Merge branch 'for-davem' of git://git.kernel.org/pub/scm/linux/kernel/git/bwh/sfc-next

Ben Hutchings says:

====================
1. Change the TX path to stop queues earlier and avoid returning
NETDEV_TX_BUSY.
2. Remove some inefficiencies in soft-TSO.
3. Fix various bugs involving device state transitions and/or reset
scheduling by error handlers.
4. Take advantage of my previous change to operstate initialisation.
5. Miscellaneous cleanup.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 410 insertions, 507 deletions

diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c
index 65a8d49106a4..a606db43c5ba 100644
--- a/drivers/net/ethernet/sfc/efx.c
+++ b/drivers/net/ethernet/sfc/efx.c
@@ -202,11 +202,21 @@ static void efx_stop_all(struct efx_nic *efx);
 
 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
 	do {						\
-		if ((efx->state == STATE_RUNNING) ||	\
+		if ((efx->state == STATE_READY) ||	\
 		    (efx->state == STATE_DISABLED))	\
 			ASSERT_RTNL();			\
 	} while (0)
 
+static int efx_check_disabled(struct efx_nic *efx)
+{
+	if (efx->state == STATE_DISABLED) {
+		netif_err(efx, drv, efx->net_dev,
+			  "device is disabled due to earlier errors\n");
+		return -EIO;
+	}
+	return 0;
+}
+
 /**************************************************************************
  *
  * Event queue processing
@@ -630,6 +640,16 @@ static void efx_start_datapath(struct efx_nic *efx)
 	efx->rx_buffer_order = get_order(efx->rx_buffer_len +
 					 sizeof(struct efx_rx_page_state));
 
+	/* We must keep at least one descriptor in a TX ring empty.
+	 * We could avoid this when the queue size does not exactly
+	 * match the hardware ring size, but it's not that important.
+	 * Therefore we stop the queue when one more skb might fill
+	 * the ring completely.  We wake it when half way back to
+	 * empty.
+	 */
+	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
+	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
+
 	/* Initialise the channels */
 	efx_for_each_channel(channel, efx) {
 		efx_for_each_channel_tx_queue(tx_queue, channel)
@@ -730,7 +750,11 @@ efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
 	u32 old_rxq_entries, old_txq_entries;
 	unsigned i, next_buffer_table = 0;
-	int rc = 0;
+	int rc;
+
+	rc = efx_check_disabled(efx);
+	if (rc)
+		return rc;
 
 	/* Not all channels should be reallocated. We must avoid
 	 * reallocating their buffer table entries.
@@ -1365,6 +1389,8 @@ static void efx_start_interrupts(struct efx_nic *efx, bool may_keep_eventq)
 {
 	struct efx_channel *channel;
 
+	BUG_ON(efx->state == STATE_DISABLED);
+
 	if (efx->legacy_irq)
 		efx->legacy_irq_enabled = true;
 	efx_nic_enable_interrupts(efx);
@@ -1382,6 +1408,9 @@ static void efx_stop_interrupts(struct efx_nic *efx, bool may_keep_eventq)
 {
 	struct efx_channel *channel;
 
+	if (efx->state == STATE_DISABLED)
+		return;
+
 	efx_mcdi_mode_poll(efx);
 
 	efx_nic_disable_interrupts(efx);
@@ -1533,22 +1562,21 @@ static int efx_probe_all(struct efx_nic *efx)
 	return rc;
 }
 
-/* Called after previous invocation(s) of efx_stop_all, restarts the port,
- * kernel transmit queues and NAPI processing, and ensures that the port is
- * scheduled to be reconfigured. This function is safe to call multiple
- * times when the NIC is in any state.
+/* If the interface is supposed to be running but is not, start
+ * the hardware and software data path, regular activity for the port
+ * (MAC statistics, link polling, etc.) and schedule the port to be
+ * reconfigured.  Interrupts must already be enabled.  This function
+ * is safe to call multiple times, so long as the NIC is not disabled.
+ * Requires the RTNL lock.
  */
 static void efx_start_all(struct efx_nic *efx)
 {
 	EFX_ASSERT_RESET_SERIALISED(efx);
+	BUG_ON(efx->state == STATE_DISABLED);
 
 	/* Check that it is appropriate to restart the interface. All
 	 * of these flags are safe to read under just the rtnl lock */
-	if (efx->port_enabled)
-		return;
-	if ((efx->state != STATE_RUNNING) && (efx->state != STATE_INIT))
-		return;
-	if (!netif_running(efx->net_dev))
+	if (efx->port_enabled || !netif_running(efx->net_dev))
 		return;
 
 	efx_start_port(efx);
@@ -1582,11 +1610,11 @@ static void efx_flush_all(struct efx_nic *efx)
 	cancel_work_sync(&efx->mac_work);
 }
 
