[PATCH net-next 0/8] s390/qeth: updates 2019-04-17

[PATCH net-next 0/8] s390/qeth: updates 2019-04-17

[Julian Wiedmann]

Hi Dave,

please apply some additional qeth patches to net-next. This patchset converts
the driver to use the kernel's multiqueue model.

Thanks,
Julian


Julian Wiedmann (8):
  s390/qeth: clarify naming for some QDIO helpers
  s390/qeth: handle error from qeth_update_from_chp_desc()
  s390/qeth: don't bother updating the last-tx time
  s390/qeth: don't keep statistics for tx timeout
  s390/qeth: add TX multiqueue support for IQD devices
  s390/qeth: add TX multiqueue support for OSA devices
  s390/qeth: simplify QoS code
  s390/qeth: stop/wake TX queues based on their fill level

 drivers/s390/net/qeth_core.h      |  36 +++---
 drivers/s390/net/qeth_core_main.c | 234 ++++++++++++++++++++++----------------
 drivers/s390/net/qeth_core_sys.c  |   7 +-
 drivers/s390/net/qeth_ethtool.c   |  17 +++
 drivers/s390/net/qeth_l2_main.c   |  38 ++++---
 drivers/s390/net/qeth_l3_main.c   |  53 +++++----
 6 files changed, 233 insertions(+), 152 deletions(-)

-- 
2.16.4

[PATCH net-next 1/8] s390/qeth: clarify naming for some QDIO helpers

[Julian Wiedmann]

The naming of several QDIO helpers doesn't match their actual
functionality, or the structures they operate on. Clean this up.

s/qeth_alloc_qdio_buffers/qeth_alloc_qdio_queues
s/qeth_free_qdio_buffers/qeth_free_qdio_queues
s/qeth_alloc_qdio_out_buf/qeth_alloc_output_queue
s/qeth_clear_outq_buffers/qeth_drain_output_queue
s/qeth_clear_qdio_buffers/qeth_drain_output_queues

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      |  2 +-
 drivers/s390/net/qeth_core_main.c | 39 +++++++++++++++++++--------------------
 drivers/s390/net/qeth_l2_main.c   |  2 +-
 drivers/s390/net/qeth_l3_main.c   |  2 +-
 4 files changed, 22 insertions(+), 23 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 4c3a2db0cf2e..41e3bc4887fe 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -988,7 +988,7 @@ void qeth_clear_ipacmd_list(struct qeth_card *);
 int qeth_qdio_clear_card(struct qeth_card *, int);
 void qeth_clear_working_pool_list(struct qeth_card *);
 void qeth_clear_cmd_buffers(struct qeth_channel *);
-void qeth_clear_qdio_buffers(struct qeth_card *);
+void qeth_drain_output_queues(struct qeth_card *card);
 void qeth_setadp_promisc_mode(struct qeth_card *);
 int qeth_setadpparms_change_macaddr(struct qeth_card *);
 void qeth_tx_timeout(struct net_device *);
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index 2b75f76f23fd..ca71632a4fa0 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -67,7 +67,7 @@ static void qeth_issue_next_read_cb(struct qeth_card *card,
 static struct qeth_cmd_buffer *qeth_get_buffer(struct qeth_channel *);
 static void qeth_free_buffer_pool(struct qeth_card *);
 static int qeth_qdio_establish(struct qeth_card *);
-static void qeth_free_qdio_buffers(struct qeth_card *);
+static void qeth_free_qdio_queues(struct qeth_card *card);
 static void qeth_notify_skbs(struct qeth_qdio_out_q *queue,
 		struct qeth_qdio_out_buffer *buf,
 		enum iucv_tx_notify notification);
@@ -1178,7 +1178,7 @@ static void qeth_clear_output_buffer(struct qeth_qdio_out_q *queue,
 	atomic_set(&buf->state, QETH_QDIO_BUF_EMPTY);
 }
 
-static void qeth_clear_outq_buffers(struct qeth_qdio_out_q *q, int free)
+static void qeth_drain_output_queue(struct qeth_qdio_out_q *q, bool free)
 {
 	int j;
 
@@ -1194,19 +1194,18 @@ static void qeth_clear_outq_buffers(struct qeth_qdio_out_q *q, int free)
 	}
 }
 
-void qeth_clear_qdio_buffers(struct qeth_card *card)
+void qeth_drain_output_queues(struct qeth_card *card)
 {
 	int i;
 
 	QETH_CARD_TEXT(card, 2, "clearqdbf");
 	/* clear outbound buffers to free skbs */
 	for (i = 0; i < card->qdio.no_out_queues; ++i) {
-		if (card->qdio.out_qs[i]) {
-			qeth_clear_outq_buffers(card->qdio.out_qs[i], 0);
-		}
+		if (card->qdio.out_qs[i])
+			qeth_drain_output_queue(card->qdio.out_qs[i], false);
 	}
 }
-EXPORT_SYMBOL_GPL(qeth_clear_qdio_buffers);
+EXPORT_SYMBOL_GPL(qeth_drain_output_queues);
 
 static void qeth_free_buffer_pool(struct qeth_card *card)
 {
@@ -1280,7 +1279,7 @@ static void qeth_set_single_write_queues(struct qeth_card *card)
 {
 	if ((atomic_read(&card->qdio.state) != QETH_QDIO_UNINITIALIZED) &&
 	    (card->qdio.no_out_queues == 4))
-		qeth_free_qdio_buffers(card);
+		qeth_free_qdio_queues(card);
 
 	card->qdio.no_out_queues = 1;
 	if (card->qdio.default_out_queue != 0)
@@ -1293,7 +1292,7 @@ static void qeth_set_multiple_write_queues(struct qeth_card *card)
 {
 	if ((atomic_read(&card->qdio.state) != QETH_QDIO_UNINITIALIZED) &&
 	    (card->qdio.no_out_queues == 1)) {
-		qeth_free_qdio_buffers(card);
+		qeth_free_qdio_queues(card);
 		card->qdio.default_out_queue = 2;
 	}
 	card->qdio.no_out_queues = 4;
@@ -2177,7 +2176,7 @@ static int qeth_update_max_mtu(struct qeth_card *card, unsigned int max_mtu)
 		/* adjust RX buffer size to new max MTU: */
 		card->qdio.in_buf_size = max_mtu + 2 * PAGE_SIZE;
 		if (dev->max_mtu && dev->max_mtu != max_mtu)
-			qeth_free_qdio_buffers(card);
+			qeth_free_qdio_queues(card);
 	} else {
 		if (dev->mtu)
 			new_mtu = dev->mtu;
@@ -2350,12 +2349,12 @@ static void qeth_free_output_queue(struct qeth_qdio_out_q *q)
 	if (!q)
 		return;
 
-	qeth_clear_outq_buffers(q, 1);
+	qeth_drain_output_queue(q, true);
 	qdio_free_buffers(q->qdio_bufs, QDIO_MAX_BUFFERS_PER_Q);
 	kfree(q);
 }
 
-static struct qeth_qdio_out_q *qeth_alloc_qdio_out_buf(void)
+static struct qeth_qdio_out_q *qeth_alloc_output_queue(void)
 {
 	struct qeth_qdio_out_q *q = kzalloc(sizeof(*q), GFP_KERNEL);
 
@@ -2369,7 +2368,7 @@ static struct qeth_qdio_out_q *qeth_alloc_qdio_out_buf(void)
 	return q;
 }
 
-static int qeth_alloc_qdio_buffers(struct qeth_card *card)
+static int qeth_alloc_qdio_queues(struct qeth_card *card)
 {
 	int i, j;
 
@@ -2390,7 +2389,7 @@ static int qeth_alloc_qdio_buffers(struct qeth_card *card)
 
 	/* outbound */
 	for (i = 0; i < card->qdio.no_out_queues; ++i) {
-		card->qdio.out_qs[i] = qeth_alloc_qdio_out_buf();
+		card->qdio.out_qs[i] = qeth_alloc_output_queue();
 		if (!card->qdio.out_qs[i])
 			goto out_freeoutq;
 		QETH_DBF_TEXT_(SETUP, 2, "outq %i", i);
@@ -2431,7 +2430,7 @@ static int qeth_alloc_qdio_buffers(struct qeth_card *card)
 	return -ENOMEM;
 }
 
-static void qeth_free_qdio_buffers(struct qeth_card *card)
+static void qeth_free_qdio_queues(struct qeth_card *card)
 {
 	int i, j;
 
