[RFC net-next v2 0/8] indirect tc block cb registration

[RFC net-next v2 0/8] indirect tc block cb registration

[John Hurley]

This patchset introduces an alternative to egdev offload by allowing a
driver to register for block updates when an external device (e.g. tunnel
netdev) is bound to a TC block. Drivers can track new netdevs or register
to existing ones to receive information on such events. Based on this,
they may register for block offload rules using already existing
functions.

The patchset also implements this new indirect block registration in the
NFP driver to allow the offloading of tunnel rules. The use of egdev
offload (which is currently only used for tunnel offload) is subsequently
removed.

John Hurley (8):
  net: sched: register callbacks for indirect tc block binds
  net: add netif_is_geneve()
  nfp: flower: include geneve as supported offload tunnel type
  nfp: flower: allow non repr netdev offload
  nfp: flower: add infastructure for indirect TC block register
  nfp: flower: offload tunnel decap rules via indirect TC blocks
  nfp: flower: remove TC egdev offloads
  nfp: flower: remove unnecessary code in flow lookup

 drivers/net/ethernet/netronome/nfp/flower/action.c |  29 +-
 drivers/net/ethernet/netronome/nfp/flower/cmsg.h   |  13 +
 drivers/net/ethernet/netronome/nfp/flower/main.c   |  25 +-
 drivers/net/ethernet/netronome/nfp/flower/main.h   |  17 +-
 drivers/net/ethernet/netronome/nfp/flower/match.c  |  38 +--
 .../net/ethernet/netronome/nfp/flower/metadata.c   |  12 +-
 .../net/ethernet/netronome/nfp/flower/offload.c    | 246 +++++++++++------
 .../ethernet/netronome/nfp/flower/tunnel_conf.c    |  11 +-
 include/net/geneve.h                               |   6 +
 include/net/pkt_cls.h                              |  56 ++++
 include/net/sch_generic.h                          |   3 +
 net/sched/cls_api.c                                | 299 ++++++++++++++++++++-
 12 files changed, 609 insertions(+), 146 deletions(-)

-- 
2.7.4

[RFC net-next v2 1/8] net: sched: register callbacks for indirect tc block binds

[John Hurley]

Currently drivers can register to receive TC block bind/unbind callbacks
by implementing the setup_tc ndo in any of their given netdevs. However,
drivers may also be interested in binds to higher level devices (e.g.
tunnel drivers) to potentially offload filters applied to them.

Introduce indirect block devs which allows drivers to register callbacks
for block binds on other devices. The calling driver is expected to
reference an 'owner' struct that it will pass to all block registrations.
This is used to track the callbacks from a given driver and free them if
the driver is removed while the upper level device is still active.
Freeing a callback will also trigger an unbind event (if necessary) to
direct the driver to remove any offloaded rules and unreg any block filter
callbacks.

Allow registering an indirect block dev callback for a device that is
already bound to a block. In this case (if it is an ingress block),
register and also trigger the callback meaning that any already installed
rules can be replayed to the calling driver.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 include/net/pkt_cls.h     |  56 +++++++++
 include/net/sch_generic.h |   3 +
 net/sched/cls_api.c       | 299 +++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 357 insertions(+), 1 deletion(-)

diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index 72ffb31..1b47837 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h
@@ -37,6 +37,7 @@ struct tcf_block_ext_info {
 };
 
 struct tcf_block_cb;
+struct tcf_indr_block_owner;
 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func);
 
 #ifdef CONFIG_NET_CLS
@@ -81,6 +82,20 @@ void __tcf_block_cb_unregister(struct tcf_block *block,
 			       struct tcf_block_cb *block_cb);
 void tcf_block_cb_unregister(struct tcf_block *block,
 			     tc_setup_cb_t *cb, void *cb_ident);
+int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+				tc_indr_block_bind_cb_t *cb, void *cb_ident,
+				struct tcf_indr_block_owner *owner);
+int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+			      tc_indr_block_bind_cb_t *cb, void *cb_ident,
+			      struct tcf_indr_block_owner *owner);
+void __tc_indr_block_cb_unregister(struct net_device *dev,
+				   tc_indr_block_bind_cb_t *cb, void *cb_ident);
+void tc_indr_block_cb_unregister(struct net_device *dev,
+				 tc_indr_block_bind_cb_t *cb,
+				 void *cb_ident);
+
+struct tcf_indr_block_owner *tc_indr_block_owner_create(void);
+void tc_indr_block_owner_clean(struct tcf_indr_block_owner *owner);
 
 int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 		 struct tcf_result *res, bool compat_mode);
@@ -183,6 +198,47 @@ void tcf_block_cb_unregister(struct tcf_block *block,
 {
 }
 
+static inline
+int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+				tc_indr_block_bind_cb_t *cb,
+				void *cb_ident,
+				struct tcf_indr_block_owner *owner)
+{
+	return 0;
+}
+
+static inline
+int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+			      tc_indr_block_bind_cb_t *cb, void *cb_ident,
+			      struct tcf_indr_block_owner *owner)
+{
+	return 0;
+}
+
+static inline
+void __tc_indr_block_cb_unregister(struct net_device *dev,
+				   tc_indr_block_bind_cb_t *cb,
+				   void *cb_ident)
+{
+}
+
+static inline
+void tc_indr_block_cb_unregister(struct net_device *dev,
+				 tc_indr_block_bind_cb_t *cb,
+				 void *cb_ident)
+{
+}
+
+static inline struct tcf_indr_block_owner *tc_indr_block_owner_create(void)
+{
+	/* NULL would mean an error, only CONFIG_NET_CLS can dereference this */
+	return (void *)1;
+}
+
+static inline void tc_indr_block_owner_clean(struct tcf_indr_block_owner *owner)
+{
+}
+
 static inline int tcf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 			       struct tcf_result *res, bool compat_mode)
 {
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 4d73642..8301581 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -24,6 +24,9 @@ struct bpf_flow_keys;
 typedef int tc_setup_cb_t(enum tc_setup_type type,
 			  void *type_data, void *cb_priv);
 
+typedef int tc_indr_block_bind_cb_t(struct net_device *dev, void *cb_priv,
+				    enum tc_setup_type type, void *type_data);
+
 struct qdisc_rate_table {
 	struct tc_ratespec rate;
 	u32		data[256];
diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
index f427a1e..79d97bf 100644
--- a/net/sched/cls_api.c
+++ b/net/sched/cls_api.c
@@ -25,6 +25,7 @@
 #include <linux/kmod.h>
 #include <linux/slab.h>
 #include <linux/idr.h>
+#include <linux/rhashtable.h>
 #include <net/net_namespace.h>
 #include <net/sock.h>
 #include <net/netlink.h>
@@ -365,6 +366,288 @@ static void tcf_chain_flush(struct tcf_chain *chain)
 	}
 }
 