-/* Quiesce hardware and software without bringing the link down.
- * Safe to call multiple times, when the nic and interface is in any
- * state. The caller is guaranteed to subsequently be in a position
- * to modify any hardware and software state they see fit without
- * taking locks. */
+/* Quiesce the hardware and software data path, and regular activity
+ * for the port without bringing the link down.  Safe to call multiple
+ * times with the NIC in almost any state, but interrupts should be
+ * enabled.  Requires the RTNL lock.
+ */
 static void efx_stop_all(struct efx_nic *efx)
 {
 	EFX_ASSERT_RESET_SERIALISED(efx);
@@ -1739,8 +1767,6 @@ static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
 	struct efx_nic *efx = netdev_priv(net_dev);
 	struct mii_ioctl_data *data = if_mii(ifr);
 
-	EFX_ASSERT_RESET_SERIALISED(efx);
-
 	/* Convert phy_id from older PRTAD/DEVAD format */
 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
 	    (data->phy_id & 0xfc00) == 0x0400)
@@ -1820,13 +1846,14 @@ static void efx_netpoll(struct net_device *net_dev)
 static int efx_net_open(struct net_device *net_dev)
 {
 	struct efx_nic *efx = netdev_priv(net_dev);
-	EFX_ASSERT_RESET_SERIALISED(efx);
+	int rc;
 
 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
 		  raw_smp_processor_id());
 
-	if (efx->state == STATE_DISABLED)
-		return -EIO;
+	rc = efx_check_disabled(efx);
+	if (rc)
+		return rc;
 	if (efx->phy_mode & PHY_MODE_SPECIAL)
 		return -EBUSY;
 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
@@ -1852,10 +1879,8 @@ static int efx_net_stop(struct net_device *net_dev)
 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
 		  raw_smp_processor_id());
 
-	if (efx->state != STATE_DISABLED) {
-		/* Stop the device and flush all the channels */
-		efx_stop_all(efx);
-	}
+	/* Stop the device and flush all the channels */
+	efx_stop_all(efx);
 
 	return 0;
 }
@@ -1915,9 +1940,11 @@ static void efx_watchdog(struct net_device *net_dev)
 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
 {
 	struct efx_nic *efx = netdev_priv(net_dev);
+	int rc;
 
-	EFX_ASSERT_RESET_SERIALISED(efx);
-
+	rc = efx_check_disabled(efx);
+	if (rc)
+		return rc;
 	if (new_mtu > EFX_MAX_MTU)
 		return -EINVAL;
 
@@ -1926,8 +1953,6 @@ static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
 
 	mutex_lock(&efx->mac_lock);
-	/* Reconfigure the MAC before enabling the dma queues so that
-	 * the RX buffers don't overflow */
 	net_dev->mtu = new_mtu;
 	efx->type->reconfigure_mac(efx);
 	mutex_unlock(&efx->mac_lock);
@@ -1942,8 +1967,6 @@ static int efx_set_mac_address(struct net_device *net_dev, void *data)
 	struct sockaddr *addr = data;
 	char *new_addr = addr->sa_data;
 
-	EFX_ASSERT_RESET_SERIALISED(efx);
-
 	if (!is_valid_ether_addr(new_addr)) {
 		netif_err(efx, drv, efx->net_dev,
 			  "invalid ethernet MAC address requested: %pM\n",
@@ -2079,11 +2102,27 @@ static int efx_register_netdev(struct efx_nic *efx)
 
 	rtnl_lock();
 
+	/* Enable resets to be scheduled and check whether any were
+	 * already requested.  If so, the NIC is probably hosed so we
+	 * abort.
+	 */
+	efx->state = STATE_READY;
+	smp_mb(); /* ensure we change state before checking reset_pending */
+	if (efx->reset_pending) {
+		netif_err(efx, probe, efx->net_dev,
+			  "aborting probe due to scheduled reset\n");
+		rc = -EIO;
+		goto fail_locked;
+	}
+
 	rc = dev_alloc_name(net_dev, net_dev->name);
 	if (rc < 0)
 		goto fail_locked;
 	efx_update_name(efx);
 
+	/* Always start with carrier off; PHY events will detect the link */
+	netif_carrier_off(net_dev);
+
 	rc = register_netdevice(net_dev);
 	if (rc)
 		goto fail_locked;
@@ -2094,9 +2133,6 @@ static int efx_register_netdev(struct efx_nic *efx)
 			efx_init_tx_queue_core_txq(tx_queue);
 	}
 
-	/* Always start with carrier off; PHY events will detect the link */
-	netif_carrier_off(net_dev);
-
 	rtnl_unlock();
 
 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
@@ -2108,14 +2144,14 @@ static int efx_register_netdev(struct efx_nic *efx)
 
 	return 0;
 
+fail_registered:
+	rtnl_lock();
+	unregister_netdevice(net_dev);
 fail_locked:
+	efx->state = STATE_UNINIT;
 	rtnl_unlock();
 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
 	return rc;
-
-fail_registered:
-	unregister_netdev(net_dev);
-	return rc;
 }
 
 static void efx_unregister_netdev(struct efx_nic *efx)
@@ -2138,7 +2174,11 @@ static void efx_unregister_netdev(struct efx_nic *efx)
 