@@ -2538,7 +2537,7 @@ static int qeth_mpc_initialize(struct qeth_card *card)
 		QETH_DBF_TEXT_(SETUP, 2, "5err%d", rc);
 		goto out_qdio;
 	}
-	rc = qeth_alloc_qdio_buffers(card);
+	rc = qeth_alloc_qdio_queues(card);
 	if (rc) {
 		QETH_DBF_TEXT_(SETUP, 2, "5err%d", rc);
 		goto out_qdio;
@@ -2546,7 +2545,7 @@ static int qeth_mpc_initialize(struct qeth_card *card)
 	rc = qeth_qdio_establish(card);
 	if (rc) {
 		QETH_DBF_TEXT_(SETUP, 2, "6err%d", rc);
-		qeth_free_qdio_buffers(card);
+		qeth_free_qdio_queues(card);
 		goto out_qdio;
 	}
 	rc = qeth_qdio_activate(card);
@@ -3460,7 +3459,7 @@ int qeth_configure_cq(struct qeth_card *card, enum qeth_cq cq)
 			goto out;
 		}
 
-		qeth_free_qdio_buffers(card);
+		qeth_free_qdio_queues(card);
 		card->options.cq = cq;
 		rc = 0;
 	}
@@ -4930,7 +4929,7 @@ static void qeth_core_free_card(struct qeth_card *card)
 	qeth_clean_channel(&card->write);
 	qeth_clean_channel(&card->data);
 	destroy_workqueue(card->event_wq);
-	qeth_free_qdio_buffers(card);
+	qeth_free_qdio_queues(card);
 	unregister_service_level(&card->qeth_service_level);
 	dev_set_drvdata(&card->gdev->dev, NULL);
 	kfree(card);
@@ -5732,7 +5731,7 @@ static void qeth_core_shutdown(struct ccwgroup_device *gdev)
 	if ((gdev->state == CCWGROUP_ONLINE) && card->info.hwtrap)
 		qeth_hw_trap(card, QETH_DIAGS_TRAP_DISARM);
 	qeth_qdio_clear_card(card, 0);
-	qeth_clear_qdio_buffers(card);
+	qeth_drain_output_queues(card);
 	qdio_free(CARD_DDEV(card));
 }
 
diff --git a/drivers/s390/net/qeth_l2_main.c b/drivers/s390/net/qeth_l2_main.c
index 5549c66c6b5d..641dc17e3b79 100644
--- a/drivers/s390/net/qeth_l2_main.c
+++ b/drivers/s390/net/qeth_l2_main.c
@@ -299,7 +299,7 @@ static void qeth_l2_stop_card(struct qeth_card *card)
 	}
 	if (card->state == CARD_STATE_HARDSETUP) {
 		qeth_qdio_clear_card(card, 0);
-		qeth_clear_qdio_buffers(card);
+		qeth_drain_output_queues(card);
 		qeth_clear_working_pool_list(card);
 		card->state = CARD_STATE_DOWN;
 	}
diff --git a/drivers/s390/net/qeth_l3_main.c b/drivers/s390/net/qeth_l3_main.c
index f804d27eb569..102be697f5db 100644
--- a/drivers/s390/net/qeth_l3_main.c
+++ b/drivers/s390/net/qeth_l3_main.c
@@ -1433,7 +1433,7 @@ static void qeth_l3_stop_card(struct qeth_card *card)
 	}
 	if (card->state == CARD_STATE_HARDSETUP) {
 		qeth_qdio_clear_card(card, 0);
-		qeth_clear_qdio_buffers(card);
+		qeth_drain_output_queues(card);
 		qeth_clear_working_pool_list(card);
 		card->state = CARD_STATE_DOWN;
 	}
-- 
2.16.4

[PATCH net-next 2/8] s390/qeth: handle error from qeth_update_from_chp_desc()

[Julian Wiedmann]

Subsequent code relies on the values that qeth_update_from_chp_desc()
reads from the CHP descriptor. Rather than dealing with weird errors
later on, just handle it properly here.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core_main.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index ca71632a4fa0..ac592d610c5e 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -1298,7 +1298,7 @@ static void qeth_set_multiple_write_queues(struct qeth_card *card)
 	card->qdio.no_out_queues = 4;
 }
 
-static void qeth_update_from_chp_desc(struct qeth_card *card)
+static int qeth_update_from_chp_desc(struct qeth_card *card)
 {
 	struct ccw_device *ccwdev;
 	struct channel_path_desc_fmt0 *chp_dsc;
@@ -1308,7 +1308,7 @@ static void qeth_update_from_chp_desc(struct qeth_card *card)
 	ccwdev = card->data.ccwdev;
 	chp_dsc = ccw_device_get_chp_desc(ccwdev, 0);
 	if (!chp_dsc)
-		goto out;
+		return -ENOMEM;
 
 	card->info.func_level = 0x4100 + chp_dsc->desc;
 	if (card->info.type == QETH_CARD_TYPE_IQD)
@@ -1323,6 +1323,7 @@ static void qeth_update_from_chp_desc(struct qeth_card *card)
 	kfree(chp_dsc);
 	QETH_DBF_TEXT_(SETUP, 2, "nr:%x", card->qdio.no_out_queues);
 	QETH_DBF_TEXT_(SETUP, 2, "lvl:%02x", card->info.func_level);
+	return 0;
 }
 
 static void qeth_init_qdio_info(struct qeth_card *card)
@@ -4978,7 +4979,9 @@ int qeth_core_hardsetup_card(struct qeth_card *card, bool *carrier_ok)
 
 	QETH_DBF_TEXT(SETUP, 2, "hrdsetup");
 	atomic_set(&card->force_alloc_skb, 0);
-	qeth_update_from_chp_desc(card);
+	rc = qeth_update_from_chp_desc(card);
+	if (rc)
+		return rc;
 retry:
 	if (retries < 3)
 		QETH_DBF_MESSAGE(2, "Retrying to do IDX activates on device %x.\n",
@@ -5635,7 +5638,9 @@ static int qeth_core_probe_device(struct ccwgroup_device *gdev)
 	}
 
 	qeth_setup_card(card);
-	qeth_update_from_chp_desc(card);
+	rc = qeth_update_from_chp_desc(card);
+	if (rc)
+		goto err_chp_desc;
 
 	card->dev = qeth_alloc_netdev(card);
 	if (!card->dev) {
@@ -5670,6 +5675,7 @@ static int qeth_core_probe_device(struct ccwgroup_device *gdev)
 	qeth_core_free_discipline(card);
 err_load:
 	free_netdev(card->dev);
+err_chp_desc:
 err_card:
 	qeth_core_free_card(card);
 err_dev:
-- 
2.16.4

[PATCH net-next 3/8] s390/qeth: don't bother updating the last-tx time

[Julian Wiedmann]

As the documentation for netif_trans_update() says, netdev_start_xmit()
already updates the last-tx time after every good xmit. So don't
duplicate that effort.

One odd case is that qeth_flush_buffers() also gets called from our
TX completion handler, to flush out any partially filled buffer when
we switch the queue to non-packing mode. But as the TX completion
handler will _always_ wake the txq, we don't have to worry about
the TX watchdog there.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core_main.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index ac592d610c5e..9e495df742cb 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -3371,7 +3371,6 @@ static void qeth_flush_buffers(struct qeth_qdio_out_q *queue, int index,
 	}
 
 	QETH_TXQ_STAT_ADD(queue, bufs, count);
-	netif_trans_update(queue->card->dev);
 	qdio_flags = QDIO_FLAG_SYNC_OUTPUT;
 	if (atomic_read(&queue->set_pci_flags_count))
 		qdio_flags |= QDIO_FLAG_PCI_OUT;
-- 
2.16.4

[PATCH net-next 4/8] s390/qeth: don't keep statistics for tx timeout

[Julian Wiedmann]

struct netdev_queue contains a counter for tx timeouts, which gets
updated by dev_watchdog(). So let's not attempt to maintain our own
statistics, in particular not by overloading the skb-error counter.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      | 1 -
 drivers/s390/net/qeth_core_main.c | 2 --
 2 files changed, 3 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 41e3bc4887fe..730930135aba 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -464,7 +464,6 @@ struct qeth_card_stats {
 	u64 rx_errors;
 	u64 rx_dropped;
 	u64 rx_multicast;
-	u64 tx_errors;
 };
 
 struct qeth_out_q_stats {
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index 9e495df742cb..09ed9e04f4ca 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -4340,7 +4340,6 @@ void qeth_tx_timeout(struct net_device *dev)
 
 	card = dev->ml_priv;
 	QETH_CARD_TEXT(card, 4, "txtimeo");
-	QETH_CARD_STAT_INC(card, tx_errors);
 	qeth_schedule_recovery(card);
 }
 EXPORT_SYMBOL_GPL(qeth_tx_timeout);
@@ -6192,7 +6191,6 @@ void qeth_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 	stats->rx_errors = card->stats.rx_errors;
 	stats->rx_dropped = card->stats.rx_dropped;
 	stats->multicast = card->stats.rx_multicast;
-	stats->tx_errors = card->stats.tx_errors;
 
 	for (i = 0; i < card->qdio.no_out_queues; i++) {
 		queue = card->qdio.out_qs[i];
-- 
2.16.4

[PATCH net-next 5/8] s390/qeth: add TX multiqueue support for IQD devices

[Julian Wiedmann]

qeth has been supporting multiple HW Output Queues for a long time. But
rather than exposing those queues to the stack, it uses its own queue
selection logic in .ndo_start_xmit... with all the drawbacks that
entails.
Start off by switching IQD devices over to a proper mqs net_device,
and converting all the netdev_queue management code.

