Skip to contentSkip to conversation

tc-gate offload for SJA1105 DSA switch

PATCH In review Vladimir Oltean 4 participants 17 messages May 3, 2020 – May 5, 2020

Subsystems (5)
lore
Share & tools
lore

Vladimir Oltean

author maintainer
#0 lore raw

cover tc-gate offload for SJA1105 DSA switch

From: Vladimir Oltean <[email protected]>

Expose the TTEthernet hardware features of the switch using standard tc-flower actions: trap, drop, redirect and gate.

Vladimir Oltean (6):
net: dsa: export dsa_slave_dev_check and dsa_slave_to_port
net: dsa: sja1105: add static tables for virtual links
net: dsa: sja1105: make room for virtual link parsing in flower
offload
net: dsa: sja1105: support flow-based redirection via virtual links
net: dsa: sja1105: implement tc-gate using time-triggered virtual
links
docs: net: dsa: sja1105: document intended usage of virtual links

 Documentation/networking/dsa/sja1105.rst      | 116 +++
 drivers/net/dsa/sja1105/Kconfig               |   9 +
 drivers/net/dsa/sja1105/Makefile              |   4 +
 drivers/net/dsa/sja1105/sja1105.h             |  59 +-
 .../net/dsa/sja1105/sja1105_dynamic_config.c  |  51 ++
 drivers/net/dsa/sja1105/sja1105_flower.c      | 215 ++++-
 drivers/net/dsa/sja1105/sja1105_main.c        |  13 +-
 drivers/net/dsa/sja1105/sja1105_ptp.h         |  13 +
 drivers/net/dsa/sja1105/sja1105_spi.c         |   2 +
 .../net/dsa/sja1105/sja1105_static_config.c   | 202 +++++
 .../net/dsa/sja1105/sja1105_static_config.h   |  65 ++
 drivers/net/dsa/sja1105/sja1105_tas.c         | 127 ++-
 drivers/net/dsa/sja1105/sja1105_tas.h         |  31 +
 drivers/net/dsa/sja1105/sja1105_vl.c          | 796 ++++++++++++++++++
 drivers/net/dsa/sja1105/sja1105_vl.h          |  72 ++
 include/net/dsa.h                             |   2 +
 net/dsa/dsa_priv.h                            |   8 -
 net/dsa/slave.c                               |   9 +
 18 files changed, 1742 insertions(+), 52 deletions(-)
 create mode 100644 drivers/net/dsa/sja1105/sja1105_vl.c
 create mode 100644 drivers/net/dsa/sja1105/sja1105_vl.h
Signature · 1 line
2.17.1
author maintainer
#1 lore raw

PATCH 1/6 net: dsa: export dsa_slave_dev_check and dsa_slave_to_port

From: Vladimir Oltean <[email protected]>

To be able to perform mirroring and redirection through tc-flower offloads (the implementation of which is given raw access to the flow_cls_offload structure), switch drivers need to be able to call these functions on act->dev.

---
Changes from RFC:
None.

 include/net/dsa.h  | 2 ++
 net/dsa/dsa_priv.h | 8 --------
 net/dsa/slave.c    | 9 +++++++++
 3 files changed, 11 insertions(+), 8 deletions(-)
include/net/dsa.h +2 -0
index fb3f9222f2a1..62beaa4c234e 100644
@@ -739,6 +739,8 @@ int dsa_port_get_phy_strings(struct dsa_port *dp, uint8_t *data);
 int dsa_port_get_ethtool_phy_stats(struct dsa_port *dp, uint64_t *data);
 int dsa_port_get_phy_sset_count(struct dsa_port *dp);
 void dsa_port_phylink_mac_change(struct dsa_switch *ds, int port, bool up);
+bool dsa_slave_dev_check(const struct net_device *dev);
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev);
 
 struct dsa_tag_driver {
 	const struct dsa_device_ops *ops;
net/dsa/dsa_priv.h +0 -8
index 6d9a1ef65fa0..32bf570fd71c 100644
@@ -173,19 +173,11 @@ extern const struct dsa_device_ops notag_netdev_ops;
 void dsa_slave_mii_bus_init(struct dsa_switch *ds);
 int dsa_slave_create(struct dsa_port *dp);
 void dsa_slave_destroy(struct net_device *slave_dev);
-bool dsa_slave_dev_check(const struct net_device *dev);
 int dsa_slave_suspend(struct net_device *slave_dev);
 int dsa_slave_resume(struct net_device *slave_dev);
 int dsa_slave_register_notifier(void);
 void dsa_slave_unregister_notifier(void);
 
-static inline struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
-{
-	struct dsa_slave_priv *p = netdev_priv(dev);
-
-	return p->dp;
-}
-
 static inline struct net_device *
 dsa_slave_to_master(const struct net_device *dev)
 {
net/dsa/slave.c +9 -0
index ba8bf90dc0cc..4eeb5b47ef99 100644
@@ -62,6 +62,14 @@ static int dsa_slave_get_iflink(const struct net_device *dev)
 	return dsa_slave_to_master(dev)->ifindex;
 }
 
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
+{
+	struct dsa_slave_priv *p = netdev_priv(dev);
+
+	return p->dp;
+}
+EXPORT_SYMBOL_GPL(dsa_slave_to_port);
+
 static int dsa_slave_open(struct net_device *dev)
 {
 	struct net_device *master = dsa_slave_to_master(dev);
@@ -1836,6 +1844,7 @@ bool dsa_slave_dev_check(const struct net_device *dev)
 {
 	return dev->netdev_ops == &dsa_slave_netdev_ops;
 }
+EXPORT_SYMBOL_GPL(dsa_slave_dev_check);
 
 static int dsa_slave_changeupper(struct net_device *dev,
 				 struct netdev_notifier_changeupper_info *info)
Signature · 1 line
2.17.1
Reviewed-by on v1 1/6
#2 lore raw

Re: [PATCH net-next 1/6] net: dsa: export dsa_slave_dev_check and dsa_slave_to_port

Quoted 59 lines · Vladimir Oltean

From: Vladimir Oltean <[email protected]>

To be able to perform mirroring and redirection through tc-flower offloads (the implementation of which is given raw access to the flow_cls_offload structure), switch drivers need to be able to call these functions on act->dev.

---
Changes from RFC:
None.

 include/net/dsa.h  | 2 ++
 net/dsa/dsa_priv.h | 8 --------
 net/dsa/slave.c    | 9 +++++++++
 3 files changed, 11 insertions(+), 8 deletions(-)
include/net/dsa.h +2 -0
index fb3f9222f2a1..62beaa4c234e 100644
@@ -739,6 +739,8 @@ int dsa_port_get_phy_strings(struct dsa_port *dp, uint8_t *data);
 int dsa_port_get_ethtool_phy_stats(struct dsa_port *dp, uint64_t *data);
 int dsa_port_get_phy_sset_count(struct dsa_port *dp);
 void dsa_port_phylink_mac_change(struct dsa_switch *ds, int port, bool up);
+bool dsa_slave_dev_check(const struct net_device *dev);
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev);
 
 struct dsa_tag_driver {
 	const struct dsa_device_ops *ops;
net/dsa/dsa_priv.h +0 -8
index 6d9a1ef65fa0..32bf570fd71c 100644
@@ -173,19 +173,11 @@ extern const struct dsa_device_ops notag_netdev_ops;
 void dsa_slave_mii_bus_init(struct dsa_switch *ds);
 int dsa_slave_create(struct dsa_port *dp);
 void dsa_slave_destroy(struct net_device *slave_dev);
-bool dsa_slave_dev_check(const struct net_device *dev);
 int dsa_slave_suspend(struct net_device *slave_dev);
 int dsa_slave_resume(struct net_device *slave_dev);
 int dsa_slave_register_notifier(void);
 void dsa_slave_unregister_notifier(void);
 
-static inline struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
-{
-	struct dsa_slave_priv *p = netdev_priv(dev);
-
-	return p->dp;
-}
-
 static inline struct net_device *
 dsa_slave_to_master(const struct net_device *dev)
 {
net/dsa/slave.c +0 -0
index ba8bf90dc0cc..4eeb5b47ef99 100644

net/dsa/slave.c static int dsa_slave_get_iflink(const struct net_device *dev) · line 71

Quoted 10 lines
net/dsa/slave.c +7 -0
@@ -62,6 +62,14 @@ static int dsa_slave_get_iflink(const struct net_device *dev)
 	return dsa_slave_to_master(dev)->ifindex;
 }
 
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
+{
+	struct dsa_slave_priv *p = netdev_priv(dev);
+
+	return p->dp;
+}
+EXPORT_SYMBOL_GPL(dsa_slave_to_port);

You could probably make this a static inline in net/dsa.h, too. With or without doing that:

Signature · 1 line
Florian
author maintainer
#3 lore raw

Re: [PATCH net-next 1/6] net: dsa: export dsa_slave_dev_check and dsa_slave_to_port

Hi Florian,

Quoted 66 lines · Florian Fainelli
Quoted 62 lines · Vladimir Oltean

From: Vladimir Oltean <[email protected]>

To be able to perform mirroring and redirection through tc-flower offloads (the implementation of which is given raw access to the flow_cls_offload structure), switch drivers need to be able to call these functions on act->dev.