 	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
-	unregister_netdev(efx->net_dev);
+
+	rtnl_lock();
+	unregister_netdevice(efx->net_dev);
+	efx->state = STATE_UNINIT;
+	rtnl_unlock();
 }
 
 /**************************************************************************
@@ -2154,9 +2194,9 @@ void efx_reset_down(struct efx_nic *efx, enum reset_type method)
 	EFX_ASSERT_RESET_SERIALISED(efx);
 
 	efx_stop_all(efx);
-	mutex_lock(&efx->mac_lock);
-
 	efx_stop_interrupts(efx, false);
+
+	mutex_lock(&efx->mac_lock);
 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
 		efx->phy_op->fini(efx);
 	efx->type->fini(efx);
@@ -2276,16 +2316,15 @@ static void efx_reset_work(struct work_struct *data)
 	if (!pending)
 		return;
 
-	/* If we're not RUNNING then don't reset. Leave the reset_pending
-	 * flags set so that efx_pci_probe_main will be retried */
-	if (efx->state != STATE_RUNNING) {
-		netif_info(efx, drv, efx->net_dev,
-			   "scheduled reset quenched. NIC not RUNNING\n");
-		return;
-	}
-
 	rtnl_lock();
-	(void)efx_reset(efx, fls(pending) - 1);
+
+	/* We checked the state in efx_schedule_reset() but it may
+	 * have changed by now.  Now that we have the RTNL lock,
+	 * it cannot change again.
+	 */
+	if (efx->state == STATE_READY)
+		(void)efx_reset(efx, fls(pending) - 1);
+
 	rtnl_unlock();
 }
 
@@ -2311,6 +2350,13 @@ void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
 	}
 
 	set_bit(method, &efx->reset_pending);
+	smp_mb(); /* ensure we change reset_pending before checking state */
+
+	/* If we're not READY then just leave the flags set as the cue
+	 * to abort probing or reschedule the reset later.
+	 */
+	if (ACCESS_ONCE(efx->state) != STATE_READY)
+		return;
 
 	/* efx_process_channel() will no longer read events once a
 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
@@ -2376,13 +2422,12 @@ static const struct efx_phy_operations efx_dummy_phy_operations = {
 /* This zeroes out and then fills in the invariants in a struct
  * efx_nic (including all sub-structures).
  */
-static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
+static int efx_init_struct(struct efx_nic *efx,
 			   struct pci_dev *pci_dev, struct net_device *net_dev)
 {
 	int i;
 
 	/* Initialise common structures */
-	memset(efx, 0, sizeof(*efx));
 	spin_lock_init(&efx->biu_lock);
 #ifdef CONFIG_SFC_MTD
 	INIT_LIST_HEAD(&efx->mtd_list);
@@ -2392,7 +2437,7 @@ static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
 	efx->pci_dev = pci_dev;
 	efx->msg_enable = debug;
-	efx->state = STATE_INIT;
+	efx->state = STATE_UNINIT;
 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
 
 	efx->net_dev = net_dev;
@@ -2409,8 +2454,6 @@ static int efx_init_struct(struct efx_nic *efx, const struct efx_nic_type *type,
 			goto fail;
 	}
 
-	efx->type = type;
-
 	EFX_BUG_ON_PARANOID(efx->type->phys_addr_channels > EFX_MAX_CHANNELS);
 
 	/* Higher numbered interrupt modes are less capable! */
@@ -2455,6 +2498,12 @@ static void efx_fini_struct(struct efx_nic *efx)
  */
 static void efx_pci_remove_main(struct efx_nic *efx)
 {
+	/* Flush reset_work. It can no longer be scheduled since we
+	 * are not READY.
+	 */
+	BUG_ON(efx->state == STATE_READY);
+	cancel_work_sync(&efx->reset_work);
+
 #ifdef CONFIG_RFS_ACCEL
 	free_irq_cpu_rmap(efx->net_dev->rx_cpu_rmap);
 	efx->net_dev->rx_cpu_rmap = NULL;
@@ -2480,24 +2529,15 @@ static void efx_pci_remove(struct pci_dev *pci_dev)
 
 	/* Mark the NIC as fini, then stop the interface */
 	rtnl_lock();
-	efx->state = STATE_FINI;
 	dev_close(efx->net_dev);
-
-	/* Allow any queued efx_resets() to complete */
+	efx_stop_interrupts(efx, false);
 	rtnl_unlock();
 
-	efx_stop_interrupts(efx, false);
 	efx_sriov_fini(efx);
 	efx_unregister_netdev(efx);
 
 	efx_mtd_remove(efx);
 
