[net-next-2.6 PATCH v7 1/2] net: implement mechanism for HW based QOS

[net-next-2.6 PATCH v7 1/2] net: implement mechanism for HW based QOS

[John Fastabend]

This patch provides a mechanism for lower layer devices to
steer traffic using skb->priority to tx queues. This allows
for hardware based QOS schemes to use the default qdisc without
incurring the penalties related to global state and the qdisc
lock. While reliably receiving skbs on the correct tx ring
to avoid head of line blocking resulting from shuffling in
the LLD. Finally, all the goodness from txq caching and xps/rps
can still be leveraged.

Many drivers and hardware exist with the ability to implement
QOS schemes in the hardware but currently these drivers tend
to rely on firmware to reroute specific traffic, a driver
specific select_queue or the queue_mapping action in the
qdisc.

By using select_queue for this drivers need to be updated for
each and every traffic type and we lose the goodness of much
of the upstream work. Firmware solutions are inherently
inflexible. And finally if admins are expected to build a
qdisc and filter rules to steer traffic this requires knowledge
of how the hardware is currently configured. The number of tx
queues and the queue offsets may change depending on resources.
Also this approach incurs all the overhead of a qdisc with filters.

With the mechanism in this patch users can set skb priority using
expected methods ie setsockopt() or the stack can set the priority
directly. Then the skb will be steered to the correct tx queues
aligned with hardware QOS traffic classes. In the normal case with
a single traffic class and all queues in this class everything
works as is until the LLD enables multiple tcs.

To steer the skb we mask out the lower 4 bits of the priority
and allow the hardware to configure upto 15 distinct classes
of traffic. This is expected to be sufficient for most applications
at any rate it is more then the 8021Q spec designates and is
equal to the number of prio bands currently implemented in
the default qdisc.

This in conjunction with a userspace application such as
lldpad can be used to implement 8021Q transmission selection
algorithms one of these algorithms being the extended transmission
selection algorithm currently being used for DCB.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---

 include/linux/netdevice.h |   65 +++++++++++++++++++++++++++++++++++++++++++++
 net/core/dev.c            |   61 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 125 insertions(+), 1 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 0f6b1c9..b1dbbed 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -646,6 +646,14 @@ struct xps_dev_maps {
     (nr_cpu_ids * sizeof(struct xps_map *)))
 #endif /* CONFIG_XPS */
 
+#define TC_MAX_QUEUE	16
+#define TC_BITMASK	15
+/* HW offloaded queuing disciplines txq count and offset maps */
+struct netdev_tc_txq {
+	u16 count;
+	u16 offset;
+};
+
 /*
  * This structure defines the management hooks for network devices.
  * The following hooks can be defined; unless noted otherwise, they are
@@ -756,6 +764,7 @@ struct xps_dev_maps {
  * int (*ndo_set_vf_port)(struct net_device *dev, int vf,
  *			  struct nlattr *port[]);
  * int (*ndo_get_vf_port)(struct net_device *dev, int vf, struct sk_buff *skb);
+ * void (*ndo_setup_tc)(struct net_device *dev, u8 tc, unsigned int txq)
  */
 #define HAVE_NET_DEVICE_OPS
 struct net_device_ops {
@@ -814,6 +823,8 @@ struct net_device_ops {
 						   struct nlattr *port[]);
 	int			(*ndo_get_vf_port)(struct net_device *dev,
 						   int vf, struct sk_buff *skb);
+	int			(*ndo_setup_tc)(struct net_device *dev, u8 tc,
+						unsigned int txq);
 #if defined(CONFIG_FCOE) || defined(CONFIG_FCOE_MODULE)
 	int			(*ndo_fcoe_enable)(struct net_device *dev);
 	int			(*ndo_fcoe_disable)(struct net_device *dev);
@@ -1146,6 +1157,9 @@ struct net_device {
 	/* Data Center Bridging netlink ops */
 	const struct dcbnl_rtnl_ops *dcbnl_ops;
 #endif
+	u8 num_tc;
+	struct netdev_tc_txq tc_to_txq[TC_MAX_QUEUE];
+	u8 prio_tc_map[TC_BITMASK + 1];
 