+static struct tcf_block *tc_dev_ingress_block(struct net_device *dev)
+{
+	const struct Qdisc_class_ops *cops;
+	struct Qdisc *qdisc;
+
+	if (!dev_ingress_queue(dev))
+		return NULL;
+
+	qdisc = dev_ingress_queue(dev)->qdisc_sleeping;
+	if (!qdisc)
+		return NULL;
+
+	cops = qdisc->ops->cl_ops;
+	if (!cops)
+		return NULL;
+
+	if (!cops->tcf_block)
+		return NULL;
+
+	return cops->tcf_block(qdisc, TC_H_MIN_INGRESS, NULL);
+}
+
+static struct rhashtable indr_setup_block_ht;
+
+struct tc_indr_block_dev {
+	struct rhash_head ht_node;
+	struct net_device *dev;
+	unsigned int refcnt;
+	struct list_head cb_list;
+	struct tcf_block *block;
+};
+
+struct tc_indr_block_cb {
+	struct tc_indr_block_dev *indr_dev;
+	struct list_head list;
+	void *cb_priv;
+	tc_indr_block_bind_cb_t *cb;
+	void *cb_ident;
+	struct list_head owner_list;
+};
+
+struct tcf_indr_block_owner {
+	struct list_head cb_list;
+};
+
+static const struct rhashtable_params tc_indr_setup_block_ht_params = {
+	.key_offset	= offsetof(struct tc_indr_block_dev, dev),
+	.head_offset	= offsetof(struct tc_indr_block_dev, ht_node),
+	.key_len	= sizeof(struct net_device *),
+};
+
+static struct tc_indr_block_dev *
+tc_indr_block_dev_lookup(struct net_device *dev)
+{
+	return rhashtable_lookup_fast(&indr_setup_block_ht, &dev,
+				      tc_indr_setup_block_ht_params);
+}
+
+static struct tc_indr_block_dev *tc_indr_block_dev_get(struct net_device *dev)
+{
+	struct tc_indr_block_dev *indr_dev;
+
+	indr_dev = tc_indr_block_dev_lookup(dev);
+	if (indr_dev)
+		goto inc_ref;
+
+	indr_dev = kzalloc(sizeof(*indr_dev), GFP_KERNEL);
+	if (!indr_dev)
+		return NULL;
+
+	INIT_LIST_HEAD(&indr_dev->cb_list);
+	indr_dev->dev = dev;
+	indr_dev->block = tc_dev_ingress_block(dev);
+	if (rhashtable_insert_fast(&indr_setup_block_ht, &indr_dev->ht_node,
+				   tc_indr_setup_block_ht_params)) {
+		kfree(indr_dev);
+		return NULL;
+	}
+
+inc_ref:
+	indr_dev->refcnt++;
+	return indr_dev;
+}
+
+static void tc_indr_block_dev_put(struct tc_indr_block_dev *indr_dev)
+{
+	if (--indr_dev->refcnt)
+		return;
+
+	rhashtable_remove_fast(&indr_setup_block_ht, &indr_dev->ht_node,
+			       tc_indr_setup_block_ht_params);
+	kfree(indr_dev);
+}
+
+static struct tc_indr_block_cb *
+tc_indr_block_cb_lookup(struct tc_indr_block_dev *indr_dev,
+			tc_indr_block_bind_cb_t *cb, void *cb_ident)
+{
+	struct tc_indr_block_cb *indr_block_cb;
+
+	list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
+		if (indr_block_cb->cb == cb &&
+		    indr_block_cb->cb_ident == cb_ident)
+			return indr_block_cb;
+	return NULL;
+}
+
+static struct tc_indr_block_cb *
+tc_indr_block_cb_add(struct tc_indr_block_dev *indr_dev,
+		     void *cb_priv, tc_indr_block_bind_cb_t *cb,
+		     void *cb_ident, struct tcf_indr_block_owner *owner)
+{
+	struct tc_indr_block_cb *indr_block_cb;
+
+	indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
+	if (indr_block_cb)
+		return ERR_PTR(-EEXIST);
+
+	indr_block_cb = kzalloc(sizeof(*indr_block_cb), GFP_KERNEL);
+	if (!indr_block_cb)
+		return ERR_PTR(-ENOMEM);
+
+	indr_block_cb->indr_dev = indr_dev;
+	indr_block_cb->cb_priv = cb_priv;
+	indr_block_cb->cb = cb;
+	indr_block_cb->cb_ident = cb_ident;
+	list_add(&indr_block_cb->list, &indr_dev->cb_list);
+	list_add(&indr_block_cb->owner_list, &owner->cb_list);
+
+	return indr_block_cb;
+}
+
+static void tc_indr_block_cb_del(struct tc_indr_block_cb *indr_block_cb)
+{
+	list_del(&indr_block_cb->list);
+	list_del(&indr_block_cb->owner_list);
+	kfree(indr_block_cb);
+}
+
+static int tc_indr_block_ing_cmd(struct tc_indr_block_dev *indr_dev,
+				 struct tc_indr_block_cb *indr_block_cb,
+				 enum tc_block_command command)
+{
+	struct tc_block_offload bo = {
+		.command	= command,
+		.binder_type	= TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS,
+		.block		= indr_dev->block,
+	};
+
+	if (!indr_dev->block)
+		return 0;
+	return indr_block_cb->cb(indr_dev->dev, indr_block_cb->cb_priv,
+				 TC_SETUP_BLOCK, &bo);
+}
+
+int __tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+				tc_indr_block_bind_cb_t *cb, void *cb_ident,
+				struct tcf_indr_block_owner *owner)
+{
+	struct tc_indr_block_cb *indr_block_cb;
+	struct tc_indr_block_dev *indr_dev;
+	int err;
+
+	indr_dev = tc_indr_block_dev_get(dev);
+	if (!indr_dev)
+		return -ENOMEM;
+
+	indr_block_cb = tc_indr_block_cb_add(indr_dev, cb_priv, cb, cb_ident,
+					     owner);
+	err = PTR_ERR_OR_ZERO(indr_block_cb);
+	if (err)
+		goto err_dev_put;
+
+	tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_BIND);
+	return 0;
+
+err_dev_put:
+	tc_indr_block_dev_put(indr_dev);
+	return err;
+}
+EXPORT_SYMBOL_GPL(__tc_indr_block_cb_register);
+
+int tc_indr_block_cb_register(struct net_device *dev, void *cb_priv,
+			      tc_indr_block_bind_cb_t *cb, void *cb_ident,
+			      struct tcf_indr_block_owner *owner)
+{
+	int err;
+
+	rtnl_lock();
+	err = __tc_indr_block_cb_register(dev, cb_priv, cb, cb_ident, owner);
+	rtnl_unlock();
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(tc_indr_block_cb_register);
+
+void __tc_indr_block_cb_unregister(struct net_device *dev,
+				   tc_indr_block_bind_cb_t *cb, void *cb_ident)
+{
+	struct tc_indr_block_cb *indr_block_cb;
+	struct tc_indr_block_dev *indr_dev;
+
+	indr_dev = tc_indr_block_dev_lookup(dev);
+	if (!indr_dev)
+		return;
+
+	indr_block_cb = tc_indr_block_cb_lookup(indr_dev, cb, cb_ident);
+	if (!indr_block_cb)
+		return;
+
+	/* Send unbind message if required to free any block cbs. */
+	tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_UNBIND);
+	tc_indr_block_cb_del(indr_block_cb);
+	tc_indr_block_dev_put(indr_dev);
+}
+EXPORT_SYMBOL_GPL(__tc_indr_block_cb_unregister);
+
+void tc_indr_block_cb_unregister(struct net_device *dev,
+				 tc_indr_block_bind_cb_t *cb, void *cb_ident)
+{
+	rtnl_lock();
+	__tc_indr_block_cb_unregister(dev, cb, cb_ident);
+	rtnl_unlock();
+}
+EXPORT_SYMBOL_GPL(tc_indr_block_cb_unregister);
+
+struct tcf_indr_block_owner *tc_indr_block_owner_create(void)
+{
+	struct tcf_indr_block_owner *owner;
+
+	owner = kzalloc(sizeof(*owner), GFP_KERNEL);
+	if (!owner)
+		return NULL;
+	INIT_LIST_HEAD(&owner->cb_list);
+	return owner;
+}
+EXPORT_SYMBOL_GPL(tc_indr_block_owner_create);
+
+void tc_indr_block_owner_clean(struct tcf_indr_block_owner *owner)
+{
+	struct tc_indr_block_cb *indr_block_cb, *store;
+	struct tc_indr_block_dev *indr_dev;
+
+	rtnl_lock();
+	list_for_each_entry_safe(indr_block_cb, store, &owner->cb_list,
+				 owner_list) {
+		indr_dev = indr_block_cb->indr_dev;
+		tc_indr_block_ing_cmd(indr_dev, indr_block_cb, TC_BLOCK_UNBIND);
+		tc_indr_block_cb_del(indr_block_cb);
+		tc_indr_block_dev_put(indr_dev);
+	}
+	rtnl_unlock();
+
+	kfree(owner);
+}
+EXPORT_SYMBOL_GPL(tc_indr_block_owner_clean);
+
+static void tc_indr_block_call(struct tcf_block *block, struct net_device *dev,
+			       struct tcf_block_ext_info *ei,
+			       enum tc_block_command command,
+			       struct netlink_ext_ack *extack)
+{
+	struct tc_indr_block_cb *indr_block_cb;
+	struct tc_indr_block_dev *indr_dev;
+	struct tc_block_offload bo = {
+		.command	= command,
+		.binder_type	= ei->binder_type,
+		.block		= block,
+		.extack		= extack,
+	};
+
+	indr_dev = tc_indr_block_dev_lookup(dev);
+	if (!indr_dev)
+		return;
+
+	indr_dev->block = command == TC_BLOCK_BIND ? block : NULL;
+
+	list_for_each_entry(indr_block_cb, &indr_dev->cb_list, list)
+		indr_block_cb->cb(dev, indr_block_cb->cb_priv, TC_SETUP_BLOCK,
+				  &bo);
+}
+
 static bool tcf_block_offload_in_use(struct tcf_block *block)
 {
 	return block->offloadcnt;
@@ -406,12 +689,17 @@ static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
 	err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_BIND, extack);
 	if (err == -EOPNOTSUPP)
 		goto no_offload_dev_inc;
-	return err;
+	if (err)
+		return err;
+
+	tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
+	return 0;
 
 no_offload_dev_inc:
 	if (tcf_block_offload_in_use(block))
 		return -EOPNOTSUPP;
 	block->nooffloaddevcnt++;
+	tc_indr_block_call(block, dev, ei, TC_BLOCK_BIND, extack);
 	return 0;
 }
 
@@ -421,6 +709,8 @@ static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
 	struct net_device *dev = q->dev_queue->dev;
 	int err;
 
+	tc_indr_block_call(block, dev, ei, TC_BLOCK_UNBIND, NULL);
+
 	if (!dev->netdev_ops->ndo_setup_tc)
 		goto no_offload_dev_dec;
 	err = tcf_block_offload_cmd(block, dev, ei, TC_BLOCK_UNBIND, NULL);
@@ -2355,6 +2645,11 @@ static int __init tc_filter_init(void)
 	if (err)
 		goto err_register_pernet_subsys;
 
+	err = rhashtable_init(&indr_setup_block_ht,
+			      &tc_indr_setup_block_ht_params);
+	if (err)
+		goto err_rhash_setup_block_ht;
+
 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL, 0);
 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL, 0);
 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
@@ -2366,6 +2661,8 @@ static int __init tc_filter_init(void)
 
 	return 0;
 
+err_rhash_setup_block_ht:
+	unregister_pernet_subsys(&tcf_net_ops);
 err_register_pernet_subsys:
 	destroy_workqueue(tc_filter_wq);
 	return err;
-- 
2.7.4

[RFC net-next v2 2/8] net: add netif_is_geneve()

[John Hurley]

Add a helper function to determine if the type of a netdev is geneve based
on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels
to check the underlying type of the device.