One oddity with IQD devices is the requirement to place all mcast
traffic on the _highest_ established HW queue. Doing so via
.ndo_select_queue seems straight-forward - but that won't work if only
some of the HW queues are active
(ie. when dev->real_num_tx_queues < dev->num_tx_queues), since
netdev_cap_txqueue() will not allow us to put skbs on the higher queues.

To make this work, we
1. let .ndo_select_queue() map all mcast traffic to netdev_queue 0, and
2. later re-map the netdev_queue and HW queue indices in
   .ndo_start_xmit and the TX completion handler.

With this patch we default to a fixed set of 1 ucast and 1 mcast queue.
Support for dynamic reconfiguration is added at a later time.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      | 18 +++++++++++++++---
 drivers/s390/net/qeth_core_main.c | 27 +++++++++++++++++++++------
 drivers/s390/net/qeth_core_sys.c  |  3 +++
 drivers/s390/net/qeth_ethtool.c   | 16 ++++++++++++++++
 drivers/s390/net/qeth_l2_main.c   | 29 +++++++++++++++++++----------
 drivers/s390/net/qeth_l3_main.c   | 30 +++++++++++++++++++++++-------
 6 files changed, 97 insertions(+), 26 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 730930135aba..836cde67f367 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -219,6 +219,9 @@ static inline int qeth_is_ipa_enabled(struct qeth_ipa_info *ipa,
 /* QDIO queue and buffer handling                                            */
 /*****************************************************************************/
 #define QETH_MAX_QUEUES 4
+#define QETH_IQD_MIN_TXQ	2	/* One for ucast, one for mcast. */
+#define QETH_IQD_MCAST_TXQ	0
+#define QETH_IQD_MIN_UCAST_TXQ	1
 #define QETH_IN_BUF_SIZE_DEFAULT 65536
 #define QETH_IN_BUF_COUNT_DEFAULT 64
 #define QETH_IN_BUF_COUNT_HSDEFAULT 128
@@ -835,6 +838,15 @@ static inline bool qeth_netdev_is_registered(struct net_device *dev)
 	return dev->netdev_ops != NULL;
 }
 
+static inline u16 qeth_iqd_translate_txq(struct net_device *dev, u16 txq)
+{
+	if (txq == QETH_IQD_MCAST_TXQ)
+		return dev->num_tx_queues - 1;
+	if (txq == dev->num_tx_queues - 1)
+		return QETH_IQD_MCAST_TXQ;
+	return txq;
+}
+
 static inline void qeth_scrub_qdio_buffer(struct qdio_buffer *buf,
 					  unsigned int elements)
 {
@@ -934,10 +946,8 @@ int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb,
 			    int ipv);
 static inline struct qeth_qdio_out_q *qeth_get_tx_queue(struct qeth_card *card,
 							struct sk_buff *skb,
-							int ipv, int cast_type)
+							int ipv)
 {
-	if (IS_IQD(card) && cast_type != RTN_UNICAST)
-		return card->qdio.out_qs[card->qdio.no_out_queues - 1];
 	if (!card->qdio.do_prio_queueing)
 		return card->qdio.out_qs[card->qdio.default_out_queue];
 	return card->qdio.out_qs[qeth_get_priority_queue(card, skb, ipv)];
@@ -1022,6 +1032,8 @@ netdev_features_t qeth_features_check(struct sk_buff *skb,
 				      struct net_device *dev,
 				      netdev_features_t features);
 void qeth_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats);
+u16 qeth_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
+			  u8 cast_type, struct net_device *sb_dev);
 int qeth_open(struct net_device *dev);
 int qeth_stop(struct net_device *dev);
 
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index 09ed9e04f4ca..68f6043f033a 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -3485,7 +3485,7 @@ static void qeth_qdio_cq_handler(struct qeth_card *card, unsigned int qdio_err,
 	QETH_CARD_TEXT_(card, 5, "qcqherr%d", qdio_err);
 
 	if (qdio_err) {
-		netif_stop_queue(card->dev);
+		netif_tx_stop_all_queues(card->dev);
 		qeth_schedule_recovery(card);
 		return;
 	}
@@ -3541,12 +3541,14 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	struct qeth_card *card        = (struct qeth_card *) card_ptr;
 	struct qeth_qdio_out_q *queue = card->qdio.out_qs[__queue];
 	struct qeth_qdio_out_buffer *buffer;
+	struct net_device *dev = card->dev;
+	u16 txq;
 	int i;
 
 	QETH_CARD_TEXT(card, 6, "qdouhdl");
 	if (qdio_error & QDIO_ERROR_FATAL) {
 		QETH_CARD_TEXT(card, 2, "achkcond");
-		netif_stop_queue(card->dev);
+		netif_tx_stop_all_queues(dev);
 		qeth_schedule_recovery(card);
 		return;
 	}
@@ -3595,7 +3597,8 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	if (card->info.type != QETH_CARD_TYPE_IQD)
 		qeth_check_outbound_queue(queue);
 
-	netif_wake_queue(queue->card->dev);
+	txq = IS_IQD(card) ? qeth_iqd_translate_txq(dev, __queue) : 0;
+	netif_wake_subqueue(dev, txq);
 }
 
 /* We cannot use outbound queue 3 for unicast packets on HiperSockets */
@@ -5557,7 +5560,8 @@ static struct net_device *qeth_alloc_netdev(struct qeth_card *card)
 
 	switch (card->info.type) {
 	case QETH_CARD_TYPE_IQD:
-		dev = alloc_netdev(0, "hsi%d", NET_NAME_UNKNOWN, ether_setup);
+		dev = alloc_netdev_mqs(0, "hsi%d", NET_NAME_UNKNOWN,
+				       ether_setup, QETH_MAX_QUEUES, 1);
 		break;
 	case QETH_CARD_TYPE_OSN:
 		dev = alloc_netdev(0, "osn%d", NET_NAME_UNKNOWN, ether_setup);
@@ -5585,8 +5589,10 @@ static struct net_device *qeth_alloc_netdev(struct qeth_card *card)
 		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 		dev->hw_features |= NETIF_F_SG;
 		dev->vlan_features |= NETIF_F_SG;
-		if (IS_IQD(card))
+		if (IS_IQD(card)) {
+			netif_set_real_num_tx_queues(dev, QETH_IQD_MIN_TXQ);
 			dev->features |= NETIF_F_SG;
+		}
 	}
 
 	return dev;
@@ -6203,6 +6209,15 @@ void qeth_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 }
 EXPORT_SYMBOL_GPL(qeth_get_stats64);
 
+u16 qeth_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
+			  u8 cast_type, struct net_device *sb_dev)
+{
+	if (cast_type != RTN_UNICAST)
+		return QETH_IQD_MCAST_TXQ;
+	return QETH_IQD_MIN_UCAST_TXQ;
+}
+EXPORT_SYMBOL_GPL(qeth_iqd_select_queue);
+
 int qeth_open(struct net_device *dev)
 {
 	struct qeth_card *card = dev->ml_priv;
@@ -6213,7 +6228,7 @@ int qeth_open(struct net_device *dev)
 		return -EIO;
 
 	card->data.state = CH_STATE_UP;
-	netif_start_queue(dev);
+	netif_tx_start_all_queues(dev);
 
 	napi_enable(&card->napi);
 	local_bh_disable();
diff --git a/drivers/s390/net/qeth_core_sys.c b/drivers/s390/net/qeth_core_sys.c
index 56deeb6f7bc0..b43d8bdf4c3e 100644
--- a/drivers/s390/net/qeth_core_sys.c
+++ b/drivers/s390/net/qeth_core_sys.c
@@ -198,6 +198,9 @@ static ssize_t qeth_dev_prioqing_store(struct device *dev,
 	if (!card)
 		return -EINVAL;
 