-	/* Wait for any scheduled resets to complete. No more will be
-	 * scheduled from this point because efx_stop_all() has been
-	 * called, we are no longer registered with driverlink, and
-	 * the net_device's have been removed. */
-	cancel_work_sync(&efx->reset_work);
-
 	efx_pci_remove_main(efx);
 
 	efx_fini_io(efx);
@@ -2617,7 +2657,6 @@ static int efx_pci_probe_main(struct efx_nic *efx)
 static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
 				   const struct pci_device_id *entry)
 {
-	const struct efx_nic_type *type = (const struct efx_nic_type *) entry->driver_data;
 	struct net_device *net_dev;
 	struct efx_nic *efx;
 	int rc;
@@ -2627,10 +2666,12 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
 				     EFX_MAX_RX_QUEUES);
 	if (!net_dev)
 		return -ENOMEM;
-	net_dev->features |= (type->offload_features | NETIF_F_SG |
+	efx = netdev_priv(net_dev);
+	efx->type = (const struct efx_nic_type *) entry->driver_data;
+	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
 			      NETIF_F_HIGHDMA | NETIF_F_TSO |
 			      NETIF_F_RXCSUM);
-	if (type->offload_features & NETIF_F_V6_CSUM)
+	if (efx->type->offload_features & NETIF_F_V6_CSUM)
 		net_dev->features |= NETIF_F_TSO6;
 	/* Mask for features that also apply to VLAN devices */
 	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
@@ -2638,10 +2679,9 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
 				   NETIF_F_RXCSUM);
 	/* All offloads can be toggled */
 	net_dev->hw_features = net_dev->features & ~NETIF_F_HIGHDMA;
-	efx = netdev_priv(net_dev);
 	pci_set_drvdata(pci_dev, efx);
 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
-	rc = efx_init_struct(efx, type, pci_dev, net_dev);
+	rc = efx_init_struct(efx, pci_dev, net_dev);
 	if (rc)
 		goto fail1;
 
@@ -2656,28 +2696,9 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev,
 		goto fail2;
 
 	rc = efx_pci_probe_main(efx);
-
-	/* Serialise against efx_reset(). No more resets will be
-	 * scheduled since efx_stop_all() has been called, and we have
-	 * not and never have been registered.
-	 */
-	cancel_work_sync(&efx->reset_work);
-
 	if (rc)
 		goto fail3;
 
-	/* If there was a scheduled reset during probe, the NIC is
-	 * probably hosed anyway.
-	 */
-	if (efx->reset_pending) {
-		rc = -EIO;
-		goto fail4;
-	}
-
-	/* Switch to the running state before we expose the device to the OS,
-	 * so that dev_open()|efx_start_all() will actually start the device */
-	efx->state = STATE_RUNNING;
-
 	rc = efx_register_netdev(efx);
 	if (rc)
 		goto fail4;
@@ -2717,12 +2738,18 @@ static int efx_pm_freeze(struct device *dev)
 {
 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
 
-	efx->state = STATE_FINI;
+	rtnl_lock();
 
-	netif_device_detach(efx->net_dev);
+	if (efx->state != STATE_DISABLED) {
+		efx->state = STATE_UNINIT;
 
-	efx_stop_all(efx);
-	efx_stop_interrupts(efx, false);
+		netif_device_detach(efx->net_dev);
+
+		efx_stop_all(efx);
+		efx_stop_interrupts(efx, false);
+	}
+
+	rtnl_unlock();
 
 	return 0;
 }
@@ -2731,21 +2758,25 @@ static int efx_pm_thaw(struct device *dev)
 {
 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
 
-	efx->state = STATE_INIT;
+	rtnl_lock();
 
-	efx_start_interrupts(efx, false);
+	if (efx->state != STATE_DISABLED) {
+		efx_start_interrupts(efx, false);
 
-	mutex_lock(&efx->mac_lock);
-	efx->phy_op->reconfigure(efx);
-	mutex_unlock(&efx->mac_lock);
+		mutex_lock(&efx->mac_lock);
+		efx->phy_op->reconfigure(efx);
+		mutex_unlock(&efx->mac_lock);
 
-	efx_start_all(efx);
+		efx_start_all(efx);
 
-	netif_device_attach(efx->net_dev);
+		netif_device_attach(efx->net_dev);
 
-	efx->state = STATE_RUNNING;
+		efx->state = STATE_READY;
 
-	efx->type->resume_wol(efx);
+		efx->type->resume_wol(efx);
+	}
+
+	rtnl_unlock();
 
 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
 	queue_work(reset_workqueue, &efx->reset_work);
diff --git a/drivers/net/ethernet/sfc/ethtool.c b/drivers/net/ethernet/sfc/ethtool.c
index 8cba2df82b18..2bd5c2d35e5d 100644
--- a/drivers/net/ethernet/sfc/ethtool.c
+++ b/drivers/net/ethernet/sfc/ethtool.c
@@ -529,9 +529,7 @@ static void efx_ethtool_self_test(struct net_device *net_dev,
 	if (!efx_tests)
 		goto fail;
 