 #if defined(CONFIG_FCOE) || defined(CONFIG_FCOE_MODULE)
 	/* max exchange id for FCoE LRO by ddp */
@@ -1162,6 +1176,57 @@ struct net_device {
 #define	NETDEV_ALIGN		32
 
 static inline
+int netdev_get_prio_tc_map(const struct net_device *dev, u32 prio)
+{
+	return dev->prio_tc_map[prio & TC_BITMASK];
+}
+
+static inline
+int netdev_set_prio_tc_map(struct net_device *dev, u8 prio, u8 tc)
+{
+	if (tc >= dev->num_tc)
+		return -EINVAL;
+
+	dev->prio_tc_map[prio & TC_BITMASK] = tc & TC_BITMASK;
+	return 0;
+}
+
+static inline
+void netdev_reset_tc(struct net_device *dev)
+{
+	dev->num_tc = 0;
+	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
+	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
+}
+
+static inline
+int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
+{
+	if (tc >= dev->num_tc)
+		return -EINVAL;
+
+	dev->tc_to_txq[tc].count = count;
+	dev->tc_to_txq[tc].offset = offset;
+	return 0;
+}
+
+static inline
+int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
+{
+	if (num_tc > TC_MAX_QUEUE)
+		return -EINVAL;
+
+	dev->num_tc = num_tc;
+	return 0;
+}
+
+static inline
+int netdev_get_num_tc(struct net_device *dev)
+{
+	return dev->num_tc;
+}
+
+static inline
 struct netdev_queue *netdev_get_tx_queue(const struct net_device *dev,
 					 unsigned int index)
 {
diff --git a/net/core/dev.c b/net/core/dev.c
index a215269..7c9b1aa 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1593,6 +1593,54 @@ static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
 	rcu_read_unlock();
 }
 
+/* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
+ * @dev: Network device
+ * @txq: number of queues available
+ *
+ * If real_num_tx_queues is changed the tc mappings may no longer be
+ * valid. To resolve this if the net_device supports ndo_setup_tc
+ * call the ops routine with the new queue number. If the ops is not
+ * available verify the tc mapping remains valid and if not NULL the
+ * mapping. With no priorities mapping to this offset/count pair it
+ * will no longer be used. In the worst case TC0 is invalid nothing
+ * can be done so disable priority mappings.
+ */
+void netif_setup_tc(struct net_device *dev, unsigned int txq)
+{
+	const struct net_device_ops *ops = dev->netdev_ops;
+
+	if (ops->ndo_setup_tc) {
+		ops->ndo_setup_tc(dev, dev->num_tc, txq);
+	} else {
+		int i;
+		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
+
+		/* If TC0 is invalidated disable TC mapping */
+		if (tc->offset + tc->count > txq) {
+			pr_warning("Number of in use tx queues changed "
+				   "invalidating tc mappings. Priority "
+				   "traffic classification disabled!\n");
+			dev->num_tc = 0;
+			return;
+		}
+
+		/* Invalidated prio to tc mappings set to TC0 */
+		for (i = 1; i < TC_BITMASK + 1; i++) {
+			int q = netdev_get_prio_tc_map(dev, i);
+
+			tc = &dev->tc_to_txq[q];
+			if (tc->offset + tc->count > txq) {
+				pr_warning("Number of in use tx queues "
+					   "changed. Priority %i to tc "
+					   "mapping %i is no longer valid "
+					   "setting map to 0\n",
+					   i, q);
+				netdev_set_prio_tc_map(dev, i, 0);
+			}
+		}
+	}
+}
+
 /*
  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
  * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
@@ -1612,6 +1660,9 @@ int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
 		if (rc)
 			return rc;
 
+		if (dev->num_tc)
+			netif_setup_tc(dev, txq);
+
 		if (txq < dev->real_num_tx_queues)
 			qdisc_reset_all_tx_gt(dev, txq);
 	}
@@ -2165,6 +2216,8 @@ u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
 		  unsigned int num_tx_queues)
 {
 	u32 hash;
+	u16 qoffset = 0;
+	u16 qcount = num_tx_queues;
 
 	if (skb_rx_queue_recorded(skb)) {
 		hash = skb_get_rx_queue(skb);
@@ -2173,13 +2226,19 @@ u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
 		return hash;
 	}
 
+	if (dev->num_tc) {
+		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
+		qoffset = dev->tc_to_txq[tc].offset;
+		qcount = dev->tc_to_txq[tc].count;
+	}
+
 	if (skb->sk && skb->sk->sk_hash)
 		hash = skb->sk->sk_hash;
 	else
 		hash = (__force u16) skb->protocol ^ skb->rxhash;
 	hash = jhash_1word(hash, hashrnd);
 
-	return (u16) (((u64) hash * num_tx_queues) >> 32);
+	return (u16) (((u64) hash * qcount) >> 32) + qoffset;
 }
 EXPORT_SYMBOL(__skb_tx_hash);
 

[net-next-2.6 PATCH v7 2/2] net_sched: implement a root container qdisc sch_mqprio

[John Fastabend]

This implements a mqprio queueing discipline that by default creates
a pfifo_fast qdisc per tx queue and provides the needed configuration
interface.

Using the mqprio qdisc the number of tcs currently in use along
with the range of queues alloted to each class can be configured. By
default skbs are mapped to traffic classes using the skb priority.
This mapping is configurable.