+	if (IS_IQD(card))
+		return -EOPNOTSUPP;
+
 	mutex_lock(&card->conf_mutex);
 	if (card->state != CARD_STATE_DOWN) {
 		rc = -EPERM;
diff --git a/drivers/s390/net/qeth_ethtool.c b/drivers/s390/net/qeth_ethtool.c
index 93a53fed4cf8..a443e5f86ab7 100644
--- a/drivers/s390/net/qeth_ethtool.c
+++ b/drivers/s390/net/qeth_ethtool.c
@@ -154,6 +154,21 @@ static void qeth_get_drvinfo(struct net_device *dev,
 		 CARD_RDEV_ID(card), CARD_WDEV_ID(card), CARD_DDEV_ID(card));
 }
 
+static void qeth_get_channels(struct net_device *dev,
+			      struct ethtool_channels *channels)
+{
+	struct qeth_card *card = dev->ml_priv;
+
+	channels->max_rx = dev->num_rx_queues;
+	channels->max_tx = card->qdio.no_out_queues;
+	channels->max_other = 0;
+	channels->max_combined = 0;
+	channels->rx_count = dev->real_num_rx_queues;
+	channels->tx_count = dev->real_num_tx_queues;
+	channels->other_count = 0;
+	channels->combined_count = 0;
+}
+
 /* Helper function to fill 'advertising' and 'supported' which are the same. */
 /* Autoneg and full-duplex are supported and advertised unconditionally.     */
 /* Always advertise and support all speeds up to specified, and only one     */
@@ -359,6 +374,7 @@ const struct ethtool_ops qeth_ethtool_ops = {
 	.get_ethtool_stats = qeth_get_ethtool_stats,
 	.get_sset_count = qeth_get_sset_count,
 	.get_drvinfo = qeth_get_drvinfo,
+	.get_channels = qeth_get_channels,
 	.get_link_ksettings = qeth_get_link_ksettings,
 };
 
diff --git a/drivers/s390/net/qeth_l2_main.c b/drivers/s390/net/qeth_l2_main.c
index 641dc17e3b79..1491281600c2 100644
--- a/drivers/s390/net/qeth_l2_main.c
+++ b/drivers/s390/net/qeth_l2_main.c
@@ -161,10 +161,8 @@ static void qeth_l2_drain_rx_mode_cache(struct qeth_card *card)
 	}
 }
 
-static int qeth_l2_get_cast_type(struct qeth_card *card, struct sk_buff *skb)
+static int qeth_l2_get_cast_type(struct sk_buff *skb)
 {
-	if (card->info.type == QETH_CARD_TYPE_OSN)
-		return RTN_UNICAST;
 	if (is_broadcast_ether_addr(skb->data))
 		return RTN_BROADCAST;
 	if (is_multicast_ether_addr(skb->data))
@@ -603,26 +601,29 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 					   struct net_device *dev)
 {
 	struct qeth_card *card = dev->ml_priv;
-	int cast_type = qeth_l2_get_cast_type(card, skb);
+	u16 txq = skb_get_queue_mapping(skb);
 	int ipv = qeth_get_ip_version(skb);
 	struct qeth_qdio_out_q *queue;
 	int tx_bytes = skb->len;
 	int rc;
 
-	queue = qeth_get_tx_queue(card, skb, ipv, cast_type);
+	if (IS_IQD(card))
+		queue = card->qdio.out_qs[qeth_iqd_translate_txq(dev, txq)];
+	else
+		queue = qeth_get_tx_queue(card, skb, ipv);
 
-	netif_stop_queue(dev);
+	netif_stop_subqueue(dev, txq);
 
 	if (IS_OSN(card))
 		rc = qeth_l2_xmit_osn(card, skb, queue);
 	else
-		rc = qeth_xmit(card, skb, queue, ipv, cast_type,
-			       qeth_l2_fill_header);
+		rc = qeth_xmit(card, skb, queue, ipv,
+			       qeth_l2_get_cast_type(skb), qeth_l2_fill_header);
 
 	if (!rc) {
 		QETH_TXQ_STAT_INC(queue, tx_packets);
 		QETH_TXQ_STAT_ADD(queue, tx_bytes, tx_bytes);
-		netif_wake_queue(dev);
+		netif_wake_subqueue(dev, txq);
 		return NETDEV_TX_OK;
 	} else if (rc == -EBUSY) {
 		return NETDEV_TX_BUSY;
@@ -630,10 +631,17 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 
 	QETH_TXQ_STAT_INC(queue, tx_dropped);
 	kfree_skb(skb);
-	netif_wake_queue(dev);
+	netif_wake_subqueue(dev, txq);
 	return NETDEV_TX_OK;
 }
 
+static u16 qeth_l2_select_queue(struct net_device *dev, struct sk_buff *skb,
+				struct net_device *sb_dev)
+{
+	return qeth_iqd_select_queue(dev, skb, qeth_l2_get_cast_type(skb),
+				     sb_dev);
+}
+
 static const struct device_type qeth_l2_devtype = {
 	.name = "qeth_layer2",
 	.groups = qeth_l2_attr_groups,
@@ -687,6 +695,7 @@ static const struct net_device_ops qeth_l2_netdev_ops = {
 	.ndo_get_stats64	= qeth_get_stats64,
 	.ndo_start_xmit		= qeth_l2_hard_start_xmit,
 	.ndo_features_check	= qeth_features_check,
+	.ndo_select_queue	= qeth_l2_select_queue,
 	.ndo_validate_addr	= qeth_l2_validate_addr,
 	.ndo_set_rx_mode	= qeth_l2_set_rx_mode,
 	.ndo_do_ioctl		= qeth_do_ioctl,
diff --git a/drivers/s390/net/qeth_l3_main.c b/drivers/s390/net/qeth_l3_main.c
index 102be697f5db..120193e90adb 100644
--- a/drivers/s390/net/qeth_l3_main.c
+++ b/drivers/s390/net/qeth_l3_main.c
@@ -2062,28 +2062,36 @@ static int qeth_l3_xmit(struct qeth_card *card, struct sk_buff *skb,
 static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 					   struct net_device *dev)
 {
-	int cast_type = qeth_l3_get_cast_type(skb);
 	struct qeth_card *card = dev->ml_priv;
+	u16 txq = skb_get_queue_mapping(skb);
 	int ipv = qeth_get_ip_version(skb);
 	struct qeth_qdio_out_q *queue;
 	int tx_bytes = skb->len;
-	int rc;
-
-	queue = qeth_get_tx_queue(card, skb, ipv, cast_type);
+	int cast_type, rc;
 
 	if (IS_IQD(card)) {
+		queue = card->qdio.out_qs[qeth_iqd_translate_txq(dev, txq)];
+
 		if (card->options.sniffer)
 			goto tx_drop;
 		if ((card->options.cq != QETH_CQ_ENABLED && !ipv) ||
 		    (card->options.cq == QETH_CQ_ENABLED &&
 		     skb->protocol != htons(ETH_P_AF_IUCV)))
 			goto tx_drop;
+
+		if (txq == QETH_IQD_MCAST_TXQ)
+			cast_type = qeth_l3_get_cast_type(skb);
+		else
+			cast_type = RTN_UNICAST;
+	} else {
+		queue = qeth_get_tx_queue(card, skb, ipv);
+		cast_type = qeth_l3_get_cast_type(skb);
 	}
 
 	if (cast_type == RTN_BROADCAST && !card->info.broadcast_capable)
 		goto tx_drop;
 