---
Changes from RFC:
None.

 include/net/dsa.h  | 2 ++
 net/dsa/dsa_priv.h | 8 --------
 net/dsa/slave.c    | 9 +++++++++
 3 files changed, 11 insertions(+), 8 deletions(-)
include/net/dsa.h +2 -0
index fb3f9222f2a1..62beaa4c234e 100644
@@ -739,6 +739,8 @@ int dsa_port_get_phy_strings(struct dsa_port *dp, uint8_t *data);
 int dsa_port_get_ethtool_phy_stats(struct dsa_port *dp, uint64_t *data);
 int dsa_port_get_phy_sset_count(struct dsa_port *dp);
 void dsa_port_phylink_mac_change(struct dsa_switch *ds, int port, bool up);
+bool dsa_slave_dev_check(const struct net_device *dev);
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev);
 
 struct dsa_tag_driver {
      const struct dsa_device_ops *ops;
net/dsa/dsa_priv.h +0 -8
index 6d9a1ef65fa0..32bf570fd71c 100644
@@ -173,19 +173,11 @@ extern const struct dsa_device_ops notag_netdev_ops;
 void dsa_slave_mii_bus_init(struct dsa_switch *ds);
 int dsa_slave_create(struct dsa_port *dp);
 void dsa_slave_destroy(struct net_device *slave_dev);
-bool dsa_slave_dev_check(const struct net_device *dev);
 int dsa_slave_suspend(struct net_device *slave_dev);
 int dsa_slave_resume(struct net_device *slave_dev);
 int dsa_slave_register_notifier(void);
 void dsa_slave_unregister_notifier(void);
 
-static inline struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
-{
-     struct dsa_slave_priv *p = netdev_priv(dev);
-
-     return p->dp;
-}
-
 static inline struct net_device *
 dsa_slave_to_master(const struct net_device *dev)
 {
net/dsa/slave.c +0 -0
index ba8bf90dc0cc..4eeb5b47ef99 100644
@@ -62,6 +62,14 @@ static int dsa_slave_get_iflink(const struct net_device *dev)
      return dsa_slave_to_master(dev)->ifindex;
 }
 
Quoted 10 lines

net/dsa/slave.c static int dsa_slave_get_iflink(const struct net_device *dev) · line 71

Quoted 7 lines
net/dsa/slave.c +7 -0
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
+{
+     struct dsa_slave_priv *p = netdev_priv(dev);
+
+     return p->dp;
+}
+EXPORT_SYMBOL_GPL(dsa_slave_to_port);

You could probably make this a static inline in net/dsa.h, too. With or without doing that:

With dereferencing dsa_slave_priv, I don't think so.

Quoted 4 lines

--
Florian

Thanks,
-Vladimir

#4 lore raw

Re: [PATCH net-next 1/6] net: dsa: export dsa_slave_dev_check and dsa_slave_to_port

Quoted 80 lines · Vladimir Oltean

Hi Florian,

Quoted 66 lines · Florian Fainelli
Quoted 62 lines · Vladimir Oltean

From: Vladimir Oltean <[email protected]>

To be able to perform mirroring and redirection through tc-flower offloads (the implementation of which is given raw access to the flow_cls_offload structure), switch drivers need to be able to call these functions on act->dev.

---
Changes from RFC:
None.

 include/net/dsa.h  | 2 ++
 net/dsa/dsa_priv.h | 8 --------
 net/dsa/slave.c    | 9 +++++++++
 3 files changed, 11 insertions(+), 8 deletions(-)
include/net/dsa.h +2 -0
index fb3f9222f2a1..62beaa4c234e 100644
@@ -739,6 +739,8 @@ int dsa_port_get_phy_strings(struct dsa_port *dp, uint8_t *data);
 int dsa_port_get_ethtool_phy_stats(struct dsa_port *dp, uint64_t *data);
 int dsa_port_get_phy_sset_count(struct dsa_port *dp);
 void dsa_port_phylink_mac_change(struct dsa_switch *ds, int port, bool up);
+bool dsa_slave_dev_check(const struct net_device *dev);
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev);
 
 struct dsa_tag_driver {
      const struct dsa_device_ops *ops;
net/dsa/dsa_priv.h +0 -8
index 6d9a1ef65fa0..32bf570fd71c 100644
@@ -173,19 +173,11 @@ extern const struct dsa_device_ops notag_netdev_ops;
 void dsa_slave_mii_bus_init(struct dsa_switch *ds);
 int dsa_slave_create(struct dsa_port *dp);
 void dsa_slave_destroy(struct net_device *slave_dev);
-bool dsa_slave_dev_check(const struct net_device *dev);
 int dsa_slave_suspend(struct net_device *slave_dev);
 int dsa_slave_resume(struct net_device *slave_dev);
 int dsa_slave_register_notifier(void);
 void dsa_slave_unregister_notifier(void);
 
-static inline struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
-{
-     struct dsa_slave_priv *p = netdev_priv(dev);
-
-     return p->dp;
-}
-
 static inline struct net_device *
 dsa_slave_to_master(const struct net_device *dev)
 {
net/dsa/slave.c +0 -0
index ba8bf90dc0cc..4eeb5b47ef99 100644
@@ -62,6 +62,14 @@ static int dsa_slave_get_iflink(const struct net_device *dev)
      return dsa_slave_to_master(dev)->ifindex;
 }
 
Quoted 10 lines

net/dsa/slave.c static int dsa_slave_get_iflink(const struct net_device *dev) · line 71

Quoted 7 lines
net/dsa/slave.c +7 -0
+struct dsa_port *dsa_slave_to_port(const struct net_device *dev)
+{
+     struct dsa_slave_priv *p = netdev_priv(dev);
+
+     return p->dp;
+}
+EXPORT_SYMBOL_GPL(dsa_slave_to_port);

You could probably make this a static inline in net/dsa.h, too. With or without doing that:

Quoted 1 line

With dereferencing dsa_slave_priv, I don't think so.

Yes, I missed that dsa_slave_priv is not visible outside of dsa_priv.h (on purpose). Thanks!

Signature · 1 line
Florian
author maintainer
#5 lore raw

PATCH 2/6 net: dsa: sja1105: add static tables for virtual links

In reply to Vladimir Oltean

From: Vladimir Oltean <[email protected]>

This patch adds the register definitions for the:
- VL Lookup Table
- VL Policing Table
- VL Forwarding Table
- VL Forwarding Parameters Table

These are needed in order to perform TTEthernet operations: QoS classification, flow-based policing and/or frame redirecting with the switch.

---
Changes from RFC:
None.

 drivers/net/dsa/sja1105/sja1105.h             |   2 +
 .../net/dsa/sja1105/sja1105_dynamic_config.c  |  51 +++++
 .../net/dsa/sja1105/sja1105_static_config.c   | 202 ++++++++++++++++++
 .../net/dsa/sja1105/sja1105_static_config.h   |  63 ++++++
 4 files changed, 318 insertions(+)
drivers/net/dsa/sja1105/sja1105.h +2 -0
index 2f62942692ec..602aa30c832f 100644
@@ -254,6 +254,8 @@ size_t sja1105pqrs_mac_config_entry_packing(void *buf, void *entry_ptr,
 					    enum packing_op op);
 size_t sja1105pqrs_avb_params_entry_packing(void *buf, void *entry_ptr,
 					    enum packing_op op);
+size_t sja1105_vl_lookup_entry_packing(void *buf, void *entry_ptr,
+				       enum packing_op op);
 
 /* From sja1105_flower.c */
 int sja1105_cls_flower_del(struct dsa_switch *ds, int port,
drivers/net/dsa/sja1105/sja1105_dynamic_config.c +51 -0

87 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_static_config.c +202 -0

316 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_static_config.h +63 -0

123 lines not on this page · raw message

Signature · 1 line
2.17.1
author maintainer
#6 lore raw

PATCH 3/6 net: dsa: sja1105: make room for virtual link parsing in flower offload

In reply to Vladimir Oltean

From: Vladimir Oltean <[email protected]>

Virtual links are a sja1105 hardware concept of executing various flow actions based on a key extracted from the frame's DMAC, VID and PCP.