Configurable parameters,

struct tc_mqprio_qopt {
        __u8    num_tc;
        __u8    prio_tc_map[TC_BITMASK + 1];
        __u8    hw;
        __u16   count[TC_MAX_QUEUE];
        __u16   offset[TC_MAX_QUEUE];
};

Here the count/offset pairing give the queue alignment and the
prio_tc_map gives the mapping from skb->priority to tc.

The hw bit determines if the hardware should configure the count
and offset values. If the hardware bit is set then the operation
will fail if the hardware does not implement the ndo_setup_tc
operation. This is to avoid undetermined states where the hardware
may or may not control the queue mapping. Also minimal bounds
checking is done on the count/offset to verify a queue does not
exceed num_tx_queues and that queue ranges do not overlap. Otherwise
it is left to user policy or hardware configuration to create
useful mappings.

It is expected that hardware QOS schemes can be implemented by
creating appropriate mappings of queues in ndo_tc_setup().

One expected use case is drivers will use the ndo_setup_tc to map
queue ranges onto 802.1Q traffic classes. This provides a generic
mechanism to map network traffic onto these traffic classes and
removes the need for lower layer drivers to know specifics about
traffic types.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
---

 include/linux/pkt_sched.h |   12 +
 net/sched/Kconfig         |   12 +
 net/sched/Makefile        |    1 
 net/sched/sch_generic.c   |    4 
 net/sched/sch_mqprio.c    |  418 +++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 447 insertions(+), 0 deletions(-)
 create mode 100644 net/sched/sch_mqprio.c

diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h
index 2cfa4bc..776cd93 100644
--- a/include/linux/pkt_sched.h
+++ b/include/linux/pkt_sched.h
@@ -481,4 +481,16 @@ struct tc_drr_stats {
 	__u32	deficit;
 };
 
+/* MQPRIO */
+#define TC_QOPT_BITMASK 15
+#define TC_QOPT_MAX_QUEUE 16
+
+struct tc_mqprio_qopt {
+	__u8	num_tc;
+	__u8	prio_tc_map[TC_QOPT_BITMASK + 1];
+	__u8	hw;
+	__u16	count[TC_QOPT_MAX_QUEUE];
+	__u16	offset[TC_QOPT_MAX_QUEUE];
+};
+
 #endif
diff --git a/net/sched/Kconfig b/net/sched/Kconfig
index a36270a..f52f5eb 100644
--- a/net/sched/Kconfig
+++ b/net/sched/Kconfig
@@ -205,6 +205,18 @@ config NET_SCH_DRR
 
 	  If unsure, say N.
 
+config NET_SCH_MQPRIO
+	tristate "Multi-queue priority scheduler (MQPRIO)"
+	help
+	  Say Y here if you want to use the Multi-queue Priority scheduler.
+	  This scheduler allows QOS to be offloaded on NICs that have support
+	  for offloading QOS schedulers.
+
+	  To compile this driver as a module, choose M here: the module will
+	  be called sch_mqprio.
+
+	  If unsure, say N.
+
 config NET_SCH_INGRESS
 	tristate "Ingress Qdisc"
 	depends on NET_CLS_ACT
diff --git a/net/sched/Makefile b/net/sched/Makefile
index 960f5db..26ce681 100644
--- a/net/sched/Makefile
+++ b/net/sched/Makefile
@@ -32,6 +32,7 @@ obj-$(CONFIG_NET_SCH_MULTIQ)	+= sch_multiq.o
 obj-$(CONFIG_NET_SCH_ATM)	+= sch_atm.o
 obj-$(CONFIG_NET_SCH_NETEM)	+= sch_netem.o
 obj-$(CONFIG_NET_SCH_DRR)	+= sch_drr.o
+obj-$(CONFIG_NET_SCH_MQPRIO)	+= sch_mqprio.o
 obj-$(CONFIG_NET_CLS_U32)	+= cls_u32.o
 obj-$(CONFIG_NET_CLS_ROUTE4)	+= cls_route.o
 obj-$(CONFIG_NET_CLS_FW)	+= cls_fw.o
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 34dc598..723b278 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -540,6 +540,7 @@ struct Qdisc_ops pfifo_fast_ops __read_mostly = {
 	.dump		=	pfifo_fast_dump,
 	.owner		=	THIS_MODULE,
 };
+EXPORT_SYMBOL(pfifo_fast_ops);
 