-
-	ASSERT_RTNL();
-	if (efx->state != STATE_RUNNING) {
+	if (efx->state != STATE_READY) {
 		rc = -EIO;
 		goto fail1;
 	}
diff --git a/drivers/net/ethernet/sfc/falcon_boards.c b/drivers/net/ethernet/sfc/falcon_boards.c
index 8687a6c3db0d..ec1e99d0dcad 100644
--- a/drivers/net/ethernet/sfc/falcon_boards.c
+++ b/drivers/net/ethernet/sfc/falcon_boards.c
@@ -380,7 +380,7 @@ static ssize_t set_phy_flash_cfg(struct device *dev,
 		new_mode = PHY_MODE_SPECIAL;
 	if (!((old_mode ^ new_mode) & PHY_MODE_SPECIAL)) {
 		err = 0;
-	} else if (efx->state != STATE_RUNNING || netif_running(efx->net_dev)) {
+	} else if (efx->state != STATE_READY || netif_running(efx->net_dev)) {
 		err = -EBUSY;
 	} else {
 		/* Reset the PHY, reconfigure the MAC and enable/disable
diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h
index cd9c0a989692..7ab1232494ef 100644
--- a/drivers/net/ethernet/sfc/net_driver.h
+++ b/drivers/net/ethernet/sfc/net_driver.h
@@ -91,29 +91,31 @@ struct efx_special_buffer {
 };
 
 /**
- * struct efx_tx_buffer - An Efx TX buffer
- * @skb: The associated socket buffer.
- *	Set only on the final fragment of a packet; %NULL for all other
- *	fragments.  When this fragment completes, then we can free this
- *	skb.
- * @tsoh: The associated TSO header structure, or %NULL if this
- *	buffer is not a TSO header.
+ * struct efx_tx_buffer - buffer state for a TX descriptor
+ * @skb: When @flags & %EFX_TX_BUF_SKB, the associated socket buffer to be
+ *	freed when descriptor completes
+ * @heap_buf: When @flags & %EFX_TX_BUF_HEAP, the associated heap buffer to be
+ *	freed when descriptor completes.
  * @dma_addr: DMA address of the fragment.
+ * @flags: Flags for allocation and DMA mapping type
  * @len: Length of this fragment.
  *	This field is zero when the queue slot is empty.
- * @continuation: True if this fragment is not the end of a packet.
- * @unmap_single: True if dma_unmap_single should be used.
  * @unmap_len: Length of this fragment to unmap
  */
 struct efx_tx_buffer {
-	const struct sk_buff *skb;
-	struct efx_tso_header *tsoh;
+	union {
+		const struct sk_buff *skb;
+		void *heap_buf;
+	};
 	dma_addr_t dma_addr;
+	unsigned short flags;
 	unsigned short len;
-	bool continuation;
-	bool unmap_single;
 	unsigned short unmap_len;
 };
+#define EFX_TX_BUF_CONT		1	/* not last descriptor of packet */
+#define EFX_TX_BUF_SKB		2	/* buffer is last part of skb */
+#define EFX_TX_BUF_HEAP		4	/* buffer was allocated with kmalloc() */
+#define EFX_TX_BUF_MAP_SINGLE	8	/* buffer was mapped with dma_map_single() */
 
 /**
  * struct efx_tx_queue - An Efx TX queue
@@ -133,6 +135,7 @@ struct efx_tx_buffer {
  * @channel: The associated channel
  * @core_txq: The networking core TX queue structure
  * @buffer: The software buffer ring
+ * @tsoh_page: Array of pages of TSO header buffers
  * @txd: The hardware descriptor ring
  * @ptr_mask: The size of the ring minus 1.
  * @initialised: Has hardware queue been initialised?
@@ -156,9 +159,6 @@ struct efx_tx_buffer {
  *	variable indicates that the queue is full.  This is to
  *	avoid cache-line ping-pong between the xmit path and the
  *	completion path.
- * @tso_headers_free: A list of TSO headers allocated for this TX queue
- *	that are not in use, and so available for new TSO sends. The list
- *	is protected by the TX queue lock.
  * @tso_bursts: Number of times TSO xmit invoked by kernel
  * @tso_long_headers: Number of packets with headers too long for standard
  *	blocks
@@ -175,6 +175,7 @@ struct efx_tx_queue {
 	struct efx_channel *channel;
 	struct netdev_queue *core_txq;
 	struct efx_tx_buffer *buffer;
+	struct efx_buffer *tsoh_page;
 	struct efx_special_buffer txd;
 	unsigned int ptr_mask;
 	bool initialised;
@@ -187,7 +188,6 @@ struct efx_tx_queue {
 	unsigned int insert_count ____cacheline_aligned_in_smp;
 	unsigned int write_count;
 	unsigned int old_read_count;
-	struct efx_tso_header *tso_headers_free;
 	unsigned int tso_bursts;
 	unsigned int tso_long_headers;
 	unsigned int tso_packets;
@@ -430,11 +430,9 @@ enum efx_int_mode {
 #define EFX_INT_MODE_USE_MSI(x) (((x)->interrupt_mode) <= EFX_INT_MODE_MSI)
 
 enum nic_state {
-	STATE_INIT = 0,
-	STATE_RUNNING = 1,
-	STATE_FINI = 2,
-	STATE_DISABLED = 3,
-	STATE_MAX,
+	STATE_UNINIT = 0,	/* device being probed/removed or is frozen */
+	STATE_READY = 1,	/* hardware ready and netdev registered */
+	STATE_DISABLED = 2,	/* device disabled due to hardware errors */
 };
 