Currently the tc-flower offload code supports only parsing the DMAC if that is the broadcast MAC address, and the VLAN PCP. Extract the key parsing logic from the L2 policers functionality and move it into its own function, after adding extra logic for matching on any DMAC and VID.

---
Changes from RFC:
None.

 drivers/net/dsa/sja1105/sja1105.h        |  28 +++++-
 drivers/net/dsa/sja1105/sja1105_flower.c | 111 +++++++++++++++++------
 2 files changed, 112 insertions(+), 27 deletions(-)
drivers/net/dsa/sja1105/sja1105.h +27 -1
index 602aa30c832f..95633ad9bfb7 100644
@@ -97,6 +97,32 @@ struct sja1105_info {
 	const char *name;
 };
 
+enum sja1105_key_type {
+	SJA1105_KEY_BCAST,
+	SJA1105_KEY_TC,
+	SJA1105_KEY_VLAN_UNAWARE_VL,
+	SJA1105_KEY_VLAN_AWARE_VL,
+};
+
+struct sja1105_key {
+	enum sja1105_key_type type;
+
+	union {
+		/* SJA1105_KEY_TC */
+		struct {
+			int pcp;
+		} tc;
+
+		/* SJA1105_KEY_VLAN_UNAWARE_VL */
+		/* SJA1105_KEY_VLAN_AWARE_VL */
+		struct {
+			u64 dmac;
+			u16 vid;
+			u16 pcp;
+		} vl;
+	};
+};
+
 enum sja1105_rule_type {
 	SJA1105_RULE_BCAST_POLICER,
 	SJA1105_RULE_TC_POLICER,
@@ -106,6 +132,7 @@ struct sja1105_rule {
 	struct list_head list;
 	unsigned long cookie;
 	unsigned long port_mask;
+	struct sja1105_key key;
 	enum sja1105_rule_type type;
 
 	union {
@@ -117,7 +144,6 @@ struct sja1105_rule {
 		/* SJA1105_RULE_TC_POLICER */
 		struct {
 			int sharindx;
-			int tc;
 		} tc_pol;
 	};
 };
drivers/net/dsa/sja1105/sja1105_flower.c +85 -26

179 lines not on this page · raw message

Signature · 1 line
2.17.1
author maintainer
#7 lore raw

PATCH 4/6 net: dsa: sja1105: support flow-based redirection via virtual links

In reply to Vladimir Oltean

From: Vladimir Oltean <[email protected]>

Implement tc-flower offloads for redirect, trap and drop using non-critical virtual links.

Commands which were tested to work are:

  # Send frames received on swp2 with a DA of 42:be:24:9b:76:20 to the
  # CPU and to swp3. This type of key (DA only) when the port's VLAN
  # awareness state is off.
  tc qdisc add dev swp2 clsact
  tc filter add dev swp2 ingress flower skip_sw dst_mac 42:be:24:9b:76:20 \
          action mirred egress redirect dev swp3 \
          action trap

  # Drop frames received on swp2 with a DA of 42:be:24:9b:76:20, a VID
  # of 100 and a PCP of 0.
  tc filter add dev swp2 ingress protocol 802.1Q flower skip_sw \
          dst_mac 42:be:24:9b:76:20 vlan_id 100 vlan_prio 0 action drop

Under the hood, all rules match on DMAC, VID and PCP, but when VLAN filtering is disabled, those are set internally by the driver to the port-based defaults. Because we would be put in an awkward situation if the user were to change the VLAN filtering state while there are active rules (packets would no longer match on the specified keys), we simply deny changing vlan_filtering unless the list of flows offloaded via virtual links is empty. Then the user can re-add new rules.

---
Changes from RFC:
None.

 drivers/net/dsa/sja1105/Kconfig          |   9 +
 drivers/net/dsa/sja1105/Makefile         |   4 +
 drivers/net/dsa/sja1105/sja1105.h        |  18 ++
 drivers/net/dsa/sja1105/sja1105_flower.c |  57 ++++-
 drivers/net/dsa/sja1105/sja1105_main.c   |  12 +-
 drivers/net/dsa/sja1105/sja1105_vl.c     | 302 +++++++++++++++++++++++
 drivers/net/dsa/sja1105/sja1105_vl.h     |  41 +++
 7 files changed, 437 insertions(+), 6 deletions(-)
 create mode 100644 drivers/net/dsa/sja1105/sja1105_vl.c
 create mode 100644 drivers/net/dsa/sja1105/sja1105_vl.h
drivers/net/dsa/sja1105/Kconfig +9 -0
index 0fe1ae173aa1..bc59c3fbab50 100644
@@ -33,3 +33,12 @@ config NET_DSA_SJA1105_TAS
 	  This enables support for the TTEthernet-based egress scheduling
 	  engine in the SJA1105 DSA driver, which is controlled using a
 	  hardware offload of the tc-tqprio qdisc.
+
+config NET_DSA_SJA1105_VL
+	bool "Support for Virtual Links on NXP SJA1105"
+	depends on NET_DSA_SJA1105_TAS
+	help
+	  This enables support for flow classification using capable devices
+	  (SJA1105T, SJA1105Q, SJA1105S). The following actions are supported:
+	  - redirect, trap, drop
+	  - time-based ingress policing, via the tc-gate action
drivers/net/dsa/sja1105/Makefile +4 -0
index 8943d8d66f2b..c88e56a29db8 100644
@@ -17,3 +17,7 @@ endif
 ifdef CONFIG_NET_DSA_SJA1105_TAS
 sja1105-objs += sja1105_tas.o
 endif
+
+ifdef CONFIG_NET_DSA_SJA1105_VL
+sja1105-objs += sja1105_vl.o
+endif
drivers/net/dsa/sja1105/sja1105.h +18 -0
index 95633ad9bfb7..1756000f6936 100644
@@ -126,6 +126,13 @@ struct sja1105_key {
 enum sja1105_rule_type {
 	SJA1105_RULE_BCAST_POLICER,
 	SJA1105_RULE_TC_POLICER,
+	SJA1105_RULE_VL,
+};
+
+enum sja1105_vl_type {
+	SJA1105_VL_NONCRITICAL,
+	SJA1105_VL_RATE_CONSTRAINED,
+	SJA1105_VL_TIME_TRIGGERED,
 };
 
 struct sja1105_rule {
@@ -135,6 +142,7 @@ struct sja1105_rule {
 	struct sja1105_key key;
 	enum sja1105_rule_type type;
 
+	/* Action */
 	union {
 		/* SJA1105_RULE_BCAST_POLICER */
 		struct {
@@ -145,12 +153,19 @@ struct sja1105_rule {
 		struct {
 			int sharindx;
 		} tc_pol;
+
+		/* SJA1105_RULE_VL */
+		struct {
+			unsigned long destports;
+			enum sja1105_vl_type type;
+		} vl;
 	};
 };
 
 struct sja1105_flow_block {
 	struct list_head rules;
 	bool l2_policer_used[SJA1105_NUM_L2_POLICERS];
+	int num_virtual_links;
 };
 
 struct sja1105_private {
@@ -187,6 +202,7 @@ enum sja1105_reset_reason {
 	SJA1105_AGEING_TIME,
 	SJA1105_SCHEDULING,
 	SJA1105_BEST_EFFORT_POLICING,
+	SJA1105_VIRTUAL_LINKS,
 };
 
 int sja1105_static_config_reload(struct sja1105_private *priv,
@@ -290,5 +306,7 @@ int sja1105_cls_flower_add(struct dsa_switch *ds, int port,
 			   struct flow_cls_offload *cls, bool ingress);
 void sja1105_flower_setup(struct dsa_switch *ds);
 void sja1105_flower_teardown(struct dsa_switch *ds);
+struct sja1105_rule *sja1105_rule_find(struct sja1105_private *priv,
+				       unsigned long cookie);
 
 #endif
drivers/net/dsa/sja1105/sja1105_flower.c +52 -5

99 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_main.c +11 -1
index 472f4eb20c49..8bb104ee73d5 100644
@@ -445,7 +445,7 @@ static int sja1105_init_general_params(struct sja1105_private *priv)
 		 */
 		.casc_port = SJA1105_NUM_PORTS,
 		/* No TTEthernet */
-		.vllupformat = 0,
+		.vllupformat = SJA1105_VL_FORMAT_PSFP,
 		.vlmarker = 0,
 		.vlmask = 0,
 		/* Only update correctionField for 1-step PTP (L2 transport) */
@@ -1589,6 +1589,7 @@ static const char * const sja1105_reset_reasons[] = {
 	[SJA1105_AGEING_TIME] = "Ageing time",
 	[SJA1105_SCHEDULING] = "Time-aware scheduling",
 	[SJA1105_BEST_EFFORT_POLICING] = "Best-effort policing",
+	[SJA1105_VIRTUAL_LINKS] = "Virtual links",
 };
 
