netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Julian Wiedmann <jwi@linux.ibm.com>
To: David Miller <davem@davemloft.net>
Cc: netdev <netdev@vger.kernel.org>,
	linux-s390 <linux-s390@vger.kernel.org>,
	Heiko Carstens <heiko.carstens@de.ibm.com>,
	Ursula Braun <ubraun@linux.ibm.com>,
	Julian Wiedmann <jwi@linux.ibm.com>
Subject: [PATCH net-next 06/11] s390/qeth: balance the TX queue selection for IQD devices
Date: Wed, 18 Mar 2020 13:54:50 +0100	[thread overview]
Message-ID: <20200318125455.5838-7-jwi@linux.ibm.com> (raw)
In-Reply-To: <20200318125455.5838-1-jwi@linux.ibm.com>

For ucast traffic, qeth_iqd_select_queue() falls back to
netdev_pick_tx(). This will potentially use skb_tx_hash() to distribute
the flow over all active TX queues - so txq 0 is a valid selection, and
qeth_iqd_select_queue() needs to check for this and put it on some other
queue. As a result, the distribution for ucast flows is unbalanced and
hits QETH_IQD_MIN_UCAST_TXQ heavier than the other queues.

Open-coding a custom variant of skb_tx_hash() isn't an option, since
netdev_pick_tx() also gives us eg. access to XPS. But we can pull a
little trick: add a single TC class that excludes the mcast txq, and
thus encourage skb_tx_hash() to not pick the mcast txq.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
---
 drivers/s390/net/qeth_core.h      |  1 +
 drivers/s390/net/qeth_core_main.c | 45 ++++++++++++++++++++++++++++++-
 drivers/s390/net/qeth_ethtool.c   |  2 +-
 3 files changed, 46 insertions(+), 2 deletions(-)

diff --git a/drivers/s390/net/qeth_core.h b/drivers/s390/net/qeth_core.h
index 94cd39631eee..b8b356aca674 100644
--- a/drivers/s390/net/qeth_core.h
+++ b/drivers/s390/net/qeth_core.h
@@ -1061,6 +1061,7 @@ 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);
+int qeth_set_real_num_tx_queues(struct qeth_card *card, unsigned int count);
 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);
diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index aa493edc0082..e1d984c29e1f 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -6025,7 +6025,7 @@ int qeth_setup_netdev(struct qeth_card *card)
 		num_tx_queues = dev->real_num_tx_queues;
 	}
 
-	return netif_set_real_num_tx_queues(dev, num_tx_queues);
+	return qeth_set_real_num_tx_queues(card, num_tx_queues);
 }
 EXPORT_SYMBOL_GPL(qeth_setup_netdev);
 
@@ -6641,6 +6641,47 @@ void qeth_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 }
 EXPORT_SYMBOL_GPL(qeth_get_stats64);
 
+#define TC_IQD_UCAST   0
+static void qeth_iqd_set_prio_tc_map(struct net_device *dev,
+				     unsigned int ucast_txqs)
+{
+	unsigned int prio;
+
+	/* IQD requires mcast traffic to be placed on a dedicated queue, and
+	 * qeth_iqd_select_queue() deals with this.
+	 * For unicast traffic, we defer the queue selection to the stack.
+	 * By installing a trivial prio map that spans over only the unicast
+	 * queues, we can encourage the stack to spread the ucast traffic evenly
+	 * without selecting the mcast queue.
+	 */
+
+	/* One traffic class, spanning over all active ucast queues: */
+	netdev_set_num_tc(dev, 1);
+	netdev_set_tc_queue(dev, TC_IQD_UCAST, ucast_txqs,
+			    QETH_IQD_MIN_UCAST_TXQ);
+
+	/* Map all priorities to this traffic class: */
+	for (prio = 0; prio <= TC_BITMASK; prio++)
+		netdev_set_prio_tc_map(dev, prio, TC_IQD_UCAST);
+}
+
+int qeth_set_real_num_tx_queues(struct qeth_card *card, unsigned int count)
+{
+	struct net_device *dev = card->dev;
+	int rc;
+
+	/* Per netif_setup_tc(), adjust the mapping first: */
+	if (IS_IQD(card))
+		qeth_iqd_set_prio_tc_map(dev, count - 1);
+
+	rc = netif_set_real_num_tx_queues(dev, count);
+
+	if (rc && IS_IQD(card))
+		qeth_iqd_set_prio_tc_map(dev, dev->real_num_tx_queues - 1);
+
+	return rc;
+}
+
 u16 qeth_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
 			  u8 cast_type, struct net_device *sb_dev)
 {
@@ -6648,6 +6689,8 @@ u16 qeth_iqd_select_queue(struct net_device *dev, struct sk_buff *skb,
 
 	if (cast_type != RTN_UNICAST)
 		return QETH_IQD_MCAST_TXQ;
+	if (dev->real_num_tx_queues == QETH_IQD_MIN_TXQ)
+		return QETH_IQD_MIN_UCAST_TXQ;
 
 	txq = netdev_pick_tx(dev, skb, sb_dev);
 	return (txq == QETH_IQD_MCAST_TXQ) ? QETH_IQD_MIN_UCAST_TXQ : txq;
diff --git a/drivers/s390/net/qeth_ethtool.c b/drivers/s390/net/qeth_ethtool.c
index 715ee0015847..079b695032ef 100644
--- a/drivers/s390/net/qeth_ethtool.c
+++ b/drivers/s390/net/qeth_ethtool.c
@@ -201,7 +201,7 @@ static int qeth_set_channels(struct net_device *dev,
 			return -EOPNOTSUPP;
 	}
 
-	return netif_set_real_num_tx_queues(dev, channels->tx_count);
+	return qeth_set_real_num_tx_queues(card, channels->tx_count);
 }
 
 static int qeth_get_tunable(struct net_device *dev,
-- 
2.17.1


  parent reply	other threads:[~2020-03-18 12:55 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-18 12:54 [PATCH net-next 00/11] s390/qeth: updates 2020-03-18 Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 01/11] s390/qeth: use memory reserves to back RX buffers Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 02/11] s390/qeth: use memory reserves in TX slow path Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 03/11] s390/qeth: remove prio-queueing support for z/VM NICs Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 04/11] s390/qeth: allow configuration of TX queues " Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 05/11] s390/qeth: allow configuration of TX queues for IQD devices Julian Wiedmann
2020-03-18 12:54 ` Julian Wiedmann [this message]
2020-03-18 12:54 ` [PATCH net-next 07/11] s390/qeth: add SW timestamping support " Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 08/11] s390/qeth: don't report hard-coded driver version Julian Wiedmann
2020-03-18 13:07   ` Leon Romanovsky
2020-03-18 12:54 ` [PATCH net-next 09/11] s390/qeth: add phys_to_virt() translation for AOB Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 10/11] s390/qeth: remove gratuitous NULL checks Julian Wiedmann
2020-03-18 12:54 ` [PATCH net-next 11/11] s390/qeth: use dev->reg_state Julian Wiedmann
2020-03-18 23:34 ` [PATCH net-next 00/11] s390/qeth: updates 2020-03-18 David Miller

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200318125455.5838-7-jwi@linux.ibm.com \
    --to=jwi@linux.ibm.com \
    --cc=davem@davemloft.net \
    --cc=heiko.carstens@de.ibm.com \
    --cc=linux-s390@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=ubraun@linux.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).