-	netif_stop_queue(dev);
+	netif_stop_subqueue(dev, txq);
 
 	if (ipv == 4 || IS_IQD(card))
 		rc = qeth_l3_xmit(card, skb, queue, ipv, cast_type);
@@ -2094,7 +2102,7 @@ static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 	if (!rc) {
 		QETH_TXQ_STAT_INC(queue, tx_packets);
 		QETH_TXQ_STAT_ADD(queue, tx_bytes, tx_bytes);
-		netif_wake_queue(dev);
+		netif_wake_subqueue(dev, txq);
 		return NETDEV_TX_OK;
 	} else if (rc == -EBUSY) {
 		return NETDEV_TX_BUSY;
@@ -2103,7 +2111,7 @@ static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 tx_drop:
 	QETH_TXQ_STAT_INC(queue, tx_dropped);
 	kfree_skb(skb);
-	netif_wake_queue(dev);
+	netif_wake_subqueue(dev, txq);
 	return NETDEV_TX_OK;
 }
 
@@ -2147,11 +2155,19 @@ static netdev_features_t qeth_l3_osa_features_check(struct sk_buff *skb,
 	return qeth_features_check(skb, dev, features);
 }
 
+static u16 qeth_l3_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
+				    struct net_device *sb_dev)
+{
+	return qeth_iqd_select_queue(dev, skb, qeth_l3_get_cast_type(skb),
+				     sb_dev);
+}
+
 static const struct net_device_ops qeth_l3_netdev_ops = {
 	.ndo_open		= qeth_open,
 	.ndo_stop		= qeth_stop,
 	.ndo_get_stats64	= qeth_get_stats64,
 	.ndo_start_xmit		= qeth_l3_hard_start_xmit,
+	.ndo_select_queue	= qeth_l3_iqd_select_queue,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_set_rx_mode	= qeth_l3_set_rx_mode,
 	.ndo_do_ioctl		= qeth_do_ioctl,
-- 
2.16.4

[PATCH net-next 6/8] s390/qeth: add TX multiqueue support for OSA devices

[Julian Wiedmann]

This adds trivial support for multiple TX queues on OSA-style devices
(both real HW and z/VM NICs). For now we expose the driver's existing
QoS mechanism via .ndo_select_queue, and adjust the number of available
TX queues when qeth_update_from_chp_desc() detects that the
HW configuration has changed.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      | 11 +------
 drivers/s390/net/qeth_core_main.c | 67 ++++++++++++++++++---------------------
 drivers/s390/net/qeth_l2_main.c   | 14 +++++---
 drivers/s390/net/qeth_l3_main.c   | 11 ++++++-
 4 files changed, 51 insertions(+), 52 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 836cde67f367..4989dc7b58c1 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -942,16 +942,7 @@ static inline int qeth_send_simple_setassparms_v6(struct qeth_card *card,
 						 data, QETH_PROT_IPV6);
 }
 
-int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb,
-			    int ipv);
-static inline struct qeth_qdio_out_q *qeth_get_tx_queue(struct qeth_card *card,
-							struct sk_buff *skb,
-							int ipv)
-{
-	if (!card->qdio.do_prio_queueing)
-		return card->qdio.out_qs[card->qdio.default_out_queue];
-	return card->qdio.out_qs[qeth_get_priority_queue(card, skb, ipv)];
-}
+int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb);
 
 extern struct qeth_discipline qeth_l2_discipline;
 extern struct qeth_discipline qeth_l3_discipline;
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index 68f6043f033a..6640616a8439 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -1275,27 +1275,25 @@ static int qeth_setup_channel(struct qeth_channel *channel, bool alloc_buffers)
 	return 0;
 }
 
-static void qeth_set_single_write_queues(struct qeth_card *card)
+static void qeth_osa_set_output_queues(struct qeth_card *card, bool single)
 {
-	if ((atomic_read(&card->qdio.state) != QETH_QDIO_UNINITIALIZED) &&
-	    (card->qdio.no_out_queues == 4))
-		qeth_free_qdio_queues(card);
+	unsigned int count = single ? 1 : card->dev->num_tx_queues;
 
-	card->qdio.no_out_queues = 1;
-	if (card->qdio.default_out_queue != 0)
-		dev_info(&card->gdev->dev, "Priority Queueing not supported\n");
+	rtnl_lock();
+	netif_set_real_num_tx_queues(card->dev, count);
+	rtnl_unlock();
 
-	card->qdio.default_out_queue = 0;
-}
+	if (card->qdio.no_out_queues == count)
+		return;
 
-static void qeth_set_multiple_write_queues(struct qeth_card *card)
-{
-	if ((atomic_read(&card->qdio.state) != QETH_QDIO_UNINITIALIZED) &&
-	    (card->qdio.no_out_queues == 1)) {
+	if (atomic_read(&card->qdio.state) != QETH_QDIO_UNINITIALIZED)
 		qeth_free_qdio_queues(card);
-		card->qdio.default_out_queue = 2;
-	}
-	card->qdio.no_out_queues = 4;
+
+	if (count == 1)
+		dev_info(&card->gdev->dev, "Priority Queueing not supported\n");
+
+	card->qdio.default_out_queue = single ? 0 : QETH_DEFAULT_QUEUE;
+	card->qdio.no_out_queues = count;
 }
 
 static int qeth_update_from_chp_desc(struct qeth_card *card)
@@ -1311,15 +1309,11 @@ static int qeth_update_from_chp_desc(struct qeth_card *card)
 		return -ENOMEM;
 
 	card->info.func_level = 0x4100 + chp_dsc->desc;
-	if (card->info.type == QETH_CARD_TYPE_IQD)
-		goto out;
 
-	/* CHPP field bit 6 == 1 -> single queue */
-	if ((chp_dsc->chpp & 0x02) == 0x02)
-		qeth_set_single_write_queues(card);
-	else
-		qeth_set_multiple_write_queues(card);
-out:
+	if (IS_OSD(card) || IS_OSX(card))
+		/* CHPP field bit 6 == 1 -> single queue */
+		qeth_osa_set_output_queues(card, chp_dsc->chpp & 0x02);
+
 	kfree(chp_dsc);
 	QETH_DBF_TEXT_(SETUP, 2, "nr:%x", card->qdio.no_out_queues);
 	QETH_DBF_TEXT_(SETUP, 2, "lvl:%02x", card->info.func_level);
@@ -1332,7 +1326,6 @@ static void qeth_init_qdio_info(struct qeth_card *card)
 	atomic_set(&card->qdio.state, QETH_QDIO_UNINITIALIZED);
 	card->qdio.do_prio_queueing = QETH_PRIOQ_DEFAULT;
 	card->qdio.default_out_queue = QETH_DEFAULT_QUEUE;
-	card->qdio.no_out_queues = QETH_MAX_QUEUES;
 
 	/* inbound */
 	card->qdio.no_in_queues = 1;
@@ -3414,7 +3407,7 @@ static void qeth_check_outbound_queue(struct qeth_qdio_out_q *queue)
 			 * do_send_packet. So, we check if there is a
 			 * packing buffer to be flushed here.
 			 */
-			netif_stop_queue(queue->card->dev);
+			netif_stop_subqueue(queue->card->dev, queue->queue_no);
 			index = queue->next_buf_to_fill;
 			q_was_packing = queue->do_pack;
 			/* queue->do_pack may change */
@@ -3597,7 +3590,7 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	if (card->info.type != QETH_CARD_TYPE_IQD)
 		qeth_check_outbound_queue(queue);
 
-	txq = IS_IQD(card) ? qeth_iqd_translate_txq(dev, __queue) : 0;
+	txq = IS_IQD(card) ? qeth_iqd_translate_txq(dev, __queue) : __queue;
 	netif_wake_subqueue(dev, txq);
 }
 
@@ -3612,8 +3605,7 @@ static inline int qeth_cut_iqd_prio(struct qeth_card *card, int queue_num)
 /**
  * Note: Function assumes that we have 4 outbound queues.
  */
-int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb,
-			    int ipv)
+int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb)
 {
 	__be16 *tci;
 	u8 tos;
@@ -3621,7 +3613,7 @@ int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb,
 	switch (card->qdio.do_prio_queueing) {
 	case QETH_PRIO_Q_ING_TOS:
 	case QETH_PRIO_Q_ING_PREC:
-		switch (ipv) {
+		switch (qeth_get_ip_version(skb)) {
 		case 4:
 			tos = ipv4_get_dsfield(ip_hdr(skb));
 			break;
@@ -5563,11 +5555,14 @@ static struct net_device *qeth_alloc_netdev(struct qeth_card *card)
 		dev = alloc_netdev_mqs(0, "hsi%d", NET_NAME_UNKNOWN,
 				       ether_setup, QETH_MAX_QUEUES, 1);
 		break;
+	case QETH_CARD_TYPE_OSM:
+		dev = alloc_etherdev(0);
+		break;
 	case QETH_CARD_TYPE_OSN:
 		dev = alloc_netdev(0, "osn%d", NET_NAME_UNKNOWN, ether_setup);
 		break;
 	default:
-		dev = alloc_etherdev(0);
+		dev = alloc_etherdev_mqs(0, QETH_MAX_QUEUES, 1);
 	}
 
 	if (!dev)
@@ -5642,16 +5637,16 @@ static int qeth_core_probe_device(struct ccwgroup_device *gdev)
 	}
 
 	qeth_setup_card(card);
-	rc = qeth_update_from_chp_desc(card);
-	if (rc)
-		goto err_chp_desc;
-
 	card->dev = qeth_alloc_netdev(card);
 	if (!card->dev) {
 		rc = -ENOMEM;
 		goto err_card;
 	}
 