 /*
@@ -654,7 +652,7 @@ struct vfdi_status;
  * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues
  * @irq_rx_moderation: IRQ moderation time for RX event queues
  * @msg_enable: Log message enable flags
- * @state: Device state flag. Serialised by the rtnl_lock.
+ * @state: Device state number (%STATE_*). Serialised by the rtnl_lock.
  * @reset_pending: Bitmask for pending resets
  * @tx_queue: TX DMA queues
  * @rx_queue: RX DMA queues
@@ -664,6 +662,8 @@ struct vfdi_status;
  *	should be allocated for this NIC
  * @rxq_entries: Size of receive queues requested by user.
  * @txq_entries: Size of transmit queues requested by user.
+ * @txq_stop_thresh: TX queue fill level at or above which we stop it.
+ * @txq_wake_thresh: TX queue fill level at or below which we wake it.
  * @tx_dc_base: Base qword address in SRAM of TX queue descriptor caches
  * @rx_dc_base: Base qword address in SRAM of RX queue descriptor caches
  * @sram_lim_qw: Qword address limit of SRAM
@@ -774,6 +774,9 @@ struct efx_nic {
 
 	unsigned rxq_entries;
 	unsigned txq_entries;
+	unsigned int txq_stop_thresh;
+	unsigned int txq_wake_thresh;
+
 	unsigned tx_dc_base;
 	unsigned rx_dc_base;
 	unsigned sram_lim_qw;
diff --git a/drivers/net/ethernet/sfc/nic.c b/drivers/net/ethernet/sfc/nic.c
index 326d799762d6..cdff40b65729 100644
--- a/drivers/net/ethernet/sfc/nic.c
+++ b/drivers/net/ethernet/sfc/nic.c
@@ -298,7 +298,7 @@ efx_free_special_buffer(struct efx_nic *efx, struct efx_special_buffer *buffer)
 /**************************************************************************
  *
  * Generic buffer handling
- * These buffers are used for interrupt status and MAC stats
+ * These buffers are used for interrupt status, MAC stats, etc.
  *
  **************************************************************************/
 
@@ -401,8 +401,10 @@ void efx_nic_push_buffers(struct efx_tx_queue *tx_queue)
 		++tx_queue->write_count;
 
 		/* Create TX descriptor ring entry */
+		BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
 		EFX_POPULATE_QWORD_4(*txd,
-				     FSF_AZ_TX_KER_CONT, buffer->continuation,
+				     FSF_AZ_TX_KER_CONT,
+				     buffer->flags & EFX_TX_BUF_CONT,
 				     FSF_AZ_TX_KER_BYTE_COUNT, buffer->len,
 				     FSF_AZ_TX_KER_BUF_REGION, 0,
 				     FSF_AZ_TX_KER_BUF_ADDR, buffer->dma_addr);
diff --git a/drivers/net/ethernet/sfc/tx.c b/drivers/net/ethernet/sfc/tx.c
index 18713436b443..ebca75ed78dc 100644
--- a/drivers/net/ethernet/sfc/tx.c
+++ b/drivers/net/ethernet/sfc/tx.c
@@ -22,14 +22,6 @@
 #include "nic.h"
 #include "workarounds.h"
 
-/*
- * TX descriptor ring full threshold
- *
- * The tx_queue descriptor ring fill-level must fall below this value
- * before we restart the netif queue
- */
-#define EFX_TXQ_THRESHOLD(_efx) ((_efx)->txq_entries / 2u)
-
 static void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
 			       struct efx_tx_buffer *buffer,
 			       unsigned int *pkts_compl,
@@ -39,67 +31,32 @@ static void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
 		struct device *dma_dev = &tx_queue->efx->pci_dev->dev;
 		dma_addr_t unmap_addr = (buffer->dma_addr + buffer->len -
 					 buffer->unmap_len);
-		if (buffer->unmap_single)
+		if (buffer->flags & EFX_TX_BUF_MAP_SINGLE)
 			dma_unmap_single(dma_dev, unmap_addr, buffer->unmap_len,
 					 DMA_TO_DEVICE);
 		else
 			dma_unmap_page(dma_dev, unmap_addr, buffer->unmap_len,
 				       DMA_TO_DEVICE);
 		buffer->unmap_len = 0;
-		buffer->unmap_single = false;
 	}
 
-	if (buffer->skb) {
+	if (buffer->flags & EFX_TX_BUF_SKB) {
 		(*pkts_compl)++;
 		(*bytes_compl) += buffer->skb->len;
 		dev_kfree_skb_any((struct sk_buff *) buffer->skb);
-		buffer->skb = NULL;
 		netif_vdbg(tx_queue->efx, tx_done, tx_queue->efx->net_dev,
 			   "TX queue %d transmission id %x complete\n",
 			   tx_queue->queue, tx_queue->read_count);
+	} else if (buffer->flags & EFX_TX_BUF_HEAP) {
+		kfree(buffer->heap_buf);
 	}
-}
 