A recent patch added a similar helper to vxlan.h

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 include/net/geneve.h | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/include/net/geneve.h b/include/net/geneve.h
index a7600ed..fc6a7e0 100644
--- a/include/net/geneve.h
+++ b/include/net/geneve.h
@@ -60,6 +60,12 @@ struct genevehdr {
 	struct geneve_opt options[];
 };
 
+static inline bool netif_is_geneve(const struct net_device *dev)
+{
+	return dev->rtnl_link_ops &&
+	       !strcmp(dev->rtnl_link_ops->kind, "geneve");
+}
+
 #ifdef CONFIG_INET
 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
 					u8 name_assign_type, u16 dst_port);
-- 
2.7.4

[RFC net-next v2 3/8] nfp: flower: include geneve as supported offload tunnel type

[John Hurley]

Offload of geneve decap rules is supported in NFP. Include geneve in the
check for supported types.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c b/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
index 8e5bec0..170f314 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
@@ -190,6 +190,8 @@ static bool nfp_tun_is_netdev_to_offload(struct net_device *netdev)
 		return true;
 	if (netif_is_vxlan(netdev))
 		return true;
+	if (netif_is_geneve(netdev))
+		return true;
 
 	return false;
 }
-- 
2.7.4

[RFC net-next v2 4/8] nfp: flower: allow non repr netdev offload

[John Hurley]

Previously the offload functions in NFP assumed that the ingress (or
egress) netdev passed to them was an nfp repr.

Modify the driver to permit the passing of non repr netdevs as the ingress
device for an offload rule candidate. This may include devices such as
tunnels. The driver should then base its offload decision on a combination
of ingress device and egress port for a rule.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/action.c | 14 ++++----
 drivers/net/ethernet/netronome/nfp/flower/main.h   |  3 +-
 drivers/net/ethernet/netronome/nfp/flower/match.c  | 38 ++++++++++++----------
 .../net/ethernet/netronome/nfp/flower/offload.c    | 33 +++++++++++--------
 4 files changed, 49 insertions(+), 39 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/action.c b/drivers/net/ethernet/netronome/nfp/flower/action.c
index 244dc26..04349c7 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/action.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/action.c
@@ -151,11 +151,12 @@ nfp_fl_output(struct nfp_app *app, struct nfp_fl_output *output,
 		/* Set action output parameters. */
 		output->flags = cpu_to_be16(tmp_flags);
 
-		/* Only offload if egress ports are on the same device as the
-		 * ingress port.
-		 */
-		if (!switchdev_port_same_parent_id(in_dev, out_dev))
-			return -EOPNOTSUPP;
+		if (nfp_netdev_is_nfp_repr(in_dev)) {
+			/* Confirm ingress and egress are on same device. */
+			if (!switchdev_port_same_parent_id(in_dev, out_dev))
+				return -EOPNOTSUPP;
+		}
+
 		if (!nfp_netdev_is_nfp_repr(out_dev))
 			return -EOPNOTSUPP;
 
@@ -728,9 +729,8 @@ nfp_flower_loop_action(struct nfp_app *app, const struct tc_action *a,
 		*a_len += sizeof(struct nfp_fl_push_vlan);
 	} else if (is_tcf_tunnel_set(a)) {
 		struct ip_tunnel_info *ip_tun = tcf_tunnel_info(a);
-		struct nfp_repr *repr = netdev_priv(netdev);
 
-		*tun_type = nfp_fl_get_tun_from_act_l4_port(repr->app, a);
+		*tun_type = nfp_fl_get_tun_from_act_l4_port(app, a);
 		if (*tun_type == NFP_FL_TUNNEL_NONE)
 			return -EOPNOTSUPP;
 
diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.h b/drivers/net/ethernet/netronome/nfp/flower/main.h
index 90045ba..a91ac52 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.h
@@ -226,7 +226,8 @@ void nfp_flower_metadata_cleanup(struct nfp_app *app);
 
 int nfp_flower_setup_tc(struct nfp_app *app, struct net_device *netdev,
 			enum tc_setup_type type, void *type_data);
-int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
+int nfp_flower_compile_flow_match(struct nfp_app *app,
+				  struct tc_cls_flower_offload *flow,
 				  struct nfp_fl_key_ls *key_ls,
 				  struct net_device *netdev,
 				  struct nfp_fl_payload *nfp_flow,
diff --git a/drivers/net/ethernet/netronome/nfp/flower/match.c b/drivers/net/ethernet/netronome/nfp/flower/match.c
index e54fb60..cdf7559 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/match.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/match.c
@@ -52,10 +52,13 @@ nfp_flower_compile_port(struct nfp_flower_in_port *frame, u32 cmsg_port,
 		return 0;
 	}
 
-	if (tun_type)
+	if (tun_type) {
 		frame->in_port = cpu_to_be32(NFP_FL_PORT_TYPE_TUN | tun_type);
-	else
+	} else {
+		if (!cmsg_port)
+			return -EOPNOTSUPP;
 		frame->in_port = cpu_to_be32(cmsg_port);
+	}
 
 	return 0;
 }
@@ -289,17 +292,21 @@ nfp_flower_compile_ipv4_udp_tun(struct nfp_flower_ipv4_udp_tun *frame,
 	}
 }
 
-int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
+int nfp_flower_compile_flow_match(struct nfp_app *app,
+				  struct tc_cls_flower_offload *flow,
 				  struct nfp_fl_key_ls *key_ls,
 				  struct net_device *netdev,
 				  struct nfp_fl_payload *nfp_flow,
 				  enum nfp_flower_tun_type tun_type)
 {
-	struct nfp_repr *netdev_repr;
+	u32 cmsg_port = 0;
 	int err;
 	u8 *ext;
 	u8 *msk;
 
+	if (nfp_netdev_is_nfp_repr(netdev))
+		cmsg_port = nfp_repr_get_port_id(netdev);
+
 	memset(nfp_flow->unmasked_data, 0, key_ls->key_size);
 	memset(nfp_flow->mask_data, 0, key_ls->key_size);
 
@@ -327,15 +334,13 @@ int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
 
 	/* Populate Exact Port data. */
 	err = nfp_flower_compile_port((struct nfp_flower_in_port *)ext,
-				      nfp_repr_get_port_id(netdev),
-				      false, tun_type);
+				      cmsg_port, false, tun_type);
 	if (err)
 		return err;
 
 	/* Populate Mask Port Data. */
 	err = nfp_flower_compile_port((struct nfp_flower_in_port *)msk,
-				      nfp_repr_get_port_id(netdev),
-				      true, tun_type);
+				      cmsg_port, true, tun_type);
 	if (err)
 		return err;
 
@@ -399,16 +404,13 @@ int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
 		msk += sizeof(struct nfp_flower_ipv4_udp_tun);
 
 		/* Configure tunnel end point MAC. */
-		if (nfp_netdev_is_nfp_repr(netdev)) {
-			netdev_repr = netdev_priv(netdev);
-			nfp_tunnel_write_macs(netdev_repr->app);
-
-			/* Store the tunnel destination in the rule data.
-			 * This must be present and be an exact match.
-			 */
-			nfp_flow->nfp_tun_ipv4_addr = tun_dst;
-			nfp_tunnel_add_ipv4_off(netdev_repr->app, tun_dst);
-		}
+		nfp_tunnel_write_macs(app);
+
+		/* Store the tunnel destination in the rule data.
+		 * This must be present and be an exact match.
+		 */
+		nfp_flow->nfp_tun_ipv4_addr = tun_dst;
+		nfp_tunnel_add_ipv4_off(app, tun_dst);
 
 		if (key_ls->key_layer_two & NFP_FLOWER_LAYER2_GENEVE_OP) {
 			err = nfp_flower_compile_geneve_opt(ext, flow, false);
diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index 29c9542..2c32edf 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -56,11 +56,10 @@
 	 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS))
 
 static int
-nfp_flower_xmit_flow(struct net_device *netdev,
-		     struct nfp_fl_payload *nfp_flow, u8 mtype)
+nfp_flower_xmit_flow(struct nfp_app *app, struct nfp_fl_payload *nfp_flow,
+		     u8 mtype)
 {
 	u32 meta_len, key_len, mask_len, act_len, tot_len;
-	struct nfp_repr *priv = netdev_priv(netdev);
 	struct sk_buff *skb;
 	unsigned char *msg;
 
@@ -78,7 +77,7 @@ nfp_flower_xmit_flow(struct net_device *netdev,
 	nfp_flow->meta.mask_len >>= NFP_FL_LW_SIZ;
 	nfp_flow->meta.act_len >>= NFP_FL_LW_SIZ;
 
-	skb = nfp_flower_cmsg_alloc(priv->app, tot_len, mtype, GFP_KERNEL);
+	skb = nfp_flower_cmsg_alloc(app, tot_len, mtype, GFP_KERNEL);
 	if (!skb)
 		return -ENOMEM;
 
@@ -96,7 +95,7 @@ nfp_flower_xmit_flow(struct net_device *netdev,
 	nfp_flow->meta.mask_len <<= NFP_FL_LW_SIZ;
 	nfp_flow->meta.act_len <<= NFP_FL_LW_SIZ;
 
-	nfp_ctrl_tx(priv->app->ctrl, skb);
+	nfp_ctrl_tx(app->ctrl, skb);
 
 	return 0;
 }
@@ -427,13 +426,16 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 		       struct tc_cls_flower_offload *flow, bool egress)
 {
 	enum nfp_flower_tun_type tun_type = NFP_FL_TUNNEL_NONE;
-	struct nfp_port *port = nfp_port_from_netdev(netdev);
 	struct nfp_flower_priv *priv = app->priv;
 	struct nfp_fl_payload *flow_pay;
 	struct nfp_fl_key_ls *key_layer;
+	struct nfp_port *port = NULL;
 	struct net_device *ingr_dev;
 	int err;
 