+	card->qdio.no_out_queues = card->dev->num_tx_queues;
+	rc = qeth_update_from_chp_desc(card);
+	if (rc)
+		goto err_chp_desc;
 	qeth_determine_capabilities(card);
 	enforced_disc = qeth_enforce_discipline(card);
 	switch (enforced_disc) {
@@ -5678,8 +5673,8 @@ static int qeth_core_probe_device(struct ccwgroup_device *gdev)
 err_disc:
 	qeth_core_free_discipline(card);
 err_load:
-	free_netdev(card->dev);
 err_chp_desc:
+	free_netdev(card->dev);
 err_card:
 	qeth_core_free_card(card);
 err_dev:
diff --git a/drivers/s390/net/qeth_l2_main.c b/drivers/s390/net/qeth_l2_main.c
index 1491281600c2..7381917f76dd 100644
--- a/drivers/s390/net/qeth_l2_main.c
+++ b/drivers/s390/net/qeth_l2_main.c
@@ -602,7 +602,6 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 {
 	struct qeth_card *card = dev->ml_priv;
 	u16 txq = skb_get_queue_mapping(skb);
-	int ipv = qeth_get_ip_version(skb);
 	struct qeth_qdio_out_q *queue;
 	int tx_bytes = skb->len;
 	int rc;
@@ -610,14 +609,14 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 	if (IS_IQD(card))
 		queue = card->qdio.out_qs[qeth_iqd_translate_txq(dev, txq)];
 	else
-		queue = qeth_get_tx_queue(card, skb, ipv);
+		queue = card->qdio.out_qs[txq];
 
 	netif_stop_subqueue(dev, txq);
 
 	if (IS_OSN(card))
 		rc = qeth_l2_xmit_osn(card, skb, queue);
 	else
-		rc = qeth_xmit(card, skb, queue, ipv,
+		rc = qeth_xmit(card, skb, queue, qeth_get_ip_version(skb),
 			       qeth_l2_get_cast_type(skb), qeth_l2_fill_header);
 
 	if (!rc) {
@@ -638,8 +637,13 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 static u16 qeth_l2_select_queue(struct net_device *dev, struct sk_buff *skb,
 				struct net_device *sb_dev)
 {
-	return qeth_iqd_select_queue(dev, skb, qeth_l2_get_cast_type(skb),
-				     sb_dev);
+	struct qeth_card *card = dev->ml_priv;
+
+	if (IS_IQD(card))
+		return qeth_iqd_select_queue(dev, skb,
+					     qeth_l2_get_cast_type(skb),
+					     sb_dev);
+	return qeth_get_priority_queue(card, skb);
 }
 
 static const struct device_type qeth_l2_devtype = {
diff --git a/drivers/s390/net/qeth_l3_main.c b/drivers/s390/net/qeth_l3_main.c
index 120193e90adb..65244da4f415 100644
--- a/drivers/s390/net/qeth_l3_main.c
+++ b/drivers/s390/net/qeth_l3_main.c
@@ -2084,7 +2084,7 @@ static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 		else
 			cast_type = RTN_UNICAST;
 	} else {
-		queue = qeth_get_tx_queue(card, skb, ipv);
+		queue = card->qdio.out_qs[txq];
 		cast_type = qeth_l3_get_cast_type(skb);
 	}
 
@@ -2162,6 +2162,14 @@ static u16 qeth_l3_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
 				     sb_dev);
 }
 
+static u16 qeth_l3_osa_select_queue(struct net_device *dev, struct sk_buff *skb,
+				    struct net_device *sb_dev)
+{
+	struct qeth_card *card = dev->ml_priv;
+
+	return qeth_get_priority_queue(card, skb);
+}
+
 static const struct net_device_ops qeth_l3_netdev_ops = {
 	.ndo_open		= qeth_open,
 	.ndo_stop		= qeth_stop,
@@ -2184,6 +2192,7 @@ static const struct net_device_ops qeth_l3_osa_netdev_ops = {
 	.ndo_get_stats64	= qeth_get_stats64,
 	.ndo_start_xmit		= qeth_l3_hard_start_xmit,
 	.ndo_features_check	= qeth_l3_osa_features_check,
+	.ndo_select_queue	= qeth_l3_osa_select_queue,
 	.ndo_validate_addr	= eth_validate_addr,
 	.ndo_set_rx_mode	= qeth_l3_set_rx_mode,
 	.ndo_do_ioctl		= qeth_do_ioctl,
-- 
2.16.4

[PATCH net-next 7/8] s390/qeth: simplify QoS code

[Julian Wiedmann]

qeth_get_priority_queue() is no longer used for IQD devices, remove the
special-casing of their mcast queue.

This effectively reverts
commit 70deb01662b1 ("qeth: omit outbound queue 3 for unicast packets in Priority Queuing on HiperSockets").

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core_main.c | 23 +++++++----------------
 drivers/s390/net/qeth_core_sys.c  |  4 ----
 2 files changed, 7 insertions(+), 20 deletions(-)

diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index 6640616a8439..d23ad6e3bb45 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -3594,20 +3594,12 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	netif_wake_subqueue(dev, txq);
 }
 
-/* We cannot use outbound queue 3 for unicast packets on HiperSockets */
-static inline int qeth_cut_iqd_prio(struct qeth_card *card, int queue_num)
-{
-	if ((card->info.type == QETH_CARD_TYPE_IQD) && (queue_num == 3))
-		return 2;
-	return queue_num;
-}
-
 /**
  * Note: Function assumes that we have 4 outbound queues.
  */
 int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb)
 {
-	__be16 *tci;
+	struct vlan_ethhdr *veth = vlan_eth_hdr(skb);
 	u8 tos;
 
 	switch (card->qdio.do_prio_queueing) {
@@ -3624,9 +3616,9 @@ int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb)
 			return card->qdio.default_out_queue;
 		}
 		if (card->qdio.do_prio_queueing == QETH_PRIO_Q_ING_PREC)
-			return qeth_cut_iqd_prio(card, ~tos >> 6 & 3);
+			return ~tos >> 6 & 3;
 		if (tos & IPTOS_MINCOST)
-			return qeth_cut_iqd_prio(card, 3);
+			return 3;
 		if (tos & IPTOS_RELIABILITY)
 			return 2;
 		if (tos & IPTOS_THROUGHPUT)
@@ -3637,12 +3629,11 @@ int qeth_get_priority_queue(struct qeth_card *card, struct sk_buff *skb)
 	case QETH_PRIO_Q_ING_SKB:
 		if (skb->priority > 5)
 			return 0;
-		return qeth_cut_iqd_prio(card, ~skb->priority >> 1 & 3);
+		return ~skb->priority >> 1 & 3;
 	case QETH_PRIO_Q_ING_VLAN:
-		tci = &((struct ethhdr *)skb->data)->h_proto;
-		if (be16_to_cpu(*tci) == ETH_P_8021Q)
-			return qeth_cut_iqd_prio(card,
-			~be16_to_cpu(*(tci + 1)) >> (VLAN_PRIO_SHIFT + 1) & 3);
+		if (veth->h_vlan_proto == htons(ETH_P_8021Q))
+			return ~ntohs(veth->h_vlan_TCI) >>
+			       (VLAN_PRIO_SHIFT + 1) & 3;
 		break;
 	default:
 		break;
diff --git a/drivers/s390/net/qeth_core_sys.c b/drivers/s390/net/qeth_core_sys.c
index b43d8bdf4c3e..cea4a0bbc303 100644
--- a/drivers/s390/net/qeth_core_sys.c
+++ b/drivers/s390/net/qeth_core_sys.c
@@ -242,10 +242,6 @@ static ssize_t qeth_dev_prioqing_store(struct device *dev,
 		card->qdio.do_prio_queueing = QETH_NO_PRIO_QUEUEING;
 		card->qdio.default_out_queue = 2;
 	} else if (sysfs_streq(buf, "no_prio_queueing:3")) {
-		if (card->info.type == QETH_CARD_TYPE_IQD) {
-			rc = -EPERM;
-			goto out;
-		}
 		card->qdio.do_prio_queueing = QETH_NO_PRIO_QUEUEING;
 		card->qdio.default_out_queue = 3;
 	} else if (sysfs_streq(buf, "no_prio_queueing")) {
-- 
2.16.4

[PATCH net-next 8/8] s390/qeth: stop/wake TX queues based on their fill level

[Julian Wiedmann]

Current xmit code only stops the txq after attempting to fill an
IO buffer that hasn't been TX-completed yet. In many-connection
scenarios, this can result in frequent rejected TX attempts, requeuing
of skbs with NETDEV_TX_BUSY and extra overhead.