 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 			  struct Qdisc_ops *ops)
@@ -674,6 +675,7 @@ struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
 
 	return oqdisc;
 }
+EXPORT_SYMBOL(dev_graft_qdisc);
 
 static void attach_one_default_qdisc(struct net_device *dev,
 				     struct netdev_queue *dev_queue,
@@ -761,6 +763,7 @@ void dev_activate(struct net_device *dev)
 		dev_watchdog_up(dev);
 	}
 }
+EXPORT_SYMBOL(dev_activate);
 
 static void dev_deactivate_queue(struct net_device *dev,
 				 struct netdev_queue *dev_queue,
@@ -840,6 +843,7 @@ void dev_deactivate(struct net_device *dev)
 	list_add(&dev->unreg_list, &single);
 	dev_deactivate_many(&single);
 }
+EXPORT_SYMBOL(dev_deactivate);
 
 static void dev_init_scheduler_queue(struct net_device *dev,
 				     struct netdev_queue *dev_queue,
diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c
new file mode 100644
index 0000000..705bdfa
--- /dev/null
+++ b/net/sched/sch_mqprio.c
@@ -0,0 +1,418 @@
+/*
+ * net/sched/sch_mqprio.c
+ *
+ * Copyright (c) 2010 John Fastabend <john.r.fastabend@intel.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * version 2 as published by the Free Software Foundation.
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/skbuff.h>
+#include <net/netlink.h>
+#include <net/pkt_sched.h>
+#include <net/sch_generic.h>
+
+struct mqprio_sched {
+	struct Qdisc		**qdiscs;
+	int hw_owned;
+};
+
+static void mqprio_destroy(struct Qdisc *sch)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	struct mqprio_sched *priv = qdisc_priv(sch);
+	unsigned int ntx;
+
+	if (!priv->qdiscs)
+		return;
+
+	for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++)
+		qdisc_destroy(priv->qdiscs[ntx]);
+
+	if (priv->hw_owned && dev->netdev_ops->ndo_setup_tc)
+		dev->netdev_ops->ndo_setup_tc(dev, 0, dev->real_num_tx_queues);
+	else
+		netdev_set_num_tc(dev, 0);
+
+	kfree(priv->qdiscs);
+}
+
+static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt)
+{
+	int i, j;
+
+	/* Verify num_tc is not out of max range */
+	if (qopt->num_tc > TC_MAX_QUEUE)
+		return -EINVAL;
+
+	/* Verify priority mapping uses valid tcs */
+	for (i = 0; i < TC_BITMASK + 1; i++) {
+		if (qopt->prio_tc_map[i] >= qopt->num_tc)
+			return -EINVAL;
+	}
+
+	/* net_device does not support requested operation */
+	if (qopt->hw && !dev->netdev_ops->ndo_setup_tc)
+		return -EINVAL;
+
+	/* if hw owned qcount and qoffset are taken from LLD so
+	 * no reason to verify them here
+	 */
+	if (qopt->hw)
+		return 0;
+
+	for (i = 0; i < qopt->num_tc; i++) {
+		unsigned int last = qopt->offset[i] + qopt->count[i];
+
+		/* Verify the queue count is in tx range being equal to the
+		 * real_num_tx_queues indicates the last queue is in use.
+		 */
+		if (qopt->offset[i] >= dev->real_num_tx_queues ||
+		    !qopt->count[i] ||
+		    last > dev->real_num_tx_queues)
+			return -EINVAL;
+
+		/* Verify that the offset and counts do not overlap */
+		for (j = i + 1; j < qopt->num_tc; j++) {
+			if (last > qopt->offset[j])
+				return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+static int mqprio_init(struct Qdisc *sch, struct nlattr *opt)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	struct mqprio_sched *priv = qdisc_priv(sch);
+	struct netdev_queue *dev_queue;
+	struct Qdisc *qdisc;
+	int i, err = -EOPNOTSUPP;
+	struct tc_mqprio_qopt *qopt = NULL;
+
+	BUILD_BUG_ON(TC_MAX_QUEUE != TC_QOPT_MAX_QUEUE);
+	BUILD_BUG_ON(TC_BITMASK != TC_QOPT_BITMASK);
+
+	if (sch->parent != TC_H_ROOT)
+		return -EOPNOTSUPP;
+
+	if (!netif_is_multiqueue(dev))
+		return -EOPNOTSUPP;
+
+	if (nla_len(opt) < sizeof(*qopt))
+		return -EINVAL;
+
+	qopt = nla_data(opt);
+	if (mqprio_parse_opt(dev, qopt))
+		return -EINVAL;
+
+	/* pre-allocate qdisc, attachment can't fail */
+	priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
+			       GFP_KERNEL);
+	if (priv->qdiscs == NULL) {