+	if (nfp_netdev_is_nfp_repr(netdev))
+		port = nfp_port_from_netdev(netdev);
+
 	ingr_dev = egress ? NULL : netdev;
 	flow_pay = nfp_flower_search_fl_table(app, flow->cookie, ingr_dev,
 					      NFP_FL_STATS_CTX_DONT_CARE);
@@ -462,8 +464,8 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 
 	flow_pay->ingress_dev = egress ? NULL : netdev;
 
-	err = nfp_flower_compile_flow_match(flow, key_layer, netdev, flow_pay,
-					    tun_type);
+	err = nfp_flower_compile_flow_match(app, flow, key_layer, netdev,
+					    flow_pay, tun_type);
 	if (err)
 		goto err_destroy_flow;
 
@@ -476,7 +478,7 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 	if (err)
 		goto err_destroy_flow;
 
-	err = nfp_flower_xmit_flow(netdev, flow_pay,
+	err = nfp_flower_xmit_flow(app, flow_pay,
 				   NFP_FLOWER_CMSG_TYPE_FLOW_ADD);
 	if (err)
 		goto err_destroy_flow;
@@ -487,7 +489,8 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 	if (err)
 		goto err_destroy_flow;
 
-	port->tc_offload_cnt++;
+	if (port)
+		port->tc_offload_cnt++;
 
 	/* Deallocate flow payload when flower rule has been destroyed. */
 	kfree(key_layer);
@@ -520,12 +523,15 @@ static int
 nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
 		       struct tc_cls_flower_offload *flow, bool egress)
 {
-	struct nfp_port *port = nfp_port_from_netdev(netdev);
 	struct nfp_flower_priv *priv = app->priv;
 	struct nfp_fl_payload *nfp_flow;
+	struct nfp_port *port = NULL;
 	struct net_device *ingr_dev;
 	int err;
 
+	if (nfp_netdev_is_nfp_repr(netdev))
+		port = nfp_port_from_netdev(netdev);
+
 	ingr_dev = egress ? NULL : netdev;
 	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, ingr_dev,
 					      NFP_FL_STATS_CTX_DONT_CARE);
@@ -539,13 +545,14 @@ nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
 	if (nfp_flow->nfp_tun_ipv4_addr)
 		nfp_tunnel_del_ipv4_off(app, nfp_flow->nfp_tun_ipv4_addr);
 
-	err = nfp_flower_xmit_flow(netdev, nfp_flow,
+	err = nfp_flower_xmit_flow(app, nfp_flow,
 				   NFP_FLOWER_CMSG_TYPE_FLOW_DEL);
 	if (err)
 		goto err_free_flow;
 
 err_free_flow:
-	port->tc_offload_cnt--;
+	if (port)
+		port->tc_offload_cnt--;
 	kfree(nfp_flow->action_data);
 	kfree(nfp_flow->mask_data);
 	kfree(nfp_flow->unmasked_data);
-- 
2.7.4

[RFC net-next v2 5/8] nfp: flower: add infastructure for indirect TC block register

[John Hurley]

Add support structures and functions that can be used by NFP to impliment
the indirect block register functionality of TC.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/main.c   |  13 +++
 drivers/net/ethernet/netronome/nfp/flower/main.h   |   8 ++
 .../net/ethernet/netronome/nfp/flower/offload.c    | 129 +++++++++++++++++++++
 3 files changed, 150 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.c b/drivers/net/ethernet/netronome/nfp/flower/main.c
index 3a54728..518006c 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.c
@@ -568,8 +568,18 @@ static int nfp_flower_init(struct nfp_app *app)
 		goto err_cleanup_metadata;
 	}
 
+	INIT_LIST_HEAD(&app_priv->indr_block_cb_priv);
+	app_priv->indr_block_owner = tc_indr_block_owner_create();
+	if (!app_priv->indr_block_owner) {
+		err = -ENOMEM;
+		goto err_lag_clean;
+	}
+
 	return 0;
 
+err_lag_clean:
+	if (app_priv->flower_ext_feats & NFP_FL_FEATS_LAG)
+		nfp_flower_lag_cleanup(&app_priv->nfp_lag);
 err_cleanup_metadata:
 	nfp_flower_metadata_cleanup(app);
 err_free_app_priv:
@@ -588,6 +598,8 @@ static void nfp_flower_clean(struct nfp_app *app)
 	if (app_priv->flower_ext_feats & NFP_FL_FEATS_LAG)
 		nfp_flower_lag_cleanup(&app_priv->nfp_lag);
 
+	nfp_flower_clean_indr_block_priv(app);
+
 	nfp_flower_metadata_cleanup(app);
 	vfree(app->priv);
 	app->priv = NULL;
@@ -678,6 +690,7 @@ static void nfp_flower_stop(struct nfp_app *app)
 		unregister_netdevice_notifier(&app_priv->nfp_lag.lag_nb);
 
 	nfp_tunnel_config_stop(app);
+	tc_indr_block_owner_clean(app_priv->indr_block_owner);
 }
 
 const struct nfp_app_type app_flower = {
diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.h b/drivers/net/ethernet/netronome/nfp/flower/main.h
index a91ac52..8b4bcf3 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.h
@@ -133,6 +133,8 @@ struct nfp_fl_lag {
  * @reify_wait_queue:	wait queue for repr reify response counting
  * @mtu_conf:		Configuration of repr MTU value
  * @nfp_lag:		Link aggregation data block
+ * @indr_block_cb_priv:	List of priv data passed to indirect block registers
+ * @indr_block_owner:	Struct required for indirect blocks
  */
 struct nfp_flower_priv {
 	struct nfp_app *app;
@@ -166,6 +168,8 @@ struct nfp_flower_priv {
 	wait_queue_head_t reify_wait_queue;
 	struct nfp_mtu_conf mtu_conf;
 	struct nfp_fl_lag nfp_lag;
+	struct list_head indr_block_cb_priv;
+	struct tcf_indr_block_owner *indr_block_owner;
 };
 
 /**
@@ -269,5 +273,9 @@ int nfp_flower_lag_populate_pre_action(struct nfp_app *app,
 				       struct nfp_fl_pre_lag *pre_act);
 int nfp_flower_lag_get_output_id(struct nfp_app *app,
 				 struct net_device *master);
+void
+nfp_flower_register_indr_block(struct nfp_app *app, struct net_device *netdev);
+void nfp_flower_unregister_indr_block(struct net_device *netdev);
+void nfp_flower_clean_indr_block_priv(struct nfp_app *app);
 
 #endif
diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index 2c32edf..f701b2e 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -693,3 +693,132 @@ int nfp_flower_setup_tc(struct nfp_app *app, struct net_device *netdev,
 		return -EOPNOTSUPP;
 	}
 }
+
+struct nfp_flower_indr_block_cb_priv {
+	struct net_device *netdev;
+	struct nfp_app *app;
+	struct list_head list;
+};
+
+static struct nfp_flower_indr_block_cb_priv *
+nfp_flower_indr_block_cb_priv_lookup(struct nfp_app *app,
+				     struct net_device *netdev)
+{
+	struct nfp_flower_indr_block_cb_priv *cb_priv;
+	struct nfp_flower_priv *priv = app->priv;
+
+	/* All callback list access should be protected by RTNL. */
+	ASSERT_RTNL();
+
+	list_for_each_entry(cb_priv, &priv->indr_block_cb_priv, list)
+		if (cb_priv->netdev == netdev)
+			return cb_priv;
+
+	return NULL;
+}
+
+void nfp_flower_clean_indr_block_priv(struct nfp_app *app)
+{
+	struct nfp_flower_indr_block_cb_priv *cb_priv, *temp;
+	struct nfp_flower_priv *priv = app->priv;
+
+	list_for_each_entry_safe(cb_priv, temp, &priv->indr_block_cb_priv, list)
+		kfree(cb_priv);
+}
+
+static int nfp_flower_setup_indr_block_cb(enum tc_setup_type type,
+					  void *type_data, void *cb_priv)
+{
+	struct nfp_flower_indr_block_cb_priv *priv = cb_priv;
+	struct tc_cls_flower_offload *flower = type_data;
+
+	if (flower->common.chain_index)
+		return -EOPNOTSUPP;
+
+	switch (type) {
+	case TC_SETUP_CLSFLOWER:
+		return nfp_flower_repr_offload(priv->app, priv->netdev,
+					       type_data, false);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int
+nfp_flower_setup_indr_tc_block(struct net_device *netdev, struct nfp_app *app,
+			       struct tc_block_offload *f)
+{
+	struct nfp_flower_indr_block_cb_priv *cb_priv;
+	struct nfp_flower_priv *priv = app->priv;
+	int err;
+
+	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
+		return -EOPNOTSUPP;
+
+	switch (f->command) {
+	case TC_BLOCK_BIND:
+		cb_priv = kmalloc(sizeof(*cb_priv), GFP_KERNEL);
+		if (!cb_priv)
+			return -ENOMEM;
+
+		cb_priv->netdev = netdev;
+		cb_priv->app = app;
+		list_add(&cb_priv->list, &priv->indr_block_cb_priv);
+
+		err = tcf_block_cb_register(f->block,
+					    nfp_flower_setup_indr_block_cb,
+					    netdev, cb_priv, f->extack);
+		if (err) {
+			list_del(&cb_priv->list);
+			kfree(cb_priv);
+		}
+
+		return err;
+	case TC_BLOCK_UNBIND:
+		tcf_block_cb_unregister(f->block,
+					nfp_flower_setup_indr_block_cb, netdev);
+		cb_priv = nfp_flower_indr_block_cb_priv_lookup(app, netdev);
+		if (cb_priv) {
+			list_del(&cb_priv->list);
+			kfree(cb_priv);
+		}
+
+		return 0;
+	default:
+		return -EOPNOTSUPP;
+	}
+	return 0;
+}
+
+static int
+nfp_flower_indr_setup_tc_cb(struct net_device *netdev, void *cb_priv,
+			    enum tc_setup_type type, void *type_data)
+{
+	switch (type) {
+	case TC_SETUP_BLOCK:
+		return nfp_flower_setup_indr_tc_block(netdev, cb_priv,
+						      type_data);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+void
+nfp_flower_register_indr_block(struct nfp_app *app, struct net_device *netdev)
+{
+	struct nfp_flower_priv *priv = app->priv;
+	int err;
+
+	err = __tc_indr_block_cb_register(netdev, app,
+					  nfp_flower_indr_setup_tc_cb, netdev,
+					  priv->indr_block_owner);
+	if (err)
+		nfp_flower_cmsg_warn(priv->app,
+				     "Failed to reg indirect block cb for %s\n", netdev->name);
+}
+
+void nfp_flower_unregister_indr_block(struct net_device *netdev)
+{
+	__tc_indr_block_cb_unregister(netdev, nfp_flower_indr_setup_tc_cb,
+				      netdev);
+}
-- 
2.7.4