-/**
- * struct efx_tso_header - a DMA mapped buffer for packet headers
- * @next: Linked list of free ones.
- *	The list is protected by the TX queue lock.
- * @dma_unmap_len: Length to unmap for an oversize buffer, or 0.
- * @dma_addr: The DMA address of the header below.
- *
- * This controls the memory used for a TSO header.  Use TSOH_DATA()
- * to find the packet header data.  Use TSOH_SIZE() to calculate the
- * total size required for a given packet header length.  TSO headers
- * in the free list are exactly %TSOH_STD_SIZE bytes in size.
- */
-struct efx_tso_header {
-	union {
-		struct efx_tso_header *next;
-		size_t unmap_len;
-	};
-	dma_addr_t dma_addr;
-};
+	buffer->len = 0;
+	buffer->flags = 0;
+}
 
 static int efx_enqueue_skb_tso(struct efx_tx_queue *tx_queue,
 			       struct sk_buff *skb);
-static void efx_fini_tso(struct efx_tx_queue *tx_queue);
-static void efx_tsoh_heap_free(struct efx_tx_queue *tx_queue,
-			       struct efx_tso_header *tsoh);
-
-static void efx_tsoh_free(struct efx_tx_queue *tx_queue,
-			  struct efx_tx_buffer *buffer)
-{
-	if (buffer->tsoh) {
-		if (likely(!buffer->tsoh->unmap_len)) {
-			buffer->tsoh->next = tx_queue->tso_headers_free;
-			tx_queue->tso_headers_free = buffer->tsoh;
-		} else {
-			efx_tsoh_heap_free(tx_queue, buffer->tsoh);
-		}
-		buffer->tsoh = NULL;
-	}
-}
-
 
 static inline unsigned
 efx_max_tx_len(struct efx_nic *efx, dma_addr_t dma_addr)
@@ -138,6 +95,56 @@ unsigned int efx_tx_max_skb_descs(struct efx_nic *efx)
 	return max_descs;
 }
 
+/* Get partner of a TX queue, seen as part of the same net core queue */
+static struct efx_tx_queue *efx_tx_queue_partner(struct efx_tx_queue *tx_queue)
+{
+	if (tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD)
+		return tx_queue - EFX_TXQ_TYPE_OFFLOAD;
+	else
+		return tx_queue + EFX_TXQ_TYPE_OFFLOAD;
+}
+
+static void efx_tx_maybe_stop_queue(struct efx_tx_queue *txq1)
+{
+	/* We need to consider both queues that the net core sees as one */
+	struct efx_tx_queue *txq2 = efx_tx_queue_partner(txq1);
+	struct efx_nic *efx = txq1->efx;
+	unsigned int fill_level;
+
+	fill_level = max(txq1->insert_count - txq1->old_read_count,
+			 txq2->insert_count - txq2->old_read_count);
+	if (likely(fill_level < efx->txq_stop_thresh))
+		return;
+
+	/* We used the stale old_read_count above, which gives us a
+	 * pessimistic estimate of the fill level (which may even
+	 * validly be >= efx->txq_entries).  Now try again using
+	 * read_count (more likely to be a cache miss).
+	 *
+	 * If we read read_count and then conditionally stop the
+	 * queue, it is possible for the completion path to race with
+	 * us and complete all outstanding descriptors in the middle,
+	 * after which there will be no more completions to wake it.
+	 * Therefore we stop the queue first, then read read_count
+	 * (with a memory barrier to ensure the ordering), then
+	 * restart the queue if the fill level turns out to be low
+	 * enough.
+	 */
+	netif_tx_stop_queue(txq1->core_txq);
+	smp_mb();
+	txq1->old_read_count = ACCESS_ONCE(txq1->read_count);
+	txq2->old_read_count = ACCESS_ONCE(txq2->read_count);
+
+	fill_level = max(txq1->insert_count - txq1->old_read_count,
+			 txq2->insert_count - txq2->old_read_count);
+	EFX_BUG_ON_PARANOID(fill_level >= efx->txq_entries);
+	if (likely(fill_level < efx->txq_stop_thresh)) {
+		smp_mb();
+		if (likely(!efx->loopback_selftest))
+			netif_tx_start_queue(txq1->core_txq);
+	}
+}
+
 /*
  * Add a socket buffer to a TX queue
  *
@@ -151,7 +158,7 @@ unsigned int efx_tx_max_skb_descs(struct efx_nic *efx)
  * This function is split out from efx_hard_start_xmit to allow the
  * loopback test to direct packets via specific TX queues.
  *
- * Returns NETDEV_TX_OK or NETDEV_TX_BUSY
+ * Returns NETDEV_TX_OK.
  * You must hold netif_tx_lock() to call this function.
  */
 netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
@@ -160,12 +167,11 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 	struct device *dma_dev = &efx->pci_dev->dev;
 	struct efx_tx_buffer *buffer;
 	skb_frag_t *fragment;
-	unsigned int len, unmap_len = 0, fill_level, insert_ptr;
+	unsigned int len, unmap_len = 0, insert_ptr;
 	dma_addr_t dma_addr, unmap_addr = 0;
 	unsigned int dma_len;
-	bool unmap_single;
-	int q_space, i = 0;
-	netdev_tx_t rc = NETDEV_TX_OK;
+	unsigned short dma_flags;
+	int i = 0;
 