+		err = -ENOMEM;
+		goto err;
+	}
+
+	for (i = 0; i < dev->num_tx_queues; i++) {
+		dev_queue = netdev_get_tx_queue(dev, i);
+		qdisc = qdisc_create_dflt(dev_queue, &pfifo_fast_ops,
+					  TC_H_MAKE(TC_H_MAJ(sch->handle),
+						    TC_H_MIN(i + 1)));
+		if (qdisc == NULL) {
+			err = -ENOMEM;
+			goto err;
+		}
+		qdisc->flags |= TCQ_F_CAN_BYPASS;
+		priv->qdiscs[i] = qdisc;
+	}
+
+	/* If the mqprio options indicate that hardware should own
+	 * the queue mapping then run ndo_setup_tc otherwise use the
+	 * supplied and verified mapping
+	 */
+	if (qopt->hw) {
+		priv->hw_owned = 1;
+		err = dev->netdev_ops->ndo_setup_tc(dev, qopt->num_tc,
+						    dev->real_num_tx_queues);
+		if (err)
+			goto err;
+	} else {
+		netdev_set_num_tc(dev, qopt->num_tc);
+		for (i = 0; i < qopt->num_tc; i++)
+			netdev_set_tc_queue(dev, i,
+					    qopt->count[i], qopt->offset[i]);
+	}
+
+	/* Always use supplied priority mappings */
+	for (i = 0; i < TC_BITMASK + 1; i++)
+		netdev_set_prio_tc_map(dev, i, qopt->prio_tc_map[i]);
+
+	sch->flags |= TCQ_F_MQROOT;
+	return 0;
+
+err:
+	mqprio_destroy(sch);
+	return err;
+}
+
+static void mqprio_attach(struct Qdisc *sch)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	struct mqprio_sched *priv = qdisc_priv(sch);
+	struct Qdisc *qdisc;
+	unsigned int ntx;
+
+	/* Attach underlying qdisc */
+	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
+		qdisc = priv->qdiscs[ntx];
+		qdisc = dev_graft_qdisc(qdisc->dev_queue, qdisc);
+		if (qdisc)
+			qdisc_destroy(qdisc);
+	}
+	kfree(priv->qdiscs);
+	priv->qdiscs = NULL;
+}
+
+static struct netdev_queue *mqprio_queue_get(struct Qdisc *sch,
+					     unsigned long cl)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	unsigned long ntx = cl - 1 - netdev_get_num_tc(dev);
+
+	if (ntx >= dev->num_tx_queues)
+		return NULL;
+	return netdev_get_tx_queue(dev, ntx);
+}
+
+static int mqprio_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
+		    struct Qdisc **old)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
+
+	if (!dev_queue)
+		return -EINVAL;
+
+	if (dev->flags & IFF_UP)
+		dev_deactivate(dev);
+
+	*old = dev_graft_qdisc(dev_queue, new);
+
+	if (dev->flags & IFF_UP)
+		dev_activate(dev);
+
+	return 0;
+}
+
+static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	struct mqprio_sched *priv = qdisc_priv(sch);
+	unsigned char *b = skb_tail_pointer(skb);
+	struct tc_mqprio_qopt opt;
+	struct Qdisc *qdisc;
+	unsigned int i;
+
+	sch->q.qlen = 0;
+	memset(&sch->bstats, 0, sizeof(sch->bstats));
+	memset(&sch->qstats, 0, sizeof(sch->qstats));
+
+	for (i = 0; i < dev->num_tx_queues; i++) {
+		qdisc = netdev_get_tx_queue(dev, i)->qdisc;
+		spin_lock_bh(qdisc_lock(qdisc));
+		sch->q.qlen		+= qdisc->q.qlen;
+		sch->bstats.bytes	+= qdisc->bstats.bytes;
+		sch->bstats.packets	+= qdisc->bstats.packets;
+		sch->qstats.qlen	+= qdisc->qstats.qlen;
+		sch->qstats.backlog	+= qdisc->qstats.backlog;
+		sch->qstats.drops	+= qdisc->qstats.drops;
+		sch->qstats.requeues	+= qdisc->qstats.requeues;
+		sch->qstats.overlimits	+= qdisc->qstats.overlimits;
+		spin_unlock_bh(qdisc_lock(qdisc));
+	}
+
+	opt.num_tc = netdev_get_num_tc(dev);
+	memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
+	opt.hw = priv->hw_owned;
+
+	for (i = 0; i < netdev_get_num_tc(dev); i++) {
+		opt.count[i] = dev->tc_to_txq[i].count;
+		opt.offset[i] = dev->tc_to_txq[i].offset;
+	}
+
+	NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
+
+	return skb->len;
+nla_put_failure:
+	nlmsg_trim(skb, b);
+	return -1;
+}
+
+static struct Qdisc *mqprio_leaf(struct Qdisc *sch, unsigned long cl)
+{
+	struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
+
+	if (!dev_queue)
+		return NULL;
+
+	return dev_queue->qdisc_sleeping;
+}
+
+static unsigned long mqprio_get(struct Qdisc *sch, u32 classid)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	unsigned int ntx = TC_H_MIN(classid);