[RFC net-next v2 6/8] nfp: flower: offload tunnel decap rules via indirect TC blocks

[John Hurley]

Previously, TC block tunnel decap rules were only offloaded when a
callback was triggered through registration of the rules egress device.
This meant that the driver had no access to the ingress netdev and so
could not verify it was the same tunnel type that the rule implied.

Register tunnel devices for indirect TC block offloads in NFP, giving
access to new rules based on the ingress device rather than egress. Use
this to verify the netdev type of VXLAN and Geneve based rules and offload
the rules to HW if applicable.

Tunnel registration is done via a netdev notifier. On notifier
registration, this is triggered for already existing netdevs. This means
that NFP can register for offloads from devices that exist before it is
loaded (filter rules will be replayed from the TC core).

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/action.c      | 15 ---------------
 drivers/net/ethernet/netronome/nfp/flower/cmsg.h        | 13 +++++++++++++
 drivers/net/ethernet/netronome/nfp/flower/offload.c     | 11 +++++++----
 drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c |  9 ++++++++-
 4 files changed, 28 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/action.c b/drivers/net/ethernet/netronome/nfp/flower/action.c
index 04349c7..1260825 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/action.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/action.c
@@ -91,21 +91,6 @@ nfp_fl_pre_lag(struct nfp_app *app, const struct tc_action *action,
 	return act_size;
 }
 
-static bool nfp_fl_netdev_is_tunnel_type(struct net_device *out_dev,
-					 enum nfp_flower_tun_type tun_type)
-{
-	if (!out_dev->rtnl_link_ops)
-		return false;
-
-	if (!strcmp(out_dev->rtnl_link_ops->kind, "vxlan"))
-		return tun_type == NFP_FL_TUNNEL_VXLAN;
-
-	if (!strcmp(out_dev->rtnl_link_ops->kind, "geneve"))
-		return tun_type == NFP_FL_TUNNEL_GENEVE;
-
-	return false;
-}
-
 static int
 nfp_fl_output(struct nfp_app *app, struct nfp_fl_output *output,
 	      const struct tc_action *action, struct nfp_fl_payload *nfp_flow,
diff --git a/drivers/net/ethernet/netronome/nfp/flower/cmsg.h b/drivers/net/ethernet/netronome/nfp/flower/cmsg.h
index 29d673a..06e2888 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/cmsg.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/cmsg.h
@@ -8,6 +8,7 @@
 #include <linux/skbuff.h>
 #include <linux/types.h>
 #include <net/geneve.h>
+#include <net/vxlan.h>
 
 #include "../nfp_app.h"
 #include "../nfpcore/nfp_cpp.h"
@@ -475,6 +476,18 @@ static inline int nfp_flower_cmsg_get_data_len(struct sk_buff *skb)
 	return skb->len - NFP_FLOWER_CMSG_HLEN;
 }
 
+static inline bool
+nfp_fl_netdev_is_tunnel_type(struct net_device *dev,
+			     enum nfp_flower_tun_type tun_type)
+{
+	if (netif_is_vxlan(dev))
+		return tun_type == NFP_FL_TUNNEL_VXLAN;
+	if (netif_is_geneve(dev))
+		return tun_type == NFP_FL_TUNNEL_GENEVE;
+
+	return false;
+}
+
 struct sk_buff *
 nfp_flower_cmsg_mac_repr_start(struct nfp_app *app, unsigned int num_ports);
 void
diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index f701b2e..1dc6044 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -128,6 +128,7 @@ nfp_flower_calc_opt_layer(struct flow_dissector_key_enc_opts *enc_opts,
 
 static int
 nfp_flower_calculate_key_layers(struct nfp_app *app,
+				struct net_device *netdev,
 				struct nfp_fl_key_ls *ret_key_ls,
 				struct tc_cls_flower_offload *flow,
 				bool egress,
@@ -186,8 +187,6 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 			skb_flow_dissector_target(flow->dissector,
 						  FLOW_DISSECTOR_KEY_ENC_CONTROL,
 						  flow->key);
-		if (!egress)
-			return -EOPNOTSUPP;
 
 		if (mask_enc_ctl->addr_type != 0xffff ||
 		    enc_ctl->addr_type != FLOW_DISSECTOR_KEY_IPV4_ADDRS)
@@ -250,6 +249,10 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 		default:
 			return -EOPNOTSUPP;
 		}
+
+		/* Ensure the ingress netdev matches the expected tun type. */
+		if (!nfp_fl_netdev_is_tunnel_type(netdev, *tun_type))
+			return -EOPNOTSUPP;
 	} else if (egress) {
 		/* Reject non tunnel matches offloaded to egress repr. */
 		return -EOPNOTSUPP;
@@ -451,8 +454,8 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 	if (!key_layer)
 		return -ENOMEM;
 
-	err = nfp_flower_calculate_key_layers(app, key_layer, flow, egress,
-					      &tun_type);
+	err = nfp_flower_calculate_key_layers(app, netdev, key_layer, flow,
+					      egress, &tun_type);
 	if (err)
 		goto err_free_key_ls;
 
diff --git a/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c b/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
index 170f314..f138f00 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
@@ -4,7 +4,6 @@
 #include <linux/etherdevice.h>
 #include <linux/inetdevice.h>
 #include <net/netevent.h>
-#include <net/vxlan.h>
 #include <linux/idr.h>
 #include <net/dst_metadata.h>
 #include <net/arp.h>
@@ -670,6 +669,10 @@ static int nfp_tun_mac_event_handler(struct notifier_block *nb,
 		/* If non-nfp netdev then free its offload index. */
 		if (nfp_tun_is_netdev_to_offload(netdev))
 			nfp_tun_del_mac_idx(app, netdev->ifindex);
+
+		if (event == NETDEV_UNREGISTER &&
+		    nfp_tun_is_netdev_to_offload(netdev))
+			nfp_flower_unregister_indr_block(netdev);
 	} else if (event == NETDEV_UP || event == NETDEV_CHANGEADDR ||
 		   event == NETDEV_REGISTER) {
 		app_priv = container_of(nb, struct nfp_flower_priv,
@@ -681,6 +684,10 @@ static int nfp_tun_mac_event_handler(struct notifier_block *nb,
 
 		/* Force a list write to keep NFP up to date. */
 		nfp_tunnel_write_macs(app);
+
+		if (event == NETDEV_REGISTER &&
+		    nfp_tun_is_netdev_to_offload(netdev))
+			nfp_flower_register_indr_block(app, netdev);
 	}
 	return NOTIFY_OK;
 }
-- 
2.7.4

[RFC net-next v2 7/8] nfp: flower: remove TC egdev offloads

[John Hurley]

Previously, only tunnel decap rules required egdev registration for
offload in NFP. These are now supported via indirect TC block callbacks.