 /* For situations where we need to change a setting at runtime that is only
@@ -1831,9 +1832,18 @@ static int sja1105_vlan_filtering(struct dsa_switch *ds, int port, bool enabled)
 	struct sja1105_general_params_entry *general_params;
 	struct sja1105_private *priv = ds->priv;
 	struct sja1105_table *table;
+	struct sja1105_rule *rule;
 	u16 tpid, tpid2;
 	int rc;
 
+	list_for_each_entry(rule, &priv->flow_block.rules, list) {
+		if (rule->type == SJA1105_RULE_VL) {
+			dev_err(ds->dev,
+				"Cannot change VLAN filtering state while VL rules are active\n");
+			return -EBUSY;
+		}
+	}
+
 	if (enabled) {
 		/* Enable VLAN filtering. */
 		tpid  = ETH_P_8021Q;
drivers/net/dsa/sja1105/sja1105_vl.c new file +302 -0

302 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_vl.h new file +41 -0

41 lines not on this page · raw message

Signature · 1 line
2.17.1
#8 lore raw

Re: [PATCH net-next 4/6] net: dsa: sja1105: support flow-based redirection via virtual links

Hi Vladimir,

drivers/net/dsa/sja1105/sja1105_flower.c int sja1105_cls_flower_add(struct dsa_switch *ds, int port, · line 352

Quoted 10 lines · Vladimir Oltean
drivers/net/dsa/sja1105/sja1105_flower.c +10 -0
+		case FLOW_ACTION_REDIRECT: {
+			struct dsa_port *to_dp;
+
+			if (!dsa_slave_dev_check(act->dev)) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "Destination not a switch port");
+				return -EOPNOTSUPP;
+			}
+
+			to_dp = dsa_slave_to_port(act->dev);

Instead of exporting two DSA core internal functions, I would rather expose a new helper for drivers, such as this one:

    struct dsa_port *dsa_dev_to_port(struct net_device *dev)
    {
        if (!dsa_slave_dev_check(dev))
            return -EOPNOTSUPP;
    
        return dsa_slave_to_port(dev);
    }

The naming might not be the best, this helper could even be mirroring-specific, I didn't really check the requirements for this functionality yet.

Thank you,

Vivien

#9 lore raw

Re: [PATCH net-next 4/6] net: dsa: sja1105: support flow-based redirection via virtual links

Quoted 17 lines · Vivien Didelot

Hi Vladimir,

drivers/net/dsa/sja1105/sja1105_flower.c int sja1105_cls_flower_add(struct dsa_switch *ds, int port, · line 352

Quoted 10 lines · Vladimir Oltean
drivers/net/dsa/sja1105/sja1105_flower.c +10 -0
+		case FLOW_ACTION_REDIRECT: {
+			struct dsa_port *to_dp;
+
+			if (!dsa_slave_dev_check(act->dev)) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "Destination not a switch port");
+				return -EOPNOTSUPP;
+			}
+
+			to_dp = dsa_slave_to_port(act->dev);

Instead of exporting two DSA core internal functions, I would rather expose a new helper for drivers, such as this one:

Quoted 4 lines
    struct dsa_port *dsa_dev_to_port(struct net_device *dev)
    {
        if (!dsa_slave_dev_check(dev))
            return -EOPNOTSUPP;

Oops, NULL, not an integer error code, but you get the idea of public helpers.

Quoted 11 lines
    
        return dsa_slave_to_port(dev);
    }

The naming might not be the best, this helper could even be mirroring-specific, I didn't really check the requirements for this functionality yet.

Thank you,

Vivien

author maintainer
#10 lore raw

Re: [PATCH net-next 4/6] net: dsa: sja1105: support flow-based redirection via virtual links

Hi Vivien,

Quoted 19 lines · Vivien Didelot
Quoted 17 lines · Vivien Didelot

Hi Vladimir,

drivers/net/dsa/sja1105/sja1105_flower.c int sja1105_cls_flower_add(struct dsa_switch *ds, int port, · line 352

Quoted 10 lines · Vladimir Oltean
drivers/net/dsa/sja1105/sja1105_flower.c +10 -0
+           case FLOW_ACTION_REDIRECT: {
+                   struct dsa_port *to_dp;
+
+                   if (!dsa_slave_dev_check(act->dev)) {
+                           NL_SET_ERR_MSG_MOD(extack,
+                                              "Destination not a switch port");
+                           return -EOPNOTSUPP;
+                   }
+
+                   to_dp = dsa_slave_to_port(act->dev);

Instead of exporting two DSA core internal functions, I would rather expose a new helper for drivers, such as this one:

Quoted 18 lines
Quoted 4 lines
    struct dsa_port *dsa_dev_to_port(struct net_device *dev)
    {
        if (!dsa_slave_dev_check(dev))
            return -EOPNOTSUPP;

Oops, NULL, not an integer error code, but you get the idea of public helpers.

Quoted 11 lines
        return dsa_slave_to_port(dev);
    }

The naming might not be the best, this helper could even be mirroring-specific, I didn't really check the requirements for this functionality yet.

Thank you,

Vivien

How about

int dsa_slave_get_port_index(struct net_device *dev)

{
    if (!dsa_slave_dev_check(dev))
        return -EINVAL;

    return dsa_slave_to_port(dev)->index;
}

EXPORT_SYMBOL_GPL(dsa_slave_get_port_index);

also, where to put it? slave.c I suppose?

Thanks,
-Vladimir

#11 lore raw

Re: [PATCH net-next 4/6] net: dsa: sja1105: support flow-based redirection via virtual links

Hi Vladimir,

Quoted 50 lines · Vladimir Oltean

Hi Vivien,

Quoted 19 lines · Vivien Didelot
Quoted 17 lines · Vivien Didelot

Hi Vladimir,

drivers/net/dsa/sja1105/sja1105_flower.c int sja1105_cls_flower_add(struct dsa_switch *ds, int port, · line 352

Quoted 10 lines · Vladimir Oltean
drivers/net/dsa/sja1105/sja1105_flower.c +10 -0
+           case FLOW_ACTION_REDIRECT: {
+                   struct dsa_port *to_dp;
+
+                   if (!dsa_slave_dev_check(act->dev)) {
+                           NL_SET_ERR_MSG_MOD(extack,
+                                              "Destination not a switch port");
+                           return -EOPNOTSUPP;
+                   }
+
+                   to_dp = dsa_slave_to_port(act->dev);

Instead of exporting two DSA core internal functions, I would rather expose a new helper for drivers, such as this one:

Quoted 18 lines
Quoted 4 lines
    struct dsa_port *dsa_dev_to_port(struct net_device *dev)
    {
        if (!dsa_slave_dev_check(dev))
            return -EOPNOTSUPP;

Oops, NULL, not an integer error code, but you get the idea of public helpers.

Quoted 11 lines
        return dsa_slave_to_port(dev);
    }

The naming might not be the best, this helper could even be mirroring-specific, I didn't really check the requirements for this functionality yet.

Thank you,

Vivien

How about

int dsa_slave_get_port_index(struct net_device *dev)

{
    if (!dsa_slave_dev_check(dev))
        return -EINVAL;

    return dsa_slave_to_port(dev)->index;
}
Quoted 3 lines

EXPORT_SYMBOL_GPL(dsa_slave_get_port_index);

also, where to put it? slave.c I suppose?

dsa.c is the place for private implementation of public functions. "slave" is a core term, no need to expose it. Public helpers exposed in dsa.h usually scope the dsa_switch structure and an optional port index. mv88e6xxx allows mirroring an external device port, so dsa_port would be preferred, but this can wait. So I'm thinking about implementing the following:

net/dsa/dsa.c:

int dsa_to_port_index(struct dsa_switch *ds, struct net_device *dev)

{
    struct dsa_port *dp;

    if (!dsa_slave_dev_check(dev))
        return -ENODEV;

    dp = dsa_slave_to_port(dev);

    if (dp->ds != ds)
        return -EINVAL;

    return dp->index;
}

include/net/dsa.h:

int dsa_to_port_index(struct dsa_switch *ds, struct net_device *dev);

What do you think?