+
+	if (ntx > dev->num_tx_queues + netdev_get_num_tc(dev))
+		return 0;
+	return ntx;
+}
+
+static void mqprio_put(struct Qdisc *sch, unsigned long cl)
+{
+}
+
+static int mqprio_dump_class(struct Qdisc *sch, unsigned long cl,
+			 struct sk_buff *skb, struct tcmsg *tcm)
+{
+	struct net_device *dev = qdisc_dev(sch);
+
+	if (cl <= netdev_get_num_tc(dev)) {
+		tcm->tcm_parent = TC_H_ROOT;
+		tcm->tcm_info = 0;
+	} else {
+		int i;
+		struct netdev_queue *dev_queue;
+
+		dev_queue = mqprio_queue_get(sch, cl);
+		tcm->tcm_parent = 0;
+		for (i = 0; i < netdev_get_num_tc(dev); i++) {
+			struct netdev_tc_txq tc = dev->tc_to_txq[i];
+			int q_idx = cl - netdev_get_num_tc(dev);
+
+			if (q_idx > tc.offset &&
+			    q_idx <= tc.offset + tc.count) {
+				tcm->tcm_parent =
+					TC_H_MAKE(TC_H_MAJ(sch->handle),
+						  TC_H_MIN(i + 1));
+				break;
+			}
+		}
+		tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
+	}
+	tcm->tcm_handle |= TC_H_MIN(cl);
+	return 0;
+}
+
+static int mqprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
+			       struct gnet_dump *d)
+{
+	struct net_device *dev = qdisc_dev(sch);
+
+	if (cl <= netdev_get_num_tc(dev)) {
+		int i;
+		struct Qdisc *qdisc;
+		struct gnet_stats_queue qstats = {0};
+		struct gnet_stats_basic_packed bstats = {0};
+		struct netdev_tc_txq tc = dev->tc_to_txq[cl - 1];
+
+		/* Drop lock here it will be reclaimed before touching
+		 * statistics this is required because the d->lock we
+		 * hold here is the look on dev_queue->qdisc_sleeping
+		 * also acquired below.
+		 */
+		spin_unlock_bh(d->lock);
+
+		for (i = tc.offset; i < tc.offset + tc.count; i++) {
+			qdisc = netdev_get_tx_queue(dev, i)->qdisc;
+			spin_lock_bh(qdisc_lock(qdisc));
+			bstats.bytes      += qdisc->bstats.bytes;
+			bstats.packets    += qdisc->bstats.packets;
+			qstats.qlen       += qdisc->qstats.qlen;
+			qstats.backlog    += qdisc->qstats.backlog;
+			qstats.drops      += qdisc->qstats.drops;
+			qstats.requeues   += qdisc->qstats.requeues;
+			qstats.overlimits += qdisc->qstats.overlimits;
+			spin_unlock_bh(qdisc_lock(qdisc));
+		}
+		/* Reclaim root sleeping lock before completing stats */
+		spin_lock_bh(d->lock);
+		if (gnet_stats_copy_basic(d, &bstats) < 0 ||
+		    gnet_stats_copy_queue(d, &qstats) < 0)
+			return -1;
+	} else {
+		struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
+
+		sch = dev_queue->qdisc_sleeping;
+		sch->qstats.qlen = sch->q.qlen;
+		if (gnet_stats_copy_basic(d, &sch->bstats) < 0 ||
+		    gnet_stats_copy_queue(d, &sch->qstats) < 0)
+			return -1;
+	}
+	return 0;
+}
+
+static void mqprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
+{
+	struct net_device *dev = qdisc_dev(sch);
+	unsigned long ntx;
+
+	if (arg->stop)
+		return;
+
+	/* Walk hierarchy with a virtual class per tc */
+	arg->count = arg->skip;
+	for (ntx = arg->skip;
+	     ntx < dev->num_tx_queues + netdev_get_num_tc(dev);
+	     ntx++) {
+		if (arg->fn(sch, ntx + 1, arg) < 0) {
+			arg->stop = 1;
+			break;
+		}
+		arg->count++;
+	}
+}
+
+static const struct Qdisc_class_ops mqprio_class_ops = {
+	.graft		= mqprio_graft,
+	.leaf		= mqprio_leaf,
+	.get		= mqprio_get,
+	.put		= mqprio_put,
+	.walk		= mqprio_walk,
+	.dump		= mqprio_dump_class,
+	.dump_stats	= mqprio_dump_class_stats,
+};
+
+struct Qdisc_ops mqprio_qdisc_ops __read_mostly = {
+	.cl_ops		= &mqprio_class_ops,
+	.id		= "mqprio",
+	.priv_size	= sizeof(struct mqprio_sched),
+	.init		= mqprio_init,
+	.destroy	= mqprio_destroy,
+	.attach		= mqprio_attach,
+	.dump		= mqprio_dump,
+	.owner		= THIS_MODULE,
+};
+
+static int __init mqprio_module_init(void)
+{
+	return register_qdisc(&mqprio_qdisc_ops);
+}
+
+static void __exit mqprio_module_exit(void)
+{
+	unregister_qdisc(&mqprio_qdisc_ops);
+}
+
+module_init(mqprio_module_init);
+module_exit(mqprio_module_exit);
+
+MODULE_LICENSE("GPL");