Remove the egdev code from NFP.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/main.c   | 12 ----
 drivers/net/ethernet/netronome/nfp/flower/main.h   |  3 -
 .../net/ethernet/netronome/nfp/flower/metadata.c   |  1 +
 .../net/ethernet/netronome/nfp/flower/offload.c    | 79 +++++-----------------
 4 files changed, 17 insertions(+), 78 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.c b/drivers/net/ethernet/netronome/nfp/flower/main.c
index 518006c..45ab4be 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.c
@@ -146,23 +146,12 @@ nfp_flower_repr_netdev_stop(struct nfp_app *app, struct nfp_repr *repr)
 	return nfp_flower_cmsg_portmod(repr, false, repr->netdev->mtu, false);
 }
 
-static int
-nfp_flower_repr_netdev_init(struct nfp_app *app, struct net_device *netdev)
-{
-	return tc_setup_cb_egdev_register(netdev,
-					  nfp_flower_setup_tc_egress_cb,
-					  netdev_priv(netdev));
-}
-
 static void
 nfp_flower_repr_netdev_clean(struct nfp_app *app, struct net_device *netdev)
 {
 	struct nfp_repr *repr = netdev_priv(netdev);
 
 	kfree(repr->app_priv);
-
-	tc_setup_cb_egdev_unregister(netdev, nfp_flower_setup_tc_egress_cb,
-				     netdev_priv(netdev));
 }
 
 static void
@@ -711,7 +700,6 @@ const struct nfp_app_type app_flower = {
 	.vnic_init	= nfp_flower_vnic_init,
 	.vnic_clean	= nfp_flower_vnic_clean,
 
-	.repr_init	= nfp_flower_repr_netdev_init,
 	.repr_preclean	= nfp_flower_repr_netdev_preclean,
 	.repr_clean	= nfp_flower_repr_netdev_clean,
 
diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.h b/drivers/net/ethernet/netronome/nfp/flower/main.h
index 8b4bcf3..d8c8f0d 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.h
@@ -213,7 +213,6 @@ struct nfp_fl_payload {
 	char *unmasked_data;
 	char *mask_data;
 	char *action_data;
-	bool ingress_offload;
 };
 
 extern const struct rhashtable_params nfp_flower_table_params;
@@ -262,8 +261,6 @@ void nfp_tunnel_del_ipv4_off(struct nfp_app *app, __be32 ipv4);
 void nfp_tunnel_add_ipv4_off(struct nfp_app *app, __be32 ipv4);
 void nfp_tunnel_request_route(struct nfp_app *app, struct sk_buff *skb);
 void nfp_tunnel_keep_alive(struct nfp_app *app, struct sk_buff *skb);
-int nfp_flower_setup_tc_egress_cb(enum tc_setup_type type, void *type_data,
-				  void *cb_priv);
 void nfp_flower_lag_init(struct nfp_fl_lag *lag);
 void nfp_flower_lag_cleanup(struct nfp_fl_lag *lag);
 int nfp_flower_lag_reset(struct nfp_fl_lag *lag);
diff --git a/drivers/net/ethernet/netronome/nfp/flower/metadata.c b/drivers/net/ethernet/netronome/nfp/flower/metadata.c
index 48729bf..9b4711c 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/metadata.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/metadata.c
@@ -287,6 +287,7 @@ int nfp_compile_flow_metadata(struct nfp_app *app,
 
 	nfp_flow->meta.host_ctx_id = cpu_to_be32(stats_cxt);
 	nfp_flow->meta.host_cookie = cpu_to_be64(flow->cookie);
+	nfp_flow->ingress_dev = netdev;
 
 	new_mask_id = 0;
 	if (!nfp_check_mask_add(app, nfp_flow->mask_data,
diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index 1dc6044..392d292 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -131,7 +131,6 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 				struct net_device *netdev,
 				struct nfp_fl_key_ls *ret_key_ls,
 				struct tc_cls_flower_offload *flow,
-				bool egress,
 				enum nfp_flower_tun_type *tun_type)
 {
 	struct flow_dissector_key_basic *mask_basic = NULL;
@@ -253,9 +252,6 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 		/* Ensure the ingress netdev matches the expected tun type. */
 		if (!nfp_fl_netdev_is_tunnel_type(netdev, *tun_type))
 			return -EOPNOTSUPP;
-	} else if (egress) {
-		/* Reject non tunnel matches offloaded to egress repr. */
-		return -EOPNOTSUPP;
 	}
 
 	if (dissector_uses_key(flow->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
@@ -376,7 +372,7 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 }
 
 static struct nfp_fl_payload *
-nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer, bool egress)
+nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer)
 {
 	struct nfp_fl_payload *flow_pay;
 
@@ -400,7 +396,6 @@ nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer, bool egress)
 
 	flow_pay->nfp_tun_ipv4_addr = 0;
 	flow_pay->meta.flags = 0;
-	flow_pay->ingress_offload = !egress;
 
 	return flow_pay;
 
@@ -418,7 +413,6 @@ nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer, bool egress)
  * @app:	Pointer to the APP handle
  * @netdev:	netdev structure.
  * @flow:	TC flower classifier offload structure.
- * @egress:	NFP netdev is the egress.
  *
  * Adds a new flow to the repeated hash structure and action payload.
  *
@@ -426,47 +420,33 @@ nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer, bool egress)
  */
 static int
 nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
-		       struct tc_cls_flower_offload *flow, bool egress)
+		       struct tc_cls_flower_offload *flow)
 {
 	enum nfp_flower_tun_type tun_type = NFP_FL_TUNNEL_NONE;
 	struct nfp_flower_priv *priv = app->priv;
 	struct nfp_fl_payload *flow_pay;
 	struct nfp_fl_key_ls *key_layer;
 	struct nfp_port *port = NULL;
-	struct net_device *ingr_dev;
 	int err;
 
 	if (nfp_netdev_is_nfp_repr(netdev))
 		port = nfp_port_from_netdev(netdev);
 
-	ingr_dev = egress ? NULL : netdev;
-	flow_pay = nfp_flower_search_fl_table(app, flow->cookie, ingr_dev,
-					      NFP_FL_STATS_CTX_DONT_CARE);
-	if (flow_pay) {
-		/* Ignore as duplicate if it has been added by different cb. */
-		if (flow_pay->ingress_offload && egress)
-			return 0;
-		else
-			return -EOPNOTSUPP;
-	}
-
 	key_layer = kmalloc(sizeof(*key_layer), GFP_KERNEL);
 	if (!key_layer)
 		return -ENOMEM;
 
 	err = nfp_flower_calculate_key_layers(app, netdev, key_layer, flow,
-					      egress, &tun_type);
+					      &tun_type);
 	if (err)
 		goto err_free_key_ls;
 
-	flow_pay = nfp_flower_allocate_new(key_layer, egress);
+	flow_pay = nfp_flower_allocate_new(key_layer);
 	if (!flow_pay) {
 		err = -ENOMEM;
 		goto err_free_key_ls;
 	}
 
-	flow_pay->ingress_dev = egress ? NULL : netdev;
-
 	err = nfp_flower_compile_flow_match(app, flow, key_layer, netdev,
 					    flow_pay, tun_type);
 	if (err)
@@ -476,8 +456,7 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 	if (err)
 		goto err_destroy_flow;
 
-	err = nfp_compile_flow_metadata(app, flow, flow_pay,
-					flow_pay->ingress_dev);
+	err = nfp_compile_flow_metadata(app, flow, flow_pay, netdev);
 	if (err)
 		goto err_destroy_flow;
 
@@ -515,7 +494,6 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
  * @app:	Pointer to the APP handle
  * @netdev:	netdev structure.
  * @flow:	TC flower classifier offload structure
- * @egress:	Netdev is the egress dev.
  *
  * Removes a flow from the repeated hash structure and clears the
  * action payload.
@@ -524,22 +502,20 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
  */
 static int
 nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
-		       struct tc_cls_flower_offload *flow, bool egress)
+		       struct tc_cls_flower_offload *flow)
 {
 	struct nfp_flower_priv *priv = app->priv;
 	struct nfp_fl_payload *nfp_flow;
 	struct nfp_port *port = NULL;
-	struct net_device *ingr_dev;
 	int err;
 
 	if (nfp_netdev_is_nfp_repr(netdev))
 		port = nfp_port_from_netdev(netdev);
 
-	ingr_dev = egress ? NULL : netdev;
-	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, ingr_dev,
+	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev,
 					      NFP_FL_STATS_CTX_DONT_CARE);
 	if (!nfp_flow)
-		return egress ? 0 : -ENOENT;
+		return -ENOENT;
 
 	err = nfp_modify_flow_metadata(app, nfp_flow);
 	if (err)
@@ -571,7 +547,6 @@ nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
  * @app:	Pointer to the APP handle
  * @netdev:	Netdev structure.
  * @flow:	TC flower classifier offload structure
- * @egress:	Netdev is the egress dev.
  *
  * Populates a flow statistics structure which which corresponds to a
  * specific flow.
@@ -580,22 +555,17 @@ nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
  */
 static int
 nfp_flower_get_stats(struct nfp_app *app, struct net_device *netdev,
-		     struct tc_cls_flower_offload *flow, bool egress)
+		     struct tc_cls_flower_offload *flow)
 {
 	struct nfp_flower_priv *priv = app->priv;
 	struct nfp_fl_payload *nfp_flow;
-	struct net_device *ingr_dev;
 	u32 ctx_id;
 
-	ingr_dev = egress ? NULL : netdev;
-	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, ingr_dev,
+	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev,
 					      NFP_FL_STATS_CTX_DONT_CARE);
 	if (!nfp_flow)
 		return -EINVAL;
 
