get
+	 * transport events (i.e new work items for the worker thread).
+	 */
+	adapter->state = VNIC_REMOVING;
+	complete(&adapter->probe_done);
+	flush_work(&adapter->ibmvnic_reset);
+	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
+
+	flush_reset_queue(adapter);
+
 	mutex_destroy(&adapter->fw_lock);
 	free_netdev(netdev);