Re: [net-next-2.6 PATCH v7 2/2] net_sched: implement a root container qdisc sch_mqprio

[Jarek Poplawski]

On Fri, Jan 07, 2011 at 02:45:49PM -0800, John Fastabend wrote:
> This implements a mqprio queueing discipline that by default creates
> a pfifo_fast qdisc per tx queue and provides the needed configuration
> interface.
> 
> Using the mqprio qdisc the number of tcs currently in use along
> with the range of queues alloted to each class can be configured. By
> default skbs are mapped to traffic classes using the skb priority.
> This mapping is configurable.
> 
> Configurable parameters,
> 
> struct tc_mqprio_qopt {
>         __u8    num_tc;
>         __u8    prio_tc_map[TC_BITMASK + 1];
>         __u8    hw;
>         __u16   count[TC_MAX_QUEUE];
>         __u16   offset[TC_MAX_QUEUE];
> };
> 
> Here the count/offset pairing give the queue alignment and the
> prio_tc_map gives the mapping from skb->priority to tc.
> 
> The hw bit determines if the hardware should configure the count
> and offset values. If the hardware bit is set then the operation
> will fail if the hardware does not implement the ndo_setup_tc
> operation. This is to avoid undetermined states where the hardware
> may or may not control the queue mapping. Also minimal bounds
> checking is done on the count/offset to verify a queue does not
> exceed num_tx_queues and that queue ranges do not overlap. Otherwise
> it is left to user policy or hardware configuration to create
> useful mappings.
> 
> It is expected that hardware QOS schemes can be implemented by
> creating appropriate mappings of queues in ndo_tc_setup().
> 
> One expected use case is drivers will use the ndo_setup_tc to map
> queue ranges onto 802.1Q traffic classes. This provides a generic
> mechanism to map network traffic onto these traffic classes and
> removes the need for lower layer drivers to know specifics about
> traffic types.
> 
> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>

Acked-by: Jarek Poplawski <jarkao2@gmail.com>

Re: [net-next-2.6 PATCH v7 1/2] net: implement mechanism for HW based QOS

[Ben Hutchings]

I'm actually having a go at implementing this, as a result of which I
have some more questions:

On Fri, 2011-01-07 at 14:45 -0800, John Fastabend wrote:
[...]
> With the mechanism in this patch users can set skb priority using
> expected methods ie setsockopt() or the stack can set the priority
> directly. Then the skb will be steered to the correct tx queues
> aligned with hardware QOS traffic classes. In the normal case with
> a single traffic class and all queues in this class everything
> works as is until the LLD enables multiple tcs.
> 
> To steer the skb we mask out the lower 4 bits of the priority
> and allow the hardware to configure upto 15 distinct classes
> of traffic. This is expected to be sufficient for most applications
> at any rate it is more then the 8021Q spec designates and is
> equal to the number of prio bands currently implemented in
> the default qdisc.

What is the meaning of a class number?  Is it simply higher number =>
higher priority?  If there are exactly 8 classes, can they be assumed to
match the 802.1q priority classes?

> This in conjunction with a userspace application such as
> lldpad can be used to implement 8021Q transmission selection
> algorithms one of these algorithms being the extended transmission
> selection algorithm currently being used for DCB.

Should an Ethernet driver/hardware insert a 802.1q priority tag if it
implements this, or should that be left to higher levels?

> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
> ---
> 
>  include/linux/netdevice.h |   65 +++++++++++++++++++++++++++++++++++++++++++++
>  net/core/dev.c            |   61 ++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 125 insertions(+), 1 deletions(-)
> 
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index 0f6b1c9..b1dbbed 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
[...]
> @@ -756,6 +764,7 @@ struct xps_dev_maps {
>   * int (*ndo_set_vf_port)(struct net_device *dev, int vf,
>   *			  struct nlattr *port[]);
>   * int (*ndo_get_vf_port)(struct net_device *dev, int vf, struct sk_buff *skb);
> + * void (*ndo_setup_tc)(struct net_device *dev, u8 tc, unsigned int txq)
[...]

This is not documentation, it is just repetition!

Please specify what the parameters mean.  In particular, what is the
purpose of the txq parameter; can it be different from
dev->real_num_tx_queues?

Is this operation allowed to change the number of TX queues?  I was
looking at scaling the number of queues according to the number of
classes.