Now that we have a proper 1-to-1 relation between stack-side txqs and
our HW Queues, overhaul the stop/wake logic so that the xmit code
stops the txq as needed.
Given that we might map multiple skbs into a single buffer, it's crucial
to ensure that the queue always provides an _entirely_ empty IO buffer.
Otherwise large skbs (eg TSO) might not fit into the last available
buffer. So whenever qeth_do_send_packet() first utilizes an _empty_
buffer, it updates & checks the used_buffers count.

This now ensures that an skb passed to qeth_xmit() can always be mapped
into an IO buffer, so remove all of the -EBUSY roll-back handling in the
TX path. We preserve the minimal safety-checks ("Is this IO buffer
really available?"), just in case some nasty future bug ever attempts to
corrupt an in-use buffer.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      |  6 +++
 drivers/s390/net/qeth_core_main.c | 81 ++++++++++++++++++++++++++++-----------
 drivers/s390/net/qeth_ethtool.c   |  1 +
 drivers/s390/net/qeth_l2_main.c   | 13 ++-----
 drivers/s390/net/qeth_l3_main.c   | 18 +--------
 5 files changed, 71 insertions(+), 48 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 4989dc7b58c1..fbaf434e2e34 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -481,6 +481,7 @@ struct qeth_out_q_stats {
 	u64 skbs_linearized_fail;
 	u64 tso_bytes;
 	u64 packing_mode_switch;
+	u64 stopped;
 
 	/* rtnl_link_stats64 */
 	u64 tx_packets;
@@ -511,6 +512,11 @@ struct qeth_qdio_out_q {
 	atomic_t set_pci_flags_count;
 };
 
+static inline bool qeth_out_queue_is_full(struct qeth_qdio_out_q *queue)
+{
+	return atomic_read(&queue->used_buffers) >= QDIO_MAX_BUFFERS_PER_Q;
+}
+
 struct qeth_qdio_info {
 	atomic_t state;
 	/* input */
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index d23ad6e3bb45..d057ead200b5 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -3367,7 +3367,6 @@ static void qeth_flush_buffers(struct qeth_qdio_out_q *queue, int index,
 	qdio_flags = QDIO_FLAG_SYNC_OUTPUT;
 	if (atomic_read(&queue->set_pci_flags_count))
 		qdio_flags |= QDIO_FLAG_PCI_OUT;
-	atomic_add(count, &queue->used_buffers);
 	rc = do_QDIO(CARD_DDEV(queue->card), qdio_flags,
 		     queue->queue_no, index, count);
 	if (rc) {
@@ -3407,7 +3406,6 @@ static void qeth_check_outbound_queue(struct qeth_qdio_out_q *queue)
 			 * do_send_packet. So, we check if there is a
 			 * packing buffer to be flushed here.
 			 */
-			netif_stop_subqueue(queue->card->dev, queue->queue_no);
 			index = queue->next_buf_to_fill;
 			q_was_packing = queue->do_pack;
 			/* queue->do_pack may change */
@@ -3535,7 +3533,7 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	struct qeth_qdio_out_q *queue = card->qdio.out_qs[__queue];
 	struct qeth_qdio_out_buffer *buffer;
 	struct net_device *dev = card->dev;
-	u16 txq;
+	struct netdev_queue *txq;
 	int i;
 
 	QETH_CARD_TEXT(card, 6, "qdouhdl");
@@ -3590,8 +3588,15 @@ static void qeth_qdio_output_handler(struct ccw_device *ccwdev,
 	if (card->info.type != QETH_CARD_TYPE_IQD)
 		qeth_check_outbound_queue(queue);
 
-	txq = IS_IQD(card) ? qeth_iqd_translate_txq(dev, __queue) : __queue;
-	netif_wake_subqueue(dev, txq);
+	if (IS_IQD(card))
+		__queue = qeth_iqd_translate_txq(dev, __queue);
+	txq = netdev_get_tx_queue(dev, __queue);
+	/* xmit may have observed the full-condition, but not yet stopped the
+	 * txq. In which case the code below won't trigger. So before returning,
+	 * xmit will re-check the txq's fill level and wake it up if needed.
+	 */
+	if (netif_tx_queue_stopped(txq) && !qeth_out_queue_is_full(queue))
+		netif_tx_wake_queue(txq);
 }
 
 /**
@@ -3845,11 +3850,13 @@ static void __qeth_fill_buffer(struct sk_buff *skb,
  *		from qeth_core_header_cache.
  * @offset:	when mapping the skb, start at skb->data + offset
  * @hd_len:	if > 0, build a dedicated header element of this size
+ * flush:	Prepare the buffer to be flushed, regardless of its fill level.
  */
 static int qeth_fill_buffer(struct qeth_qdio_out_q *queue,
 			    struct qeth_qdio_out_buffer *buf,
 			    struct sk_buff *skb, struct qeth_hdr *hdr,
-			    unsigned int offset, unsigned int hd_len)
+			    unsigned int offset, unsigned int hd_len,
+			    bool flush)
 {
 	struct qdio_buffer *buffer = buf->buffer;
 	bool is_first_elem = true;
@@ -3878,8 +3885,8 @@ static int qeth_fill_buffer(struct qeth_qdio_out_q *queue,
 
 		QETH_TXQ_STAT_INC(queue, skbs_pack);
 		/* If the buffer still has free elements, keep using it. */
-		if (buf->next_element_to_fill <
-		    QETH_MAX_BUFFER_ELEMENTS(queue->card))
+		if (!flush && buf->next_element_to_fill <
+			      QETH_MAX_BUFFER_ELEMENTS(queue->card))
 			return 0;
 	}
 
@@ -3896,15 +3903,31 @@ static int qeth_do_send_packet_fast(struct qeth_qdio_out_q *queue,
 {
 	int index = queue->next_buf_to_fill;
 	struct qeth_qdio_out_buffer *buffer = queue->bufs[index];
+	struct netdev_queue *txq;
+	bool stopped = false;
 
-	/*
-	 * check if buffer is empty to make sure that we do not 'overtake'
-	 * ourselves and try to fill a buffer that is already primed
+	/* Just a sanity check, the wake/stop logic should ensure that we always
+	 * get a free buffer.
 	 */
 	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY)
 		return -EBUSY;
-	qeth_fill_buffer(queue, buffer, skb, hdr, offset, hd_len);
+
+	txq = netdev_get_tx_queue(queue->card->dev, skb_get_queue_mapping(skb));
+
+	if (atomic_inc_return(&queue->used_buffers) >= QDIO_MAX_BUFFERS_PER_Q) {
+		/* If a TX completion happens right _here_ and misses to wake
+		 * the txq, then our re-check below will catch the race.
+		 */
+		QETH_TXQ_STAT_INC(queue, stopped);
+		netif_tx_stop_queue(txq);
+		stopped = true;
+	}
+
+	qeth_fill_buffer(queue, buffer, skb, hdr, offset, hd_len, stopped);
 	qeth_flush_buffers(queue, index, 1);
+
+	if (stopped && !qeth_out_queue_is_full(queue))
+		netif_tx_start_queue(txq);
 	return 0;
 }
 