Vivien

author maintainer
#12 lore raw

Re: [PATCH net-next 4/6] net: dsa: sja1105: support flow-based redirection via virtual links

Quoted 56 lines · Vivien Didelot

Hi Vladimir,

Quoted 52 lines · Vladimir Oltean

Hi Vivien,

Quoted 19 lines · Vivien Didelot
Quoted 17 lines · Vivien Didelot

Hi Vladimir,

drivers/net/dsa/sja1105/sja1105_flower.c int sja1105_cls_flower_add(struct dsa_switch *ds, int port, · line 352

Quoted 10 lines · Vladimir Oltean
drivers/net/dsa/sja1105/sja1105_flower.c +10 -0
+           case FLOW_ACTION_REDIRECT: {
+                   struct dsa_port *to_dp;
+
+                   if (!dsa_slave_dev_check(act->dev)) {
+                           NL_SET_ERR_MSG_MOD(extack,
+                                              "Destination not a switch port");
+                           return -EOPNOTSUPP;
+                   }
+
+                   to_dp = dsa_slave_to_port(act->dev);

Instead of exporting two DSA core internal functions, I would rather expose a new helper for drivers, such as this one:

Quoted 18 lines
Quoted 4 lines
    struct dsa_port *dsa_dev_to_port(struct net_device *dev)
    {
        if (!dsa_slave_dev_check(dev))
            return -EOPNOTSUPP;

Oops, NULL, not an integer error code, but you get the idea of public helpers.

Quoted 11 lines
        return dsa_slave_to_port(dev);
    }

The naming might not be the best, this helper could even be mirroring-specific, I didn't really check the requirements for this functionality yet.

Thank you,

Vivien

How about

int dsa_slave_get_port_index(struct net_device *dev)

{
    if (!dsa_slave_dev_check(dev))
        return -EINVAL;

    return dsa_slave_to_port(dev)->index;
}

EXPORT_SYMBOL_GPL(dsa_slave_get_port_index);

Quoted 6 lines
Quoted 1 line

also, where to put it? slave.c I suppose?

dsa.c is the place for private implementation of public functions. "slave" is a core term, no need to expose it. Public helpers exposed in dsa.h usually scope the dsa_switch structure and an optional port index. mv88e6xxx allows mirroring an external device port,

For mirroring an entire port (via tc-matchall), the tc structures are already parsed by DSA core and a simple API is given to drivers. The discussion we're having is for flow-based mirroring (via tc-flower) where that is not the case.

Quoted 6 lines

so dsa_port would be preferred, but this
can wait. So I'm thinking about implementing the following:

net/dsa/dsa.c:

int dsa_to_port_index(struct dsa_switch *ds, struct net_device *dev)

But let's assume for a second that mv88e6xxx supports flow-based mirroring/redirection too.
Aren't we limiting ourselves uselessly here, by requiring the caller to pass a ds pointer just to perform validation on it? I think it's a valid use case to want to support cross-chip mirroring/redirection sometime in the future. Both sja1105 and mv88e6xxx support that kind of setup, you just need to set the destination port to dsa_towards_port() in case the dp->ds found by dsa_slave_to_port does not coincide with ours. But surprise, using the syntactic sugar API we're introducing here, we'd get -EINVAL and we would have to somehow try again and guess with a ds pointer we don't have.

Quoted 14 lines
{
    struct dsa_port *dp;

    if (!dsa_slave_dev_check(dev))
        return -ENODEV;

    dp = dsa_slave_to_port(dev);

    if (dp->ds != ds)
        return -EINVAL;

    return dp->index;
}
Quoted 6 lines

include/net/dsa.h:

int dsa_to_port_index(struct dsa_switch *ds, struct net_device *dev);

What do you think?

I'm actually not convinced about this idea. I think the function that should be called should be named dsa_slave_to_port, and it should return a struct dsa_port. Quite conveniently, that function already exists. I'm not actually sure what are the issues of exposing the existing functions.

Quoted 2 lines

Vivien

Thanks,
-Vladimir

author maintainer
#13 lore raw

PATCH 5/6 net: dsa: sja1105: implement tc-gate using time-triggered virtual links

In reply to Vladimir Oltean

From: Vladimir Oltean <[email protected]>

Restrict the TTEthernet hardware support on this switch to operate as closely as possible to IEEE 802.1Qci as possible. This means that it can perform PTP-time-based ingress admission control on streams identified by {DMAC, VID, PCP}, which is useful when trying to ensure the determinism of traffic scheduled via IEEE 802.1Qbv.

The oddity comes from the fact that in hardware (and in TTEthernet at large), virtual links always need a full-blown action, including not only the type of policing, but also the list of destination ports. So in practice, a single tc-gate action will result in all packets getting dropped. Additional actions (either "trap" or "redirect") need to be specified in the same filter rule such that the conforming packets are actually forwarded somewhere.

Apart from the VL Lookup, Policing and Forwarding tables which need to be programmed for each flow (virtual link), the Schedule engine also needs to be told to open/close the admission gates for each individual virtual link. A fairly accurate (and detailed) description of how that works is already present in sja1105_tas.c, since it is already used to trigger the egress gates for the tc-taprio offload (IEEE 802.1Qbv). Key point here, we remember that the schedule engine supports 8 "subschedules" (execution threads that iterate through the global schedule in parallel, and that no 2 hardware threads must execute a schedule entry at the same time). For tc-taprio, each egress port used one of these 8 subschedules, leaving a total of 4 subschedules unused. In principle we could have allocated 1 subschedule for the tc-gate offload of each ingress port, but actually the schedules of all virtual links installed on each ingress port would have needed to be merged together, before they could have been programmed to hardware. So simplify our life and just merge the entire tc-gate configuration, for all virtual links on all ingress ports, into a single subschedule. Be sure to check that against the usual hardware scheduling conflicts, and program it to hardware alongside any tc-taprio subschedule that may be present.

The following scenarios were tested:

1. Quantitative testing:

   tc qdisc add dev swp2 clsact
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate index 1 base-time 0 \
           sched-entry OPEN 1200 -1 -1 \
           sched-entry CLOSE 1200 -1 -1 \
           action trap

ping 192.168.1.2 -f
PING 192.168.1.2 (192.168.1.2) 56(84) bytes of data.
.............................
--- 192.168.1.2 ping statistics ---
948 packets transmitted, 467 received, 50.7384% packet loss, time 9671ms

2. Qualitative testing (with a phase-aligned schedule - the clocks are synchronized by ptp4l, not shown here):

Receiver (sja1105):

   tc qdisc add dev swp2 clsact
   now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc filter add dev swp2 ingress flower skip_sw \
           dst_mac 42:be:24:9b:76:20 \
           action gate base-time ${base_time} \
           sched-entry OPEN  60000 -1 -1 \
           sched-entry CLOSE 40000 -1 -1 \
           action trap

   Sender (enetc):
   now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
           sec=$(echo $now | awk -F. '{print $1}') && \
           base_time="$(((sec + 2) * 1000000000))" && \
           echo "base time ${base_time}"
   tc qdisc add dev eno0 parent root taprio \
           num_tc 8 \
           map 0 1 2 3 4 5 6 7 \
           queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
           base-time ${base_time} \
           sched-entry S 01  50000 \
           sched-entry S 00  50000 \
           flags 2

ping -A 192.168.1.1
PING 192.168.1.1 (192.168.1.1): 56 data bytes
...
^C
--- 192.168.1.1 ping statistics ---
1425 packets transmitted, 1424 packets received, 0% packet loss round-trip min/avg/max = 0.322/0.361/0.990 ms

And just for comparison, with the tc-taprio schedule deleted:

ping -A 192.168.1.1
PING 192.168.1.1 (192.168.1.1): 56 data bytes
...
^C
--- 192.168.1.1 ping statistics ---
33 packets transmitted, 19 packets received, 42% packet loss round-trip min/avg/max = 0.336/0.464/0.597 ms

---
Changes from RFC:
* Deny gating actions in absence of any routing action. Without this check, the default behavior would be to drop everything, which is probably not what the user expects.
* Deny offloading IntervalOctetMax, which I initially mistook for meaning per-flow MTU. It actually means a sort of flow metering within each time slot. Not supported.

 drivers/net/dsa/sja1105/sja1105.h             |  13 +-
 drivers/net/dsa/sja1105/sja1105_flower.c      |  57 +-
 drivers/net/dsa/sja1105/sja1105_main.c        |   1 +
 drivers/net/dsa/sja1105/sja1105_ptp.h         |  13 +
 drivers/net/dsa/sja1105/sja1105_spi.c         |   2 +
 .../net/dsa/sja1105/sja1105_static_config.h   |   2 +
 drivers/net/dsa/sja1105/sja1105_tas.c         | 127 ++++-
 drivers/net/dsa/sja1105/sja1105_tas.h         |  31 ++
 drivers/net/dsa/sja1105/sja1105_vl.c          | 494 ++++++++++++++++++
 drivers/net/dsa/sja1105/sja1105_vl.h          |  31 ++
 10 files changed, 754 insertions(+), 17 deletions(-)
drivers/net/dsa/sja1105/sja1105.h +12 -1
index 1756000f6936..8df2a5c53b02 100644
@@ -36,6 +36,7 @@ struct sja1105_regs {
 	u64 status;
 	u64 port_control;
 	u64 rgu;
+	u64 vl_status;
 	u64 config;
 	u64 sgmii;
 	u64 rmii_pll1;
@@ -156,8 +157,16 @@ struct sja1105_rule {
 
 		/* SJA1105_RULE_VL */
 		struct {
-			unsigned long destports;
 			enum sja1105_vl_type type;
+			unsigned long destports;
+			int sharindx;
+			int maxlen;
+			int ipv;
+			u64 base_time;
+			u64 cycle_time;
+			int num_entries;
+			struct action_gate_entry *entries;
+			struct flow_stats stats;
 		} vl;
 	};
 };
@@ -304,6 +313,8 @@ int sja1105_cls_flower_del(struct dsa_switch *ds, int port,
 			   struct flow_cls_offload *cls, bool ingress);
 int sja1105_cls_flower_add(struct dsa_switch *ds, int port,
 			   struct flow_cls_offload *cls, bool ingress);
+int sja1105_cls_flower_stats(struct dsa_switch *ds, int port,
+			     struct flow_cls_offload *cls, bool ingress);
 void sja1105_flower_setup(struct dsa_switch *ds);
 void sja1105_flower_teardown(struct dsa_switch *ds);
 struct sja1105_rule *sja1105_rule_find(struct sja1105_private *priv,
drivers/net/dsa/sja1105/sja1105_flower.c +56 -1

95 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_main.c +1 -0
index 8bb104ee73d5..666e54565df0 100644
@@ -2369,6 +2369,7 @@ static const struct dsa_switch_ops sja1105_switch_ops = {
 	.port_policer_del	= sja1105_port_policer_del,
 	.cls_flower_add		= sja1105_cls_flower_add,
 	.cls_flower_del		= sja1105_cls_flower_del,
+	.cls_flower_stats	= sja1105_cls_flower_stats,
 };
 
 static int sja1105_check_device_id(struct sja1105_private *priv)
drivers/net/dsa/sja1105/sja1105_ptp.h +13 -0
index 43480b24f1f0..6408d1158f2d 100644
@@ -48,6 +48,19 @@ static inline s64 future_base_time(s64 base_time, s64 cycle_time, s64 now)
 	return base_time + n * cycle_time;
 }
 
+/* This is not a preprocessor macro because the "ns" argument may or may not be
+ * s64 at caller side. This ensures it is properly type-cast before div_s64.
+ */
+static inline s64 ns_to_sja1105_delta(s64 ns)
+{
+	return div_s64(ns, 200);
+}
+
+static inline s64 sja1105_delta_to_ns(s64 delta)
+{
+	return delta * 200;
+}
+
 struct sja1105_ptp_cmd {
 	u64 startptpcp;		/* start toggling PTP_CLK pin */
 	u64 stopptpcp;		/* stop toggling PTP_CLK pin */
drivers/net/dsa/sja1105/sja1105_spi.c +2 -0
index 43f14a5c2718..0be75c49e6c3 100644
@@ -439,6 +439,7 @@ static struct sja1105_regs sja1105et_regs = {
 	.prod_id = 0x100BC3,
 	.status = 0x1,
 	.port_control = 0x11,
+	.vl_status = 0x10000,
 	.config = 0x020000,
 	.rgu = 0x100440,
 	/* UM10944.pdf, Table 86, ACU Register overview */
@@ -472,6 +473,7 @@ static struct sja1105_regs sja1105pqrs_regs = {
 	.prod_id = 0x100BC3,
 	.status = 0x1,
 	.port_control = 0x12,
+	.vl_status = 0x10000,
 	.config = 0x020000,
 	.rgu = 0x100440,
 	/* UM10944.pdf, Table 86, ACU Register overview */
drivers/net/dsa/sja1105/sja1105_static_config.h +2 -0
index 1a8fcbbb57b6..b569e3de3590 100644
@@ -302,6 +302,8 @@ struct sja1105_vl_lookup_entry {
 			u64 vlid;
 		};
 	};
+	/* Not part of hardware structure */
+	unsigned long flow_cookie;
 };
 
 struct sja1105_vl_policing_entry {
drivers/net/dsa/sja1105/sja1105_tas.c +112 -15

190 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_tas.h +31 -0
index b226c3dfd5b1..2dc1856d403d 100644
@@ -6,6 +6,8 @@
 
 #include <net/pkt_sched.h>
 
+#define SJA1105_TAS_MAX_DELTA		BIT(18)
+
 #if IS_ENABLED(CONFIG_NET_DSA_SJA1105_TAS)
 
 enum sja1105_tas_state {
@@ -20,8 +22,23 @@ enum sja1105_ptp_op {
 	SJA1105_PTP_ADJUSTFREQ,
 };
 
+struct sja1105_gate_entry {
+	struct list_head list;
+	struct sja1105_rule *rule;
+	s64 interval;
+	u8 gate_state;
+};
+
+struct sja1105_gating_config {
+	u64 cycle_time;
+	s64 base_time;
+	int num_entries;
+	struct list_head entries;
+};
+
 struct sja1105_tas_data {
 	struct tc_taprio_qopt_offload *offload[SJA1105_NUM_PORTS];
+	struct sja1105_gating_config gating_cfg;
 	enum sja1105_tas_state state;
 	enum sja1105_ptp_op last_op;
 	struct work_struct tas_work;
@@ -31,6 +48,8 @@ struct sja1105_tas_data {
 	bool enabled;
 };
 
+struct sja1105_private;
+
 int sja1105_setup_tc_taprio(struct dsa_switch *ds, int port,
 			    struct tc_taprio_qopt_offload *admin);
 
@@ -42,6 +61,11 @@ void sja1105_tas_clockstep(struct dsa_switch *ds);
 
 void sja1105_tas_adjfreq(struct dsa_switch *ds);
 
+bool sja1105_gating_check_conflicts(struct sja1105_private *priv, int port,
+				    struct netlink_ext_ack *extack);
+
+int sja1105_init_scheduling(struct sja1105_private *priv);
+
 #else
 
 /* C doesn't allow empty structures, bah! */
@@ -63,6 +87,13 @@ static inline void sja1105_tas_clockstep(struct dsa_switch *ds) { }
 
 static inline void sja1105_tas_adjfreq(struct dsa_switch *ds) { }
 
+static inline bool
+sja1105_gating_check_conflicts(struct dsa_switch *ds, int port,
+			       struct netlink_ext_ack *extack)
+{
+	return true;
+}
+
 #endif /* IS_ENABLED(CONFIG_NET_DSA_SJA1105_TAS) */
 
 #endif /* _SJA1105_TAS_H */
drivers/net/dsa/sja1105/sja1105_vl.c +494 -0

548 lines not on this page · raw message

drivers/net/dsa/sja1105/sja1105_vl.h +31 -0
index 08ee5557b463..323fa0535af7 100644
@@ -15,6 +15,16 @@ int sja1105_vl_delete(struct sja1105_private *priv, int port,
 		      struct sja1105_rule *rule,
 		      struct netlink_ext_ack *extack);
 
+int sja1105_vl_gate(struct sja1105_private *priv, int port,
+		    struct netlink_ext_ack *extack, unsigned long cookie,
+		    struct sja1105_key *key, u32 index, s32 prio,
+		    u64 base_time, u64 cycle_time, u64 cycle_time_ext,
+		    u32 num_entries, struct action_gate_entry *entries);
+
+int sja1105_vl_stats(struct sja1105_private *priv, int port,
+		     struct sja1105_rule *rule, struct flow_stats *stats,
+		     struct netlink_ext_ack *extack);
+
 #else
 
 static inline int sja1105_vl_redirect(struct sja1105_private *priv, int port,
@@ -36,6 +46,27 @@ static inline int sja1105_vl_delete(struct sja1105_private *priv,
 	return -EOPNOTSUPP;
 }
 
+static inline int sja1105_vl_gate(struct sja1105_private *priv, int port,
+				  struct netlink_ext_ack *extack,
+				  unsigned long cookie,
+				  struct sja1105_key *key, u32 index, s32 prio,
+				  u64 base_time, u64 cycle_time,
+				  u64 cycle_time_ext, u32 num_entries,
+				  struct action_gate_entry *entries)
+{
+	NL_SET_ERR_MSG_MOD(extack, "Virtual Links not compiled in");
+	return -EOPNOTSUPP;
+}
+
+static inline int sja1105_vl_stats(struct sja1105_private *priv, int port,
+				   struct sja1105_rule *rule,
+				   struct flow_stats *stats,
+				   struct netlink_ext_ack *extack)
+{
+	NL_SET_ERR_MSG_MOD(extack, "Virtual Links not compiled in");
+	return -EOPNOTSUPP;
+}
+
 #endif /* IS_ENABLED(CONFIG_NET_DSA_SJA1105_VL) */
 
 #endif /* _SJA1105_VL_H */
Signature · 1 line
2.17.1
#14 lore raw

Re: [PATCH net-next 5/6] net: dsa: sja1105: implement tc-gate using time-triggered virtual links

I love your patch! Yet something to improve:

Hi Vladimir,

I love your patch! Yet something to improve:

[auto build test ERROR on net-next/master]
[also build test ERROR on net/master linus/master v5.7-rc4 next-20200505] [if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Vladimir-Oltean/tc-gate-offload-for-SJA1105-DSA-switch/20200505-040345
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git 627642f07b3093f501495d226c7a0b9d56a0c870
config: i386-randconfig-h001-20200503 (attached as .config)
compiler: gcc-7 (Ubuntu 7.5.0-6ubuntu2) 7.5.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

drivers/net/dsa/sja1105/sja1105_flower.c: In function 'sja1105_cls_flower_add':

>> drivers/net/dsa/sja1105/sja1105_flower.c:406:9: error: implicit declaration of function 'sja1105_init_scheduling'; did you mean 'sja1105_get_strings'? [-Werror=implicit-function-declaration]
       rc = sja1105_init_scheduling(priv);
            ^~~~~~~~~~~~~~~~~~~~~~~
            sja1105_get_strings

cc1: some warnings being treated as errors

vim +406 drivers/net/dsa/sja1105/sja1105_flower.c

303
304 int sja1105_cls_flower_add(struct dsa_switch *ds, int port,
305 struct flow_cls_offload *cls, bool ingress)
306 {
307 struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
308 struct netlink_ext_ack *extack = cls->common.extack;
309 struct sja1105_private *priv = ds->priv;
310 const struct flow_action_entry *act;
311 unsigned long cookie = cls->cookie;
312 bool routing_rule = false;
313 struct sja1105_key key;
314 bool gate_rule = false;
315 bool vl_rule = false;
316 int rc, i;
317
318 rc = sja1105_flower_parse_key(priv, extack, cls, &key);
319 if (rc)
320 return rc;
321
322 rc = -EOPNOTSUPP;
323
324 flow_action_for_each(i, act, &rule->action) {
325 switch (act->id) {
326 case FLOW_ACTION_POLICE:
327 rc = sja1105_flower_policer(priv, port, extack, cookie,

   328							    &key,
   329							    act->police.rate_bytes_ps,
   330							    act->police.burst);
   331				if (rc)
   332					goto out;
   333				break;
   334			case FLOW_ACTION_TRAP: {
   335				int cpu = dsa_upstream_port(ds, port);
   336	
   337				routing_rule = true;
   338				vl_rule = true;
   339	
   340				rc = sja1105_vl_redirect(priv, port, extack, cookie,
   341							 &key, BIT(cpu), true);
   342				if (rc)
   343					goto out;
   344				break;
   345			}
   346			case FLOW_ACTION_REDIRECT: {
   347				struct dsa_port *to_dp;
   348	
   349				if (!dsa_slave_dev_check(act->dev)) {
   350					NL_SET_ERR_MSG_MOD(extack,
   351							   "Destination not a switch port");
   352					return -EOPNOTSUPP;
   353				}
   354	
   355				to_dp = dsa_slave_to_port(act->dev);
   356				routing_rule = true;
   357				vl_rule = true;
   358	
   359				rc = sja1105_vl_redirect(priv, port, extack, cookie,
   360							 &key, BIT(to_dp->index), true);
   361				if (rc)
   362					goto out;
   363				break;
   364			}
   365			case FLOW_ACTION_DROP:
   366				vl_rule = true;
   367	
   368				rc = sja1105_vl_redirect(priv, port, extack, cookie,
   369							 &key, 0, false);
   370				if (rc)
   371					goto out;
   372				break;
   373			case FLOW_ACTION_GATE:
   374				gate_rule = true;
   375				vl_rule = true;
   376	
   377				rc = sja1105_vl_gate(priv, port, extack, cookie,
   378						     &key, act->gate.index,
   379						     act->gate.prio,
   380						     act->gate.basetime,
   381						     act->gate.cycletime,
   382						     act->gate.cycletimeext,
   383						     act->gate.num_entries,
   384						     act->gate.entries);
   385				if (rc)
   386					goto out;
   387				break;
   388			default:
   389				NL_SET_ERR_MSG_MOD(extack,
   390						   "Action not supported");
   391				rc = -EOPNOTSUPP;
   392				goto out;
   393			}
   394		}
   395	
   396		if (vl_rule && !rc) {
   397			/* Delay scheduling configuration until DESTPORTS has been
   398			 * populated by all other actions.
   399			 */
   400			if (gate_rule) {
   401				if (!routing_rule) {
   402					NL_SET_ERR_MSG_MOD(extack,
   403							   "Can only offload gate action together with redirect or trap");
   404					return -EOPNOTSUPP;
   405				}
 > 406				rc = sja1105_init_scheduling(priv);

407 if (rc)
408 goto out;
409 }
410
411 rc = sja1105_static_config_reload(priv, SJA1105_VIRTUAL_LINKS);
412 }
413
414 out:
415 return rc;
416 }
417

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

author maintainer
#15 lore raw

Re: [PATCH net-next 5/6] net: dsa: sja1105: implement tc-gate using time-triggered virtual links

Quoted 149 lines · kbuild test robot

Hi Vladimir,

I love your patch! Yet something to improve:

[auto build test ERROR on net-next/master]
[also build test ERROR on net/master linus/master v5.7-rc4 next-20200505] [if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Vladimir-Oltean/tc-gate-offload-for-SJA1105-DSA-switch/20200505-040345
base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git 627642f07b3093f501495d226c7a0b9d56a0c870
config: i386-randconfig-h001-20200503 (attached as .config)
compiler: gcc-7 (Ubuntu 7.5.0-6ubuntu2) 7.5.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

   drivers/net/dsa/sja1105/sja1105_flower.c: In function 'sja1105_cls_flower_add':
Quoted 1 line
Quoted 1 line

drivers/net/dsa/sja1105/sja1105_flower.c:406:9: error: implicit declaration of function 'sja1105_init_scheduling'; did you mean 'sja1105_get_strings'? [-Werror=implicit-function-declaration]

       rc = sja1105_init_scheduling(priv);
            ^~~~~~~~~~~~~~~~~~~~~~~
            sja1105_get_strings

cc1: some warnings being treated as errors

vim +406 drivers/net/dsa/sja1105/sja1105_flower.c

   303
   304  int sja1105_cls_flower_add(struct dsa_switch *ds, int port,
   305                             struct flow_cls_offload *cls, bool ingress)
   306  {
   307          struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
   308          struct netlink_ext_ack *extack = cls->common.extack;
   309          struct sja1105_private *priv = ds->priv;
   310          const struct flow_action_entry *act;
   311          unsigned long cookie = cls->cookie;
   312          bool routing_rule = false;
   313          struct sja1105_key key;
   314          bool gate_rule = false;
   315          bool vl_rule = false;
   316          int rc, i;
   317
   318          rc = sja1105_flower_parse_key(priv, extack, cls, &key);
   319          if (rc)
   320                  return rc;
   321
   322          rc = -EOPNOTSUPP;
   323
   324          flow_action_for_each(i, act, &rule->action) {
   325                  switch (act->id) {
   326                  case FLOW_ACTION_POLICE:
   327                          rc = sja1105_flower_policer(priv, port, extack, cookie,
   328                                                      &key,
   329                                                      act->police.rate_bytes_ps,
   330                                                      act->police.burst);
   331                          if (rc)
   332                                  goto out;
   333                          break;
   334                  case FLOW_ACTION_TRAP: {
   335                          int cpu = dsa_upstream_port(ds, port);
   336
   337                          routing_rule = true;
   338                          vl_rule = true;
   339
   340                          rc = sja1105_vl_redirect(priv, port, extack, cookie,
   341                                                   &key, BIT(cpu), true);
   342                          if (rc)
   343                                  goto out;
   344                          break;
   345                  }
   346                  case FLOW_ACTION_REDIRECT: {
   347                          struct dsa_port *to_dp;
   348
   349                          if (!dsa_slave_dev_check(act->dev)) {
   350                                  NL_SET_ERR_MSG_MOD(extack,
   351                                                     "Destination not a switch port");
   352                                  return -EOPNOTSUPP;
   353                          }
   354
   355                          to_dp = dsa_slave_to_port(act->dev);
   356                          routing_rule = true;
   357                          vl_rule = true;
   358
   359                          rc = sja1105_vl_redirect(priv, port, extack, cookie,
   360                                                   &key, BIT(to_dp->index), true);
   361                          if (rc)
   362                                  goto out;
   363                          break;
   364                  }
   365                  case FLOW_ACTION_DROP:
   366                          vl_rule = true;
   367
   368                          rc = sja1105_vl_redirect(priv, port, extack, cookie,
   369                                                   &key, 0, false);
   370                          if (rc)
   371                                  goto out;
   372                          break;
   373                  case FLOW_ACTION_GATE:
   374                          gate_rule = true;
   375                          vl_rule = true;
   376
   377                          rc = sja1105_vl_gate(priv, port, extack, cookie,
   378                                               &key, act->gate.index,
   379                                               act->gate.prio,
   380                                               act->gate.basetime,
   381                                               act->gate.cycletime,
   382                                               act->gate.cycletimeext,
   383                                               act->gate.num_entries,
   384                                               act->gate.entries);
   385                          if (rc)
   386                                  goto out;
   387                          break;
   388                  default:
   389                          NL_SET_ERR_MSG_MOD(extack,
   390                                             "Action not supported");
   391                          rc = -EOPNOTSUPP;
   392                          goto out;
   393                  }
   394          }
   395
   396          if (vl_rule && !rc) {
   397                  /* Delay scheduling configuration until DESTPORTS has been
   398                   * populated by all other actions.
   399                   */
   400                  if (gate_rule) {
   401                          if (!routing_rule) {
   402                                  NL_SET_ERR_MSG_MOD(extack,
   403                                                     "Can only offload gate action together with redirect or trap");
   404                                  return -EOPNOTSUPP;
   405                          }
Quoted 1 line

406 rc = sja1105_init_scheduling(priv);

407 if (rc)

   408                                  goto out;
   409                  }
   410
   411                  rc = sja1105_static_config_reload(priv, SJA1105_VIRTUAL_LINKS);
   412          }
   413
   414  out:
   415          return rc;
   416  }
   417
Quoted 3 lines

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/[email protected]

Whoops, I forgot to provide a shim implementation sja1105_init_scheduling for the case where NET_DSA_SJA1105_TAS is not enabled.
If there are no other comments I'll send a v3 soon with just this change.

-Vladimir

author maintainer
#16 lore raw

PATCH 6/6 docs: net: dsa: sja1105: document intended usage of virtual links

In reply to Vladimir Oltean

From: Vladimir Oltean <[email protected]>

Add some verbiage describing how the hardware features of the switch are exposed to users through tc-flower.

---
Changes from RFC:
Patch is new.

 Documentation/networking/dsa/sja1105.rst | 116 +++++++++++++++++++++++
 1 file changed, 116 insertions(+)
Documentation/networking/dsa/sja1105.rst +116 -0
index 64553d8d91cb..34581629dd3f 100644
@@ -230,6 +230,122 @@ simultaneously on two ports. The driver checks the consistency of the schedules
 against this restriction and errors out when appropriate. Schedule analysis is
 needed to avoid this, which is outside the scope of the document.
 
+Routing actions (redirect, trap, drop)
+--------------------------------------
+
+The switch is able to offload flow-based redirection of packets to a set of
+destination ports specified by the user. Internally, this is implemented by
+making use of Virtual Links, a TTEthernet concept.
+
+The driver supports 2 types of keys for Virtual Links:
+
+- VLAN-aware virtual links: these match on destination MAC address, VLAN ID and
+  VLAN PCP.
+- VLAN-unaware virtual links: these match on destination MAC address only.
+
+The VLAN awareness state of the bridge (vlan_filtering) cannot be changed while
+there are virtual link rules installed.
+
+Composing multiple actions inside the same rule is supported. When only routing
+actions are requested, the driver creates a "non-critical" virtual link. When
+the action list also contains tc-gate (more details below), the virtual link
+becomes "time-critical" (draws frame buffers from a reserved memory partition,
+etc).
+
+The 3 routing actions that are supported are "trap", "drop" and "redirect".
+
+Example 1: send frames received on swp2 with a DA of 42:be:24:9b:76:20 to the
+CPU and to swp3. This type of key (DA only) when the port's VLAN awareness
+state is off::
+
+  tc qdisc add dev swp2 clsact
+  tc filter add dev swp2 ingress flower skip_sw dst_mac 42:be:24:9b:76:20 \
+          action mirred egress redirect dev swp3 \
+          action trap
+
+Example 2: drop frames received on swp2 with a DA of 42:be:24:9b:76:20, a VID
+of 100 and a PCP of 0::
+
+  tc filter add dev swp2 ingress protocol 802.1Q flower skip_sw \
+          dst_mac 42:be:24:9b:76:20 vlan_id 100 vlan_prio 0 action drop
+
+Time-based ingress policing
+---------------------------
+
+The TTEthernet hardware abilities of the switch can be constrained to act
+similarly to the Per-Stream Filtering and Policing (PSFP) clause specified in
+IEEE 802.1Q-2018 (formerly 802.1Qci). This means it can be used to perform
+tight timing-based admission control for up to 1024 flows (identified by a
+tuple composed of destination MAC address, VLAN ID and VLAN PCP). Packets which
+are received outside their expected reception window are dropped.
+
+This capability can be managed through the offload of the tc-gate action. As
+routing actions are intrinsic to virtual links in TTEthernet (which performs
+explicit routing of time-critical traffic and does not leave that in the hands
+of the FDB, flooding etc), the tc-gate action may never appear alone when
+asking sja1105 to offload it. One (or more) redirect or trap actions must also
+follow along.
+
+Example: create a tc-taprio schedule that is phase-aligned with a tc-gate
+schedule (the clocks must be synchronized by a 1588 application stack, which is
+outside the scope of this document). No packet delivered by the sender will be
+dropped. Note that the reception window is larger than the transmission window
+(and much more so, in this example) to compensate for the packet propagation
+delay of the link (which can be determined by the 1588 application stack).
+
+Receiver (sja1105)::
+
+  tc qdisc add dev swp2 clsact
+  now=$(phc_ctl /dev/ptp1 get | awk '/clock time is/ {print $5}') && \
+          sec=$(echo $now | awk -F. '{print $1}') && \
+          base_time="$(((sec + 2) * 1000000000))" && \
+          echo "base time ${base_time}"
+  tc filter add dev swp2 ingress flower skip_sw \
+          dst_mac 42:be:24:9b:76:20 \
+          action gate base-time ${base_time} \
+          sched-entry OPEN  60000 -1 -1 \
+          sched-entry CLOSE 40000 -1 -1 \
+          action trap
+
+Sender::
+
+  now=$(phc_ctl /dev/ptp0 get | awk '/clock time is/ {print $5}') && \
+          sec=$(echo $now | awk -F. '{print $1}') && \
+          base_time="$(((sec + 2) * 1000000000))" && \
+          echo "base time ${base_time}"
+  tc qdisc add dev eno0 parent root taprio \
+          num_tc 8 \
+          map 0 1 2 3 4 5 6 7 \
+          queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
+          base-time ${base_time} \
+          sched-entry S 01  50000 \
+          sched-entry S 00  50000 \
+          flags 2
+
+The engine used to schedule the ingress gate operations is the same that the
+one used for the tc-taprio offload. Therefore, the restrictions regarding the
+fact that no two gate actions (either tc-gate or tc-taprio gates) may fire at
+the same time (during the same 200 ns slot) still apply.
+
+To come in handy, it is possible to share time-triggered virtual links across
+more than 1 ingress port, via flow blocks. In this case, the restriction of
+firing at the same time does not apply because there is a single schedule in
+the system, that of the shared virtual link::
+
+  tc qdisc add dev swp2 ingress_block 1 clsact
+  tc qdisc add dev swp3 ingress_block 1 clsact
+  tc filter add block 1 flower skip_sw dst_mac 42:be:24:9b:76:20 \
+          action gate index 2 \
+          base-time 0 \
+          sched-entry OPEN 50000000 -1 -1 \
+          sched-entry CLOSE 50000000 -1 -1 \
+          action trap
+
+Hardware statistics for each flow are also available ("pkts" counts the number
+of dropped frames, which is a sum of frames dropped due to timing violations,
+lack of destination ports and MTU enforcement checks). Byte-level counters are
+not available.
+
 Device Tree bindings and board design
 =====================================
 
Signature · 1 line
2.17.1