-	if (nfp_flow->ingress_offload && egress)
-		return 0;
-
 	ctx_id = be32_to_cpu(nfp_flow->meta.host_ctx_id);
 
 	spin_lock_bh(&priv->stats_lock);
@@ -612,35 +582,18 @@ nfp_flower_get_stats(struct nfp_app *app, struct net_device *netdev,
 
 static int
 nfp_flower_repr_offload(struct nfp_app *app, struct net_device *netdev,
-			struct tc_cls_flower_offload *flower, bool egress)
+			struct tc_cls_flower_offload *flower)
 {
 	if (!eth_proto_is_802_3(flower->common.protocol))
 		return -EOPNOTSUPP;
 
 	switch (flower->command) {
 	case TC_CLSFLOWER_REPLACE:
-		return nfp_flower_add_offload(app, netdev, flower, egress);
+		return nfp_flower_add_offload(app, netdev, flower);
 	case TC_CLSFLOWER_DESTROY:
-		return nfp_flower_del_offload(app, netdev, flower, egress);
+		return nfp_flower_del_offload(app, netdev, flower);
 	case TC_CLSFLOWER_STATS:
-		return nfp_flower_get_stats(app, netdev, flower, egress);
-	default:
-		return -EOPNOTSUPP;
-	}
-}
-
-int nfp_flower_setup_tc_egress_cb(enum tc_setup_type type, void *type_data,
-				  void *cb_priv)
-{
-	struct nfp_repr *repr = cb_priv;
-
-	if (!tc_cls_can_offload_and_chain0(repr->netdev, type_data))
-		return -EOPNOTSUPP;
-
-	switch (type) {
-	case TC_SETUP_CLSFLOWER:
-		return nfp_flower_repr_offload(repr->app, repr->netdev,
-					       type_data, true);
+		return nfp_flower_get_stats(app, netdev, flower);
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -657,7 +610,7 @@ static int nfp_flower_setup_tc_block_cb(enum tc_setup_type type,
 	switch (type) {
 	case TC_SETUP_CLSFLOWER:
 		return nfp_flower_repr_offload(repr->app, repr->netdev,
-					       type_data, false);
+					       type_data);
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -741,7 +694,7 @@ static int nfp_flower_setup_indr_block_cb(enum tc_setup_type type,
 	switch (type) {
 	case TC_SETUP_CLSFLOWER:
 		return nfp_flower_repr_offload(priv->app, priv->netdev,
-					       type_data, false);
+					       type_data);
 	default:
 		return -EOPNOTSUPP;
 	}
-- 
2.7.4

[RFC net-next v2 8/8] nfp: flower: remove unnecessary code in flow lookup

[John Hurley]

Recent changes to NFP mean that stats updates from fw to driver no longer
require a flow lookup and (because egdev offload has been removed) the
ingress netdev for a lookup is now always known.

Remove obsolete code in a flow lookup that matches on host context and
that allows for a netdev to be NULL.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/flower/main.h     |  3 +--
 drivers/net/ethernet/netronome/nfp/flower/metadata.c | 11 +++--------
 drivers/net/ethernet/netronome/nfp/flower/offload.c  |  6 ++----
 3 files changed, 6 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.h b/drivers/net/ethernet/netronome/nfp/flower/main.h
index d8c8f0d..3d3a13f 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.h
@@ -20,7 +20,6 @@ struct nfp_fl_pre_lag;
 struct net_device;
 struct nfp_app;
 
-#define NFP_FL_STATS_CTX_DONT_CARE	cpu_to_be32(0xffffffff)
 #define NFP_FL_STATS_ELEM_RS		FIELD_SIZEOF(struct nfp_fl_stats_id, \
 						     init_unalloc)
 #define NFP_FLOWER_MASK_ENTRY_RS	256
@@ -248,7 +247,7 @@ int nfp_modify_flow_metadata(struct nfp_app *app,
 
 struct nfp_fl_payload *
 nfp_flower_search_fl_table(struct nfp_app *app, unsigned long tc_flower_cookie,
-			   struct net_device *netdev, __be32 host_ctx);
+			   struct net_device *netdev);
 struct nfp_fl_payload *
 nfp_flower_remove_fl_table(struct nfp_app *app, unsigned long tc_flower_cookie);
 
diff --git a/drivers/net/ethernet/netronome/nfp/flower/metadata.c b/drivers/net/ethernet/netronome/nfp/flower/metadata.c
index 9b4711c..573a440 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/metadata.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/metadata.c
@@ -21,7 +21,6 @@ struct nfp_mask_id_table {
 struct nfp_fl_flow_table_cmp_arg {
 	struct net_device *netdev;
 	unsigned long cookie;
-	__be32 host_ctx;
 };
 
 static int nfp_release_stats_entry(struct nfp_app *app, u32 stats_context_id)
@@ -76,14 +75,13 @@ static int nfp_get_stats_entry(struct nfp_app *app, u32 *stats_context_id)
 /* Must be called with either RTNL or rcu_read_lock */
 struct nfp_fl_payload *
 nfp_flower_search_fl_table(struct nfp_app *app, unsigned long tc_flower_cookie,
-			   struct net_device *netdev, __be32 host_ctx)
+			   struct net_device *netdev)
 {
 	struct nfp_fl_flow_table_cmp_arg flower_cmp_arg;
 	struct nfp_flower_priv *priv = app->priv;
 
 	flower_cmp_arg.netdev = netdev;
 	flower_cmp_arg.cookie = tc_flower_cookie;
-	flower_cmp_arg.host_ctx = host_ctx;
 
 	return rhashtable_lookup_fast(&priv->flow_table, &flower_cmp_arg,
 				      nfp_flower_table_params);
@@ -307,8 +305,7 @@ int nfp_compile_flow_metadata(struct nfp_app *app,
 	priv->stats[stats_cxt].bytes = 0;
 	priv->stats[stats_cxt].used = jiffies;
 
-	check_entry = nfp_flower_search_fl_table(app, flow->cookie, netdev,
-						 NFP_FL_STATS_CTX_DONT_CARE);
+	check_entry = nfp_flower_search_fl_table(app, flow->cookie, netdev);
 	if (check_entry) {
 		if (nfp_release_stats_entry(app, stats_cxt))
 			return -EINVAL;
@@ -353,9 +350,7 @@ static int nfp_fl_obj_cmpfn(struct rhashtable_compare_arg *arg,
 	const struct nfp_fl_flow_table_cmp_arg *cmp_arg = arg->key;
 	const struct nfp_fl_payload *flow_entry = obj;
 
-	if ((!cmp_arg->netdev || flow_entry->ingress_dev == cmp_arg->netdev) &&
-	    (cmp_arg->host_ctx == NFP_FL_STATS_CTX_DONT_CARE ||
-	     flow_entry->meta.host_ctx_id == cmp_arg->host_ctx))
+	if (flow_entry->ingress_dev == cmp_arg->netdev)
 		return flow_entry->tc_flower_cookie != cmp_arg->cookie;
 
 	return 1;
diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index 392d292..07ff728 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -512,8 +512,7 @@ nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
 	if (nfp_netdev_is_nfp_repr(netdev))
 		port = nfp_port_from_netdev(netdev);
 
-	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev,
-					      NFP_FL_STATS_CTX_DONT_CARE);
+	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev);
 	if (!nfp_flow)
 		return -ENOENT;
 
@@ -561,8 +560,7 @@ nfp_flower_get_stats(struct nfp_app *app, struct net_device *netdev,
 	struct nfp_fl_payload *nfp_flow;
 	u32 ctx_id;
 
-	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev,
-					      NFP_FL_STATS_CTX_DONT_CARE);
+	nfp_flow = nfp_flower_search_fl_table(app, flow->cookie, netdev);
 	if (!nfp_flow)
 		return -EINVAL;
 
-- 
2.7.4

Re: [RFC net-next v2 0/8] indirect tc block cb registration

[Jiri Pirko]

Thu, Oct 25, 2018 at 02:26:49PM CEST, john.hurley@netronome.com wrote:
>This patchset introduces an alternative to egdev offload by allowing a
>driver to register for block updates when an external device (e.g. tunnel
>netdev) is bound to a TC block. Drivers can track new netdevs or register
>to existing ones to receive information on such events. Based on this,
>they may register for block offload rules using already existing
>functions.
>
>The patchset also implements this new indirect block registration in the
>NFP driver to allow the offloading of tunnel rules. The use of egdev
>offload (which is currently only used for tunnel offload) is subsequently
>removed.

John, I'm missing v1->v2 changelog. Could you please add it?

Thanks!

Re: [RFC net-next v2 2/8] net: add netif_is_geneve()

[Jiri Pirko]

Thu, Oct 25, 2018 at 02:26:51PM CEST, john.hurley@netronome.com wrote:
>Add a helper function to determine if the type of a netdev is geneve based
>on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels
>to check the underlying type of the device.
>
>A recent patch added a similar helper to vxlan.h
>
>Signed-off-by: John Hurley <john.hurley@netronome.com>
>Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>

I don't understand why this and the next patch are part of this
patchset. They don't seem directly related.

Re: [RFC net-next v2 0/8] indirect tc block cb registration

[John Hurley]

On Thu, Oct 25, 2018 at 1:58 PM Jiri Pirko <jiri@resnulli.us> wrote:
>
> Thu, Oct 25, 2018 at 02:26:49PM CEST, john.hurley@netronome.com wrote:
> >This patchset introduces an alternative to egdev offload by allowing a
> >driver to register for block updates when an external device (e.g. tunnel
> >netdev) is bound to a TC block. Drivers can track new netdevs or register
> >to existing ones to receive information on such events. Based on this,
> >they may register for block offload rules using already existing
> >functions.
> >
> >The patchset also implements this new indirect block registration in the
> >NFP driver to allow the offloading of tunnel rules. The use of egdev
> >offload (which is currently only used for tunnel offload) is subsequently
> >removed.
>
> John, I'm missing v1->v2 changelog. Could you please add it?
>
> Thanks!

Hi Jiri,
There's little change outside the NFP in v2 but here's short changelog:

v1->v2:
- free allocated owner struct in block_owner_clean function
- add geneve type helper function
- move test stub in NFP (v1 patch 2) to full tunnel offload
implementation via indirect blocks (v2 patches 3-8)

Re: [RFC net-next v2 2/8] net: add netif_is_geneve()

[John Hurley]

On Thu, Oct 25, 2018 at 2:00 PM Jiri Pirko <jiri@resnulli.us> wrote:
>
> Thu, Oct 25, 2018 at 02:26:51PM CEST, john.hurley@netronome.com wrote:
> >Add a helper function to determine if the type of a netdev is geneve based
> >on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels
> >to check the underlying type of the device.
> >
> >A recent patch added a similar helper to vxlan.h
> >
> >Signed-off-by: John Hurley <john.hurley@netronome.com>
> >Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
>
> I don't understand why this and the next patch are part of this
> patchset. They don't seem directly related.