@@ -3914,6 +3937,8 @@ int qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 			int elements_needed)
 {
 	struct qeth_qdio_out_buffer *buffer;
+	struct netdev_queue *txq;
+	bool stopped = false;
 	int start_index;
 	int flush_count = 0;
 	int do_pack = 0;
@@ -3925,14 +3950,17 @@ int qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 			      QETH_OUT_Q_LOCKED) != QETH_OUT_Q_UNLOCKED);
 	start_index = queue->next_buf_to_fill;
 	buffer = queue->bufs[queue->next_buf_to_fill];
-	/*
-	 * check if buffer is empty to make sure that we do not 'overtake'
-	 * ourselves and try to fill a buffer that is already primed
+
+	/* Just a sanity check, the wake/stop logic should ensure that we always
+	 * get a free buffer.
 	 */
 	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY) {
 		atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
 		return -EBUSY;
 	}
+
+	txq = netdev_get_tx_queue(card->dev, skb_get_queue_mapping(skb));
+
 	/* check if we need to switch packing state of this queue */
 	qeth_switch_to_packing_if_needed(queue);
 	if (queue->do_pack) {
@@ -3947,8 +3975,8 @@ int qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 				(queue->next_buf_to_fill + 1) %
 				QDIO_MAX_BUFFERS_PER_Q;
 			buffer = queue->bufs[queue->next_buf_to_fill];
-			/* we did a step forward, so check buffer state
-			 * again */
+
+			/* We stepped forward, so sanity-check again: */
 			if (atomic_read(&buffer->state) !=
 			    QETH_QDIO_BUF_EMPTY) {
 				qeth_flush_buffers(queue, start_index,
@@ -3961,8 +3989,18 @@ int qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 		}
 	}
 
-	flush_count += qeth_fill_buffer(queue, buffer, skb, hdr, offset,
-					hd_len);
+	if (buffer->next_element_to_fill == 0 &&
+	    atomic_inc_return(&queue->used_buffers) >= QDIO_MAX_BUFFERS_PER_Q) {
+		/* If a TX completion happens right _here_ and misses to wake
+		 * the txq, then our re-check below will catch the race.
+		 */
+		QETH_TXQ_STAT_INC(queue, stopped);
+		netif_tx_stop_queue(txq);
+		stopped = true;
+	}
+
+	flush_count += qeth_fill_buffer(queue, buffer, skb, hdr, offset, hd_len,
+					stopped);
 	if (flush_count)
 		qeth_flush_buffers(queue, start_index, flush_count);
 	else if (!atomic_read(&queue->set_pci_flags_count))
@@ -3993,6 +4031,8 @@ int qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 	if (do_pack)
 		QETH_TXQ_STAT_ADD(queue, bufs_pack, flush_count);
 
+	if (stopped && !qeth_out_queue_is_full(queue))
+		netif_tx_start_queue(txq);
 	return rc;
 }
 EXPORT_SYMBOL_GPL(qeth_do_send_packet);
@@ -4079,9 +4119,6 @@ int qeth_xmit(struct qeth_card *card, struct sk_buff *skb,
 	} else {
 		if (!push_len)
 			kmem_cache_free(qeth_core_header_cache, hdr);
-		if (rc == -EBUSY)
-			/* roll back to ETH header */
-			skb_pull(skb, push_len);
 	}
 	return rc;
 }
diff --git a/drivers/s390/net/qeth_ethtool.c b/drivers/s390/net/qeth_ethtool.c
index a443e5f86ab7..4166eb29f0bd 100644
--- a/drivers/s390/net/qeth_ethtool.c
+++ b/drivers/s390/net/qeth_ethtool.c
@@ -38,6 +38,7 @@ static const struct qeth_stats txq_stats[] = {
 	QETH_TXQ_STAT("linearized+error skbs", skbs_linearized_fail),
 	QETH_TXQ_STAT("TSO bytes", tso_bytes),
 	QETH_TXQ_STAT("Packing mode switches", packing_mode_switch),
+	QETH_TXQ_STAT("Queue stopped", stopped),
 };
 
 static const struct qeth_stats card_stats[] = {
diff --git a/drivers/s390/net/qeth_l2_main.c b/drivers/s390/net/qeth_l2_main.c
index 7381917f76dd..e26a6dff286f 100644
--- a/drivers/s390/net/qeth_l2_main.c
+++ b/drivers/s390/net/qeth_l2_main.c
@@ -607,11 +607,8 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 	int rc;
 
 	if (IS_IQD(card))
-		queue = card->qdio.out_qs[qeth_iqd_translate_txq(dev, txq)];
-	else
-		queue = card->qdio.out_qs[txq];
-
-	netif_stop_subqueue(dev, txq);
+		txq = qeth_iqd_translate_txq(dev, txq);
+	queue = card->qdio.out_qs[txq];
 
 	if (IS_OSN(card))
 		rc = qeth_l2_xmit_osn(card, skb, queue);
@@ -622,15 +619,11 @@ static netdev_tx_t qeth_l2_hard_start_xmit(struct sk_buff *skb,
 	if (!rc) {
 		QETH_TXQ_STAT_INC(queue, tx_packets);
 		QETH_TXQ_STAT_ADD(queue, tx_bytes, tx_bytes);
-		netif_wake_subqueue(dev, txq);
 		return NETDEV_TX_OK;
-	} else if (rc == -EBUSY) {
-		return NETDEV_TX_BUSY;
-	} /* else fall through */
+	}
 
 	QETH_TXQ_STAT_INC(queue, tx_dropped);
 	kfree_skb(skb);
-	netif_wake_subqueue(dev, txq);
 	return NETDEV_TX_OK;
 }
 
diff --git a/drivers/s390/net/qeth_l3_main.c b/drivers/s390/net/qeth_l3_main.c
index 65244da4f415..4c9394105138 100644
--- a/drivers/s390/net/qeth_l3_main.c
+++ b/drivers/s390/net/qeth_l3_main.c
@@ -2036,7 +2036,6 @@ static void qeth_l3_fixup_headers(struct sk_buff *skb)
 static int qeth_l3_xmit(struct qeth_card *card, struct sk_buff *skb,
 			struct qeth_qdio_out_q *queue, int ipv, int cast_type)
 {
-	unsigned char eth_hdr[ETH_HLEN];
 	unsigned int hw_hdr_len;
 	int rc;
 
@@ -2046,17 +2045,10 @@ static int qeth_l3_xmit(struct qeth_card *card, struct sk_buff *skb,
 	rc = skb_cow_head(skb, hw_hdr_len - ETH_HLEN);
 	if (rc)
 		return rc;
-	skb_copy_from_linear_data(skb, eth_hdr, ETH_HLEN);
 	skb_pull(skb, ETH_HLEN);
 
 	qeth_l3_fixup_headers(skb);
-	rc = qeth_xmit(card, skb, queue, ipv, cast_type, qeth_l3_fill_header);
-	if (rc == -EBUSY) {
-		/* roll back to ETH header */
-		skb_push(skb, ETH_HLEN);
-		skb_copy_to_linear_data(skb, eth_hdr, ETH_HLEN);
-	}
-	return rc;
+	return qeth_xmit(card, skb, queue, ipv, cast_type, qeth_l3_fill_header);
 }
 
 static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
@@ -2091,8 +2083,6 @@ static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 	if (cast_type == RTN_BROADCAST && !card->info.broadcast_capable)
 		goto tx_drop;
 
-	netif_stop_subqueue(dev, txq);
-
 	if (ipv == 4 || IS_IQD(card))
 		rc = qeth_l3_xmit(card, skb, queue, ipv, cast_type);
 	else
@@ -2102,16 +2092,12 @@ static netdev_tx_t qeth_l3_hard_start_xmit(struct sk_buff *skb,
 	if (!rc) {
 		QETH_TXQ_STAT_INC(queue, tx_packets);
 		QETH_TXQ_STAT_ADD(queue, tx_bytes, tx_bytes);
-		netif_wake_subqueue(dev, txq);
 		return NETDEV_TX_OK;
-	} else if (rc == -EBUSY) {
-		return NETDEV_TX_BUSY;
-	} /* else fall through */
+	}
 
 tx_drop:
 	QETH_TXQ_STAT_INC(queue, tx_dropped);
 	kfree_skb(skb);
-	netif_wake_subqueue(dev, txq);
 	return NETDEV_TX_OK;
 }
 
-- 
2.16.4

Re: [PATCH net-next 0/8] s390/qeth: updates 2019-04-17

[David Miller]

From: Julian Wiedmann <jwi@linux.ibm.com>
Date: Wed, 17 Apr 2019 18:17:27 +0200

> please apply some additional qeth patches to net-next. This patchset converts
> the driver to use the kernel's multiqueue model.

Series applied, thanks.