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: [net-next-2.6 PATCH v7 1/2] net: implement mechanism for HW based QOS

[John Fastabend]

On 1/13/2011 2:30 PM, Ben Hutchings wrote:
> I'm actually having a go at implementing this, as a result of which I
> have some more questions:
> 

Nice.

> On Fri, 2011-01-07 at 14:45 -0800, John Fastabend wrote:
> [...]
>> With the mechanism in this patch users can set skb priority using
>> expected methods ie setsockopt() or the stack can set the priority
>> directly. Then the skb will be steered to the correct tx queues
>> aligned with hardware QOS traffic classes. In the normal case with
>> a single traffic class and all queues in this class everything
>> works as is until the LLD enables multiple tcs.
>>
>> To steer the skb we mask out the lower 4 bits of the priority
>> and allow the hardware to configure upto 15 distinct classes
>> of traffic. This is expected to be sufficient for most applications
>> at any rate it is more then the 8021Q spec designates and is
>> equal to the number of prio bands currently implemented in
>> the default qdisc.
> 
> What is the meaning of a class number?  Is it simply higher number =>
> higher priority?  If there are exactly 8 classes, can they be assumed to
> match the 802.1q priority classes?

The exact meaning will depend on how the hardware is configured. With DCB
configured the class number is going to have a different meaning then when
DCB is disabled. Paring any specific configuration using 'strict priority'
should be expected. This would simply be higher number => higher priority.

Merely having 8 classes should not be assumed to match the 802.1q priority
classes. As soon as you say 802.1q priority classes I expect to see 802.1q
priority tagged frames. This patch doesn't enforce any sort of tagging.
However if you set up 8 traffic classes and built a qos egress map with
'ip link' that matched then this would be fair to say. I intentionally
tried to leave this up to user space policy to configure no reason to try
and enforce configurations IMHO. Notice you could go wrong with this and
create configurations that probably make little sense but again I think this
is a user space problem and at least in the DCB case we can have 'lldpad'
manage this.

By the way defaulting to 'strict priority' is the 802.1Q spec suggestion
for bridges that do not support the credit-based shaper algorithm. And the
default mapping for 8 traffic classes is priorities map 1:1 with classes
so this all lines up to get what you assumed.

> 
>> This in conjunction with a userspace application such as
>> lldpad can be used to implement 8021Q transmission selection
>> algorithms one of these algorithms being the extended transmission
>> selection algorithm currently being used for DCB.
> 
> Should an Ethernet driver/hardware insert a 802.1q priority tag if it
> implements this, or should that be left to higher levels?

I am not so sure about this. I tend to like using the mechanisms already
in place, the egress qos mapping, to add 802.1q priority tag. But it looks
like some drivers add the tags explicitly. And also there appears to be
some expectation that the 802.1q priority tag is added to all packets even
those sent on non-vlan devices at least in the DCB case.

> 
>> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
>> ---
>>
>>  include/linux/netdevice.h |   65 +++++++++++++++++++++++++++++++++++++++++++++
>>  net/core/dev.c            |   61 ++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 125 insertions(+), 1 deletions(-)
>>
>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
>> index 0f6b1c9..b1dbbed 100644
>> --- a/include/linux/netdevice.h
>> +++ b/include/linux/netdevice.h
> [...]
>> @@ -756,6 +764,7 @@ struct xps_dev_maps {
>>   * int (*ndo_set_vf_port)(struct net_device *dev, int vf,
>>   *			  struct nlattr *port[]);
>>   * int (*ndo_get_vf_port)(struct net_device *dev, int vf, struct sk_buff *skb);
>> + * void (*ndo_setup_tc)(struct net_device *dev, u8 tc, unsigned int txq)
> [...]
> 
> This is not documentation, it is just repetition!
> 

Point taken. It does seem to be the trend though.

> Please specify what the parameters mean.  In particular, what is the
> purpose of the txq parameter; can it be different from
> dev->real_num_tx_queues?

'tc' is the number of traffic classes to enable and 'txq' is the number
of queues in use. This should be dev->real_num_tx_queues except when
called from netif_set_real_num_tx_queues where the new number of queues
is used. This is done to prevent steering skb's to rings that may no
longer exist.

> 
> Is this operation allowed to change the number of TX queues?  I was
> looking at scaling the number of queues according to the number of
> classes.


As is no. Changing the number of TX queues can't be done cleanly because
calling netif_set_real_num_tx_queues() calls ndo_tc_setup(). I was
expecting other netlink messages would change the tx queue numbers.

However I think I like this better... In order to make this possible
I can remove the ndo_tc_setup() call from netif_set_real_num_tc_queues().
Then the 'txq' param can be removed making this call a bit neater as well.
The downside of this is drivers will need to ensure that calling
netif_set_real_num_tx_queues() will not break the mapping otherwise
the mapping gets invalidated. I'll have to see but this might be OK.

Thanks,
John.


> 
> Ben.
>