 	EFX_BUG_ON_PARANOID(tx_queue->write_count != tx_queue->insert_count);
 
@@ -183,14 +189,11 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 			return NETDEV_TX_OK;
 	}
 
-	fill_level = tx_queue->insert_count - tx_queue->old_read_count;
-	q_space = efx->txq_entries - 1 - fill_level;
-
 	/* Map for DMA.  Use dma_map_single rather than dma_map_page
 	 * since this is more efficient on machines with sparse
 	 * memory.
 	 */
-	unmap_single = true;
+	dma_flags = EFX_TX_BUF_MAP_SINGLE;
 	dma_addr = dma_map_single(dma_dev, skb->data, len, PCI_DMA_TODEVICE);
 
 	/* Process all fragments */
@@ -205,39 +208,10 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 
 		/* Add to TX queue, splitting across DMA boundaries */
 		do {
-			if (unlikely(q_space-- <= 0)) {
-				/* It might be that completions have
-				 * happened since the xmit path last
-				 * checked.  Update the xmit path's
-				 * copy of read_count.
-				 */
-				netif_tx_stop_queue(tx_queue->core_txq);
-				/* This memory barrier protects the
-				 * change of queue state from the access
-				 * of read_count. */
-				smp_mb();
-				tx_queue->old_read_count =
-					ACCESS_ONCE(tx_queue->read_count);
-				fill_level = (tx_queue->insert_count
-					      - tx_queue->old_read_count);
-				q_space = efx->txq_entries - 1 - fill_level;
-				if (unlikely(q_space-- <= 0)) {
-					rc = NETDEV_TX_BUSY;
-					goto unwind;
-				}
-				smp_mb();
-				if (likely(!efx->loopback_selftest))
-					netif_tx_start_queue(
-						tx_queue->core_txq);
-			}
-
 			insert_ptr = tx_queue->insert_count & tx_queue->ptr_mask;
 			buffer = &tx_queue->buffer[insert_ptr];
-			efx_tsoh_free(tx_queue, buffer);
-			EFX_BUG_ON_PARANOID(buffer->tsoh);
-			EFX_BUG_ON_PARANOID(buffer->skb);
+			EFX_BUG_ON_PARANOID(buffer->flags);
 			EFX_BUG_ON_PARANOID(buffer->len);
-			EFX_BUG_ON_PARANOID(!buffer->continuation);
 			EFX_BUG_ON_PARANOID(buffer->unmap_len);
 
 			dma_len = efx_max_tx_len(efx, dma_addr);
@@ -247,13 +221,14 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 			/* Fill out per descriptor fields */
 			buffer->len = dma_len;
 			buffer->dma_addr = dma_addr;
+			buffer->flags = EFX_TX_BUF_CONT;
 			len -= dma_len;
 			dma_addr += dma_len;
 			++tx_queue->insert_count;
 		} while (len);
 
 		/* Transfer ownership of the unmapping to the final buffer */
-		buffer->unmap_single = unmap_single;
+		buffer->flags = EFX_TX_BUF_CONT | dma_flags;
 		buffer->unmap_len = unmap_len;
 		unmap_len = 0;
 
@@ -264,20 +239,22 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 		len = skb_frag_size(fragment);
 		i++;
 		/* Map for DMA */
-		unmap_single = false;
+		dma_flags = 0;
 		dma_addr = skb_frag_dma_map(dma_dev, fragment, 0, len,
 					    DMA_TO_DEVICE);
 	}
 
 	/* Transfer ownership of the skb to the final buffer */
 	buffer->skb = skb;
-	buffer->continuation = false;
+	buffer->flags = EFX_TX_BUF_SKB | dma_flags;
 
 	netdev_tx_sent_queue(tx_queue->core_txq, skb->len);
 
 	/* Pass off to hardware */
 	efx_nic_push_buffers(tx_queue);
 
+	efx_tx_maybe_stop_queue(tx_queue);
+
 	return NETDEV_TX_OK;
 
  dma_err:
@@ -289,7 +266,6 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 	/* Mark the packet as transmitted, and free the SKB ourselves */
 	dev_kfree_skb_any(skb);
 
- unwind:
 	/* Work backwards until we hit the original insert pointer value */
 	while (tx_queue->insert_count != tx_queue->write_count) {
 		unsigned int pkts_compl = 0, bytes_compl = 0;
@@ -297,12 +273,11 @@ netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff