1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/dsa/dsa2.c - Hardware switch handling, binding version 2
4 * Copyright (c) 2008-2009 Marvell Semiconductor
5 * Copyright (c) 2013 Florian Fainelli <florian@openwrt.org>
6 * Copyright (c) 2016 Andrew Lunn <andrew@lunn.ch>
9 #include <linux/device.h>
10 #include <linux/err.h>
11 #include <linux/list.h>
12 #include <linux/netdevice.h>
13 #include <linux/slab.h>
14 #include <linux/rtnetlink.h>
16 #include <linux/of_mdio.h>
17 #include <linux/of_net.h>
18 #include <net/devlink.h>
19 #include <net/sch_generic.h>
23 static DEFINE_MUTEX(dsa2_mutex);
24 LIST_HEAD(dsa_tree_list);
26 /* Track the bridges with forwarding offload enabled */
27 static unsigned long dsa_fwd_offloading_bridges;
30 * dsa_tree_notify - Execute code for all switches in a DSA switch tree.
31 * @dst: collection of struct dsa_switch devices to notify.
32 * @e: event, must be of type DSA_NOTIFIER_*
33 * @v: event-specific value.
35 * Given a struct dsa_switch_tree, this can be used to run a function once for
36 * each member DSA switch. The other alternative of traversing the tree is only
37 * through its ports list, which does not uniquely list the switches.
39 int dsa_tree_notify(struct dsa_switch_tree *dst, unsigned long e, void *v)
41 struct raw_notifier_head *nh = &dst->nh;
44 err = raw_notifier_call_chain(nh, e, v);
46 return notifier_to_errno(err);
50 * dsa_broadcast - Notify all DSA trees in the system.
51 * @e: event, must be of type DSA_NOTIFIER_*
52 * @v: event-specific value.
54 * Can be used to notify the switching fabric of events such as cross-chip
55 * bridging between disjoint trees (such as islands of tagger-compatible
56 * switches bridged by an incompatible middle switch).
58 * WARNING: this function is not reliable during probe time, because probing
59 * between trees is asynchronous and not all DSA trees might have probed.
61 int dsa_broadcast(unsigned long e, void *v)
63 struct dsa_switch_tree *dst;
66 list_for_each_entry(dst, &dsa_tree_list, list) {
67 err = dsa_tree_notify(dst, e, v);
76 * dsa_lag_map() - Map LAG structure to a linear LAG array
77 * @dst: Tree in which to record the mapping.
78 * @lag: LAG structure that is to be mapped to the tree's array.
80 * dsa_lag_id/dsa_lag_by_id can then be used to translate between the
81 * two spaces. The size of the mapping space is determined by the
82 * driver by setting ds->num_lag_ids. It is perfectly legal to leave
83 * it unset if it is not needed, in which case these functions become
86 void dsa_lag_map(struct dsa_switch_tree *dst, struct dsa_lag *lag)
90 for (id = 1; id <= dst->lags_len; id++) {
91 if (!dsa_lag_by_id(dst, id)) {
92 dst->lags[id - 1] = lag;
98 /* No IDs left, which is OK. Some drivers do not need it. The
99 * ones that do, e.g. mv88e6xxx, will discover that dsa_lag_id
100 * returns an error for this device when joining the LAG. The
101 * driver can then return -EOPNOTSUPP back to DSA, which will
102 * fall back to a software LAG.
107 * dsa_lag_unmap() - Remove a LAG ID mapping
108 * @dst: Tree in which the mapping is recorded.
109 * @lag: LAG structure that was mapped.
111 * As there may be multiple users of the mapping, it is only removed
112 * if there are no other references to it.
114 void dsa_lag_unmap(struct dsa_switch_tree *dst, struct dsa_lag *lag)
118 dsa_lags_foreach_id(id, dst) {
119 if (dsa_lag_by_id(dst, id) == lag) {
120 dst->lags[id - 1] = NULL;
127 struct dsa_lag *dsa_tree_lag_find(struct dsa_switch_tree *dst,
128 const struct net_device *lag_dev)
132 list_for_each_entry(dp, &dst->ports, list)
133 if (dsa_port_lag_dev_get(dp) == lag_dev)
139 struct dsa_bridge *dsa_tree_bridge_find(struct dsa_switch_tree *dst,
140 const struct net_device *br)
144 list_for_each_entry(dp, &dst->ports, list)
145 if (dsa_port_bridge_dev_get(dp) == br)
151 static int dsa_bridge_num_find(const struct net_device *bridge_dev)
153 struct dsa_switch_tree *dst;
155 list_for_each_entry(dst, &dsa_tree_list, list) {
156 struct dsa_bridge *bridge;
158 bridge = dsa_tree_bridge_find(dst, bridge_dev);
166 unsigned int dsa_bridge_num_get(const struct net_device *bridge_dev, int max)
168 unsigned int bridge_num = dsa_bridge_num_find(bridge_dev);
170 /* Switches without FDB isolation support don't get unique
177 /* First port that requests FDB isolation or TX forwarding
178 * offload for this bridge
180 bridge_num = find_next_zero_bit(&dsa_fwd_offloading_bridges,
181 DSA_MAX_NUM_OFFLOADING_BRIDGES,
183 if (bridge_num >= max)
186 set_bit(bridge_num, &dsa_fwd_offloading_bridges);
192 void dsa_bridge_num_put(const struct net_device *bridge_dev,
193 unsigned int bridge_num)
195 /* Since we refcount bridges, we know that when we call this function
196 * it is no longer in use, so we can just go ahead and remove it from
199 clear_bit(bridge_num, &dsa_fwd_offloading_bridges);
202 struct dsa_switch *dsa_switch_find(int tree_index, int sw_index)
204 struct dsa_switch_tree *dst;
207 list_for_each_entry(dst, &dsa_tree_list, list) {
208 if (dst->index != tree_index)
211 list_for_each_entry(dp, &dst->ports, list) {
212 if (dp->ds->index != sw_index)
221 EXPORT_SYMBOL_GPL(dsa_switch_find);
223 static struct dsa_switch_tree *dsa_tree_find(int index)
225 struct dsa_switch_tree *dst;
227 list_for_each_entry(dst, &dsa_tree_list, list)
228 if (dst->index == index)
234 static struct dsa_switch_tree *dsa_tree_alloc(int index)
236 struct dsa_switch_tree *dst;
238 dst = kzalloc(sizeof(*dst), GFP_KERNEL);
244 INIT_LIST_HEAD(&dst->rtable);
246 INIT_LIST_HEAD(&dst->ports);
248 INIT_LIST_HEAD(&dst->list);
249 list_add_tail(&dst->list, &dsa_tree_list);
251 kref_init(&dst->refcount);
256 static void dsa_tree_free(struct dsa_switch_tree *dst)
259 dsa_tag_driver_put(dst->tag_ops);
260 list_del(&dst->list);
264 static struct dsa_switch_tree *dsa_tree_get(struct dsa_switch_tree *dst)
267 kref_get(&dst->refcount);
272 static struct dsa_switch_tree *dsa_tree_touch(int index)
274 struct dsa_switch_tree *dst;
276 dst = dsa_tree_find(index);
278 return dsa_tree_get(dst);
280 return dsa_tree_alloc(index);
283 static void dsa_tree_release(struct kref *ref)
285 struct dsa_switch_tree *dst;
287 dst = container_of(ref, struct dsa_switch_tree, refcount);
292 static void dsa_tree_put(struct dsa_switch_tree *dst)
295 kref_put(&dst->refcount, dsa_tree_release);
298 static struct dsa_port *dsa_tree_find_port_by_node(struct dsa_switch_tree *dst,
299 struct device_node *dn)
303 list_for_each_entry(dp, &dst->ports, list)
310 static struct dsa_link *dsa_link_touch(struct dsa_port *dp,
311 struct dsa_port *link_dp)
313 struct dsa_switch *ds = dp->ds;
314 struct dsa_switch_tree *dst;
319 list_for_each_entry(dl, &dst->rtable, list)
320 if (dl->dp == dp && dl->link_dp == link_dp)
323 dl = kzalloc(sizeof(*dl), GFP_KERNEL);
328 dl->link_dp = link_dp;
330 INIT_LIST_HEAD(&dl->list);
331 list_add_tail(&dl->list, &dst->rtable);
336 static bool dsa_port_setup_routing_table(struct dsa_port *dp)
338 struct dsa_switch *ds = dp->ds;
339 struct dsa_switch_tree *dst = ds->dst;
340 struct device_node *dn = dp->dn;
341 struct of_phandle_iterator it;
342 struct dsa_port *link_dp;
346 of_for_each_phandle(&it, err, dn, "link", NULL, 0) {
347 link_dp = dsa_tree_find_port_by_node(dst, it.node);
349 of_node_put(it.node);
353 dl = dsa_link_touch(dp, link_dp);
355 of_node_put(it.node);
363 static bool dsa_tree_setup_routing_table(struct dsa_switch_tree *dst)
365 bool complete = true;
368 list_for_each_entry(dp, &dst->ports, list) {
369 if (dsa_port_is_dsa(dp)) {
370 complete = dsa_port_setup_routing_table(dp);
379 static struct dsa_port *dsa_tree_find_first_cpu(struct dsa_switch_tree *dst)
383 list_for_each_entry(dp, &dst->ports, list)
384 if (dsa_port_is_cpu(dp))
390 struct net_device *dsa_tree_find_first_master(struct dsa_switch_tree *dst)
392 struct device_node *ethernet;
393 struct net_device *master;
394 struct dsa_port *cpu_dp;
396 cpu_dp = dsa_tree_find_first_cpu(dst);
397 ethernet = of_parse_phandle(cpu_dp->dn, "ethernet", 0);
398 master = of_find_net_device_by_node(ethernet);
399 of_node_put(ethernet);
404 /* Assign the default CPU port (the first one in the tree) to all ports of the
405 * fabric which don't already have one as part of their own switch.
407 static int dsa_tree_setup_default_cpu(struct dsa_switch_tree *dst)
409 struct dsa_port *cpu_dp, *dp;
411 cpu_dp = dsa_tree_find_first_cpu(dst);
413 pr_err("DSA: tree %d has no CPU port\n", dst->index);
417 list_for_each_entry(dp, &dst->ports, list) {
421 if (dsa_port_is_user(dp) || dsa_port_is_dsa(dp))
428 static struct dsa_port *
429 dsa_switch_preferred_default_local_cpu_port(struct dsa_switch *ds)
431 struct dsa_port *cpu_dp;
433 if (!ds->ops->preferred_default_local_cpu_port)
436 cpu_dp = ds->ops->preferred_default_local_cpu_port(ds);
440 if (WARN_ON(!dsa_port_is_cpu(cpu_dp) || cpu_dp->ds != ds))
446 /* Perform initial assignment of CPU ports to user ports and DSA links in the
447 * fabric, giving preference to CPU ports local to each switch. Default to
448 * using the first CPU port in the switch tree if the port does not have a CPU
449 * port local to this switch.
451 static int dsa_tree_setup_cpu_ports(struct dsa_switch_tree *dst)
453 struct dsa_port *preferred_cpu_dp, *cpu_dp, *dp;
455 list_for_each_entry(cpu_dp, &dst->ports, list) {
456 if (!dsa_port_is_cpu(cpu_dp))
459 preferred_cpu_dp = dsa_switch_preferred_default_local_cpu_port(cpu_dp->ds);
460 if (preferred_cpu_dp && preferred_cpu_dp != cpu_dp)
463 /* Prefer a local CPU port */
464 dsa_switch_for_each_port(dp, cpu_dp->ds) {
465 /* Prefer the first local CPU port found */
469 if (dsa_port_is_user(dp) || dsa_port_is_dsa(dp))
474 return dsa_tree_setup_default_cpu(dst);
477 static void dsa_tree_teardown_cpu_ports(struct dsa_switch_tree *dst)
481 list_for_each_entry(dp, &dst->ports, list)
482 if (dsa_port_is_user(dp) || dsa_port_is_dsa(dp))
486 static int dsa_port_devlink_setup(struct dsa_port *dp)
488 struct devlink_port *dlp = &dp->devlink_port;
489 struct dsa_switch_tree *dst = dp->ds->dst;
490 struct devlink_port_attrs attrs = {};
491 struct devlink *dl = dp->ds->devlink;
492 struct dsa_switch *ds = dp->ds;
493 const unsigned char *id;
497 memset(dlp, 0, sizeof(*dlp));
498 devlink_port_init(dl, dlp);
500 if (ds->ops->port_setup) {
501 err = ds->ops->port_setup(ds, dp->index);
506 id = (const unsigned char *)&dst->index;
507 len = sizeof(dst->index);
509 attrs.phys.port_number = dp->index;
510 memcpy(attrs.switch_id.id, id, len);
511 attrs.switch_id.id_len = len;
514 case DSA_PORT_TYPE_UNUSED:
515 attrs.flavour = DEVLINK_PORT_FLAVOUR_UNUSED;
517 case DSA_PORT_TYPE_CPU:
518 attrs.flavour = DEVLINK_PORT_FLAVOUR_CPU;
520 case DSA_PORT_TYPE_DSA:
521 attrs.flavour = DEVLINK_PORT_FLAVOUR_DSA;
523 case DSA_PORT_TYPE_USER:
524 attrs.flavour = DEVLINK_PORT_FLAVOUR_PHYSICAL;
528 devlink_port_attrs_set(dlp, &attrs);
529 err = devlink_port_register(dl, dlp, dp->index);
531 if (ds->ops->port_teardown)
532 ds->ops->port_teardown(ds, dp->index);
539 static void dsa_port_devlink_teardown(struct dsa_port *dp)
541 struct devlink_port *dlp = &dp->devlink_port;
542 struct dsa_switch *ds = dp->ds;
544 devlink_port_unregister(dlp);
546 if (ds->ops->port_teardown)
547 ds->ops->port_teardown(ds, dp->index);
549 devlink_port_fini(dlp);
552 static int dsa_port_setup(struct dsa_port *dp)
554 struct devlink_port *dlp = &dp->devlink_port;
555 bool dsa_port_link_registered = false;
556 struct dsa_switch *ds = dp->ds;
557 bool dsa_port_enabled = false;
563 err = dsa_port_devlink_setup(dp);
568 case DSA_PORT_TYPE_UNUSED:
569 dsa_port_disable(dp);
571 case DSA_PORT_TYPE_CPU:
573 err = dsa_shared_port_link_register_of(dp);
576 dsa_port_link_registered = true;
579 "skipping link registration for CPU port %d\n",
583 err = dsa_port_enable(dp, NULL);
586 dsa_port_enabled = true;
589 case DSA_PORT_TYPE_DSA:
591 err = dsa_shared_port_link_register_of(dp);
594 dsa_port_link_registered = true;
597 "skipping link registration for DSA port %d\n",
601 err = dsa_port_enable(dp, NULL);
604 dsa_port_enabled = true;
607 case DSA_PORT_TYPE_USER:
608 of_get_mac_address(dp->dn, dp->mac);
609 err = dsa_slave_create(dp);
613 devlink_port_type_eth_set(dlp, dp->slave);
617 if (err && dsa_port_enabled)
618 dsa_port_disable(dp);
619 if (err && dsa_port_link_registered)
620 dsa_shared_port_link_unregister_of(dp);
622 dsa_port_devlink_teardown(dp);
631 static void dsa_port_teardown(struct dsa_port *dp)
633 struct devlink_port *dlp = &dp->devlink_port;
638 devlink_port_type_clear(dlp);
641 case DSA_PORT_TYPE_UNUSED:
643 case DSA_PORT_TYPE_CPU:
644 dsa_port_disable(dp);
646 dsa_shared_port_link_unregister_of(dp);
648 case DSA_PORT_TYPE_DSA:
649 dsa_port_disable(dp);
651 dsa_shared_port_link_unregister_of(dp);
653 case DSA_PORT_TYPE_USER:
655 dsa_slave_destroy(dp->slave);
661 dsa_port_devlink_teardown(dp);
666 static int dsa_port_setup_as_unused(struct dsa_port *dp)
668 dp->type = DSA_PORT_TYPE_UNUSED;
669 return dsa_port_setup(dp);
672 static int dsa_devlink_info_get(struct devlink *dl,
673 struct devlink_info_req *req,
674 struct netlink_ext_ack *extack)
676 struct dsa_switch *ds = dsa_devlink_to_ds(dl);
678 if (ds->ops->devlink_info_get)
679 return ds->ops->devlink_info_get(ds, req, extack);
684 static int dsa_devlink_sb_pool_get(struct devlink *dl,
685 unsigned int sb_index, u16 pool_index,
686 struct devlink_sb_pool_info *pool_info)
688 struct dsa_switch *ds = dsa_devlink_to_ds(dl);
690 if (!ds->ops->devlink_sb_pool_get)
693 return ds->ops->devlink_sb_pool_get(ds, sb_index, pool_index,
697 static int dsa_devlink_sb_pool_set(struct devlink *dl, unsigned int sb_index,
698 u16 pool_index, u32 size,
699 enum devlink_sb_threshold_type threshold_type,
700 struct netlink_ext_ack *extack)
702 struct dsa_switch *ds = dsa_devlink_to_ds(dl);
704 if (!ds->ops->devlink_sb_pool_set)
707 return ds->ops->devlink_sb_pool_set(ds, sb_index, pool_index, size,
708 threshold_type, extack);
711 static int dsa_devlink_sb_port_pool_get(struct devlink_port *dlp,
712 unsigned int sb_index, u16 pool_index,
715 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
716 int port = dsa_devlink_port_to_port(dlp);
718 if (!ds->ops->devlink_sb_port_pool_get)
721 return ds->ops->devlink_sb_port_pool_get(ds, port, sb_index,
722 pool_index, p_threshold);
725 static int dsa_devlink_sb_port_pool_set(struct devlink_port *dlp,
726 unsigned int sb_index, u16 pool_index,
728 struct netlink_ext_ack *extack)
730 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
731 int port = dsa_devlink_port_to_port(dlp);
733 if (!ds->ops->devlink_sb_port_pool_set)
736 return ds->ops->devlink_sb_port_pool_set(ds, port, sb_index,
737 pool_index, threshold, extack);
741 dsa_devlink_sb_tc_pool_bind_get(struct devlink_port *dlp,
742 unsigned int sb_index, u16 tc_index,
743 enum devlink_sb_pool_type pool_type,
744 u16 *p_pool_index, u32 *p_threshold)
746 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
747 int port = dsa_devlink_port_to_port(dlp);
749 if (!ds->ops->devlink_sb_tc_pool_bind_get)
752 return ds->ops->devlink_sb_tc_pool_bind_get(ds, port, sb_index,
754 p_pool_index, p_threshold);
758 dsa_devlink_sb_tc_pool_bind_set(struct devlink_port *dlp,
759 unsigned int sb_index, u16 tc_index,
760 enum devlink_sb_pool_type pool_type,
761 u16 pool_index, u32 threshold,
762 struct netlink_ext_ack *extack)
764 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
765 int port = dsa_devlink_port_to_port(dlp);
767 if (!ds->ops->devlink_sb_tc_pool_bind_set)
770 return ds->ops->devlink_sb_tc_pool_bind_set(ds, port, sb_index,
772 pool_index, threshold,
776 static int dsa_devlink_sb_occ_snapshot(struct devlink *dl,
777 unsigned int sb_index)
779 struct dsa_switch *ds = dsa_devlink_to_ds(dl);
781 if (!ds->ops->devlink_sb_occ_snapshot)
784 return ds->ops->devlink_sb_occ_snapshot(ds, sb_index);
787 static int dsa_devlink_sb_occ_max_clear(struct devlink *dl,
788 unsigned int sb_index)
790 struct dsa_switch *ds = dsa_devlink_to_ds(dl);
792 if (!ds->ops->devlink_sb_occ_max_clear)
795 return ds->ops->devlink_sb_occ_max_clear(ds, sb_index);
798 static int dsa_devlink_sb_occ_port_pool_get(struct devlink_port *dlp,
799 unsigned int sb_index,
800 u16 pool_index, u32 *p_cur,
803 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
804 int port = dsa_devlink_port_to_port(dlp);
806 if (!ds->ops->devlink_sb_occ_port_pool_get)
809 return ds->ops->devlink_sb_occ_port_pool_get(ds, port, sb_index,
810 pool_index, p_cur, p_max);
814 dsa_devlink_sb_occ_tc_port_bind_get(struct devlink_port *dlp,
815 unsigned int sb_index, u16 tc_index,
816 enum devlink_sb_pool_type pool_type,
817 u32 *p_cur, u32 *p_max)
819 struct dsa_switch *ds = dsa_devlink_port_to_ds(dlp);
820 int port = dsa_devlink_port_to_port(dlp);
822 if (!ds->ops->devlink_sb_occ_tc_port_bind_get)
825 return ds->ops->devlink_sb_occ_tc_port_bind_get(ds, port,
831 static const struct devlink_ops dsa_devlink_ops = {
832 .info_get = dsa_devlink_info_get,
833 .sb_pool_get = dsa_devlink_sb_pool_get,
834 .sb_pool_set = dsa_devlink_sb_pool_set,
835 .sb_port_pool_get = dsa_devlink_sb_port_pool_get,
836 .sb_port_pool_set = dsa_devlink_sb_port_pool_set,
837 .sb_tc_pool_bind_get = dsa_devlink_sb_tc_pool_bind_get,
838 .sb_tc_pool_bind_set = dsa_devlink_sb_tc_pool_bind_set,
839 .sb_occ_snapshot = dsa_devlink_sb_occ_snapshot,
840 .sb_occ_max_clear = dsa_devlink_sb_occ_max_clear,
841 .sb_occ_port_pool_get = dsa_devlink_sb_occ_port_pool_get,
842 .sb_occ_tc_port_bind_get = dsa_devlink_sb_occ_tc_port_bind_get,
845 static int dsa_switch_setup_tag_protocol(struct dsa_switch *ds)
847 const struct dsa_device_ops *tag_ops = ds->dst->tag_ops;
848 struct dsa_switch_tree *dst = ds->dst;
851 if (tag_ops->proto == dst->default_proto)
855 err = ds->ops->change_tag_protocol(ds, tag_ops->proto);
858 dev_err(ds->dev, "Unable to use tag protocol \"%s\": %pe\n",
859 tag_ops->name, ERR_PTR(err));
864 if (tag_ops->connect) {
865 err = tag_ops->connect(ds);
870 if (ds->ops->connect_tag_protocol) {
871 err = ds->ops->connect_tag_protocol(ds, tag_ops->proto);
874 "Unable to connect to tag protocol \"%s\": %pe\n",
875 tag_ops->name, ERR_PTR(err));
883 if (tag_ops->disconnect)
884 tag_ops->disconnect(ds);
889 static void dsa_switch_teardown_tag_protocol(struct dsa_switch *ds)
891 const struct dsa_device_ops *tag_ops = ds->dst->tag_ops;
893 if (tag_ops->disconnect)
894 tag_ops->disconnect(ds);
897 static int dsa_switch_setup(struct dsa_switch *ds)
899 struct dsa_devlink_priv *dl_priv;
900 struct device_node *dn;
906 /* Initialize ds->phys_mii_mask before registering the slave MDIO bus
907 * driver and before ops->setup() has run, since the switch drivers and
908 * the slave MDIO bus driver rely on these values for probing PHY
911 ds->phys_mii_mask |= dsa_user_ports(ds);
913 /* Add the switch to devlink before calling setup, so that setup can
917 devlink_alloc(&dsa_devlink_ops, sizeof(*dl_priv), ds->dev);
920 dl_priv = devlink_priv(ds->devlink);
923 err = dsa_switch_register_notifier(ds);
927 ds->configure_vlan_while_not_filtering = true;
929 err = ds->ops->setup(ds);
931 goto unregister_notifier;
933 err = dsa_switch_setup_tag_protocol(ds);
937 if (!ds->slave_mii_bus && ds->ops->phy_read) {
938 ds->slave_mii_bus = mdiobus_alloc();
939 if (!ds->slave_mii_bus) {
944 dsa_slave_mii_bus_init(ds);
946 dn = of_get_child_by_name(ds->dev->of_node, "mdio");
948 err = of_mdiobus_register(ds->slave_mii_bus, dn);
951 goto free_slave_mii_bus;
955 devlink_register(ds->devlink);
959 if (ds->slave_mii_bus && ds->ops->phy_read)
960 mdiobus_free(ds->slave_mii_bus);
962 if (ds->ops->teardown)
963 ds->ops->teardown(ds);
965 dsa_switch_unregister_notifier(ds);
967 devlink_free(ds->devlink);
972 static void dsa_switch_teardown(struct dsa_switch *ds)
978 devlink_unregister(ds->devlink);
980 if (ds->slave_mii_bus && ds->ops->phy_read) {
981 mdiobus_unregister(ds->slave_mii_bus);
982 mdiobus_free(ds->slave_mii_bus);
983 ds->slave_mii_bus = NULL;
986 dsa_switch_teardown_tag_protocol(ds);
988 if (ds->ops->teardown)
989 ds->ops->teardown(ds);
991 dsa_switch_unregister_notifier(ds);
994 devlink_free(ds->devlink);
1001 /* First tear down the non-shared, then the shared ports. This ensures that
1002 * all work items scheduled by our switchdev handlers for user ports have
1003 * completed before we destroy the refcounting kept on the shared ports.
1005 static void dsa_tree_teardown_ports(struct dsa_switch_tree *dst)
1007 struct dsa_port *dp;
1009 list_for_each_entry(dp, &dst->ports, list)
1010 if (dsa_port_is_user(dp) || dsa_port_is_unused(dp))
1011 dsa_port_teardown(dp);
1013 dsa_flush_workqueue();
1015 list_for_each_entry(dp, &dst->ports, list)
1016 if (dsa_port_is_dsa(dp) || dsa_port_is_cpu(dp))
1017 dsa_port_teardown(dp);
1020 static void dsa_tree_teardown_switches(struct dsa_switch_tree *dst)
1022 struct dsa_port *dp;
1024 list_for_each_entry(dp, &dst->ports, list)
1025 dsa_switch_teardown(dp->ds);
1028 /* Bring shared ports up first, then non-shared ports */
1029 static int dsa_tree_setup_ports(struct dsa_switch_tree *dst)
1031 struct dsa_port *dp;
1034 list_for_each_entry(dp, &dst->ports, list) {
1035 if (dsa_port_is_dsa(dp) || dsa_port_is_cpu(dp)) {
1036 err = dsa_port_setup(dp);
1042 list_for_each_entry(dp, &dst->ports, list) {
1043 if (dsa_port_is_user(dp) || dsa_port_is_unused(dp)) {
1044 err = dsa_port_setup(dp);
1046 err = dsa_port_setup_as_unused(dp);
1056 dsa_tree_teardown_ports(dst);
1061 static int dsa_tree_setup_switches(struct dsa_switch_tree *dst)
1063 struct dsa_port *dp;
1066 list_for_each_entry(dp, &dst->ports, list) {
1067 err = dsa_switch_setup(dp->ds);
1069 dsa_tree_teardown_switches(dst);
1077 static int dsa_tree_setup_master(struct dsa_switch_tree *dst)
1079 struct dsa_port *cpu_dp;
1084 dsa_tree_for_each_cpu_port(cpu_dp, dst) {
1085 struct net_device *master = cpu_dp->master;
1086 bool admin_up = (master->flags & IFF_UP) &&
1087 !qdisc_tx_is_noop(master);
1089 err = dsa_master_setup(master, cpu_dp);
1093 /* Replay master state event */
1094 dsa_tree_master_admin_state_change(dst, master, admin_up);
1095 dsa_tree_master_oper_state_change(dst, master,
1096 netif_oper_up(master));
1104 static void dsa_tree_teardown_master(struct dsa_switch_tree *dst)
1106 struct dsa_port *cpu_dp;
1110 dsa_tree_for_each_cpu_port(cpu_dp, dst) {
1111 struct net_device *master = cpu_dp->master;
1113 /* Synthesizing an "admin down" state is sufficient for
1114 * the switches to get a notification if the master is
1115 * currently up and running.
1117 dsa_tree_master_admin_state_change(dst, master, false);
1119 dsa_master_teardown(master);
1125 static int dsa_tree_setup_lags(struct dsa_switch_tree *dst)
1127 unsigned int len = 0;
1128 struct dsa_port *dp;
1130 list_for_each_entry(dp, &dst->ports, list) {
1131 if (dp->ds->num_lag_ids > len)
1132 len = dp->ds->num_lag_ids;
1138 dst->lags = kcalloc(len, sizeof(*dst->lags), GFP_KERNEL);
1142 dst->lags_len = len;
1146 static void dsa_tree_teardown_lags(struct dsa_switch_tree *dst)
1151 static int dsa_tree_setup(struct dsa_switch_tree *dst)
1157 pr_err("DSA: tree %d already setup! Disjoint trees?\n",
1162 complete = dsa_tree_setup_routing_table(dst);
1166 err = dsa_tree_setup_cpu_ports(dst);
1170 err = dsa_tree_setup_switches(dst);
1172 goto teardown_cpu_ports;
1174 err = dsa_tree_setup_ports(dst);
1176 goto teardown_switches;
1178 err = dsa_tree_setup_master(dst);
1180 goto teardown_ports;
1182 err = dsa_tree_setup_lags(dst);
1184 goto teardown_master;
1188 pr_info("DSA: tree %d setup\n", dst->index);
1193 dsa_tree_teardown_master(dst);
1195 dsa_tree_teardown_ports(dst);
1197 dsa_tree_teardown_switches(dst);
1199 dsa_tree_teardown_cpu_ports(dst);
1204 static void dsa_tree_teardown(struct dsa_switch_tree *dst)
1206 struct dsa_link *dl, *next;
1211 dsa_tree_teardown_lags(dst);
1213 dsa_tree_teardown_master(dst);
1215 dsa_tree_teardown_ports(dst);
1217 dsa_tree_teardown_switches(dst);
1219 dsa_tree_teardown_cpu_ports(dst);
1221 list_for_each_entry_safe(dl, next, &dst->rtable, list) {
1222 list_del(&dl->list);
1226 pr_info("DSA: tree %d torn down\n", dst->index);
1231 static int dsa_tree_bind_tag_proto(struct dsa_switch_tree *dst,
1232 const struct dsa_device_ops *tag_ops)
1234 const struct dsa_device_ops *old_tag_ops = dst->tag_ops;
1235 struct dsa_notifier_tag_proto_info info;
1238 dst->tag_ops = tag_ops;
1240 /* Notify the switches from this tree about the connection
1243 info.tag_ops = tag_ops;
1244 err = dsa_tree_notify(dst, DSA_NOTIFIER_TAG_PROTO_CONNECT, &info);
1245 if (err && err != -EOPNOTSUPP)
1246 goto out_disconnect;
1248 /* Notify the old tagger about the disconnection from this tree */
1249 info.tag_ops = old_tag_ops;
1250 dsa_tree_notify(dst, DSA_NOTIFIER_TAG_PROTO_DISCONNECT, &info);
1255 info.tag_ops = tag_ops;
1256 dsa_tree_notify(dst, DSA_NOTIFIER_TAG_PROTO_DISCONNECT, &info);
1257 dst->tag_ops = old_tag_ops;
1262 /* Since the dsa/tagging sysfs device attribute is per master, the assumption
1263 * is that all DSA switches within a tree share the same tagger, otherwise
1264 * they would have formed disjoint trees (different "dsa,member" values).
1266 int dsa_tree_change_tag_proto(struct dsa_switch_tree *dst,
1267 const struct dsa_device_ops *tag_ops,
1268 const struct dsa_device_ops *old_tag_ops)
1270 struct dsa_notifier_tag_proto_info info;
1271 struct dsa_port *dp;
1274 if (!rtnl_trylock())
1275 return restart_syscall();
1277 /* At the moment we don't allow changing the tag protocol under
1278 * traffic. The rtnl_mutex also happens to serialize concurrent
1279 * attempts to change the tagging protocol. If we ever lift the IFF_UP
1280 * restriction, there needs to be another mutex which serializes this.
1282 dsa_tree_for_each_user_port(dp, dst) {
1283 if (dsa_port_to_master(dp)->flags & IFF_UP)
1286 if (dp->slave->flags & IFF_UP)
1290 /* Notify the tag protocol change */
1291 info.tag_ops = tag_ops;
1292 err = dsa_tree_notify(dst, DSA_NOTIFIER_TAG_PROTO, &info);
1294 goto out_unwind_tagger;
1296 err = dsa_tree_bind_tag_proto(dst, tag_ops);
1298 goto out_unwind_tagger;
1305 info.tag_ops = old_tag_ops;
1306 dsa_tree_notify(dst, DSA_NOTIFIER_TAG_PROTO, &info);
1312 static void dsa_tree_master_state_change(struct dsa_switch_tree *dst,
1313 struct net_device *master)
1315 struct dsa_notifier_master_state_info info;
1316 struct dsa_port *cpu_dp = master->dsa_ptr;
1318 info.master = master;
1319 info.operational = dsa_port_master_is_operational(cpu_dp);
1321 dsa_tree_notify(dst, DSA_NOTIFIER_MASTER_STATE_CHANGE, &info);
1324 void dsa_tree_master_admin_state_change(struct dsa_switch_tree *dst,
1325 struct net_device *master,
1328 struct dsa_port *cpu_dp = master->dsa_ptr;
1329 bool notify = false;
1331 /* Don't keep track of admin state on LAG DSA masters,
1332 * but rather just of physical DSA masters
1334 if (netif_is_lag_master(master))
1337 if ((dsa_port_master_is_operational(cpu_dp)) !=
1338 (up && cpu_dp->master_oper_up))
1341 cpu_dp->master_admin_up = up;
1344 dsa_tree_master_state_change(dst, master);
1347 void dsa_tree_master_oper_state_change(struct dsa_switch_tree *dst,
1348 struct net_device *master,
1351 struct dsa_port *cpu_dp = master->dsa_ptr;
1352 bool notify = false;
1354 /* Don't keep track of oper state on LAG DSA masters,
1355 * but rather just of physical DSA masters
1357 if (netif_is_lag_master(master))
1360 if ((dsa_port_master_is_operational(cpu_dp)) !=
1361 (cpu_dp->master_admin_up && up))
1364 cpu_dp->master_oper_up = up;
1367 dsa_tree_master_state_change(dst, master);
1370 static struct dsa_port *dsa_port_touch(struct dsa_switch *ds, int index)
1372 struct dsa_switch_tree *dst = ds->dst;
1373 struct dsa_port *dp;
1375 dsa_switch_for_each_port(dp, ds)
1376 if (dp->index == index)
1379 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1386 mutex_init(&dp->addr_lists_lock);
1387 mutex_init(&dp->vlans_lock);
1388 INIT_LIST_HEAD(&dp->fdbs);
1389 INIT_LIST_HEAD(&dp->mdbs);
1390 INIT_LIST_HEAD(&dp->vlans);
1391 INIT_LIST_HEAD(&dp->list);
1392 list_add_tail(&dp->list, &dst->ports);
1397 static int dsa_port_parse_user(struct dsa_port *dp, const char *name)
1402 dp->type = DSA_PORT_TYPE_USER;
1408 static int dsa_port_parse_dsa(struct dsa_port *dp)
1410 dp->type = DSA_PORT_TYPE_DSA;
1415 static enum dsa_tag_protocol dsa_get_tag_protocol(struct dsa_port *dp,
1416 struct net_device *master)
1418 enum dsa_tag_protocol tag_protocol = DSA_TAG_PROTO_NONE;
1419 struct dsa_switch *mds, *ds = dp->ds;
1420 unsigned int mdp_upstream;
1421 struct dsa_port *mdp;
1423 /* It is possible to stack DSA switches onto one another when that
1424 * happens the switch driver may want to know if its tagging protocol
1425 * is going to work in such a configuration.
1427 if (dsa_slave_dev_check(master)) {
1428 mdp = dsa_slave_to_port(master);
1430 mdp_upstream = dsa_upstream_port(mds, mdp->index);
1431 tag_protocol = mds->ops->get_tag_protocol(mds, mdp_upstream,
1432 DSA_TAG_PROTO_NONE);
1435 /* If the master device is not itself a DSA slave in a disjoint DSA
1436 * tree, then return immediately.
1438 return ds->ops->get_tag_protocol(ds, dp->index, tag_protocol);
1441 static int dsa_port_parse_cpu(struct dsa_port *dp, struct net_device *master,
1442 const char *user_protocol)
1444 const struct dsa_device_ops *tag_ops = NULL;
1445 struct dsa_switch *ds = dp->ds;
1446 struct dsa_switch_tree *dst = ds->dst;
1447 enum dsa_tag_protocol default_proto;
1449 /* Find out which protocol the switch would prefer. */
1450 default_proto = dsa_get_tag_protocol(dp, master);
1451 if (dst->default_proto) {
1452 if (dst->default_proto != default_proto) {
1454 "A DSA switch tree can have only one tagging protocol\n");
1458 dst->default_proto = default_proto;
1461 /* See if the user wants to override that preference. */
1462 if (user_protocol) {
1463 if (!ds->ops->change_tag_protocol) {
1464 dev_err(ds->dev, "Tag protocol cannot be modified\n");
1468 tag_ops = dsa_find_tagger_by_name(user_protocol);
1469 if (IS_ERR(tag_ops)) {
1471 "Failed to find a tagging driver for protocol %s, using default\n",
1478 tag_ops = dsa_tag_driver_get(default_proto);
1480 if (IS_ERR(tag_ops)) {
1481 if (PTR_ERR(tag_ops) == -ENOPROTOOPT)
1482 return -EPROBE_DEFER;
1484 dev_warn(ds->dev, "No tagger for this switch\n");
1485 return PTR_ERR(tag_ops);
1489 if (dst->tag_ops != tag_ops) {
1491 "A DSA switch tree can have only one tagging protocol\n");
1493 dsa_tag_driver_put(tag_ops);
1497 /* In the case of multiple CPU ports per switch, the tagging
1498 * protocol is still reference-counted only per switch tree.
1500 dsa_tag_driver_put(tag_ops);
1502 dst->tag_ops = tag_ops;
1505 dp->master = master;
1506 dp->type = DSA_PORT_TYPE_CPU;
1507 dsa_port_set_tag_protocol(dp, dst->tag_ops);
1510 /* At this point, the tree may be configured to use a different
1511 * tagger than the one chosen by the switch driver during
1512 * .setup, in the case when a user selects a custom protocol
1515 * This is resolved by syncing the driver with the tree in
1516 * dsa_switch_setup_tag_protocol once .setup has run and the
1517 * driver is ready to accept calls to .change_tag_protocol. If
1518 * the driver does not support the custom protocol at that
1519 * point, the tree is wholly rejected, thereby ensuring that the
1520 * tree and driver are always in agreement on the protocol to
1526 static int dsa_port_parse_of(struct dsa_port *dp, struct device_node *dn)
1528 struct device_node *ethernet = of_parse_phandle(dn, "ethernet", 0);
1529 const char *name = of_get_property(dn, "label", NULL);
1530 bool link = of_property_read_bool(dn, "link");
1535 struct net_device *master;
1536 const char *user_protocol;
1538 master = of_find_net_device_by_node(ethernet);
1539 of_node_put(ethernet);
1541 return -EPROBE_DEFER;
1543 user_protocol = of_get_property(dn, "dsa-tag-protocol", NULL);
1544 return dsa_port_parse_cpu(dp, master, user_protocol);
1548 return dsa_port_parse_dsa(dp);
1550 return dsa_port_parse_user(dp, name);
1553 static int dsa_switch_parse_ports_of(struct dsa_switch *ds,
1554 struct device_node *dn)
1556 struct device_node *ports, *port;
1557 struct dsa_port *dp;
1561 ports = of_get_child_by_name(dn, "ports");
1563 /* The second possibility is "ethernet-ports" */
1564 ports = of_get_child_by_name(dn, "ethernet-ports");
1566 dev_err(ds->dev, "no ports child node found\n");
1571 for_each_available_child_of_node(ports, port) {
1572 err = of_property_read_u32(port, "reg", ®);
1578 if (reg >= ds->num_ports) {
1579 dev_err(ds->dev, "port %pOF index %u exceeds num_ports (%u)\n",
1580 port, reg, ds->num_ports);
1586 dp = dsa_to_port(ds, reg);
1588 err = dsa_port_parse_of(dp, port);
1600 static int dsa_switch_parse_member_of(struct dsa_switch *ds,
1601 struct device_node *dn)
1603 u32 m[2] = { 0, 0 };
1606 /* Don't error out if this optional property isn't found */
1607 sz = of_property_read_variable_u32_array(dn, "dsa,member", m, 2, 2);
1608 if (sz < 0 && sz != -EINVAL)
1613 ds->dst = dsa_tree_touch(m[0]);
1617 if (dsa_switch_find(ds->dst->index, ds->index)) {
1619 "A DSA switch with index %d already exists in tree %d\n",
1620 ds->index, ds->dst->index);
1624 if (ds->dst->last_switch < ds->index)
1625 ds->dst->last_switch = ds->index;
1630 static int dsa_switch_touch_ports(struct dsa_switch *ds)
1632 struct dsa_port *dp;
1635 for (port = 0; port < ds->num_ports; port++) {
1636 dp = dsa_port_touch(ds, port);
1644 static int dsa_switch_parse_of(struct dsa_switch *ds, struct device_node *dn)
1648 err = dsa_switch_parse_member_of(ds, dn);
1652 err = dsa_switch_touch_ports(ds);
1656 return dsa_switch_parse_ports_of(ds, dn);
1659 static int dsa_port_parse(struct dsa_port *dp, const char *name,
1662 if (!strcmp(name, "cpu")) {
1663 struct net_device *master;
1665 master = dsa_dev_to_net_device(dev);
1667 return -EPROBE_DEFER;
1671 return dsa_port_parse_cpu(dp, master, NULL);
1674 if (!strcmp(name, "dsa"))
1675 return dsa_port_parse_dsa(dp);
1677 return dsa_port_parse_user(dp, name);
1680 static int dsa_switch_parse_ports(struct dsa_switch *ds,
1681 struct dsa_chip_data *cd)
1683 bool valid_name_found = false;
1684 struct dsa_port *dp;
1690 for (i = 0; i < DSA_MAX_PORTS; i++) {
1691 name = cd->port_names[i];
1692 dev = cd->netdev[i];
1693 dp = dsa_to_port(ds, i);
1698 err = dsa_port_parse(dp, name, dev);
1702 valid_name_found = true;
1705 if (!valid_name_found && i == DSA_MAX_PORTS)
1711 static int dsa_switch_parse(struct dsa_switch *ds, struct dsa_chip_data *cd)
1717 /* We don't support interconnected switches nor multiple trees via
1718 * platform data, so this is the unique switch of the tree.
1721 ds->dst = dsa_tree_touch(0);
1725 err = dsa_switch_touch_ports(ds);
1729 return dsa_switch_parse_ports(ds, cd);
1732 static void dsa_switch_release_ports(struct dsa_switch *ds)
1734 struct dsa_port *dp, *next;
1736 dsa_switch_for_each_port_safe(dp, next, ds) {
1737 WARN_ON(!list_empty(&dp->fdbs));
1738 WARN_ON(!list_empty(&dp->mdbs));
1739 WARN_ON(!list_empty(&dp->vlans));
1740 list_del(&dp->list);
1745 static int dsa_switch_probe(struct dsa_switch *ds)
1747 struct dsa_switch_tree *dst;
1748 struct dsa_chip_data *pdata;
1749 struct device_node *np;
1755 pdata = ds->dev->platform_data;
1756 np = ds->dev->of_node;
1762 err = dsa_switch_parse_of(ds, np);
1764 dsa_switch_release_ports(ds);
1766 err = dsa_switch_parse(ds, pdata);
1768 dsa_switch_release_ports(ds);
1778 err = dsa_tree_setup(dst);
1780 dsa_switch_release_ports(ds);
1787 int dsa_register_switch(struct dsa_switch *ds)
1791 mutex_lock(&dsa2_mutex);
1792 err = dsa_switch_probe(ds);
1793 dsa_tree_put(ds->dst);
1794 mutex_unlock(&dsa2_mutex);
1798 EXPORT_SYMBOL_GPL(dsa_register_switch);
1800 static void dsa_switch_remove(struct dsa_switch *ds)
1802 struct dsa_switch_tree *dst = ds->dst;
1804 dsa_tree_teardown(dst);
1805 dsa_switch_release_ports(ds);
1809 void dsa_unregister_switch(struct dsa_switch *ds)
1811 mutex_lock(&dsa2_mutex);
1812 dsa_switch_remove(ds);
1813 mutex_unlock(&dsa2_mutex);
1815 EXPORT_SYMBOL_GPL(dsa_unregister_switch);
1817 /* If the DSA master chooses to unregister its net_device on .shutdown, DSA is
1818 * blocking that operation from completion, due to the dev_hold taken inside
1819 * netdev_upper_dev_link. Unlink the DSA slave interfaces from being uppers of
1820 * the DSA master, so that the system can reboot successfully.
1822 void dsa_switch_shutdown(struct dsa_switch *ds)
1824 struct net_device *master, *slave_dev;
1825 struct dsa_port *dp;
1827 mutex_lock(&dsa2_mutex);
1834 dsa_switch_for_each_user_port(dp, ds) {
1835 master = dsa_port_to_master(dp);
1836 slave_dev = dp->slave;
1838 netdev_upper_dev_unlink(master, slave_dev);
1841 /* Disconnect from further netdevice notifiers on the master,
1842 * since netdev_uses_dsa() will now return false.
1844 dsa_switch_for_each_cpu_port(dp, ds)
1845 dp->master->dsa_ptr = NULL;
1849 mutex_unlock(&dsa2_mutex);
1851 EXPORT_SYMBOL_GPL(dsa_switch_shutdown);