This is used in later patches that implement the indirect block
offload but I suppose it is not directly related.
We can probably move it to a separate patchset.
Thanks

Re: [RFC net-next v2 2/8] net: add netif_is_geneve()

[Sergei Shtylyov]

Hello!

On 25.10.2018 15:26, John Hurley wrote:

> Add a helper function to determine if the type of a netdev is geneve based
> on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels

    Offload?

> to check the underlying type of the device.
>
> A recent patch added a similar helper to vxlan.h
>
> Signed-off-by: John Hurley <john.hurley@netronome.com>
> Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
[...]

MBR, Sergei

Re: [RFC net-next v2 1/8] net: sched: register callbacks for indirect tc block binds

[Or Gerlitz]

On Thu, Oct 25, 2018 at 3:28 PM John Hurley <john.hurley@netronome.com> wrote:
> Currently drivers can register to receive TC block bind/unbind callbacks
> by implementing the setup_tc ndo in any of their given netdevs. However,
> drivers may also be interested in binds to higher level devices (e.g.
> tunnel drivers) to potentially offload filters applied to them.

> Introduce indirect block devs which allows drivers to register callbacks
> for block binds on other devices. The calling driver is expected to
> reference an 'owner' struct that it will pass to all block registrations.
> This is used to track the callbacks from a given driver and free them if
> the driver is removed while the upper level device is still active.

Hi John,

Maybe it would be better to follow the trusted environment model of the kernel
and not protect the core from driver bugs? If the driver does things right they
will unregister before bailing out and if not, they will have to fix..

> Freeing a callback will also trigger an unbind event (if necessary) to
> direct the driver to remove any offloaded rules and unreg any block filter
> callbacks.

> Allow registering an indirect block dev callback for a device that is
> already bound to a block. In this case (if it is an ingress block),
> register and also trigger the callback meaning that any already installed
> rules can be replayed to the calling driver.

not just can be replayed.. they will be replayed, but through an
existing (tc re-offload?)
facility, correct?

Or.

Re: [RFC net-next v2 2/8] net: add netif_is_geneve()

[John Hurley]

On Fri, Oct 26, 2018 at 9:52 AM Sergei Shtylyov
<sergei.shtylyov@cogentembedded.com> wrote:
>
> Hello!
>
> On 25.10.2018 15:26, John Hurley wrote:
>
> > Add a helper function to determine if the type of a netdev is geneve based
> > on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels
>
>     Offload?
>

offload encap/decap to a hardware device such as a smartNIC.
Sorry, should have made this clearer

> > to check the underlying type of the device.
> >
> > A recent patch added a similar helper to vxlan.h
> >
> > Signed-off-by: John Hurley <john.hurley@netronome.com>
> > Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> [...]
>
> MBR, Sergei
>
>

Re: [RFC net-next v2 1/8] net: sched: register callbacks for indirect tc block binds

[John Hurley]

On Sun, Oct 28, 2018 at 11:10 AM Or Gerlitz <gerlitz.or@gmail.com> wrote:
>
> On Thu, Oct 25, 2018 at 3:28 PM John Hurley <john.hurley@netronome.com> wrote:
> > Currently drivers can register to receive TC block bind/unbind callbacks
> > by implementing the setup_tc ndo in any of their given netdevs. However,
> > drivers may also be interested in binds to higher level devices (e.g.
> > tunnel drivers) to potentially offload filters applied to them.
>
> > Introduce indirect block devs which allows drivers to register callbacks
> > for block binds on other devices. The calling driver is expected to
> > reference an 'owner' struct that it will pass to all block registrations.
> > This is used to track the callbacks from a given driver and free them if
> > the driver is removed while the upper level device is still active.
>
> Hi John,
>
> Maybe it would be better to follow the trusted environment model of the kernel
> and not protect the core from driver bugs? If the driver does things right they
> will unregister before bailing out and if not, they will have to fix..
>

Hi Or,
The owner stuff just makes it easier for a driver to track the blocks
it has registered for and, in turn, release these when exiting.
We could just leave this up to the driver to ensure it properly cleans
up after itself.
I don't feel that strongly either way.

> > Freeing a callback will also trigger an unbind event (if necessary) to
> > direct the driver to remove any offloaded rules and unreg any block filter
> > callbacks.
>
> > Allow registering an indirect block dev callback for a device that is
> > already bound to a block. In this case (if it is an ingress block),
> > register and also trigger the callback meaning that any already installed
> > rules can be replayed to the calling driver.
>
> not just can be replayed.. they will be replayed, but through an
> existing (tc re-offload?)
> facility, correct?
>

Yes, currently in TC, when you register for rule callbacks to a block
that already has rules, these rules are replayed.
With the indirect block approach we still use the same mechanism for
requesting rule callbacks,

> Or.

Re: [RFC net-next v2 2/8] net: add netif_is_geneve()

[Sergei Shtylyov]

On 10/29/2018 03:06 PM, John Hurley wrote:

>>> Add a helper function to determine if the type of a netdev is geneve based
>>> on its rtnl_link_ops. This allows drivers that may wish to ofload tunnels
>>
>>     Offload?
>>
> 
> offload encap/decap to a hardware device such as a smartNIC.
> Sorry, should have made this clearer

   I was only reacting to your typo. :-)

>>> to check the underlying type of the device.
>>>
>>> A recent patch added a similar helper to vxlan.h
>>>
>>> Signed-off-by: John Hurley <john.hurley@netronome.com>
>>> Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
>> [...]

MBR, Sergei

Re: [RFC net-next v2 1/8] net: sched: register callbacks for indirect tc block binds

[Or Gerlitz]

On Mon, Oct 29, 2018 at 2:54 PM John Hurley <john.hurley@netronome.com> wrote:
> On Sun, Oct 28, 2018 at 11:10 AM Or Gerlitz <gerlitz.or@gmail.com> wrote:
> >
> > On Thu, Oct 25, 2018 at 3:28 PM John Hurley <john.hurley@netronome.com> wrote:
> > > Currently drivers can register to receive TC block bind/unbind callbacks
> > > by implementing the setup_tc ndo in any of their given netdevs. However,
> > > drivers may also be interested in binds to higher level devices (e.g.
> > > tunnel drivers) to potentially offload filters applied to them.
> >
> > > Introduce indirect block devs which allows drivers to register callbacks
> > > for block binds on other devices. The calling driver is expected to
> > > reference an 'owner' struct that it will pass to all block registrations.
> > > This is used to track the callbacks from a given driver and free them if
> > > the driver is removed while the upper level device is still active.

>> Maybe it would be better to follow the trusted environment model of the kernel
>> and not protect the core from driver bugs? If the driver does things right they
>> will unregister before bailing out and if not, they will have to fix..

> The owner stuff just makes it easier for a driver to track the blocks
> it has registered for and, in turn, release these when exiting.
> We could just leave this up to the driver to ensure it properly cleans
> up after itself.

If it makes the life of the driver easier and doesn't add notable complexity,
then I think I am good to leave it

> I don't feel that strongly either way.

m2

So lets see if other comment here, if not, we can just leave it, I guess

> > > Allow registering an indirect block dev callback for a device that is
> > > already bound to a block. In this case (if it is an ingress block),
> > > register and also trigger the callback meaning that any already installed
> > > rules can be replayed to the calling driver.

> > not just can be replayed.. they will be replayed, but through an
> > existing (tc re-offload?) facility, correct?

> Yes, currently in TC, when you register for rule callbacks to a block
> that already has rules, these rules are replayed.
> With the indirect block approach we still use the same mechanism for
> requesting rule callbacks,

sounds good

Re: [RFC net-next v2 1/8] net: sched: register callbacks for indirect tc block binds

[Jakub Kicinski]

On Mon, 29 Oct 2018 17:12:27 +0200, Or Gerlitz wrote:
> >> Maybe it would be better to follow the trusted environment model of the kernel
> >> and not protect the core from driver bugs? If the driver does things right they
> >> will unregister before bailing out and if not, they will have to fix..  
> 
> > The owner stuff just makes it easier for a driver to track the blocks
> > it has registered for and, in turn, release these when exiting.
> > We could just leave this up to the driver to ensure it properly cleans
> > up after itself.  
> 
> If it makes the life of the driver easier and doesn't add notable complexity,
> then I think I am good to leave it
> 
> > I don't feel that strongly either way.  
> 
> m2
> 
> So lets see if other comment here, if not, we can just leave it, I guess

To be honest big part of why we retained this mechanism was to keep the
per-driver core structure in existence (struct tcf_indr_block_owner).
In my experience it is way easier to move common functionality into the
core if there is a place where core can track offload-related state.

Growing core structures just for offloads is not super advisable, so
unless there is a separate structure core allocates - all state lands in
the drivers.  This lesson comes from BPF offload, which started off as
mostly stateless from core's perspective where all operations were muxed
via a single NDO, but that became increasingly awkward to use.  We are
gradually moving to a "offload device + ops" form.

I'm not 100% sure the indirect callbacks are a good place for a core
structure, given we didn't seem to need such a thing for normal TC
blocks.  So yes, perhaps we should drop that code.

Hope that explanation makes sense.